@Proceedings{MLHC2026, title = {Proceedings of the 11th Machine Learning for Healthcare Conference}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, editor = {Rahul G. Krishnan and Wouter A. C. Amsterdam and Sumit Chopra and Shauna Overgaard and Michael Hughes and Erkin Ötleş and Yiqiu Shen and Divya Shanmugam and Madhur Nayan and Matthew Engelhard and Jim Fackler and Michael Oberst}, publisher = {PMLR}, series = {Proceedings of Machine Learning Research}, volume = 340 } @InProceedings{pmlr-v340-abbasi26a, title = {NAST: Improving Negation Handling in Medical Vision–Language Models through Negation-Aware Selective Training}, author = {Abbasi, Ali and Taghipour, Mehdi and Beheshti, Rahmatollah}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1--29}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/abbasi26a/abbasi26a.pdf}, url = {https://proceedings.mlr.press/v340/abbasi26a.html}, abstract = {Negation is a fundamental linguistic operation in clinical reporting, yet vision–language models (VLMs) frequently fail to distinguish affirmative from negated medical statements. To systematically characterize this limitation, we introduce MedNega-Bench, a radiology-specific diagnostic benchmark that evaluates polarity sensitivity under controlled clinical conditions, revealing that common medical VLMs consistently confuse negated and non-negated findings. To enable learning beyond simple condition absence, we further construct MedNega-FT, a contextual clinical negation dataset that encodes structured claims and supports attributelevel negations involving location and severity. Building on these resources, we propose Negation-Aware Selective Training (NAST), an interpretability-guided adaptation method that uses causal tracing effects (CTEs) to modulate layer-wise gradient updates during fine-tuning. NAST scales each layer’s update according to its causal contribution to negation processing, transforming mechanistic interpretability signals into a principled optimization rule. Experiments demonstrate improved discrimination of affirmative and negated clinical statements without degrading general vision–language alignment, highlighting the value of causal interpretability for targeted model adaptation in safety-critical medical settings. Code and resources are available at https://github.com/healthylaife/NAST.} } @InProceedings{pmlr-v340-ahmed26a, title = {FSA-Bench: Benchmarking Federated Survival Models}, author = {Ahmed, Sultan and Purushotham, Sanjay}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {30--54}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/ahmed26a/ahmed26a.pdf}, url = {https://proceedings.mlr.press/v340/ahmed26a.html}, abstract = {Clinical time-to-event data are routinely collected across multiple healthcare institutions but are rarely pooled due to privacy regulations such as GDPR and HIPAA. This fragmentation limits the development of robust and generalizable survival models for clinical decision-making. Federated Survival Analysis (FSA) enables collaborative modeling without sharing patient-level data, but the reliability, calibration, and robustness of survival models in federated settings remain insufficiently understood. We present FSA-Bench, a comprehensive benchmark for evaluating survival models under realistic federated and multi-institutional healthcare conditions. FSA-Bench includes classical statistical methods (Kaplan Meier, Weibull Accelerated Failure Time, and Cox Proportional Hazards), machine learning models, and modern deep learning approaches, including attention-based federated architectures such as FedPAttn. The benchmark evaluates these models across diverse clinical datasets under realistic challenges, including heterogeneous patient populations, varying censoring rates, and non-IID data distributed across multiple institutions. Across diverse clinical benchmark datasets, we evaluate model performance using clinically meaningful metrics, including discrimination (time-dependent C-index), calibration (Integrated Brier Score and negative log-likelihood), and ranking consistency using Spearman’s correlation, Kendall’s rank correlation, stability, and Jaccard similarity. An aggregated cross-dataset analysis further provides a comprehensive assessment of ranking robustness, revealing that Fed CoxPH offers the strongest overall ranking consistency, while FedPLSTM and FedPAttn achieve the highest stability and top-model agreement across heterogeneous federated settings. Collec- tively, FSA-Bench establishes a standardized and reproducible evaluation framework for benchmarking federated survival models and offers practical guidance for selecting reliable survival modeling approaches in privacy-preserving, multi-institutional healthcare environments.} } @InProceedings{pmlr-v340-aldeia26a, title = {A foundation-model approach to pediatric headache classification from resting-state fMRI}, author = {Aldeia, Guilherme Seidyo Imai and Moon, Clara and Shulman, Julie M. and Sethna, Navil and Smith, Allison M and LeBel, Alyssa and Cava, William La and Holmes, Scott}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {55--76}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/aldeia26a/aldeia26a.pdf}, url = {https://proceedings.mlr.press/v340/aldeia26a.html}, abstract = {Headache is the most common neurological disorder in children and substantially affects quality of life. We investigated whether resting-state functional MRI (rs-fMRI) can support pediatric headache classification using machine learning. We encoded rs-fMRI data using NeuroSTORM, a recent foundation model, and fine-tuned it to distinguish healthy controls from children with headache and subsequently classify headache subtypes. We then compared NeuroSTORM with a standard neuroscience approach that uses functional connectivity (FC) matrices derived from brain activity as predictors. Using 189 rs-fMRI scans from 110 individuals collected across two visits (prevalence of any headache: 74%), NeuroSTORM achieved an area under the receiver operating characteristic curve (AUROC) of 0.82 (95% CI, 0.82-0.82) and an area under the precision-recall curve (AUPRC) of 0.93 (95% CI, 0.93-0.94) in discriminating headache from non-headache. In contrast, models trained on FC matrices showed limited performance (AUROC, 0.67 [95% CI, 0.67-0.67]; AUPRC, 0.85 [95% CI, 0.85-0.85]). In a multiclass setting, when tasked with classifying individuals as healthy controls, individuals with chronic migraine, or individuals with non-chronic headaches (e.g., post-viral headache, new daily persistent headache, post-traumatic headache), NeuroSTORM achieved a macro-AUROC of 0.69 (95% CI, 0.68-0.69). The results suggest this approach can distinguish chronic migraine, the most common headache syndrome, but has difficulty differentiating other headache subtypes from chronic migraine. Overall, under limited-data conditions, NeuroSTORM appears to capture latent rs-fMRI representations that transfer to headache-related tasks. The findings provide proof-of-concept for fMRI-based prediction of pediatric headache using a foundation model without relying on functional connectivity features and highlight the potential of this approach for subtype identification. Further development of such tools may help clinicians improve diagnosis based on brain activity and tailor treatment strategies to individual patients.} } @InProceedings{pmlr-v340-arno26a, title = {Annotation-Assisted Learning of Treatment Policies From Multimodal Electronic Health Records}, author = {Arno, Henri and Demeester, Thomas}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {77--110}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/arno26a/arno26a.pdf}, url = {https://proceedings.mlr.press/v340/arno26a.html}, abstract = {We study how to learn treatment policies from multimodal electronic health records (EHRs) that consist of tabular data and clinical text. These policies can help physicians make better treatment decisions and allocate healthcare resources more efficiently. Causal policy learning methods prioritize patients with the largest expected treatment benefit. Yet, existing estimators are designed for tabular covariates under causal assumptions that may be hard to justify in the multimodal setting. A pragmatic alternative is to apply causal estimators directly to multimodal representations, but this can produce biased treatment effect estimates when the representations do not preserve the relevant confounding information. As a result, predictive models of baseline risk are commonly used in practice to guide treatment decisions, although they are not designed to identify which patients benefit most from treatment. We propose AACE (Annotation-Assisted Coarsened Effects), an *annotation-assisted* approach to causal policy learning for multimodal EHRs. The method uses expert-provided annotations during training to support confounding adjustment, and then predicts treatment benefit from only multimodal representations at inference. We show that the proposed method achieves strong empirical performance across synthetic, semi-synthetic, and real-world EHR datasets, outperforming risk-based and representation-based causal baselines, and offering practical insights for applying causal machine learning in clinical practice.} } @InProceedings{pmlr-v340-ashhad26a, title = {GRAFT: Decoupling Ranking and Calibration for Survival Analysis}, author = {Ashhad, Mohammad and Hoehndorf, Robert and Henao, Ricardo}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {111--140}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/ashhad26a/ashhad26a.pdf}, url = {https://proceedings.mlr.press/v340/ashhad26a.html}, abstract = {Survival analysis is complicated by censored data, high-dimensional features, and non-linear interactions. Classical models offer interpretability and superior calibration but are restricted to linear or predefined functional forms, while deep learning models are flexible and achieve strong discriminative performance, but tend to produce poorly calibrated survival estimates. To address this trade-off, we propose GRAFT (Gated Residual Accelerated Failure Time), a novel AFT model that decouples prognostic ranking from survival calibration. GRAFT’s hybrid architecture combines a linear AFT model with a non-linear residual neural network, and it also integrates stochastic gates for automatic feature selection. The model is trained by optimizing a differentiable, C-index-aligned ranking loss using stochastic conditional imputation from local Kaplan-Meier estimators, while calibrated survival estimates are obtained through simple post-training calibration. In public benchmarks, GRAFT outperforms baselines in discrimination and calibration, while remaining robust and sparse in high-noise settings.} } @InProceedings{pmlr-v340-aslam26a, title = {Scalable Clinical Data Infrastructure and Comparative ML Evaluation for Hospitalisation Risk Prediction in Elderly Patients with Multiple Long-Term Conditions using CPRD}, author = {Aslam, Asra and Chapman, Volodymyr and O'Connell, Maurice M. and Abuzour, Aseel S and Abaho, Micheal and Bollegala, Danushka and Leeming, Gary and Shantsila, Eduard and Clegg, Andrew and Walker, Lauren E and Buchan, Iain Edward and Relton, Samuel D}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {141--167}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/aslam26a/aslam26a.pdf}, url = {https://proceedings.mlr.press/v340/aslam26a.html}, abstract = {Deep learning architectures are increasingly proposed for patient trajectory modeling in electronic health records (EHRs), yet their advantage over simpler, more interpretable models is rarely subjected to rigorous empirical scrutiny in real-world clinical settings. We present a comprehensive patient timeline pipeline applied to elderly patients in CPRD Aurum, incorporating 260 clinical conditions classified via a three-tier automated framework including specialised detection logic for 17 complex conditions. Using this infrastructure, we benchmark Temporal Graph Convolutional Neural Networks (TG-CNN) against Logistic Regression with LASSO regularisation and Random Forests for predicting 12-month all-cause emergency hospitalisation risk, motivated by (but not filtered to) the elevated risk of adverse drug reactions. Under cross-validation, TG-CNN achieves a marginally higher mean AUC-ROC than LASSO (0.712 vs. 0.705), whereas on the held-out test set LASSO achieves the highest discrimination of three models (AUC-ROC 0.733, versus 0.710 for Random Forest and 0.702 for TG-CNN). We show, that discrimination alone is an incomplete criterion for clinical deployment: after Platt calibration, LASSO is the only model with an acceptable calibration slope (0.817), while Random Forest (0.759) and, TG-CNN (0.391) remain substantially miscalibrated. We argue that LASSO, not the highest-discriminating model, is the model best suited to direct clinical deployment. We present lessons for the machine learning and healthcare community regarding data infrastructure, model selection, and value of calibration and interpretability in high-stakes decision support.} } @InProceedings{pmlr-v340-baitairian26a, title = {Doubly valid and doubly sharp sensitivity analysis to unobserved confounding for survival outcomes}, author = {Baitairian, Jean-Baptiste and Sebastien, Bernard and JREICH, Rana and Katsahian, Sandrine and Guilloux, Agathe}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {168--213}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/baitairian26a/baitairian26a.pdf}, url = {https://proceedings.mlr.press/v340/baitairian26a.html}, abstract = {Time-to-event outcomes are central in oncology and rare diseases, where treatment effects are often summarized by differences in survival curves or Restricted Mean Survival Time (RMST). In real-world data, estimating these causal effects relies on the absence of unobserved confounding, an assumption that is rarely satisfied. We develop a sensitivity analysis framework for causal treatment effects with survival outcomes under the Marginal Sensitivity Model (MSM). We introduce doubly valid and doubly sharp (DVDS) bounds for differences in survival functions and RMST, extending recent DVDS results to the time-to-event setting while accounting for informative censoring. In practice, our method yields tighter bounds and improved computational efficiency compared to a previous approach from the literature, on simulated and real data. For tractability, we assume independence between censoring and unobserved confounding, a limit that should be addressed in future works.} } @InProceedings{pmlr-v340-bakumenko26a, title = {Early-Horizon Multimodal ICU Mortality Prediction Without Retraining}, author = {Bakumenko, Alexander and Smith, Dane Hudson and Hoelscher, Janine}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {214--240}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/bakumenko26a/bakumenko26a.pdf}, url = {https://proceedings.mlr.press/v340/bakumenko26a.html}, abstract = {Earlier ICU mortality prediction is more clinically useful because it can identify high-risk patients while treatment decisions can still change. Yet most models are trained on data from a fixed time window, so it is unclear whether a model trained on the first 48 hours of ICU data remains reliable when used earlier in the ICU stay. We evaluated a multimodal ICU mortality model trained once at 48 hours and then applied unchanged at 6, 12, 24, and 48 hours on MIMIC-III. The model combines an LSTM for physiological time-series data, a finetuned ClinicalModernBERT model for clinical notes, and a logistic regression fusion layer. Performance remained strong at earlier time points, suggesting that useful mortality prediction is possible earlier in the ICU stay even without retraining. At 6 hours, the model achieved AUROC 0.777 and remained well-calibrated (expected calibration error, ECE 0.038) without any recalibration, and it outperformed both single-modality models at every horizon. The benefit of combining both modalities was most evident at earlier horizons, when physiological data were sparse: agreement between the two specialists dropped by more than half from 48 to 6 hours, while the median contribution from clinical notes increased from 37% to 49%. A Bayesian version of the fusion layer showed that uncertainty decreased for survivors as more data accumulated but remained high for non-survivors; the most uncertain cases were up to 4.9 times more likely to be non-surviving patients. Continuous hourly analyses further showed that clinical notes provide stable context between documentation events. Simply carrying forward the most recent note matched or outperformed note-decay and documentation-gap alternatives. These results suggest that a multimodal ICU mortality model trained on 48 hours of data can provide trustworthy earlier predictions without retraining, while also identifying the cases that remain hardest to interpret.} } @InProceedings{pmlr-v340-banerjee26a, title = {ReXavatars: Evaluating Generated Patient Images for Clinical Realism and Reliability}, author = {Banerjee, Oishi and Willauer, Alexandra N. and Rajpurkar, Pranav}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {241--261}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/banerjee26a/banerjee26a.pdf}, url = {https://proceedings.mlr.press/v340/banerjee26a.html}, abstract = {Generative image models enable the creation of realistic “patient avatars" that can undergo virtual examinations, with potential applications in medical education and AI agent benchmarking. We evaluate Gemini-3’s ability to generate such avatars using clinician-verified prompts and clinician assessment of photorealism, composition, and clinical accuracy. We find that images are highly realistic and well-composed, often achieving an impressive level of structural and textural detail. We further find that Gemini-3 can maintain anatomical and environmental consistency across viewpoints or poses in 60 and 70% of cases respectively, a key requirement for interactive scenarios such as virtual examinations. Despite these strengths, clinical accuracy is inconsistent. When testing without additional constraints, 45 of 80 pathologies tested are generated without errors; performance degrades when we increase task complexity by requiring specific presentations or combinations of pathologies. Additionally, we find notable weaknesses in numerical reasoning tasks, with a 45% success rate on tasks requiring accurate counting. These results highlight the promise of patient avatars as a new interface for education and evaluation, while showing directions to improve clinical accuracy.} } @InProceedings{pmlr-v340-basit26a, title = {Beyond Dice: Clinically Structured Bias Analysis of Multiple Sclerosis Lesion Segmentation Models}, author = {Basit, Abdul and Hassan, Namik and Shafique, Muhammad}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {262--282}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/basit26a/basit26a.pdf}, url = {https://proceedings.mlr.press/v340/basit26a.html}, abstract = {Automated multiple sclerosis (MS) lesion segmentation is commonly benchmarked with voxel-overlap metrics such as Dice, although clinical interpretation also depends on lesion size, anatomical location, and the preservation of subject-level dissemination evidence. We present *Beyond-Dice*, a reproducible MS-specific evaluation framework spanning voxel fidelity, lesion detection, anatomical location, subject-level evidence, case-type stress testing, and split/merge topology. We compare nnU-Net, SegResNet, and SwinUNETR on a separate 24-subject MSSEG-1 evaluation cohort containing 1,172 lesions. Subject-bootstrap 95% confidence intervals (CIs) show statistically tied Dice: 0.713 [0.648, 0.772] for nnU-Net, 0.713 [0.669, 0.759] for SegResNet, and 0.708 [0.656, 0.756] for SwinUNETR. Non-overlap endpoints nevertheless separate clinically relevant behavior: nnU-Net improves subject-mean lesion recall over SegResNet by 0.051 [0.016, 0.088], whereas SwinUNETR produces 10.67 additional false-positive lesions per scan relative to nnU-Net [5.88, 16.38]. Across all models, 238 lesions are missed by all three and 226 have model-dependent detection; lesion volume is the dominant detection predictor (odds ratio 3.86 per log-volume unit [3.45, 4.74]). A modality-matched MSLesSeg-to-MSSEG-1 experiment further shows similar matched-lesion Dice across locations (0.584–0.659) despite markedly lower infratentorial recall/F1 (0.339/0.412). These results operationalize how Dice-tied models can preserve different clinical evidence and support endpoint-specific, uncertainty-aware model selection rather than a single-score leaderboard.} } @InProceedings{pmlr-v340-chan26a, title = {A Cloud–Edge System for Multimodal Clinical Screening in Resource-Constrained Rural Settings}, author = {Chan, Hei Ting Una and Wu, Chenwei and Liu, Xueshen and Zhao, Zesen and Zheng, Boyuan and Nakayama, Luis Filipe and Morley, Michael G and Shen, Liyue and Chen, Jiasi and Mao, Z. Morley}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {283--313}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/chan26a/chan26a.pdf}, url = {https://proceedings.mlr.press/v340/chan26a.html}, abstract = {Medical AI has demonstrated specialist-level diagnostic accuracy, yet these capabilities remain largely inaccessible in resource-constrained rural settings where bandwidth is scarce, compute is limited, and clinical decision-making requires integrating heterogeneous modalities. We introduce a cloud–edge collaborative architecture that addresses these constraints: lightweight, domain-specific models on the edge transform raw medical data into compact structured outputs, while a cloud LLM synthesizes these outputs into clinical summaries. An LLM-based orchestrator dynamically selects diagnostic tools based on patient context, promoting relevant modality coverage without processing irrelevant inputs. We evaluate on 100 multimodal clinical cases spanning cardiac, obstetric, trauma, ophthalmology, and screening scenarios — including sparse-input presentations with missing modalities and dense-input presentations with many overlapping inputs under three simulated network profiles (500 kbps–5 Mbps), reporting 95% confidence intervals throughout. The hybrid system attains the highest oracle accuracy (0.87–0.90) and the strongest factual grounding (KG precision up to 0.96), together with high coverage precision (0.95–0.99), while transmitting only $\tilde$6.5 KB of structured evidence to the cloud — three orders of magnitude less than cloud-only baselines. It maintains bandwidth-invariant latency (25–38 s) at up to 15$\times$lower token cost. These results highlight the role of architectural design in improving evidence selectivity and factual grounding, rather than merely reducing upload size, under deployment constraints.} } @InProceedings{pmlr-v340-chen26a, title = {Uncertainty-Aware mRMR for Transferable Feature Selection in High-Dimensional, Low-Sample-Size Biomedical Data}, author = {Chen, Xuehan and Davison, Brian D.}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {314--335}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/chen26a/chen26a.pdf}, url = {https://proceedings.mlr.press/v340/chen26a.html}, abstract = {In biomedical and clinical research, high-dimensional data are often collected from multiple heterogeneous source groups, such as different hospitals, patient populations, cancer types, or experimental conditions. Researchers often seek a compact set of features, such as genes, that is informative across source groups and remains useful in newly collected, unseen target groups. Despite its practical importance, this transferable feature selection problem in the high-dimensional, low-sample-size (HDLSS) setting remains underexplored. In this work, we propose Uncertainty-Aware mRMR (UA-mRMR), a training-free filter method that extends the minimum Redundancy Maximum Relevance (mRMR) feature selection method by accounting for sample-size-dependent statistical uncertainty across heterogeneous source groups. Specifically, UA-mRMR uses Fisher z-transformed confidence bounds to stabilize correlation-based relevance and redundancy estimates under unequal sample sizes. Experiments on real-world multi-source biomedical datasets show that features selected by UA-mRMR transfer effectively to unseen target groups and achieve strong downstream predictive performance.} } @InProceedings{pmlr-v340-chen26b, title = {Do LLMs Have a Sense of Time? Zero-Shot Survival Curve Prediction with Frontier Language Models}, author = {Chen, Emma and Yuan, Grace Chang and Zhou, Christine Yang and Kim, David A and Rajpurkar, Pranav}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {336--393}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/chen26b/chen26b.pdf}, url = {https://proceedings.mlr.press/v340/chen26b.html}, abstract = {Survival curve prediction is an underexplored formulation for time-to-event prediction with large language models (LLMs). We benchmark four frontier LLMs on zero-shot survival curve prediction for emergency department (ED) revisit and hospital readmission, comparing against regression and ordinal classification as alternative prediction formulations. Survival prediction is the top-performing formulation for every model on readmission prediction and three of four on ED revisit prediction, with Harrell’s C-index ranging from 0.703–0.732 and 0.713–0.802 respectively. Frontier LLMs can produce largely well-formed survival curves zero-shot, with all predictions being valid probabilities and monotonicity holds on 99–100% of predicted curves without post-hoc correction; the initial value condition $S(0)=1$ is satisfied exactly by two of four models on both tasks and by a third model on readmission, and approximately otherwise. Beyond structural validity, the best LLMs approach a supervised Random Survival Forest baseline (C-index 0.732 vs. 0.741) on readmission prediction with no training data. Compared to direct binary prompting, survival-derived probabilities are substantially better calibrated at every horizon tested, with ECE reductions of up to 7.5$\times$ at short horizons. Structured multi-agent deliberation achieved C-indices of 0.734 for readmission and 0.790 for ED revisit; it significantly improved on the weakest baseline for ED Revisit ($\Delta=+0.031$, two-sided $p=0.049$ for GPT-5) but did not consistently outperform the strongest individual model. These results establish survival curve prediction as the preferred formulation for LLM-based clinical time-to-event prediction.} } @InProceedings{pmlr-v340-cho26a, title = {Tree of Concepts: Interpretable Continual Learners in Non-Stationary Clinical Domains}, author = {Cho, Dongkyu and Li, Xiyue and Adhikari, Samrachana and Chunara, Rumi}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {394--409}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/cho26a/cho26a.pdf}, url = {https://proceedings.mlr.press/v340/cho26a.html}, abstract = {Continual learning updates models under distribution shift while limiting performance loss on previously seen data, but clinical use also motivates inspectable interfaces that can be compared across model versions. We propose Tree of Concepts, which uses a shallow decision tree to define a fixed vocabulary of root-to-leaf rules and sequentially updates a neural concept predictor and label head with replay. The association between each concept ID and its rule is fixed; predicted concept assignments and the concept-to-label mapping remain adaptive. Across the evaluated health-related tabular protocols, Tree of Concepts attains the highest or tied mean past- and current-slice scores among the compared methods. These comparisons are descriptive point estimates rather than evidence of statistical superiority. Leaf-assignment agreement with the frozen tree ranges from 0.81 to 0.91, indicating that a fixed vocabulary does not imply invariant instance-level explanations. The results motivate further evaluation of fixed rule vocabularies as longitudinally comparable interfaces for continually updated tabular models.} } @InProceedings{pmlr-v340-dronavajjala26a, title = {Decision-Informed Online Conformal Prediction for ICU Resource Allocation}, author = {Dronavajjala, Chandra}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {410--434}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/dronavajjala26a/dronavajjala26a.pdf}, url = {https://proceedings.mlr.press/v340/dronavajjala26a.html}, abstract = {Hospitals allocate scarce ICU resources—nurses, beds, discharge slots—using predictions of patient length of stay (LOS), but prediction errors under shifting patient populations lead to costly misallocations. Online conformal prediction provides adaptive uncertainty sets that track a target coverage rate (e.g., 90%) without distributional assumptions, but the resulting robust decisions incur a Price of Coverage—the excess cost of hedging against uncertainty. We observe that standard conformal methods distribute uncertainty budgets uniformly, including on resources the optimizer assigns to minimum levels. We propose Decision-Informed Conformal Adaptation (DICA), which uses the downstream optimization’s resource allocation as feedback to reshape uncertainty margins: tighter where allocation is minimal (saving cost), wider where allocation is high (strengthening protection). DICA preserves the same adaptive quantile update and scalar coverage tracking as standard online conformal; the reshaped radii do not carry a formal per-component guarantee, but decision-level analysis shows that coverage misses are confined to lower-bound dimensions where their clinical cost is negligible. Across 328K patient stays from three real clinical datasets (MIMIC-IV, eICU, and a general hospital cohort), DICA reduces the Price of Coverage by 43–54% relative to uniform conformal while maintaining approximately 90% coverage. Under chronological patient ordering, static calibration methods degrade to 78–86% coverage, while online conformal methods remain near the target rate.} } @InProceedings{pmlr-v340-du26a, title = {Possible or Definite? A Benchmark for Evaluating Diagnostic Uncertainty Preservation in Clinical Text}, author = {Du, Hongbo and Lu, Zixin and Qu, Jiaming}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {435--460}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/du26a/du26a.pdf}, url = {https://proceedings.mlr.press/v340/du26a.html}, abstract = {Large language models (LLMs) are increasingly used for clinical text tasks such as summarization and revision. While most studies evaluate the fluency and coherence of LLM-generated text, whether LLMs correctly preserve diagnostic uncertainty remains underexplored. In clinical practice, phrases such as “possible pneumonia” communicate the strength of available evidence and directly guide decisions about follow-up testing and treatment. Altering these uncertainty expressions can change the clinical meaning entirely. In this paper, we systematically evaluated this problem in two steps. First, we constructed a benchmark of 1,200 clinical documents with 9,184 uncertainty annotations across five levels. Second, we evaluated three LLMs on this benchmark. Our results show that (1) LLMs preserve the original uncertainty cues poorly, often less than half the time; (2) LLMs struggle with nuanced distinctions between adjacent levels. This work reveals a failure mode not captured by standard evaluation metrics and provides implications for the safe deployment of LLMs in clinical workflows.} } @InProceedings{pmlr-v340-du26b, title = {Explainable Transformer Models for Clinical Prediction Tasks on Structured Electronic Health Records}, author = {Du, Jun Ni and Adamek, Lukas and Kryukov, Maxim and Dormont, Flavio and Bar-Joseph, Ziv and Jager, Sven and Rufino, Brandon}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {461--494}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/du26b/du26b.pdf}, url = {https://proceedings.mlr.press/v340/du26b.html}, abstract = {Predictive models over structured electronic health records (EHRs) remain central to machine learning for healthcare, but few have jointly emphasized quantitative laboratory information and interpretability with respect to input medical events. We present BERT-LER, a BERT-style model for coded EHR timelines pretrained and fine-tuned from a de-identified EHR dataset of 75 million patients, that encodes laboratory test results as discrete tokens while retaining graded information through percentile-based binning, paired with Integrated Gradients for token-level attributions grounded in the input EHR sequence. We benchmark our approach on the public EHRShot benchmark suite and on an asthma severity progression study based on real-world data. This addresses a methodological gap in EHR foundation-style modeling by unifying laboratory value representation and explainability in a single framework, while assessing whether both predictive performance and explanations generalize beyond standard clinical prediction tasks. Across EHRShot and asthma tasks, BERT-LER achieves predictive performance that is competitive with, and on laboratory-related tasks often exceeds, publicly available benchmark models, and provides attributions that align with clinically known risk factors. Our architecture and explainability approach can be applied to many therapeutic areas and prediction tasks using language models trained on structured EHRs.} } @InProceedings{pmlr-v340-dzikunu26a, title = {Learning Prostate Anatomy at Test Time for Cancer Detection in Micro-Ultrasound}, author = {Dzikunu, Obed Korshie and Abootorabi, Mohammad Mahdi and Harmanani, Mohamed and Wilson, Paul F R and Willis, Emma and Luger, Ferdinand and Kinnaird, Adam and Wodlinger, Brian and Mousavi, Parvin and Abolmaesumi, Purang}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {495--520}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/dzikunu26a/dzikunu26a.pdf}, url = {https://proceedings.mlr.press/v340/dzikunu26a.html}, abstract = {Domain shift across clinical centers using different imaging hardware or acquisition protocols remains a fundamental barrier to deploying deep learning models for prostate cancer (PCa) detection. Existing test-time adaptation (TTA) methods address distribution shift through entropy minimization or augmentation-based self-supervision, correcting for statistical differences in image appearance but ignoring the anatomical structure of the target domain. We propose ANT, a segmentation-guided TTA framework that adapts a pretrained cancer detection encoder to the target domain by solving an auxiliary prostate segmentation task at test time, supervised by pseudo-masks from a frozen pretrained segmentation network. By aligning encoder representations to prostate anatomy in the target domain, ANT corrects domain-specific feature drift while preserving cancer-discriminative structure. The model was trained on 693 patients imaged with an earlier-generation micro-ultrasound scanner in a multi-center clinical trial, and evaluated on 118 patients acquired with a newer-generation system across two centers in another clinical trial. Under a leave-one-center-out protocol with identical evaluation conditions across all methods, ANT improves mean AUC by 2.9% and 3.6% at the biopsy-core and patient levels, respectively, over no adaptation, outperforming TTA baselines. Code is available at: https://github.com/ObedDzik/ant.git.} } @InProceedings{pmlr-v340-garg26a, title = {A Data-Driven Approach towards Improving Deceased-Donor Kidney Utilization}, author = {Garg, Aditya Shankar and Feigenbaum, Itai and Yu, Miko E. and Mohan, Sumit and Sethuraman, Jay}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {521--549}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/garg26a/garg26a.pdf}, url = {https://proceedings.mlr.press/v340/garg26a.html}, abstract = {Deceased donor kidneys in the United States remain substantially underutilized: Approximately 30% of recovered kidneys are discarded, and many other kidneys are transplanted in suboptimal condition due to delays. A major contributing factor is the allocation procedure, which offers kidneys sequentially to patients in priority order (“match run"). Offers are often declined, and repeated declines delay transplant, prolong cold ischemia time, and increase the risk of discarding the kidney. We propose a machine-learning framework to identify kidneys at risk of underutilization. Such a framework can be used to expedite the offering process and shorten the time lost due to strings of declines. We use only medical and historical information available upon arrival of the donor to the system. On the dataset of U.S. kidney offers, we predict discards well (AUROC 0.993, AUPRC 0.977). Furthermore, we are able to accurately identify 86% of the kidneys at the risk of underutilization. Our framework consists of three models, which are each an ensemble of decision trees. The first model, OfferPred, directly evaluates the chances of an offer acceptance for a particular kidney-patient pair; however, applying this model directly to evaluate an entire match run leads to error accumulation. Therefore, we train models DiscardPred and LocationPred, which indirectly use OfferPred to predict the chance of discard and identify some of the initial string of declines. Remarkably, DiscardPred and LocationPred do not directly use granular information about patients and transplant centers in the offer sequence, but rather replace these with aggregate data from OfferPred’s input. Our models shows that underutilization risk is not solely a function of donor characteristics but is strongly correlated with the structure of the match run and recent historical center behavior.} } @InProceedings{pmlr-v340-han26a, title = {ELF: A Family of Encoder-Free ECG-Language Models}, author = {Han, William and Chen, Tony and Duan, Chaojing and Song, Xiaoyu and Yao, Yihang and Yang, Yuzhe and Rosenberg, Michael and Liu, Emerson and Zhao, Ding}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {550--583}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/han26a/han26a.pdf}, url = {https://proceedings.mlr.press/v340/han26a.html}, abstract = {ECG–Language Models (ELMs) extend recent advances in Multimodal Large Language Models (MLLMs) to automated ECG interpretation. However, most existing ELMs inherit Vision–Language Model (VLM) design choices and rely on pretrained ECG encoders, introducing substantial architectural and training complexity. Inspired by encoder-free VLMs, we introduce ELF, a family of three encoder-free ELMs that remain competitive with, and often outperform, prior state-of-the-art ELMs across two datasets despite substantially simpler architectures and training pipelines. All code and data are available at https://github.com/ELM-Research/ECG-Language-Models.} } @InProceedings{pmlr-v340-huang26a, title = {Amortized Data Borrowing with Exchangeability-Aware Neural Posterior Estimation}, author = {Huang, Chin-Hung and Lee, JooChul and He, Huan}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {584--610}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/huang26a/huang26a.pdf}, url = {https://proceedings.mlr.press/v340/huang26a.html}, abstract = {Augmenting small concurrent studies with external or historical cohorts is attractive in drug development, where enrollment is slow, follow-up is expensive, and closely related trial or real-world data are often already available. Bayesian dynamic borrowing (BDB) provides a principled framework for adaptively controlling the influence of external data, but classical implementations often depend on hand-specified priors and MCMC-based inference, which can be computationally expensive and not generalizable. In this work, we study amortized neural posterior estimation (NPE) as a flexible alternative. A single network is pretrained on simulated current/external dataset pairs spanning covariate shift, outcome drift, and joint non-exchangeability, and then returns an approximate posterior for a scalar current-study target in a single forward pass. Through simulation studies, we find that NPE is most useful under outcome drift and joint mismatch: in the harder outcome-drift regimes, it gives up to about five-fold lower absolute bias than the best classical baseline and keeps Type I error close to nominal. After pretraining, posterior summaries are obtained in about 8 ms per dataset, roughly $10^3\times$ faster than MCMC-based borrowing baselines in our timing experiment. We further analyze Alzheimer’s Disease Neuroimaging Initiative (ADNI) data and show that, when mild cognitive impairment outcomes differ across cohorts, the NPE formulation recovers the later-cohort risk level in this example without claiming greater precision. Code is available at \url{https://github.com/ChinHungScott/NPE-for-Bayesian-Dynamic-Borrowing-MLHC-}.} } @InProceedings{pmlr-v340-huang26b, title = {Synthesizing Post-Acetazolamide Cerebral Blood Flow Maps from Baseline MRI in Moyamoya Using 3D Generative AI}, author = {Huang, Julia and Gonzalez, Camila and Goyal, Rydham and Zou, Aja and Alexander, Sasha and Moseley, Michael and Zhao, Moss Y. and Steinberg, Gary K.}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {611--635}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/huang26b/huang26b.pdf}, url = {https://proceedings.mlr.press/v340/huang26b.html}, abstract = {For patients with Moyamoya disease, impaired cerebrovascular reserve (CVR) is an important hemodynamic criterion for recommending extracranial-to-intracranial bypass surgery, making reliable CVR assessment central to treatment planning. The reference protocol used in this cohort requires paired arterial spin labeling (ASL) perfusion MRI acquired before and after administration of the vasodilator acetazolamide (ACZ). ACZ may be contraindicated or avoided in patients with substantial renal dysfunction, relevant hypersensitivity, marked electrolyte derangement, or pregnancy, depending on clinical circumstances and institutional protocol. For affected patients in whom ACZ is not administered, the standard two-scan protocol cannot be completed as intended, and the post-ACZ cerebral blood flow (CBF) data used for hemodynamic assessment and bypass planning are unavailable. We propose CAE3D, a deterministic 3D conditional autoencoder that synthesizes post-ACZ CBF maps directly from pre-ACZ ASL input. Among eleven evaluated models (CAE3D, seven directly comparable full-volume in-house baselines spanning deterministic and diffusion-style variants, one middle-slice 2D contextual baseline (CAE 2D), and two frozen-encoder foundation adapters reported separately as contextual references), CAE3D achieves the lowest held-out MAE (MAE 0.066, SSIM 0.80, PSNR 24.0 dB) with near-zero full-brain mean bias, though its regional $\Delta$CBF predictions compress the dynamic range in high-response territories. Its MAE advantage was statistically significant (paired Wilcoxon, Holm-adjusted) over seven of the eight other trained-from-scratch baselines, with the exception of the 2D middle-slice CAE 2D comparator; its SSIM and PSNR advantages were significant over all eight. These results establish the retrospective feasibility of post-ACZ CBF synthesis in patients who completed the standard two-scan protocol; extension to ACZ-contraindicated patients, who were not represented in this cohort, awaits external and prospective validation.} } @InProceedings{pmlr-v340-iqbal26a, title = {Shortcut-Aware Modeling of Serial ECGs for Cardiac Sarcoidosis}, author = {Iqbal, Humza A. and Birnie, David and Sun, Christopher LF}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {636--660}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/iqbal26a/iqbal26a.pdf}, url = {https://proceedings.mlr.press/v340/iqbal26a.html}, abstract = {Cardiac sarcoidosis is difficult to diagnose, and serial ECG data may encode both physiologic signal and observation-process shortcuts. Here, observation-process shortcuts are predictive signals from ECG count and inter-ECG intervals rather than ECG waveform physiology. We study patient-level prediction of future CS diagnosis from serial ECG data collected before clinical presentation at a tertiary cardiac center. We compare single-ECG models, shortcut-only controls, shortcut-vulnerable serial ECG models, and shortcut-aware serial ECG models under unmatched evaluation and matched shortcut-control evaluation. We introduce RECAP, a shortcut-aware joint architecture-and-training design that represents prior ECGs relative to the patient’s latest eligible ECG, conditions these relationships on elapsed time, and uses matched-pair loss to reduce reliance on the measured shortcut variables. In the unmatched evaluation, shortcut-only controls are strongly predictive, showing that apparent serial gains can be driven by shortcut variables. In the matched shortcut-control evaluation, shortcut-only controls collapse to chance, and RECAP achieves the highest matched AUROC among the evaluated configurations. These results show that serial ECG waveforms retain predictive signal after targeted shortcut control and that rigorous evaluation in serial ECG settings requires explicit shortcut-only controls and matched shortcut-control evaluation.} } @InProceedings{pmlr-v340-issa26a, title = {A Structure-Constrained Neural Simulator for Population PK under Regimen Shift (ConstrainNODE-PK)}, author = {Issa, Ali and Sahota, Tarjinder and Buil-Bruna, N\'{u}ria and Standing, Joseph F and Kloprogge, Frank}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {661--689}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/issa26a/issa26a.pdf}, url = {https://proceedings.mlr.press/v340/issa26a.html}, abstract = {Prospective dose selection requires simulating concentration distributions under regimens for which no concentrations have yet been observed. In this setting, success is not defined by low individual prediction error alone but by calibrated population simulation under regimen shift: a model can fit held-out individual observations yet still miscalibrate exposure distributions and underestimate high-exposure risk when the dosing schedule changes. We present ConstrainNODE-PK, a structure-constrained neural population pharmacokinetic simulator for intravenous bolus dosing under linear kinetics. The model expresses concentration as an exact superposition of dose events applied to a learned unit-dose impulse response, so transfer to a new regimen changes only the event sequence. The response is parameterized as a positive exponential mixture with a monotone residual neural ordinary differential equation, enforcing non-negativity, stability and monotone post-bolus decay; together with exact superposition, this gives exact dose proportionality while remaining flexible enough to capture one- to three-compartment-like behavior without per-compound compartment selection. We evaluate on six simulated settings and two public cohorts, using held-out subjects throughout and assessing both individualized prediction and population-level simulation calibration. The primary stress test is strict transfer from once-daily to three-times-daily dosing, with no target-regimen concentrations available at inference time. In this setting, ConstrainNODE-PK maintains near-nominal 95% prediction-interval coverage (95.6%), whereas a matched unconstrained dose-aware neural ordinary differential equation falls to 45.8%. Across settings, the model remains competitive with matched nonlinear mixed-effects references on individualized prediction and preserves near-nominal prior-predictive population calibration. Strict regimen-transfer evidence comes from controlled simulated intravenous-bolus settings; the African Research on Kidney Disease iohexol and Tobramycin cohorts provide complementary held-out within-cohort population-calibration checks rather than strict real-world transfer validation. More broadly, when a healthcare model is used to simulate a future intervention, the relevant validation target is calibration under the induced intervention shift, not only in-regimen fit.} } @InProceedings{pmlr-v340-janizek26a, title = {Visual concept ranking uncovers medical shortcuts used by large multimodal models}, author = {Janizek, Joseph David and Xu, Sonnet and Lateef, Junayd and Daneshjou, Roxana}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {690--729}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/janizek26a/janizek26a.pdf}, url = {https://proceedings.mlr.press/v340/janizek26a.html}, abstract = {Ensuring the reliability of machine learning models in safety-critical domains such as healthcare requires auditing methods that can uncover model shortcomings. We introduce a method for identifying important visual concepts within large multimodal models (LMMs) and use it to investigate the behaviors these models exhibit when prompted with medical tasks. We primarily focus on the task of classifying malignant skin lesions from clinical dermatology images. After showing how LMMs display unexpected gaps in performance between different demographic subgroups when prompted with demonstrating examples, we apply our method, Visual Concept Ranking (VCR), to these models and prompts. VCR generates hypotheses related to different visual feature dependencies, which we are then able to validate with manual interventions.} } @InProceedings{pmlr-v340-jia26a, title = {Composable Parameter-Level Reinforcement Learning for Dual-Chamber Pacemaker Programming}, author = {Jia, Leqi}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {730--754}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/jia26a/jia26a.pdf}, url = {https://proceedings.mlr.press/v340/jia26a.html}, abstract = {Unnecessary right ventricular pacing (VP) in dual-chamber (DDD) pacemakers is associated with pacing-induced cardiomyopathy, heart failure, and atrial fibrillation. Commercial VP-minimization algorithms such as Medtronic’s Managed Ventricular Pacing (MVP) operate at the beat level and sharply reduce VP burden in patients with intact atrioventricular (AV) conduction (e.g., sick sinus syndrome), but are limited in patients with intermittent AV block, where the algorithm cannot give conduction more time to complete. We propose a hierarchical reinforcement learning framework that operates at the parameter level (AV delay and activity-adaptive target rate) and composes cleanly on top of any certified beat-level controller. Trained via PPO on a clinically parameterized event-level surrogate, the RL policy reduces VP burden from 91.8–95.2% to 5.9–7.8% on Mobitz Type I and Type II when layered on top of standard DDDR. When layered on top of a Casavant-2021-specified MVP controller, it further reduces VP burden from 3.4% to 0.4% on first-degree AV block (driven by AV-delay widening, the same lever as commercial Search-AV) and from 48.6% to 9.2% on Mobitz Type II, while preserving MVP’s near-zero VP (0.2% to 0.1%) on sick sinus syndrome. The MVP implementation is specified against the Casavant–Belk 2021 technical reference and calibrated per subgroup against the IDEAL RVP and COMPARE trials. The parameter-level and beat-level optimizations act on separable timing primitives and combine additively, suggesting a composable deployment path in which a learned parameter-programming layer sits on top of, rather than replaces, certified commercial firmware.} } @InProceedings{pmlr-v340-jiang26a, title = {KMGen: A Skill-based Approach for Synthetic Individual Patient Data Generation}, author = {Jiang, Jalen and Gao, Chufan and Rasmussen, Ethan and Xie, Stephen Z. and Sun, Jimeng}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {755--814}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/jiang26a/jiang26a.pdf}, url = {https://proceedings.mlr.press/v340/jiang26a.html}, abstract = {Individual patient data (IPD) from clinical trials is the substrate for survival modeling, meta-analysis, and safety research, yet IPD is rarely released. Prior work has addressed only half of this gap: reconstructing Kaplan–Meier (KM) curves from published plots — typically requiring manual digitization or human-in-the-loop correction — while offering no mechanism for generating the adverse-event (AE) streams that constitute the other half of a patient record. We introduce KMGen, the first end-to-end framework that (i) fully automates KM curve extraction at accuracy competitive with human-guided tools, and (ii) generates synthetic per-patient AE trajectories from public trial registry records. The extraction stage is a fully automated agentic pipeline — an agent generates code to extract each step in the KM curve — achieving a mean Integrated Absolute Error (IAE) of 0.0151 on a 32-plot benchmark spanning clean, edge-case, and adversarial conditions. The IPD generation stage decouples patient archetype extraction from statistical sampling: an LLM distills the trial record into arm-specific statistics, adverse events, patient demographics, and risk multipliers. A mechanistic sampler generates patient events via clinical archetypes, bootstrap rank-correlation coupling to the empirical KM curve (preserving the marginal survival distribution exactly), and cycle-based AE scheduling with an induction/maintenance split. Across three held-out oncology trials spanning an order of magnitude in cohort size and 30 independent regenerations per trial, KMGen achieves mean integrated KM absolute difference $\Delta_{\mathrm{KM}} \leq 0.051$, sex/ECOG JSD $\leq 0.013$ on 5 of 6 demographic slots, and recovers $\geq 71%$ of the top-15 AEs by exact MedDRA term under a single fixed parameter set. The pipeline is interpretable end-to-end and released as open source at https://github.com/chufangao/kmgen.} } @InProceedings{pmlr-v340-khan26a, title = {SHIFT-M3: Pre-fusion Alignment-based Consistency Screening for Multimodal ECG Record Integrity}, author = {Khan, Md Ashik and Siddique, Md Nahid}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {815--834}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/khan26a/khan26a.pdf}, url = {https://proceedings.mlr.press/v340/khan26a.html}, abstract = {Multimodal clinical AI typically assumes that the waveform, report, metadata, and downstream predictions attached to a record belong to the same patient. In practice, linkage failures can silently assemble individually plausible but cross-patient components, creating a safety problem that standard predictive models are not designed to detect. We study this problem as multimodal record integrity triage: given an assembled record, should its modalities be trusted to belong together? We introduce SHIFT-M3, a lightweight text-based pre-fusion screen that measures alignment-based consistency between two separately produced ECG text views: an LLM-generated interpretation and a clinical report summary. On 784,680 MEETI ECG records, SHIFT-M3 achieves 97.6% TPR@5% FPR for full text-view swaps (AUROC 0.996), 90.3% for partial swaps (AUROC 0.974), and 97.7% for label-matched hard negatives (AUROC 0.996) with only 573,569 parameters. Compared with same-dataset lexical baselines, the gains are largest on partial swaps and hard negatives, suggesting that the model is learning more than surface overlap. We also introduce the CMST (Conflict-type Multimodal Stress Test) evaluation taxonomy, a three-seed stability study, a loss ablation, a temporal-tolerance sweep, and a shared-token masking control. The main remaining failure mode is longitudinal ambiguity: at the default operating point, same-patient cross-visit pairs still produce 87.0% Type-II false positives.} } @InProceedings{pmlr-v340-kim26a, title = {Enhancing LLMs’ Clinical Reasoning with Real-World Data from a Nationwide Sepsis Registry}, author = {Kim, Junu and Shim, Chaeeun and Park, Sungjin and Lee, Su Yeon and Suh, Gee Young and Choi, Seong Jin and Moon, Song Mi and Song, Kyoung-Ho and Kim, Eu Suk and Kim, Hong Bin and Kim, Sejoong and Im, Chami and Kang, Dong-Wan and Kim, Yong Soo and Bae, Hee-Joon and Lim, Sung Yoon and Jeong, Han-Gil and Choi, Edward}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {835--885}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/kim26a/kim26a.pdf}, url = {https://proceedings.mlr.press/v340/kim26a.html}, abstract = {Although large language models (LLMs) have demonstrated impressive reasoning capabilities across general domains, their effectiveness in real-world clinical practice remains limited. This limitation is likely due to insufficient exposure to real-world clinical data during training, as such data are typically excluded because of privacy concerns. To address this gap, we trained an LLM on real-world clinical data from a nationwide sepsis registry and evaluated the reasoning improvements across diverse datasets and tasks. The trained model demonstrated strong clinical reasoning performance on in-domain test sets, supported by both quantitative metrics and expert evaluations. Moreover, these enhanced reasoning capabilities generalized to an external sepsis dataset involving different tasks and patient cohorts, an open-ended antibiotic consultation task, and a disease beyond sepsis. Future research should focus on training LLMs on large-scale, multi-disease clinical datasets to enable more powerful and general-purpose clinical reasoning models.} } @InProceedings{pmlr-v340-kim26b, title = {Quantifying Boundary Reliability in Endoscopic Ultrasound Pancreas Segmentation via Evidential Deep Learning}, author = {Kim, Hana and Husain, Khalid Ahmed and Li, Jianing and Akshintala, Venkata S and Vedula, S. Swaroop and Canto, Marcia Irene and Jones, Craig}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {886--912}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/kim26b/kim26b.pdf}, url = {https://proceedings.mlr.press/v340/kim26b.html}, abstract = {Pancreatic cancer has a very high mortality rate, making its early detection essential for improving patient survival. Endoscopic Ultrasound (EUS) offers spatial resolution necessary for early-stage screening, but it is sensitive to artifacts, noise, and low contrast, which make automated image analysis challenging. We present an uncertainty-aware framework for EUS pancreas segmentation, benchmarking four U-Net variants across multi-institutional datasets. nnU-Net achieved the strongest performance with a Dice score of $0.811 \pm 0.122$ on our internal data. We introduce Inside Ground Truth Annulus Percentage (IGTP), a clinically-motivated metric revealing that models systematically fail to adhere to distal pancreatic margins where acoustic shadowing is severe. To localize these failures, we trained an evidential nnU-Net that parameterizes pixel-wise class probabilities using a Dirichlet distribution and learns evidence through an expected cross-entropy objective with KL regularization. The learned Dirichlet parameters enable the estimation of aleatoric and epistemic uncertainty in a single forward pass. Quantitative analysis showed that higher epistemic uncertainty was associated with poorer segmentation and boundary adherence and provided a predictive signal to identify low-performing cases. Together, the proposed IGTP metric and trained evidential framework establish a trustworthy foundation for AI-assisted EUS screening.} } @InProceedings{pmlr-v340-krishnamoorthy26a, title = {Survival Analysis with Limited Overlap and Censoring Distribution Shift}, author = {Krishnamoorthy, Meera and Tjandra, Donna and Shanmugam, Divya M and Kowalski, Amanda E. and Wiens, Jenna}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {913--947}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/krishnamoorthy26a/krishnamoorthy26a.pdf}, url = {https://proceedings.mlr.press/v340/krishnamoorthy26a.html}, abstract = {Survival analysis methods are often used to predict the time until the onset of an event in settings when the true time-to-event (TTE) may be censored during training. Such approaches typically assume uncensored data are representative of censored data and that the probability of censoring conditioned on the covariates remains constant over time, i.e., there is no censoring distribution shift. However, both assumptions can fail in practice when censoring results from interventions targeted to individuals with particular comorbidities or genetic markers (such as prophylactic surgery when predicting time to cancer onset, or scheduled cesarean delivery and induction when predicting time to spontaneous labor) and changes in clinical policies alter which individuals are targeted for these interventions over time. To address this, we propose a new approach, cluster-weighted inference of time-to-event (CWITE), that remains accurate when these assumptions do not hold. Unlike existing approaches that ignore times-to-censoring (TTC) or treat them only as a lower bound of the TTE, CWITE leverages the insight that a subset of censored individuals are likely censored close to their true TTEs, and uses a novel mechanism to learn from such individuals. On the task of predicting time to spontaneous labor using real-world data, CWITE improves TTE accuracy for individuals similar to censored training data (mean absolute error: 6.50 days, 95% CI: [5.55, 7.40] vs. 7.82 days, [6.82, 8.82]) while maintaining comparable performance for those similar to uncensored training data (6.50 days, [5.54, 7.61] vs. 6.63 days, [5.67, 7.69]). Our results demonstrate that incorporating more specific supervision from censored training data can significantly improve TTE predictions in settings with limited overlap and censoring distribution shift, challenges common in real-world clinical data. Code to implement CWITE and reproduce all experiments in the paper is available at https://github.com/MLD3/CWITE.} } @InProceedings{pmlr-v340-lee26a, title = {Test-Time Adaptation for EEG Foundation Models: A Systematic Study under Real-World Distribution Shifts}, author = {Lee, Gabriel Jason and Pradeepkumar, Jathurshan and Sun, Jimeng}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {948--984}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/lee26a/lee26a.pdf}, url = {https://proceedings.mlr.press/v340/lee26a.html}, abstract = {Electroencephalography (EEG) foundation models have shown strong potential for learning generalizable representations from large-scale neural data, yet their clinical deployment is hindered by distribution shifts across clinical settings, devices, and populations. Test-time adaptation (TTA) offers a promising solution by enabling models to adapt to unlabeled target data during inference without access to source data, a valuable property in healthcare settings constrained by privacy regulations and limited labeled data. However, its effectiveness for EEG remains largely underexplored. In this work, we introduce NeuroAdapt-Bench, a systematic benchmark for evaluating test-time adaptation methods on EEG foundation models under realistic distribution shifts. We evaluate representative TTA approaches from other domains across multiple pretrained foundation models, diverse downstream tasks, and heterogeneous datasets spanning in-distribution, out-of-distribution, and extreme modality shifts (e.g., Ear-EEG). Our results show that the evaluated TTA methods yield inconsistent gains and often degrade performance, with gradient-based approaches particularly prone to heavy degradation and optimization-free methods showing greater stability. For the evaluated EEG foundation models and representative TTA methods, these findings highlight the limitations of directly applying existing TTA techniques to EEG and underscore the need for domain-specific adaptation strategies. Code is available at https://github.com/leegabriel/NeuroAdapt-Bench.} } @InProceedings{pmlr-v340-li26a, title = {SPARK: Lightweight Adaptation of Healthcare Foundation Models via Steering Knowledge Circuits}, author = {Li, Bing and Fu, Qiang and Long, Yuwei and Yang, Huiyuan}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {985--1012}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/li26a/li26a.pdf}, url = {https://proceedings.mlr.press/v340/li26a.html}, abstract = {Large-scale foundation models (FMs) have demonstrated immense potential in medical time-series analysis, yet their static, “one-size-fits-all" nature limits their efficacy in patient-specific clinical settings. Traditional adaptation paradigms, such as full fine-tuning or standard parameter-efficient fine-tuning (PEFT), are computationally prohibitive for edge-device deployment and often overfit when faced with the chronic data scarcity and long-tail distributions characteristic of healthcare. In this work, we reveal that pre-trained medical FMs inherently possess sparse, multi-layer “Physiological Knowledge Circuits" dedicated to representing distinct clinical patterns. Leveraging this mechanistic insight, we propose SPARK (Steering Personalized Adaptation via Routing Knowledge), a novel, ultra-lightweight adaptation framework. Instead of updating the model’s global weights, SPARK utilizes a Multi-Layer Personalized Steering Hub (M-PSH) to surgically inject targeted perturbations into the hidden state activations of these specific knowledge circuits. Extensive experiments across multiple clinical ECG benchmarks demonstrate that SPARK achieves a superior parameter-to-performance ratio and excels in extreme data-scarce environments, yielding an approximately 6.8% accuracy improvement in 1-shot adaptation scenarios. Furthermore, by modulating internal representations rather than relearning them from scratch, SPARK significantly enhances model robustness against rare diseases, paving the way for privacy-preserving, dynamic, and on-device personalization in continuous health monitoring.} } @InProceedings{pmlr-v340-li26b, title = {Handling onset age inconsistencies in longitudinal healthcare survey data}, author = {Li, Wanxin and YUAN, MING and Park, Yongjin P and Duc, Khanh Dao}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1013--1055}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/li26b/li26b.pdf}, url = {https://proceedings.mlr.press/v340/li26b.html}, abstract = {Longitudinal healthcare surveys frequently contain inconsistencies in self-reported onset ages, where participants report different ages for the same condition between enrollment and follow-up surveys. We propose two methods to handle this challenge. First, we introduce a procedure that aggregates inconsistency patterns to construct participant-level reliability scores, enabling researchers to stratify participants and prioritize analysis on high-reliability cohorts. Second, we present a Bayesian adjustment method that models enrollment and follow-up reports as noisy observations of a latent true onset age, producing adjusted estimates for the inconsistent observations that account for age-dependent and inter-survey-time effects. We evaluate both methods using data from the Canadian Partnership for Tomorrow’s Health, where 57.1% of participants exhibit onset age inconsistencies. In general, both methods substantially strengthen correlations between biologically related conditions and improve predictive performance across classification and regression tasks. In addition, high-reliability cohorts from reliability score-based stratification reveal more coherent and interpretable disease clustering networks, and Bayesian adjustment shows particularly notable gains when multiple inconsistent variables are adjusted simultaneously. Finally, we provide guidance on choosing between these methods for healthcare practitioners.} } @InProceedings{pmlr-v340-li26c, title = {Deep Poisson Case Time Series Model for Complex Environmental Exposure Modeling on Acute Health Outcome}, author = {Li, Keyu and Lemoto, Elaona and Calhoun, Zachary D. and Wood, Charles T and Bhavsar, Nrupen A. and Carlson, David}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1056--1088}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/li26c/li26c.pdf}, url = {https://proceedings.mlr.press/v340/li26c.html}, abstract = {Modern environmental health studies increasingly collect rich spatio-temporal exposure data, yet existing methods for assessing acute health effects still rely on low-dimensional summaries that discard informative structure. Self-matched designs, such as the case time series, effectively control for stable between-area confounding but do not readily extend to complex, high-dimensional exposure histories. To address this limitation, we introduce the Deep Poisson Case Time Series (DPCTS), a self-matched deep learning framework for area-level health outcomes that integrates within-unit stratification with end-to-end representation learning from complex exposures. DPCTS employs a neural exposure encoder trained with a Poisson objective, incorporates temporal context and multi-scale supervision to capture lagged effects across multiple resolutions, and supports tabular, image-based, and multimodal inputs. In addition, we propose novel evaluation metrics that isolate exposure-driven performance from between-area baseline differences by measuring within-stratum fit, predictive accuracy, and discrimination. Across synthetic, semi-synthetic, and real-world experiments, DPCTS recovers nonlinear exposure-response relationships and fine-scale risk patterns more accurately than classical baselines. These findings establish that representation learning can be embedded within self-matched epidemiologic study designs to analyze complex exposure data while preserving the within-unit comparisons essential for studying short-term health risk.} } @InProceedings{pmlr-v340-liang26a, title = {OC-Distill: Ontology-aware Contrastive Learning with Cross-Modal Distillation for ICU Risk Prediction}, author = {Liang, Zhongyuan and Jo, Junhyung and Lee, Hyang-Jung and Kim, Sang Kyu and Chen, Irene Y.}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1089--1116}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/liang26a/liang26a.pdf}, url = {https://proceedings.mlr.press/v340/liang26a.html}, abstract = {Early prediction of severe clinical deterioration and remaining length of stay can enable timely intervention and better resource allocation in high-acuity settings such as the ICU. This has driven the development of machine learning models that leverage continuous streams of vital signs and other physiological signals for real-time risk prediction. Despite their promise, existing methods have important limitations. Contrastive pretraining treats all patients as equally strong negatives, failing to capture clinically meaningful similarity between patients with related diagnoses. Meanwhile, downstream fine-tuning typically ignores complementary modalities such as clinical notes, which provide rich contextual information unavailable in physiological signals alone. To address these challenges, we propose OC-Distill, a two-stage framework that leverages multimodal supervision during training while requiring only vital signs at inference. In the first stage, we introduce an ontology-aware contrastive objective that exploits the ICD hierarchy to quantify patient similarity and learn clinically grounded representations. In the second stage, we fine-tune the pretrained encoder via cross-modal knowledge distillation, transferring complementary information from clinical notes into the model. Across multiple ICU prediction tasks on MIMIC, OC-Distill demonstrates improved label efficiency and achieves state-of-the-art performance among methods that use only vital signs at inference.} } @InProceedings{pmlr-v340-lin26a, title = {Scaling up fine-grained intracranial vessel annotations in computed tomography angiography}, author = {Lin, Chu-Hsuan and Arroyo, Alberto Mario Ceballos and Kim, Jisoo and Yadav, Shrikanth and Jiang, Huaizu and Qin, Lei and Young, Geoffrey}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1117--1139}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/lin26a/lin26a.pdf}, url = {https://proceedings.mlr.press/v340/lin26a.html}, abstract = {In this work, we present SemanticVessel, a dataset for fine-grained brain vessel segmentation in computed tomography angiography scans. Based on the detailed contrast provided by dynamic 4D-CTA scans, we generate segmentation traces for arteries and veins. We then use intensity-guided region growing to obtain segmentations of the majority of vascular territories in the human brain, which are refined and annotated with 20 unique arterial classes by an expert radiologist. Unlike existing datasets, where minor arteries are discarded as background content, we merge these minor arteries into a generic arterial class. Due to the multiple-phase acquisition of dynamic 4D-CTA, labels for a single phase can be re-used for other phases in the same series, greatly increasing the size of our dataset with no additional annotation cost. The results show that models trained with the additional generic artery class produce better fine-grained segmentations across the board. Code and weights, as well as instructions to access the data, are available on: https://github.com/alceballosa/robust-vessel-segmentation} } @InProceedings{pmlr-v340-liu26a, title = {Adaptive Multi-Agent Feature Selection for Personalized Fall Risk Prevention}, author = {Liu, Chang and Thiamwong, Ladda and Fu, Yanjie and Xie, Rui}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1140--1177}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/liu26a/liu26a.pdf}, url = {https://proceedings.mlr.press/v340/liu26a.html}, abstract = {Falls among older adults represent a major public health challenge driven by complex, time-varying interactions across multiple risk domains. Effective fall risk factor identification requires learning from heterogeneous longitudinal data while accounting for sparse and delayed fall-related outcome events. However, existing approaches are largely static and fail to adaptively model evolving, individualized risk factors across modalities and time. We propose \textbf{PAFIR}, a \textbf{P}ersonalized and \textbf{A}daptive \textbf{F}eature selection framework for fall risk \textbf{I}dentification and p\textbf{R}evention, which formulates adaptive feature selection as a reinforcement learning problem over longitudinal multimodal health data. PAFIR jointly models structural dependencies among correlated assessment variables and temporal dynamics in wearable-derived physical activity data, and learns adaptive selection policies across repeated study visits using reward signals derived from sparse fall incidence outcomes. We apply PAFIR to data from the Physio fEedback Exercise pRogram (PEER) cluster-randomized trial. Experimental results demonstrate that PAFIR more effectively captures longitudinal and structural patterns of feature relevance than state-of-the-art baselines, and enables dynamic, subject-specific feature selection. By adapting selected features over time, PAFIR supports more timely and personalized fall prevention strategies.} } @InProceedings{pmlr-v340-liu26b, title = {LLM-Extracted Covariates for Clinical Causal Inference: Rethinking Integration Strategies}, author = {Liu, Lei and Chen, Jialin and Macropol, Kathy}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1178--1207}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/liu26b/liu26b.pdf}, url = {https://proceedings.mlr.press/v340/liu26b.html}, abstract = {Causal inference from electronic health records (EHR) is fundamentally limited by unmeasured confounding: critical clinical states such as frailty, goals of care, and mental status are documented in free-text notes but absent from structured data. Large language models can extract these latent confounders as interpretable, structured covariates, yet how to effectively integrate them into causal estimation pipelines has not been systematically studied. Using the MIMIC-IV database with 21,859 sepsis patients, we compare seven covariate-integration strategies for estimating the effect of early vasopressor initiation on 28-day mortality, spanning tabular-only baselines, traditional NLP representations, and three LLM-augmented approaches. A central finding is that not all integration strategies are equally effective: directly augmenting the propensity score model with LLM covariates achieves the best performance, while dual-caliper matching on text-derived categorical distances restricts the donor pool and degrades estimation. In semi-synthetic experiments with known ground-truth effects, LLM-augmented propensity scores reduce estimation bias from 0.0143 to 0.0003 relative to tabular-only methods, and this advantage persists under substantial simulated extraction error. On real data, incorporating LLM-extracted covariates reduces the estimated treatment effect from 0.055 to 0.027, directionally consistent with the CLOVERS randomized trial, and a doubly robust estimator yielding 0.031 confirms the robustness of this finding. Our results offer practical guidance on when and how text-derived covariates improve causal estimation in critical care.} } @InProceedings{pmlr-v340-liu26c, title = {GPT-Driven Drug Optimization with Structured Policy Optimization Post-training}, author = {Liu, Xuefeng and Jiang, Songhao and Chen, Siyu and Yang, Zhuoran and Chen, Yuxin and Foster, Ian T. and Stevens, Rick L.}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1208--1242}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/liu26c/liu26c.pdf}, url = {https://proceedings.mlr.press/v340/liu26c.html}, abstract = {Post-training is essential for steering generative models toward specific objectives. However, despite the growing importance of drug optimization, reinforcement learning algorithms tailored to this setting remain underexplored. In this work, we study the problem of drug optimization and propose a novel reinforcement learning framework for post-training a drug-optimization Generative Pre-trained Transformer (GPT). Our approach improves candidate molecules with respect to target objectives while preserving the desirable chemical properties of the original compounds. This work consists of two main components. (1) DrugImproverGPT, a framework designed to enhance the robustness and efficiency of drug optimization. It couples a GPT-based generative model with a theoretically grounded Structured Policy Optimization (SPO) algorithm. SPO provides a principled perspective on post-training generative models by explicitly aligning improvements in generated molecules with their corresponding input molecules under specified objectives. (2) A large-scale dataset comprising one million compounds, each annotated with OEDOCK docking scores across five human cancer-related proteins and 24 binding sites from the SARS-CoV-2 virus. Extensive in-silico experiments demonstrate that SPO generates candidates with improved values under the specified computational objectives. DrugImproverGPT is intended as a computational hit-to-lead prioritization framework whose generated candidates require subsequent experimental validation.} } @InProceedings{pmlr-v340-maddikunta26a, title = {WILLIE: A Unified Framework and Benchmark for Wound Classification, Segmentation, and Localization.}, author = {Maddikunta, Gopi Trinadh and Qian, Peizhu}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1243--1263}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/maddikunta26a/maddikunta26a.pdf}, url = {https://proceedings.mlr.press/v340/maddikunta26a.html}, abstract = {Chronic wound management affects over 8.2 million patients in the United States and imposes substantial clinical and economic burden. Clinical wound assessment commonly involves three coupled tasks: identifying wound type, delineating wound boundaries, and localizing the wound region for measurement and monitoring. Despite this clinical coupling, existing machine learning approaches typically address wound classification, segmentation, and localization using separate models. We present WILLIE, a unified framework and benchmark for wound classification, segmentation, and localization that enables systematic evaluation of multi-task wound analysis under a common protocol. WILLIE harmonizes three public wound datasets into a shared benchmark and compares unified models across three scaling configurations against 10 single-task baselines. The best model achieves 91.88% classification accuracy, 91.41% Dice, and 96.23% AP@0.5 while producing all three outputs in a single forward pass. Beyond aggregate performance, our results show that segmentation derived localization outperforms dedicated detection baselines in this benchmark, suggesting that box-based localization may be unnecessary for spatially coherent wound targets. Our findings highlight that effective multi-task learning in healthcare imaging depends not only on shared representations, but also on task formulation, compatibility, and benchmark design.} } @InProceedings{pmlr-v340-marshall26a, title = {The Severity Trap: Benchmarking Sequential Representation Learning for Oxygenation Trajectory Phenotyping in Acute Hypoxaemic Respiratory Failure}, author = {Marshall, Dominic C and Narayanan, Nikita and Antcliffe, David B and Parbhoo, Sonali}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1264--1299}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/marshall26a/marshall26a.pdf}, url = {https://proceedings.mlr.press/v340/marshall26a.html}, abstract = {Patients with acute hypoxaemic respiratory failure (AHRF) presenting with similar early physiology can follow markedly different trajectories, motivating trajectory phenotyping for mechanistic discovery and trial enrichment. Statistical models such as competing-risk latent class mixed models identify clinically meaningful classes but scale poorly to high-dimensional, irregular ICU time series, motivating sequential representation learning with variational autoencoders (VAEs). We characterise a systematic failure mode of learned trajectory representations, the \emph{severity trap}: representations organise primarily along baseline illness severity rather than trajectory shape, even with auxiliary survival objectives or disentanglement constraints. This is an instance of shortcut learning, or nuisance-variable dominance, that we characterise for clinical trajectory phenotyping and show evades the unsupervised metrics commonly used to select such models. Using engineered baselines of simple trajectory features that match, and once given the same outcome signal exceed, every deep model, together with controlled synthetic experiments, we trace the failure to reconstruction-based objectives preferentially encoding the highest-variance factor of the input, here baseline severity. We introduce a multi-cohort benchmark evaluating sequential representation learners (VAE families as the primary object of study; masked-transformer, diffusion, and contrastive encoders as stress tests) against four externally validated oxygenation trajectory archetypes across three international ICU cohorts (MIMIC-IV, Imperial College Healthcare NHS Trust, and Amsterdam UMCdb). Every reconstruction- or level-preserving objective falls into the trap. Only an explicitly level-invariant contrastive objective partially escapes, and clustering stability and reconstruction error correlate little with clinical validity. These findings highlight the need for evaluation grounded in externally validated clinical phenotypes, with nuisance-aware baselines, when developing representation learning for healthcare time series.} } @InProceedings{pmlr-v340-matsuura26a, title = {Paired Recipient-based Evaluation of Survival Prediction for Deceased Donor Kidney Transplants}, author = {Matsuura, Misaki and Nemati, Mohammadreza and Bekbolsynov, Dulat and Stepkowski, Stanislaw and Xu, Kevin S.}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1300--1320}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/matsuura26a/matsuura26a.pdf}, url = {https://proceedings.mlr.press/v340/matsuura26a.html}, abstract = {There has been significant interest in using machine learning algorithms to predict kidney transplant outcomes, such as the number of years until a graft inevitably fails. These prediction algorithms could possibly be used for pre-transplant donor-recipient matching to identify more compatible donors and recipients and thus improve post-transplant outcomes. In this study, we explore the use of survival prediction models trained on deceased donor kidney transplant data from the Scientific Registry of Transplant Recipients (SRTR). We propose a novel paired recipient-based evaluation framework that compares graft outcomes between two recipients who received kidneys from the same deceased donor, allowing us to evaluate the \emph{counterfactual benefit} of changing the recipient for a certain donor. We find that five different survival prediction models, ranging in complexity from linear to deep learning-based models, all result in $\sim 60%$ paired recipient-based accuracy. We further translate this accuracy into an interpretable quantity of post-transplant years gained. We also highlight major limitations of the commonly used concordance index (C-index) metric for evaluating survival prediction accuracy in this setting and demonstrate that our proposed paired recipient-based accuracy metric is more clinically relevant and better reflects real-world allocation settings.} } @InProceedings{pmlr-v340-mukherjee26a, title = {Retrieve, Then Classify: Corpus-Grounded Automation of Clinical Value Set Authoring}, author = {Mukherjee, Sumit and Kernell, O. Tate and Mazumder, Nairwita and Shu, Juan and Wheeler, Celena A. and Gibbons, Chris}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1321--1339}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/mukherjee26a/mukherjee26a.pdf}, url = {https://proceedings.mlr.press/v340/mukherjee26a.html}, abstract = {Clinical value set authoring — the task of identifying all codes in a standardized vocabulary that define a clinical concept — is a recurring bottleneck in clinical quality measurement and phenotyping. A natural approach is to prompt a large language model (LLM) to generate the required codes directly, but structured clinical vocabularies are large, version-controlled, and not reliably memorized during pretraining. We propose Retrieval-Augmented Set Completion (RASC): retrieve the $K$ most similar existing value sets from a curated corpus to form a candidate pool, then apply a classifier to each candidate code. Theoretically, retrieve-and-select can reduce statistical complexity by shrinking the effective output space from the full vocabulary to a much smaller retrieved candidate pool. We demonstrate the utility of RASC on 11,803 publicly available VSAC value sets, constructing the first large-scale benchmark for this task. A cross-encoder fine-tuned on SAPBert achieves AUROC 0.852 and value-set-level F1 0.298, outperforming a simpler three-layer Multilayer Perceptron (AUROC 0.799, F1 0.250) and both reduce the number of irrelevant candidates per true positive from 12.3 (retrieval-only) to approximately 3.2 and 4.4 respectively. Zero-shot GPT-4o achieves value-set-level F1 0.105, with 48.6% of returned codes absent from VSAC entirely. This performance gap widens with increasing value set size, consistent with RASC’s theoretical advantage. We observe similar performance gains across two other classifier model types, namely a cross-encoder initialized from pre-trained SAPBert and a LightGBM model, demonstrating that RASC’s benefits extend beyond a single model class.} } @InProceedings{pmlr-v340-munnangi26a, title = {Evaluating Large Language Models on Misconceptions in Multi-Turn Medical Conversations}, author = {Munnangi, Monica and Savage, Saiph}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1340--1366}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/munnangi26a/munnangi26a.pdf}, url = {https://proceedings.mlr.press/v340/munnangi26a.html}, abstract = {Patients seeking medical information often ask questions that embed incorrect assumptions or misconceptions. In such cases, safe medical communication requires not only answering the question, but identifying and correcting the underlying false belief. These interactions naturally unfold over multiple turns, a pattern now mirrored in interactions with LLMs. Yet current evaluation frameworks do not capture model behavior in these settings, where misconceptions can emerge, persist, or evolve over the course of a conversation. Whether LLMs can reliably correct such misconceptions over time remains largely unexamined. To study this, we introduce \datasetname, a multi-turn medical dialogue dataset of 2,437 patient–physician conversation threads comprising 8,204 question–answer pairs, derived from real patient interactions on r/AskDocs. This dataset enables systematic evaluation of whether models can detect and correct misconceptions under a multi-turn context. We evaluate five state-of-the-art LLMs using a rubric-based LLM-as-a-Judge framework that scores responses based on their ability to identify and correct misconceptions. Our experiments reveal a consistent pattern: even frontier models that can address misconceptions in a single interaction degrade substantially over subsequent turns. GPT-5 and Claude-Haiku correct these false presuppositions around 85% on initial questions but drop to approximately 50% within two follow-ups. Additionally, GPT-4o exhibits a sharper decline, falling from 65% to 21%, indicating a failure to sustain safe reasoning across dialogue. An oracle analysis replacing prior model outputs with physician responses shows that much of the degradation is driven by error propagation, while performance remains imperfect even under correct context. These findings reveal a critical reliability gap in LLMs. Even when models tend to correct misconceptions initially, their performance degrades substantially over subsequent turns, leading to inconsistent and potentially unsafe guidance in patient-facing settings and highlighting the need for evaluation frameworks that capture multi-turn behavior.} } @InProceedings{pmlr-v340-murali26a, title = {Human–AI Deferral under Limited Expert Availability: A Study in Intraoperative Ischemia Detection}, author = {Murali, Nihal and Elzokm, Karim and Thirumala, Parthasarathy and Batmanghelich, kayhan and Visweswaran, Shyam}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1367--1402}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/murali26a/murali26a.pdf}, url = {https://proceedings.mlr.press/v340/murali26a.html}, abstract = {Intraoperative neuromonitoring (IONM) during carotid endarterectomy (CEA) is used to monitor cerebral ischemia, but reliable interpretation depends on scarce expert neurophysiologists. Because many medical tasks are safety-critical, relying solely on artificial intelligence (AI) is impractical. We study how human–AI deferral methods perform in this setting and also introduce Learning to Augment (L2A), a simple method that selectively incorporates human input into AI predictions. We evaluate these methods on intraoperative electroencephalographic data from 400 CEA cases in a large U.S. academic health system. Unlike prior deferral work that focuses on non-clinical tasks and relies on expert humans, we use novice monitors with no prior IONM experience and only brief training. We use these novices as a conservative lower-bound test of whether non-expert human input contains complementary information, not as a proposed replacement for clinically trained personnel. Across methods, human–AI deferral shows that substantial gains over the AI alone can be achieved with minimal human involvement, even when the human is a novice. With just 5–10% novice involvement, L2A improves precision and sensitivity by $\sim$40%, and the area under the precision-recall curve (AUPRC) by $\sim$20% compared to the AI model alone. Importantly, while some deferral methods degrade with less experienced human input, others such as L2A remain robust with better calibration. We further observe that these methods produce sparse and temporally clustered deferral decisions, enabling long periods without human intervention. These properties make human–AI deferral approaches practical in settings where expert neurophysiologists are not available.} } @InProceedings{pmlr-v340-ng26a, title = {Colormaps Matter: Evaluating Their Impact on CNN-Based Clinical Thermal Imaging}, author = {Ng, Allison and Asare-Baiden, Miriam and Sonenblum, Sharon Eve and Jordan, Kathleen and Gichoya, Judy and Hertzberg, Vicki and Ho, Joyce C.}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1403--1431}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/ng26a/ng26a.pdf}, url = {https://proceedings.mlr.press/v340/ng26a.html}, abstract = {Thermal imaging is an increasingly used modality for clinical deep learning, yet a key preprocessing step in convolutional neural network (CNN)-based pipelines has gone largely unexamined. Each thermal image encodes scalar temperature values rendered into RGB via a colormap, determining how temperature information is visually represented. While existing work shows colormap selection influences human interpretation of thermal images, its impact on CNN-based classification remains unstudied. We address this gap through an evaluation of five colormaps across three pretrained CNN architectures and two clinical prediction tasks. Our results show that colormap choice produces consistent differences in predictive performance and substantially shifts the spatial regions CNNs attend to when making predictions. These findings suggest that colormap selection must be treated and reported as a factor shaping model behavior, with direct implications for how CNN-based pipelines are developed, validated, and deployed in clinical thermal imaging.} } @InProceedings{pmlr-v340-oufattole26a, title = {From Forecasting to Features: Zero-Shot Forecasting for Feature Extraction}, author = {Oufattole, Nassim and McDermott, Matthew B.A. and Stultz, Collin}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1432--1487}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/oufattole26a/oufattole26a.pdf}, url = {https://proceedings.mlr.press/v340/oufattole26a.html}, abstract = {Generative EHR models can estimate clinical risk in a zero-shot fashion by sampling future trajectories and computing outcome probabilities from generated rollouts. Yet in practice, direct use of these zero-shot risk estimates often underperforms supervised tabular baselines built from summaries of the observed past. In this work, we ask a narrower question: can future trajectories sampled from a generative foundation model be used to construct features that enhance the performance of downstream predictors? To test this, we convert model rollouts into horizon-specific \emph{generated future features} (GFFs) that summarize, for each patient, the probability of all future clinical events under the model, and train a simple XGBoost classifier that uses these features for downstream classification tasks. Across post-discharge prediction tasks in MIMIC-IV and a large private heart failure cohort, GFFs consistently achieve higher AUROC than direct zero-shot prediction and supervised learning on features derived from patient history. In our experiments we observe that stronger zero-shot forecasting performance tends to yield more performant GFFs. These results suggest that the value of current zero-shot generative EHR models is not only in their ability to generate future clinical trajectories, but also in their role as probabilistic feature generators that can be used for simple supervised risk models.} } @InProceedings{pmlr-v340-pickard26a, title = {EHR-MPC: Inference-Time Control for Sepsis Treatment with Generative Patient Digital Twins}, author = {Pickard, Joshua and Qi, Wei and Li, Na and Woolley, Ann and Cosimi, Lisa A. and Kishony, Roy and Hung, Deborah}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1488--1516}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/pickard26a/pickard26a.pdf}, url = {https://proceedings.mlr.press/v340/pickard26a.html}, abstract = {Sepsis is a leading cause of mortality, yet optimal treatment policies remain contested. Existing reinforcement learning (RL) approaches learn fixed strategies for sepsis treatment, limiting adaptability to changing clinical objectives during inference. We propose EHR-MPC, a framework that decouples learning patient dynamics from optimizing treatment by training a patient digital twin in the form of a generative electronic health record (EHR) model. The digital twin predicts clinical trajectories under interventions and enables model predictive control (MPC) to optimize treatments via inference-time planning over simulations. We evaluate EHR-MPC on a multicenter ICU sepsis cohort spanning 8 hospitals in the Mass General Brigham health system using both off-policy importance sampling and on-policy simulation-based evaluation. Relative to RL baselines, EHR-MPC achieves comparable off-policy performance and improved simulation performance. Unlike RL, this work frames sepsis treatment optimization as inference-time control over learned patient dynamics, establishing a general framework for decision making with generative clinical models.} } @InProceedings{pmlr-v340-pollet26a, title = {Privacy Audits for Clinical Large Language Models}, author = {Pollet, Florent and Nikitin, Kirill and Wang, Tong and Gupta, Rahul and Elhadad, No\'{e}mie and Gursoy, Gamze}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1517--1536}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/pollet26a/pollet26a.pdf}, url = {https://proceedings.mlr.press/v340/pollet26a.html}, abstract = {Large language models (LLMs) fine-tuned on de-identified clinical notes raise privacy concerns because automated de-identification can leave residual patient identifiers in the training data. We study whether such identifiers can be recovered from a fine-tuned model using query access alone as a function of query budget. We introduce Verified Extraction, an auditing framework that distinguishes identifiers attributable to fine-tuning data from spurious or prior-driven outputs and quantifies recoverable leakage under explicit query budgets. Using MIMIC-IV-Note as the fine-tuning dataset, we find that verified leakage is negligible at small query budgets but becomes practically significant under repeated querying, even when only a small fraction of identifiers remains in the training data. These results highlight the importance of privacy evaluations that account for repeated-query access rather than one-off prompt tests.} } @InProceedings{pmlr-v340-porya26a, title = {CARE-X: Towards Clinically Useful Radiology VLMs with Auxiliary Supervision, Reward-Aligned Learning, and Tool-Augmented Measurement}, author = {Porya, Anirban and Ranjit, Mercy Prasanna and Vadlamudi, Niharika and Eathamukkala, Nikhilesh Chowdary and V, Prasanth V and Joel, Sathvik and Swamy, Abhyuday Kumara and Umredkar, Pranay Narhari and Narayan, Pradeep and Rajagopal, Vivek and Ganu, Tanuja}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1537--1580}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/porya26a/porya26a.pdf}, url = {https://proceedings.mlr.press/v340/porya26a.html}, abstract = {A clinically useful chest X-ray system must go beyond fluent report generation: it should classify findings with tunable decision thresholds, localize them spatially, and derive the anatomical measurements on which many diagnoses depend. Today’s Vision-Language Models (VLMs) treat these as separate problems, if they address them at all — leaving a gap between what radiologists need and what generative models provide. We introduce CARE-X, a chest X-ray VLM that narrows this gap by unifying auxiliary discriminative supervision with reward-aligned generation. CARE-X augments its generative backbone with focal-loss classification and composite-loss grounding heads, co-trained with the language-modeling objective. This auxiliary supervision produces discriminative diagnostic predictions with tunable decision thresholds and precise spatial localization while also improving report quality — evidence that structured prediction and generation reinforce one another. Building on this foundation, Decoupled Clip and Dynamic sAmpling Policy Optimization (DAPO) leverages task-specific reward signals for report generation, VQA, and spatial grounding, directly optimizing the clinical quality metrics that matter in practice. The result is state-of-the-art performance on the majority of metrics across four report generation benchmarks, 94.0% VQA accuracy on ReXVQA (+6.0 pp over the next-best baseline), and generative spatial decoding that reaches near-parity with dedicated detection heads. Separately, to address measurement-dependent diagnoses, we couple Qwen3-VL-4B-Instruct an off-the-shelf VLM with native tool-calling capabilities — with deterministic measurement tools while retaining full visual access to the image. This hybrid inference yields +43.6 pp average F1 over perception-only baselines across five measurement-dependent conditions. We validate on rare, high-acuity ICU pathologies using clinical data from Narayana Health (NH), India, and on organ-enlargement conditions with CT-confirmed ground truth, showing that measurement-augmented CXR screening can identify high-risk cases who may require confirmatory imaging.} } @InProceedings{pmlr-v340-pugh26a, title = {Detecting Clinical Discrepancies in Health Coaching Agents: A Dual-Stream Memory and Reconciliation Architecture}, author = {Pugh, Samuel L and Yang, Eric and Sutherland, Alexander Muir and Breschi, Alessandra}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1581--1612}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/pugh26a/pugh26a.pdf}, url = {https://proceedings.mlr.press/v340/pugh26a.html}, abstract = {As Large Language Model (LLM) agents transition from single-session tools to persistent systems managing longitudinal healthcare journeys, their memory architectures face a critical challenge: reconciling two imperfect sources of truth. The patient’s evolving self-report is current but prone to recall bias, while the Electronic Health Record (EHR) is medically validated but frequently stale. General-purpose agent memory systems optimize for coherence by overwriting older facts with the user’s latest statement, a pattern that risks safety failures when applied to clinical data. We introduce a Dual-Stream Memory Architecture that strictly separates the patient narrative from the structured clinical record (FHIR), governed by a dedicated Reconciliation Engine that evaluates every extracted memory against the patient’s FHIR record and classifies discrepancies by type, severity, and the specific FHIR resources involved. We evaluate this architecture on 26 patients across 675 longitudinal wellness coaching sessions, using a hybrid dataset that interleaves real provider-patient transcripts with synthetic, FHIR-grounded clinical scenarios. In isolated testing, the engine detects 84.4% of designed clinical discrepancies with 86.7% safety-critical recall. By coupling extraction and reconciliation evaluation on the same data, we directly quantify a 13.6% error cascade, tracing the degradation to clinical details lost during memory extraction from unstructured conversation rather than to downstream classification errors. These findings establish that validating patient-reported memories against clinical records is both feasible and necessary for safe deployment of longitudinal health agents.} } @InProceedings{pmlr-v340-raehse26a, title = {Benchmarking Machine Learning Architectures for Antimicrobial Stewardship in Pediatric ICUs}, author = {Raehse, Niklas and Schlapbach, Luregn J and Chopard, Daphn\'{e}}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1613--1666}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/raehse26a/raehse26a.pdf}, url = {https://proceedings.mlr.press/v340/raehse26a.html}, abstract = {Antimicrobial stewardship (AMS) is critical for reducing unnecessary antibiotic exposure, particularly in pediatric intensive care units (PICUs), where clinical uncertainty leads to frequent broad-spectrum use, and overuse amplifies antimicrobial resistance which have long-term consequences. Machine learning has been proposed to support AMS by identifying patient-level opportunities for intervention from electronic health record data. However, prior work focuses on adult populations and predominantly relies on static tabular representations, leaving open questions about target design, temporal modeling, and generalizability in pediatric settings. In this work, we present a systematic benchmarking study of AMS intervention prediction in the PICU. Using the publicly available Paediatric Intensive Care database from China and a private PICU cohort from the University Children’s Hospital Zurich, Switzerland, we define four clinically relevant proxy targets for reducing antibiotic use: intravenous-to-oral switching, de-escalation, discontinuation, and short-course therapy. We then compare tabular, sequence-based, and graph-based temporal models under a unified evaluation framework. We find that model performance is primarily driven by target prevalence and data characteristics rather than model complexity. Sequence models provide improvements in precision-recall trade-off over tabular approaches at coarse (24-hour) resolution, with limited additional gains when finer temporal structure is incorporated. However, these gains come at the cost of poorer calibration, with simpler tabular models producing more reliable probability estimates. Our results highlight the importance of target selection, temporal representation, and calibration in clinical machine learning, and provide practical guidance for developing reliable decision support systems for AMS in pediatric critical care.} } @InProceedings{pmlr-v340-ren26a, title = {Attention-Enhanced Deep Generative Modeling Enables Interpretable Prediction of Cancer Drug Sensitivity}, author = {Ren, Shuangxia and QIU, AODONG and Lu, Mengyao and Lu, Xinghua}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1667--1688}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/ren26a/ren26a.pdf}, url = {https://proceedings.mlr.press/v340/ren26a.html}, abstract = {Predicting how tumors respond to specific drugs is a fundamental obstacle in precision oncology, largely because of the vast molecular diversity observed across cancers. Deep learning approaches have demonstrated potential for modeling complex relationships between molecular profiles and drug responses, yet their adoption in clinical settings has been hindered by limited transparency and poor generalization across different datasets. In this work, we introduce Residual Attention Variational Autoencoder with Elastic Net (ResAttnVAE-EN), a deep generative model augmented with attention mechanisms that combines somatic genomic alterations (SGAs) and transcriptomic profiles to learn biologically interpretable cellular representations for predicting drug sensitivity. Leveraging large-scale pharmacogenomic datasets from the Genomics of Drug Sensitivity in Cancer (GDSC) database, we show that ResAttnVAE-EN consistently surpasses standard variational autoencoder and regression-based benchmarks across a broad panel of therapeutic compounds. The attention layers highlight genomically coherent pathway-level drivers, and the learned latent spaces encode cellular states that generalize across multiple cancer types for response prediction. Notably, models derived from cell line experiments effectively distinguish survival trajectories and treatment outcomes in independent The Cancer Genome Atlas (TCGA) lung cancer patient cohorts. These findings position attention-augmented deep generative approaches as a reliable and interpretable framework for clinically translatable drug sensitivity modeling.} } @InProceedings{pmlr-v340-rhee26a, title = {Confounding Masquerading as Improvement: A Systematic Evaluation of Offline Reinforcement Learning for Stroke Antithrombotic Treatment in a 129,000-Patient Registry}, author = {Rhee, Kihun}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1689--1725}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/rhee26a/rhee26a.pdf}, url = {https://proceedings.mlr.press/v340/rhee26a.html}, abstract = {Recent offline reinforcement learning (RL) studies report data-driven policies outperform physician decisions by 10–32% on clinical outcomes. We conduct a systematic, partially crossed evaluation of five offline RL algorithm families and 14 reward designs on 44,894 post-2018 acute ischemic stroke patients from the nationwide stroke registry ($N = 129{,}033$). Standard Fitted Q-Evaluation yields $V(\mathrm{imp}) = +0.0069$ ($p = 0.048$); adding an Early Neurological Deterioration penalty strengthens the signal to $+0.0101$ ($p = 0.0002$, approximate E-value $= 18.37$)—estimates that could support a premature positive policy-improvement claim despite relying on a confounded reward function. We identify reward-embedded confounding, where the proxy terminal reward encodes baseline severity/prognosis information in addition to any treatment efficacy signal. E-value analysis does not evaluate this pathway because the confounding is carried through the reward channel. A $2 \times 2$ factorial experiment reveals that terminal reward confounding alone more than eliminates the apparent improvement, accounting for 218.6% of the $r_{\mathrm{END}} \to r_{\mathrm{full}}$ signal change (i.e., removing terminal confounding overshoots null). After DML-inspired GBM reward residualization, $V(\mathrm{imp})$ attenuates to $+0.0033$ ($p = 0.132$); full deconfounding yields $+0.0025$ ($p = 0.291$). Three independent approaches converge away from a clinically meaningful aggregate policy-improvement claim: FQE-based reward deconfounding, a T-learner showing 75–90% attenuation with a clinically small residual ($<6%$ MCID), and direct ischemic-stroke recurrence analysis (IPW/AIPW). A within-registry 1- year mRS factorial ($N = 35{,}744$) replicates the pattern: $V(\mathrm{imp}) = +0.0133$ attenuates to $-0.0004$ after reward residualization (103.3% attenuation), indicating that the finding is not specific to the 3-month mRS horizon in this registry. We present an empirically motivated six-step evaluation checklist; applied retrospectively, three highly cited clinical RL papers do not report the FQE confirmation specified by Step 1. Despite the overall null, National Institutes of Health Stroke Scale (NIHSS)-stratified heterogeneity (T-learner conditional average treatment effect (CATE) ratio $4.6\times$, corroborated by Q-evaluation $3.4\times$ gradient) identifies a hypothesis-generating subgroup for prospective trial design; hospital-level disagreement did not persist after full reward deconfounding.} } @InProceedings{pmlr-v340-shah26a, title = {Causal Discovery of Radiation Response Mechanisms in Human Cells}, author = {Shah, Ashka and Stevens, Rick}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1726--1750}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/shah26a/shah26a.pdf}, url = {https://proceedings.mlr.press/v340/shah26a.html}, abstract = {Next-generation sequencing technologies, including RNA-sequencing, provide genome-wide measurements of gene expression and enable broad explorations of biomarkers and mechanisms underlying disease and treatment response. Bionformatics tools for processing this data, such as differential expression analysis, are largely univariate, linear, and rely on predefined pathway knowledge annotations, which limits their ability to capture nonlinear and multivariate gene interactions. This paper explores the application of causal discovery to characterizing transcriptional responses to radiation as a function of dose rate in human cells. By jointly modeling radiation perturbations and gene expression, we learn directed gene networks that capture important regulatory relationships beyond correlation and exhibit significant enrichment of known radiation response pathways compared to baseline approaches. We find that inferred causal graphs reveal structured network features such as high in-degree housekeeping genes and high out-degree transcription factors. Further analysis suggests a hierarchical organization of stress response pathways and triggered cell death pathways. This work highlights the potential of causal discovery in healthcare settings with applications to understanding treatment response, identifying regulatory targets, and improving interpretation of complex genomic data. Code is available at https://github.com/shahashka/lucid_cd.} } @InProceedings{pmlr-v340-shen26a, title = {Learning Under Treatment-Induced Label Indeterminacy with Expert Annotations of Counterfactual Outcomes: A Case Study in Neurological Prognostication}, author = {Shen, Xiaobin and Huang, Chloe Y.H. and Elmer, Jonathan and Chen, George H.}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1751--1790}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/shen26a/shen26a.pdf}, url = {https://proceedings.mlr.press/v340/shen26a.html}, abstract = {Clinical prediction models are often developed as if the outcome of interest were cleanly observed for every patient. This assumption fails when treatment decisions make the clinically relevant outcome permanently unobservable. As a case study of this problem, we consider post-cardiac-arrest neurological prognostication using a cohort of 2,497 patients, including 1,429 patients whose outcomes were rendered indeterminate by treatment decisions(e.g., withdrawing or limiting life-sustaining therapies, which immediately led to death, so we do not know what would have happened otherwise). These patients with indeterminate outcomes were reviewed by independent clinical experts, who provided their guesses of counterfactual outcomes about what would have happened to the patients. We refer to these patients as uncertain cases. We also have patients for whom we observe their clinically relevant outcomes (e.g., regaining consciousness); we refer to these patients as certain cases. We propose a framework for evaluating prediction models that explicitly splits the evaluation between certain and uncertain cases. Here, we cannot easily evaluate both types of cases in a uniform manner as the available target labels differ (we have known outcomes for certain cases, and guesses of counterfactual outcomes for uncertain cases). We then propose a simple prediction model that uses target labels from both certain and uncertain cases in a manner that allows us to trade off between them. Across the proposed neural model and a collection of tabular baselines, models with similar certain-case AUROC can nevertheless differ substantially in both certain-case Brier score and their probability estimates for uncertain cases. Improving alignment with target labels of uncertain cases for our proposed model generally comes at the cost of worse accuracy on certain cases, highlighting an explicit tradeoff that standard evaluation (focusing only on certain cases) conceals. These results show that when treatment decisions determine whether clinically meaningful outcomes remain observable, conventional evaluation metrics can miss important failure modes in the very patients for whom prognostic support matters most. Code is available at https://github.com/xiaobin-xs/learning-under-label-indeterminacy.} } @InProceedings{pmlr-v340-shin26a, title = {Characterizing Population Gaps in Clinical Decision-Making: A Guideline-Based Benchmark and Population-Aware Retrieval Analysis in Anesthesiology}, author = {Shin, Chaiho and Park, Jung-Bin and Kim, Kwangsoo and Kim, Hee-Soo}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1791--1833}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/shin26a/shin26a.pdf}, url = {https://proceedings.mlr.press/v340/shin26a.html}, abstract = {Large language models (LLMs) have shown strong performance on medical question answering, yet their ability to support population-specific, action-oriented clinical decision-making remains underexplored. Our analysis of five widely used medical QA benchmarks shows that pediatric anesthesia accounts for only 0.2% of all questions, revealing a critical evaluation gap. To address this, we construct PopAnesQA, a population-aware anesthesiology question-answering benchmark grounded in clinical guidelines, and introduce Prof-RAG, a diagnostic framework for isolating the effects of patient-specific context on retrieval and downstream reasoning. Across diverse LLMs, we observe a consistent pediatric performance gap relative to the general subset, with differences exceeding 15 percentage points even in proprietary models like GPT-4o. We find that incorporating patient profile information leads to highly model-dependent effects, improving performance in some models while degrading it in others. Retrieval trajectory analysis shows that incorporating profile information increases the proportion of population-aligned documents, but does not consistently improve relevance to the clinical question. Overall, our findings highlight that effective clinical retrieval requires not only population awareness, but also careful alignment between patient-specific context and clinical intent.} } @InProceedings{pmlr-v340-shukla26a, title = {Auditing Measurement Processes in Clinical Registries: Detecting Label–Environment Confounding Before Model Development}, author = {Shukla, Apurv and Mohanty, Pravansu and McCaslin, Devin L}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1834--1864}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/shukla26a/shukla26a.pdf}, url = {https://proceedings.mlr.press/v340/shukla26a.html}, abstract = {Clinical prediction models trained on registry data routinely achieve high in-distribution accuracy, yet their transportability across institutions, clinicians, and time periods remains poorly understood. We formalize this problem through a measurement process model in which the observed label $Y^{\text{obs}}$ depends on both a true clinical state $Y^$*$$ and a reporting indicator $R(e,t)$ that varies with environment $e$ and era $t$: when reporting is inactive, the true label is unobserved and may be imputed by registry convention, introducing structured bias. We propose a measurement audita battery of three permutation-calibrated statistical tests that detects label–environment entanglement before any predictive model is builtand an evaluation-based attribution framework that estimates how much of a model’s apparent discrimination is due to environment confounding versus genuine clinical signal. Applied to a cochlear implant registry of 3,584 patients, the audit identifies severe, degenerate confounding (Cramér’s $V = 0.39$ for label–era association, 4 of 8 eras with zero negative labels, permutation $p<0.002$). The attribution reveals that approximately 10% of the IID AUROC ($0.978$, 95% CI $[0.955,0.991]$) traces to era confounding; the era-blocked AUROC of $0.875$ ($[0.668,0.925]$) better estimates transportable clinical discrimination. External validation on the MIMIC-IV ICU mortality cohort (n=67,224; 7 care-unit environments; 5 eras) shows that the audit distinguishes qualitatively different confounding regimes: it detects diffuse, unit-driven confounding ($V=0.145$, permutation $p<0.002$; confounding share 3.4% of IID AUROC) while correctly passing era-level tests on a registry with no zero-negative strata. A semi-synthetic calibration study confirms favorable operating properties (power rising monotonically with confounding strength; 4% false-positive rate; zero empirical family-wise error under label permutation). We discuss concrete remediation strategies for registries that fail the audit, and propose environment-aware evaluation as a candidate pre-modeling diagnostic for registry-trained clinical ML.} } @InProceedings{pmlr-v340-sondhi26a, title = {Bias-corrected Cox regression with AI-extracted covariates via calibration summary statistics}, author = {Sondhi, Arjun}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1865--1891}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/sondhi26a/sondhi26a.pdf}, url = {https://proceedings.mlr.press/v340/sondhi26a.html}, abstract = {Large-scale observational studies increasingly rely on AI pipelines to extract structured variables from unstructured clinical records. A common workflow separates the data vendor, who validates extraction accuracy with a gold-standard sample, from the downstream researcher, who receives only the extracted dataset and summary accuracy statistics. We develop a bias-correction framework for the Cox proportional hazards model when covariates are subject to AI extraction error. Within a unified multivariate calibration framework, we show that the naive Cox estimator’s bias decomposes into a leading-order calibration term and a second-order residual that vanishes as extraction accuracy improves. The leading-order term yields a corrected estimator that operates as a post-hoc matrix multiplication on the output of any standard Cox software. We further derive bias-adjusted confidence intervals that incorporate calibration uncertainty and a sensitivity diagnostic for assessing whether the neglected residual could materially affect inference. Simulations with cross-dependent extraction errors and controlled nonlinear calibration violations confirm that the correction substantially reduces bias and achieves near-nominal coverage even under mild violations of the linear calibration assumption. The framework yields a concrete reporting specification: a short list of summary statistics that data vendors should provide alongside any AI-extracted covariate dataset used in survival analysis.} } @InProceedings{pmlr-v340-song26a, title = {MotionAge: A Deep Learning Framework for Biological Age Prediction from Wearable Activity}, author = {Song, Yilin and Yin, Jiaqi and Hu, Yirou and Gu, Tian}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1892--1920}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/song26a/song26a.pdf}, url = {https://proceedings.mlr.press/v340/song26a.html}, abstract = {Population aging is a major public health challenge, driving increasing burden from chronic disease and mortality. To better characterize individual health beyond chronological age, there is growing interest in biological age as a quantitative measure for risk stratification and longitudinal monitoring. Existing biological age models largely rely on laboratory or structured clinical data, limiting scalability and their ability to capture dynamic aspects of health. We propose MotionAge, a deep learning framework that learns a mortality-calibrated biological age directly from high-frequency wearable activity data. The framework combines deep sequence models with a wear-aware modeling strategy that explicitly represents observation reliability, allowing the model to distinguish device non-wear from observed inactivity in noisy and irregularly observed time series. Unlike prior wearable-based methods that predict chronological age or construct unsupervised biomarkers, MotionAge is trained to predict 5-year mortality risk and subsequently maps that risk onto an interpretable age scale. In NHANES accelerometry data, MotionAge achieved stronger mortality discrimination over chronological age and established benchmarks. Within fixed chronological-age bands, higher MotionAge acceleration was associated with lower activity and worse survival, with the largest separation observed among older adults. These results demonstrate that reliability-aware deep learning of wearable time series can yield scalable and interpretable aging phenotypes aligned with clinically relevant health outcomes.} } @InProceedings{pmlr-v340-tae26a, title = {Change Point–Aware Evaluation and Re-Calibration of PPG-Based Blood Pressure Estimation}, author = {Tae, Yunwon and Park, Minje and Rho, Gyunho and Yoo, Dongjoon and Joo, Sunghoon}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1921--1949}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/tae26a/tae26a.pdf}, url = {https://proceedings.mlr.press/v340/tae26a.html}, abstract = {Non-invasive continuous blood pressure (BP) monitoring using photoplethysmography (PPG) is a promising alternative to cuff-based measurements. However, existing PPG-based BP estimation studies predominantly rely on aggregated performance metrics (e.g., mean absolute error) computed over entire evaluation intervals, which can obscure model failures during rapid BP fluctuations and limit clinical relevance. In this work, we propose a fluctuation-aware evaluation framework for PPG-based BP estimation based on time-series change point detection. Instead of heuristic BP thresholding (e.g., $\Delta\mathrm{BP} > 10\mathrm{mmHg}$), we identify BP change points by capturing abrupt distributional shifts in BP trajectories and evaluate estimation performance specifically during these fluctuation periods. Our analysis shows that several state-of-the-art models exhibit substantial performance degradation around BP change points, and that periodic test-time calibration is insufficient to handle such dynamic BP variations. To address this limitation, we introduce a targeted re-calibration framework triggered by detected BP change points, improving robustness without modifying model architectures. To the best of our knowledge, this is the first systematic evaluation of PPG-based BP estimation from a BP change point perspective, highlighting the importance of fluctuation-aware evaluation and calibration for real-world continuous BP monitoring.} } @InProceedings{pmlr-v340-talbot26a, title = {Evidence Against Homogeneity: Identifying Process Mismatch for Copy Number Variation Detection}, author = {Talbot, Austin and Kotlar, Alex V. and Ke, Yue}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1950--1975}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/talbot26a/talbot26a.pdf}, url = {https://proceedings.mlr.press/v340/talbot26a.html}, abstract = {Batch effects represent a major confounder in genomic diagnostics. In copy number variant (CNV) detection from next-generation sequencing, many algorithms compare read depth between test samples and a reference derived from the processing batch, assuming samples are process-matched. When this assumption is violated, with causes ranging from reagent lot changes and sample quality differences to multi-site processing, the reference becomes inappropriate, introducing false CNV calls or masking true pathogenic variants. Detecting such heterogeneity before downstream analysis is critical for reliable clinical interpretation. Existing batch effect detection methods either cluster samples based on raw features, risking conflation of biological signal with technical variation, or require known batch labels that are frequently unavailable. We introduce a method that addresses both limitations by clustering samples according to their Bayesian model evidence. The central insight is that evidence quantifies compatibility between data and model assumptions, technical artifacts violate assumptions and reduce evidence, whereas biological variation, including CNV status, is anticipated by the model and yields high evidence. This asymmetry provides a discriminative signal that separates batch effects from biology. We formalize heterogeneity detection as a likelihood ratio test for mixture structure in evidence space, using parametric bootstrap calibration to ensure conservative false positive rates. We validate our approach on synthetic data demonstrating proper Type I error control, three clinical targeted sequencing panels (liquid biopsy, BRCA, and thalassemia) exhibiting distinct batch effect mechanisms, and mouse electrophysiology recordings demonstrating cross-modality generalization. Our method achieves superior clustering accuracy compared to standard correlation-based and dimensionality-reduction approaches while maintaining the conservativeness required for clinical usage.} } @InProceedings{pmlr-v340-talbot26b, title = {Combining Bayesian and Frequentist Inference for Laboratory-Specific Performance Guarantees in Copy Number Variation Detection}, author = {Talbot, Austin and Kotlar, Alex V. and Ke, Yue}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1976--1997}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/talbot26b/talbot26b.pdf}, url = {https://proceedings.mlr.press/v340/talbot26b.html}, abstract = {Targeted amplicon panels are widely used in oncology diagnostics, but providing per-gene performance guarantees for copy number variant (CNV) detection remains challenging due to amplification artifacts, process-mismatch heterogeneity, and limited validation sample sizes. While Bayesian CNV callers naturally quantify per-sample uncertainty, translating this into the frequentist population-level guarantees required for clinical validation, coverage rates, false-positive bounds, and minimum detectable copy-number changes, is a fundamentally different inferential problem. We show empirically that even robust Bayesian credible intervals, including coarsened posteriors and sandwich-adjusted intervals, are severely miscalibrated on panels with small amplicon counts per gene. To address this, we propose a hybrid framework that evaluates Bayesian posterior functionals on validation samples and models the resulting squared losses with a Gamma distribution, yielding tolerance intervals with valid frequentist coverage. Three components make the method practical under real-world constraints: (1) imputation that removes the influence of true CNV-positive samples without requiring known ground truth, (2) regularization to address small sample variability, and (3) evidence-based stratification on the log model evidence to accommodate non-exchangeable noise profiles arising from process mismatch. Evaluated on two targeted amplicon panels using leave-one-out cross-validation, the proposed method achieves single-digit mean absolute coverage error across all genes under both process-matched and unmatched conditions, whereas Bayesian comparators exhibit mean absolute errors exceeding 60% on clinically relevant genes such as ERBB2.} } @InProceedings{pmlr-v340-thukral26a, title = {Hierarchical Modeling of ICD Codes in EHR Foundation Models}, author = {Thukral, Megha and Kang, Dong Gyun and Singh, Rudra Pratap and Hiremath, Shruthi Kashinath and H\"{a}nsel, Katrin and Ploetz, Thomas}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {1998--2029}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/thukral26a/thukral26a.pdf}, url = {https://proceedings.mlr.press/v340/thukral26a.html}, abstract = {Electronic health record foundation models typically treat ICD diagnosis codes as flat tokens, overlooking the clinically meaningful hierarchical structure that captures disease families, subcategories, and fine-grained diagnostic detail. As a result, existing EHR representation learning methods do not explicitly exploit the hierarchical structure already present in the coding system. In this work, we study ICD-10-CM hierarchy as a \emph{general inductive bias} for clinical representation learning. We investigate two complementary mechanisms for incorporating hierarchy: first, by augmenting diagnosis sequences in a BERT-style transformer with tokens corresponding to different levels of the ICD hierarchy, and second, by injecting hierarchy into graph-based code representations through hierarchy-aware edges combined with diagnosis co-occurrence structure. Across these settings, we evaluate whether explicit hierarchy improves downstream prediction, which levels of the hierarchy are most useful, whether hierarchy encoding improves transfer across datasets, and how hierarchy reshapes embedding similarity structure. We conduct experiments on two large-scale real-world clinical datasets: MIMIC-IV, used for pretraining and four in-domain prediction tasks, and eICU, used to assess cross-dataset transfer via frozen encoder probing. In our experiments, we find that explicitly encoding ICD hierarchy improves over flat code representations in both in-domain and cross-dataset settings, while revealing that the most useful level of hierarchy depends on both the task and the modeling approach. Broadly, we focus on hierarchy-aware EHR representation learning and show that the benefits of encoding hierarchy are generalizable across modeling settings and hierarchy levels.} } @InProceedings{pmlr-v340-vandenhirtz26a, title = {Bowel Obstruction Detection and Localization on Abdominal CT with Deep Learning}, author = {Vandenhirtz, Moritz and Agostini, Andrea and Belde, Dana and Roschewitz, M\'{e}lanie and Bakri, Ismaiel Chikh and Niemann, Tilo and Euler, Andr\'{e} and Vogt, Julia E}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {2030--2054}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/vandenhirtz26a/vandenhirtz26a.pdf}, url = {https://proceedings.mlr.press/v340/vandenhirtz26a.html}, abstract = {Bowel obstruction is a common and potentially life-threatening gastrointestinal condition. In the face of rising diagnostic workloads, the automated diagnosis of bowel obstruction on CT scans supports radiologists by accelerating detection and improving patient outcomes. In this work, we propose a deep learning framework with a multi-task objective that jointly detects bowel obstruction and localizes its transition zone. Additionally, we extend the method with an inherently interpretable classification method that locates the suspected transition point within a slice. It does so by learning a probabilistic selection mask that faithfully bases the classifier’s prediction solely on a small image region. The proposed method is evaluated on an internal dataset comprising 1,427 abdominal CTs. Here, the model achieves an obstruction detection test accuracy of 93% and a Hit@10 transition zone localization of 95%. As the first method to reliably localize the transition zone, this marks a significant step towards the automated identification of this critical clinical landmark.} } @InProceedings{pmlr-v340-vedula26a, title = {Real-world sleep phenotyping in autism: compliance, domain shift, and scalable wearable biomarkers}, author = {Vedula, Sanketh and Kim, Andrew and Kumar, Manoj and Troyanskaya, Olga and Sapiro, Guillermo}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {2055--2108}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/vedula26a/vedula26a.pdf}, url = {https://proceedings.mlr.press/v340/vedula26a.html}, abstract = {Sleep is a scalable target for digital health, but real-world sleep phenotyping remains challenging, especially in autism spectrum disorder (ASD), where laboratory polysomnography (PSG) is difficult to deploy longitudinally. We study ambulatory sleep sensing in the Simons Sleep Project, a multimodal dataset of 200 children recorded over 14 nights using dry-electrode electroencephalography (EEG) headbands and wrist-worn sensors (photoplethysmography (PPG), inertial measurement units (IMU), temperature). We first show that EEG headband compliance declines systematically overnight and is associated with behavioral measures, introducing structured missingness and biasing sleep estimates. In contrast, wrist-worn sensors maintain near-complete adherence. We then demonstrate a substantial domain shift: sleep-staging models trained on clinical-grade data show limited off-the-shelf transfer to wearable recordings. However, this shift is learnable with limited supervision. For EEG, zero-shot models perform near chance, while fine-tuning on 30–50 participants (270-450 nights) recovers performance; fine-tuning the public SleepFM foundation model on 102 participants ($\sim$9 nights per participant) achieves a macro F1 of 0.780. For PPG, the public PaPaGei foundation model achieves only 0.375 macro F1 out-of-the-box, whereas supervised models trained on PPG features reach 0.710, further indicating limited off-the-shelf transfer from clinical-grade pretraining to wearable recordings. Watch sensor ablations identify PPG as the most informative standalone peripheral signal, with PPG+IMU capturing most staging information from the wrist. EEG channel analysis further shows that frontal–occipital derivations outperform prefrontal channels, informing minimal-channel designs. Together, these results establish a modality hierarchy for ambulatory sleep phenotyping: wrist-worn PPG+IMU provides a scalable backbone for longitudinal monitoring, while EEG remains valuable for micro-architectural analysis when tolerated. We anticipate these lessons on compliance-aware design, domain adaptation, and principled sensor selection to generalize to other vulnerable populations and at-home deployment of digital health studies.} } @InProceedings{pmlr-v340-wang26a, title = {Key Coverage Matters: Semi-Structured Extraction of OCR Clinical Reports}, author = {Wang, Yu and Li, Yingyun and QIN, Ying and Qian, Haiyang}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {2109--2130}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/wang26a/wang26a.pdf}, url = {https://proceedings.mlr.press/v340/wang26a.html}, abstract = {This work addresses a cross-institution workflow in which patients present paper or scanned reports from prior visits because relevant records are not available in the receiving hospital’s EHR. This hinders not only electronic health record (EHR) integration and longitudinal review, but also downstream workflows that depend on more complete patient records, including cross-institution EHR back-fill, longitudinal follow-up for chronic disease and oncology, and report-derived clinical-trial eligibility screening. Although optical character recognition (OCR) can digitize such reports, reliable extraction remains challenging because clinical documents are heterogeneous, OCR text is noisy, and many healthcare settings require low-cost on-premise deployment. We formulate this problem as canonical key-conditioned extractive question answering and maintain a canonical inventory and alias mapping through iterative key mining, normalization, clustering, and incremental human verification. On the 849-report development set, inventory expansion increases end-to-end Exact recall from 0.3534 to 0.8042, with Exact F1 peaking at 0.8232 for Top-95. A supplementary gold-key-conditioned comparison shows that a compact 0.2B BERT–BiLSTM–CRF extractor runs substantially faster than LoRA-adapted Qwen3 models on the same GPU, with a modest trade-off in Exact F1. An independent downstream study found task-dependent effects of structured fields, including gains of 3.27 percentage points for histologic grade and 3.52 points for treatment response, but the effects were not uniformly positive across tasks; this study evaluates representation utility rather than the proposed extractor. Although the corpus is Chinese, the method is based on the language-agnostic key–value organization of semi-structured clinical reports and can be adapted to other settings with an appropriate canonical inventory and alias mapping.} } @InProceedings{pmlr-v340-wang26b, title = {Imitation Learning for Robot Assistance in Open Surgery: A Multi-Policy Evaluation on Suture Following}, author = {Wang, Xucheng and Yang, Zhizhou and Zhang, Xiaoman and Kim, Sung Eun and Hardy, Romain and Rajpurkar, Pranav}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {2131--2150}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/wang26b/wang26b.pdf}, url = {https://proceedings.mlr.press/v340/wang26b.html}, abstract = {This study presents the first evaluation of general-purpose imitation learning for surgeon-robot collaborative assistance in open surgery, targeting suture following: the grab-pull-release motion an assistant performs at every stitch. We collect 160 teleoperated demonstrations (32,374 frames) on an open-source robot arm, benchmark four architecturally diverse imitation learning policies (ACT, Diffusion Policy, SmolVLA, $\pi_0$) across 28 trained models evaluated in 32 configurations along three clinically motivated dimensions: dataset size, camera viewpoint, and background variation. Our results demonstrate that under ideal conditions, the four policies achieve 50–75% task success, with depth error as the dominant failure mode across all architectures. Among all policies, $\pi_0$ achieves the strongest results with a pretrained vision-language backbone, demonstrating superior data efficiency, greater robustness to background variation, and smoother trajectories compatible with surgical workflow. When deployed in a surgeon-robot suturing trial, $\pi_0$ yields a 92% stitch completion rate. These findings establish collaborative robotic assistance in open surgery as a feasible target for imitation learning and highlight depth perception and end-effector design as key priorities for clinical translation. All code, data, and model weights are released at https://github.com/rajpurkarlab/Suture_Follow_Evaluation.} } @InProceedings{pmlr-v340-weng26a, title = {Censoring-Aware Reinforcement Learning to Optimize Early Risk Alerts from Longitudinal Clinical Data}, author = {Weng, Qin and Goldstein, Benjamin and Engelhard, Matthew M.}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {2151--2174}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/weng26a/weng26a.pdf}, url = {https://proceedings.mlr.press/v340/weng26a.html}, abstract = {Early recognition of chronic conditions is critical to ensure patients receive timely interventions and support. Passive surveillance of routine electronic health records (EHRs) provides information about longitudinal health trajectories that can support prompt recognition and inform associated early actions. However, relevant information is acquired at a different rate for each patient, and there is an inherent trade-off between the earliness versus the specificity of diagnosis and related actions. Therefore, determining when to alert providers about a likely chronic condition requires us to weigh the predicted risk at the given time against the anticipated value of future information. To address this challenge, we analyze the optimal timing of early alerts using a Partially Observable Markov Decision Process (POMDP) with asymmetric reward. To learn an optimal alerting policy, we then propose a model-free reinforcement learning (RL) framework tailored to long-term clinical event surveillance from EHRs. Our proposed framework overcomes the pervasive issue of right-censoring in offline EHRs by leveraging a pseudo-label imputation approach. We also analytically demonstrate that entropy-regularized RL enables post-hoc threshold calibration to adapt the learned policy to specific preferences regarding the importance of earliness versus specificity without retraining. Systematic evaluations in synthetic data reveal that the advantage of RL-based look-ahead planning is maximized when diagnostic evidence emerges in predictable information bursts. Finally, real-world validations on two clinical cohorts show that our policy achieves an actionable lead time of 20.9 months prior to Alzheimer’s disease diagnosis and 8.7 months prior to autism diagnosis in a pediatric cohort, both at 90% specificity. Our method and results provide a generalizable blueprint for optimal surveillance of chronic disease processes.} } @InProceedings{pmlr-v340-xiang26a, title = {Learning Representations from Incomplete EHR Data with Dual-Masked Autoencoding}, author = {Xiang, Xiao and Restrepo, David and Jeong, Hyewon and Jia, Yugang and Celi, Leo Anthony}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {2175--2201}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/xiang26a/xiang26a.pdf}, url = {https://proceedings.mlr.press/v340/xiang26a.html}, abstract = {Electronic health records (EHR) arrive masked. Clinicians order measurements selectively, and any patient table thus contains only a subset of the values that characterize the underlying physiological state. Prior masked modeling approaches on EHR data either impute the table before learning, represent missingness through a dedicated placeholder signal, or optimize solely for imputation, which limits the representations they learn for downstream clinical tasks and carries every unobserved entry through the encoder. We introduce AID-MAE, an Augmented-Intrinsic Dual-Masked Autoencoder that learns directly from incomplete tables by combining the intrinsic mask the record already carries with an augmented mask that hides a subset of observed values for reconstruction during pretraining. Neither type of masked entry enters the encoder, so attention operates only over what was observed. AID-MAE achieves consistent improvements over strong baselines across multiple clinical tasks on two datasets. Across experiments, we discuss that recovering the missing entries is not a prerequisite for learning and show that the representations learned carry clinical structure without supervision.} } @InProceedings{pmlr-v340-yadav26a, title = {The Cost of Binarizing Survival Outcomes in Clinical Prognostic Modeling}, author = {Yadav, Shashank and Routman, David M. and Foong, Andrew Y. K.}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {2202--2226}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/yadav26a/yadav26a.pdf}, url = {https://proceedings.mlr.press/v340/yadav26a.html}, abstract = {Survival analysis is an established framework for analyzing time-to-event data, yet many clinical machine learning studies still binarize the outcome before model training. This practice excludes censored patients, collapses temporal information into a single threshold, and can affect which features are selected as prognostically relevant. We examine the cost of this binarization in the context of Bayesian network (BN) feature selection, using two recent publications as case studies: one that applies BN-based feature selection to a head-and-neck cancer cohort and a second surgical cohort study that, while not BN-based, likewise binarizes its survival endpoint. We replace the binary scoring function with the Cox partial log-likelihood for feature-to-outcome edges, a modification we call the Survival-Aware Bayesian network, and recover prognostic features that binarization misses. Our ablation experiment confirms that the improvement is driven by the time-to-event scoring formulation rather than by retaining more patients. The results generalize across five endpoint-cohort combinations in head-and-neck cancer and extend to three further cancer types (breast, colorectal, and kidney). We propose that clinical studies with survival outcomes should use time-to-event methods by default, as binarization discards the prognostic signal retained by survival analysis.} } @InProceedings{pmlr-v340-zapzalka26a, title = {Estimating Upcoding in Medicare Advantage: Identifying Contributors, Costs, and Mechanisms}, author = {Zapzalka, Dylan and Mittal, Muskaan and Wiens, Jenna and Makar, Maggie}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {2227--2256}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/zapzalka26a/zapzalka26a.pdf}, url = {https://proceedings.mlr.press/v340/zapzalka26a.html}, abstract = {Strategic misreporting – the manipulation of features to obtain a better outcome – undermines the integrity of machine learning models used for healthcare resource allocation. In the U.S. Medicare Advantage program, private insurers are reimbursed based on the beneficiary diagnoses they report, creating financial incentives that encourage over-reporting, leading to billions of dollars in annual overpayments. Despite the magnitude of the problem, estimating insurers’ misreporting rates is problematic due to the lack of ground-truth diagnoses. Past work has used causality to try to estimate misreporting rates, but assumes access to an unmanipulated dataset and fails to account for unobserved confounders. To mitigate these issues, we introduce the Stitched Causal Misreporting Estimator (SCaMEr), a causally motivated auditing approach to estimate insurer-specific misreporting rates. SCaMEr addresses two key limitations of prior work: (1) it avoids the need for an unmanipulated reference dataset, and (2) it gives reliable upper and lower bounds on misreporting estimates in the presence of hidden confounders. To overcome the lack of an unmanipulated reference, SCaMEr "stitches" multiple datasets together with varying misreporting rates to recover unbiased estimates. To account for unobserved confounding, it incorporates causal sensitivity analysis to produce uncertainty bounds. We validate SCaMEr on both semi-synthetic and real-world Medicare Advantage data, where it achieves lower estimation error than baselines. Our results show that SCaMEr can enable auditing by identifying diagnoses, insurance plans, and reporting mechanisms that are susceptible to misreporting.} } @InProceedings{pmlr-v340-zhang26a, title = {Toward Better Assessment of LLMs’ Performance in Clinical Error Detection}, author = {Zhang, Yifan and Beheshti, Rahmatollah}, booktitle = {Proceedings of the 11th Machine Learning for Healthcare Conference}, pages = {2257--2290}, year = {2026}, editor = {Krishnan, Rahul G. and van Amsterdam, Wouter A. C. and Chopra, Sumit and Overgaard, Shauna and Hughes, Michael and Ötleş, Erkin and Shen, Yiqiu and Shanmugam, Divya and Nayan, Madhur and Engelhard, Matthew and Fackler, Jim and Oberst, Michael}, volume = {340}, series = {Proceedings of Machine Learning Research}, month = {12--14 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v340/main/assets/zhang26a/zhang26a.pdf}, url = {https://proceedings.mlr.press/v340/zhang26a.html}, abstract = {Automated detection of errors in clinical documentation is a promising application of large language models (LLMs), yet decisions to deploy such models rest on benchmarks that evaluate each clinical note in isolation. Error-detection benchmarks are typically constructed by injecting errors into notes, such that each erroneous note has a natural counterpart. Aggregate discriminative metrics (e.g., balanced accuracy or F1) do not exploit this structure. We show that this omission is consequential. In particular, evaluating 15 diverse LLMs on 4 standardized clinical error-detection test sets across 3 languages, we find that 13 of 15 models fall below the level of random pairwise discrimination, even while achieving F1 scores that standard practice would read as moderate. We also observe that the underlying bias patterns differ across languages: the same model can default to “no error” on one language and over-flag errors on another. To diagnose where discrimination breaks down, we further introduce a procedure to score the evidence models cite in their outputs. We find that while models consistently locate error-relevant content, they fail to produce the corresponding correct verdict on the clean counterpart. Finally, we show that F1 and pairwise accuracy are driven in opposite directions by the same underlying bias, so that ranking models by F1 may systematically promote the weakest discriminators. For safety-critical clinical NLP applications, we advocate for supplementing aggregate metrics with paired evaluations in benchmark reporting. Code and analysis scripts are available at https://github.com/healthylaife/paired-clinical-eval.} }