Proceedings of the 3rd Machine Learning for Healthcare Conference Held in Palo Alto, California on 17-18 August 2018 Published as Volume 85 by the Proceedings of Machine Learning Research on 29 November 2018. Volume Edited by: Finale Doshi-Velez Jim Fackler Ken Jung David Kale Rajesh Ranganath Byron Wallace Jenna Wiens Series Editors: Neil D. Lawrence Mark Reid https://proceedings.mlr.press/v85/ Wed, 08 Feb 2023 10:43:48 +0000 Wed, 08 Feb 2023 10:43:48 +0000 Jekyll v3.9.3 Predicting diagnoses from Electronic Health Records (EHRs) is an important medical application of multi-label learning. We propose a convolutional residual model for multi-label classification from doctor notes in EHR data. A given patient may have multiple diagnoses, and therefore multi-label learning is required. We employ a Convolutional Neural Network (CNN) to encode plain text into a fixed-length sentence embedding vector. Since diagnoses are typically correlated, a deep residual network is employed on top of the CNN encoder, to capture label (diagnosis) dependencies and incorporate information directly from the encoded sentence vector. A real EHR dataset is considered, and we compare the proposed model with several well-known baselines, to predict diagnoses based on doctor notes. Experimental results demonstrate the superiority of the proposed convolutional residual model. Thu, 29 Nov 2018 00:00:00 +0000 https://proceedings.mlr.press/v85/zhang18a.html https://proceedings.mlr.press/v85/zhang18a.html Unstructured learning problems without well-defined rewards are unsuitable for current reinforcement learning (RL) approaches. Action-derived rewards can allow RL agents to fully explore state and action trade-offs in scenarios that require specific outcomes yet are unstructured by external reward. Clinical trial dosing choice is an example of such a problem. We report the successful formulation of clinical trial dosing choice as an RL problem using action-based rewards and learning of dosing regimens to reduce mean tumor diameters (MTD) in patients undergoing simulated temozolomide (TMZ) and procarbazine, 1-(2-chloroethyl)-3-cyclohexyl-l-nitrosourea, and vincristine (PCV) chemo- and radiotherapy clinical trials. The use of action-derived rewards as partial proxies for outcomes is described for the first time. Novel dosing regimens learned by an RL agent in the presence of action-derived rewards achieve significant reduction in MTD for cohorts and individual patients in simulated TMZ and PCV clinical trials while reducing treatment cycle administrations and dosage concentrations compared to human-expert dosing regimens. Our approach can be easily adapted for other learning tasks where outcome-based learning is not practical. Thu, 29 Nov 2018 00:00:00 +0000 https://proceedings.mlr.press/v85/yauney18a.html https://proceedings.mlr.press/v85/yauney18a.html As robots become more affordable and more common in everyday life, particularly in assistive contexts, there will be an ever-increasing demand for adaptive behavior that is personalized to the individual needs of users. To accomplish this, robots will need to learn about their users’ unique preferences through interaction. Current preference learning techniques lack the ability to infer long-term, task-independent preferences in realistic, interactive, incomplete-information settings. To address this gap, we introduce a novel preference-inference formulation, inspired by assistive robotics applications, in which a robot must infer these kinds of preferences based only on observing the user’s behavior in various tasks. We then propose a candidate inference algorithm based on maximum-margin methods, and evaluate ts performance in the context of robot-assisted prehabilitation. We find that the algorithm learns to predict aspects of the user’s behavior as it is given more data, and that it shows strong convergence properties after a small number of iterations. Thu, 29 Nov 2018 00:00:00 +0000 https://proceedings.mlr.press/v85/woodworth18a.html https://proceedings.mlr.press/v85/woodworth18a.html Cardiac arrest is a rare but devastating event in critically ill children associated with death, disability and significant healthcare costs. When a cardiac arrest occurs, the limited interventions available to save patient lives are associated with poor patient outcomes. The most effective way of improving patient outcomes and decreasing the associated healthcare costs would be to prevent cardiac arrest from occurring. This observation highlights the importance of prediction models that consistently identify high risk individuals and assist health care providers in providing targeted care to the right patient at the right time. In this paper, we took advantage of the power of convolutional neural networks (CNN) to extract information from high resolution temporal data, and combine this with a recurrent network (LSTM) to model time dependencies that exist in these temporal signals. We trained this CNN+LSTM model on high-frequency physiological measurements that are recorded in the ICU to facilitate early detection of a potential cardiac arrest at the level of the individual patient. Our model results in an F1 value of 0.61 to 0.83 across six different physiological signals, the most predictive single signal being the heart rate. To address the issue of instances of missing data in the recorded physiological signals, we have also implemented an ensemble model that combines predictors for the signals that were collected for a given patient. The ensemble achieves 0.83 average F1 score on a held-out test set, on par with the best performing signal, even in the absence of a number of signals. The results of our model are clinically relevant. We intend to explore implementation of this model at the point of care as a means of providing precise, personalized, predictive care to an at-risk cohort of patients. Thu, 29 Nov 2018 00:00:00 +0000 https://proceedings.mlr.press/v85/tonekaboni18a.html https://proceedings.mlr.press/v85/tonekaboni18a.html Given the wide success of convolutional neural networks (CNNs) applied to natural images, researchers have begun to apply them to neuroimaging data. To date, however, exploration of novel CNN architectures tailored to neuroimaging data has been limited. Several recent works fail to leverage the 3D structure of the brain, instead treating the brain as a set of independent 2D slices. Approaches that do utilize 3D convolutions rely on architectures developed for object recognition tasks in natural 2D images. Such architectures make assumptions about the input that may not hold for neuroimaging. For example, existing architectures assume that patterns in the brain exhibit translation invariance. However, a pattern in the brain may have different meaning depending on where in the brain it is located. There is a need to explore novel architectures that are tailored to brain images. We present two simple modifications to existing CNN architectures based on brain image structure. Applied to the task of brain age prediction, our network achieves a mean absolute error (MAE) of 1.4 years and trains 30% faster than a CNN baseline that achieves a MAE of 1.6 years. Our results suggest that lessons learned from developing models on natural images may not directly transfer to neuroimaging tasks. Instead, there remains a large space of unexplored questions regarding model development in this area, whose answers may differ from conventional wisdom. Thu, 29 Nov 2018 00:00:00 +0000 https://proceedings.mlr.press/v85/sturmfels18a.html https://proceedings.mlr.press/v85/sturmfels18a.html Biomedical named entity recognition (NER) is a fundamental task in text mining of medical documents and has many applications. Deep learning based approaches to this task have been gaining increasing attention in recent years as their parameters can be learned end-to-end without the need for hand-engineered features. However, these approaches rely on high-quality labeled data, which is expensive to obtain. To address this issue, we investigate how to use unlabeled text data to improve the performance of NER models. Specifically, we train a bidirectional language model (BiLM) on unlabeled data and transfer its weights to ?pretrain? an NER model with the same architecture as the BiLM, which results in a better parameter initialization of the NER model. We evaluate our approach on four benchmark datasets for biomedical NER and show that it leads to a substantial improvement in the F1 scores compared with the state-of-the-art approaches. We also show that BiLM weight transfer leads to a faster model training and the pretrained model requires fewer training examples to achieve a particular F1 score. Thu, 29 Nov 2018 00:00:00 +0000 https://proceedings.mlr.press/v85/sachan18a.html https://proceedings.mlr.press/v85/sachan18a.html Expert diagnostic support systems have been extensively studied. The practical applications of these systems in real-world scenarios have been somewhat limited due to well-understood shortcomings, such as lack of extensibility. More recently, machine-learned models for medical diagnosis have gained momentum, since they can learn and generalize patterns found in very large datasets like electronic health records. These models also have shortcomings - in particular, there is no easy way to incorporate prior knowledge from existing literature or experts. In this paper, we present a method to merge both approaches by using expert systems as generative models that create simulated data on which models can be learned. We demonstrate that such a learned model not only preserves the original properties of the expert systems but also addresses some of their limitations. Furthermore, we show how this approach can also be used as the starting point to combine expert knowledge with knowledge extracted from other data sources, such as electronic health records. Thu, 29 Nov 2018 00:00:00 +0000 https://proceedings.mlr.press/v85/ravuri18a.html https://proceedings.mlr.press/v85/ravuri18a.html Recently, researchers have started applying convolutional neural networks (CNNs) with one-dimensional convolutions to clinical tasks involving time-series data. This is due, in part, to their computational efficiency, relative to recurrent neural networks and their ability to efficiently exploit certain temporal invariances, (\textit{e.g.}, phase invariance). However, it is well-established that clinical data may exhibit many other types of invariances (\textit{e.g.}, scaling). While preprocessing techniques, (\textit{e.g.,} dynamic time warping) may successfully transform and align inputs, their use often requires one to identify the types of invariances in advance. In contrast, we propose the use of Sequence Transformer Networks, an end-to-end trainable architecture that learns to identify and account for invariances in clinical time-series data. Applied to the task of predicting in-hospital mortality, our proposed approach achieves an improvement in the area under the receiver operating characteristic curve (AUROC) relative to a baseline CNN (AUROC=0.851 vs. AUROC=0.838). Our results suggest that a variety of valuable invariances can be learned directly from the data. Thu, 29 Nov 2018 00:00:00 +0000 https://proceedings.mlr.press/v85/oh18a.html https://proceedings.mlr.press/v85/oh18a.html Clinical care requires understanding the time to medical events. Medical events include the time to a disease like chronic kidney disease progressing or the time to a complication as in stroke for high blood pressure. Models for event times live in the framework provided by survival analysis. We expand on deep survival analysis, a deep generative model for survival analysis in the presence of missing data, where the survival times are modeled using Weibull distributions. We develop methods to relax the distributional assumptions in deep survival analysis using survival distributions that can approximate any true survival function. We show that the model structure mimics the information-optimal procedure in the presence of missing data. Our experiments demonstrate that moving to flexible survival functions yields better likelihoods and concordances for coronary heart disease prediction from electronic health records. Thu, 29 Nov 2018 00:00:00 +0000 https://proceedings.mlr.press/v85/miscouridou18a.html https://proceedings.mlr.press/v85/miscouridou18a.html We combine predictive modeling techniques from machine learning and optimization methods to design coordinated imaging protocols for detection of metastatic cancer. Our approach considers different combinations of imaging tests to reduce imaging while also ensuring that the average risk of missing a metastatic cancer in the population does not exceed a desirable threshold. To account for the imperfect calibration of probability estimates obtained from predictive models, we formulate the decision problem of determining the optimal assignment of patients to imaging protocols as a robust mixed-integer program. Furthermore, we propose fast, easy-to-understand and clinically motivated approximation algorithms that can mitigate the effects of statistical error in predictions. We illustrate the practical performance of the proposed approximation algorithms and optimization models based on medical data collected by a large state-wide prostate cancer collaborative. The work presented in this article will help lay the groundwork to improve medical decision making by integrating machine learning and optimization in other disease areas. Thu, 29 Nov 2018 00:00:00 +0000 https://proceedings.mlr.press/v85/merdan18a.html https://proceedings.mlr.press/v85/merdan18a.html Knowledge-based planning (KBP) is an automated approach to radiation therapy treatment planning that involves predicting desirable treatment plans before they are then corrected to deliverable ones. We propose a generative adversarial network (GAN) approach for predicting desirable 3D dose distributions that eschews the previous paradigms of site-specific feature engineering and predicting low-dimensional representations of the plan. Experiments on a dataset of oropharyngeal cancer patients show that our approach significantly outperforms previous methods on several clinical satisfaction criteria and similarity metrics. Thu, 29 Nov 2018 00:00:00 +0000 https://proceedings.mlr.press/v85/mahmood18a.html https://proceedings.mlr.press/v85/mahmood18a.html Hypertension is a heterogeneous syndrome in need of improved subtyping using phenotypic and genetic measurements so that patients in different subtypes share similar pathophysiologic mechanisms and respond more uniformly to targeted treatments. Existing machine learning approaches often face challenges in integrating phenotype and genotype information and presenting to clinicians an interpretable model. We aim to provide informed patient stratification by introducing Hybrid Non-negative Matrix Factorization (HNMF) on phenotype and genotype matrices. HNMF simultaneously approximates the phenotypic and genetic matrices using different appropriate loss functions, and generates patient subtypes, phenotypic groups and genetic groups. Unlike previous methods, HNMF approximates phenotypic matrix under Frobenius loss, and genetic matrix under Kullback-Leibler (KL) loss. We propose an alternating projected gradient method to solve the approximation problem. Simulation shows HNMF converges fast and accurately to the true factor matrices. On real-world clinical dataset, we used the patient factor matrix as features to predict main cardiac mechanistic outcomes. We compared HNMF with six different models using phenotype or genotype features alone, with or without NMF, or using joint NMF with only one type of loss. HNMF significantly outperforms all comparison models. HNMF also reveals intuitive phenotype-genotype interactions that characterize cardiac abnormalities. Thu, 29 Nov 2018 00:00:00 +0000 https://proceedings.mlr.press/v85/luo18a.html https://proceedings.mlr.press/v85/luo18a.html The high prevalence of spinal stenosis results in a large volume of MRI imaging, yet interpretation can be time-consuming with high inter-reader variability even among the most specialized radiologists. In this paper, we develop an efficient methodology to leverage the subject-matter-expertise stored in large-scale archival reporting and image data for a deep-learning approach to fully-automated lumbar spinal stenosis grading. Specifically, we introduce three major contributions: (1) a natural-language-processing scheme to extract level-by-level ground-truth labels from free-text radiology reports for the various types and grades of spinal stenosis (2) accurate vertebral segmentation and disc-level localization using a U-Net architecture combined with a spine-curve fitting method, and (3) a multi-input, multi-task, and multi-class convolutional neural network to perform central canal and foraminal stenosis grading on both axial and sagittal imaging series inputs with the extracted report-derived labels applied to corresponding imaging level segments. This study uses a large dataset of 22796 disc-levels extracted from 4075 patients. We achieve state-of-the-art performance on lumbar spinal stenosis classification and expect the technique will increase both radiology workflow efficiency and the perceived value of radiology reports for referring clinicians and patients. Thu, 29 Nov 2018 00:00:00 +0000 https://proceedings.mlr.press/v85/lu18a.html https://proceedings.mlr.press/v85/lu18a.html Early detection of preventable diseases is important for better disease management, improved interventions, and more efficient health-care resource allocation. Various machine learning approaches have been developed to exploit information in Electronic Health Record (EHR) for this task. Majority of previous attempts, however, focus on structured fields and lose the vast amount of information in the unstructured notes. In this work we propose a general multi-task framework for disease onset prediction that combines both free-text medical notes and structured information. We compare performance of different deep learning architectures including CNN, LSTM and hierarchical models. In contrast to traditional text-based prediction models, our approach does not require disease specific feature engineering, and can handle negations and numerical values that exist in the text. Our results on a cohort of about 1 million patients show that models using text outperform models using just structured data, and that models capable of using numerical values and negations in the text, in addition to the raw text, further improve performance. Additionally, we compare different visualization methods for medical professionals to interpret model predictions. Thu, 29 Nov 2018 00:00:00 +0000 https://proceedings.mlr.press/v85/liu18b.html https://proceedings.mlr.press/v85/liu18b.html Intensive Care Units (ICUs) are some of the highest intensity areas of patient care activities in hospitals, yet documentation and understanding of the occurrence of these activities remains sub-optimal due in part to already-demanding patient care workloads of nursing staff. Recently, computer vision based methods operating over color and depth data collected from passive mounted sensors have been developed for automated activity recognition, but have been limited to coarse or simple activities due to the complex environments in ICUs, where fast-changing activities and severe occlusion occurs. In this work, we introduce an approach for tackling more challenging activities in ICUs by combining depth data from multiple sensors to form a single 3D point cloud representation, and using a neural network-based model to reason over this 3D representation. We demonstrate the effectiveness of this approach using a dataset of mobility-related patient care activities collected in a clinician-guided simulation setting. Thu, 29 Nov 2018 00:00:00 +0000 https://proceedings.mlr.press/v85/liu18a.html https://proceedings.mlr.press/v85/liu18a.html Joint models for longitudinal and time-to-event data are commonly used in longitudinal studies to forecast disease trajectories over time. While there are many advantages to joint modeling, the standard forms suffer from limitations that arise from a fixed model specification and computational difficulties when applied to high-dimensional datasets. In this paper, we propose a deep learning approach to address these limitations, enhancing existing methods with the inherent flexibility and scalability of deep neural networks while retaining the benefits of joint modeling. Using longitudinal data from the UK Cystic Fibrosis Trust, we demonstrate improvements in performance and scalability, as well as robustness in the presence of irregularly sampled data. Thu, 29 Nov 2018 00:00:00 +0000 https://proceedings.mlr.press/v85/lim18a.html https://proceedings.mlr.press/v85/lim18a.html The high rate of intensive care unit false arrhythmia alarms can lead to disruption of care and slow response time due to desensitization of clinical staff. We study the use of machine learning models to detect false ventricular tachycardia (v-tach) alarms using ECG waveform recordings. We propose using a Superv sed Denoising Autoencoder (SDAE) to detect false alarms using a low-dimensional representation of ECG dynamics learned by minimizing a combined reconstruct on and classification loss. We evaluate our algorithms on the PhysioNet Challenge 2015 dataset, containing over 500 records (over 300 training and 200 testing) with v-tach alarms. Our results indicate that using the SDAE on Fast Fourier Transformed (FFT) ECG at a beat-by-beat level outperforms several competitive baselines on the task of v-tach false alarm classification. We show that it is important to exploit the underlying known physiological structure using beat-by-beat frequency distribution from multiple cardiac cycles of the ECG waveforms to obtain competitive results and improvement over previous entries from the 2015 PhysioNet Challenge. Thu, 29 Nov 2018 00:00:00 +0000 https://proceedings.mlr.press/v85/lehman18a.html https://proceedings.mlr.press/v85/lehman18a.html Cytometry is an important single cell analysis technology in furthering our understanding of cellular biological processes and in supporting clinical diagnoses across a variety hematological and immunological conditions. Current data analysis workflows for cytometry data rely on a manual process called gating to classify cells into canonical types. This dependence on human annotation significantly limits the rate, reproducibility, and scope of cytometry’s use in both biological research and clinical practice. We develop a novel Bayesian approach for automated gating that classifies cells into different types by combining cell-level marker measurements with an informative prior. The Bayesian approach allows for the incorporation of biologically-meaningful prior information that captures the domain expertise of human experts. The inference algorithm results in a hierarchically-structured classification of individual cells in a manner that mimics the tree-structured recursive process of manual gating, making the results readily interpretable. The approach can be extended in a natural fashion to handle data from multiple different samples by the incorporation of random effects in the Bayesian model. The proposed approach is evaluated using mass cytometry data, on the problems of unsupervised cell classification and supervised clinical diagnosis, illustrating the benefits of both incorporating prior knowledge and sharing information across multiple samples. Thu, 29 Nov 2018 00:00:00 +0000 https://proceedings.mlr.press/v85/ji18a.html https://proceedings.mlr.press/v85/ji18a.html Survival prediction for cancer patients can increase the prognostic accuracy and might ultimately lead to better informed decision making. To this end, many studies apply machine learning to cancer data of the Surveillance, Epidemiology, and End Results (SEER) program. The first part of this report contains a literature review to obtain a systematic overview of these studies. We identify 34 publications and extract information about experimental setups and efforts to ensure reproducibility. The review shows that only one of the identified studies mentions reproducibility and no study contains straightforward reproducible results. This motivates the second part of this work. We demonstrate the feasibility of reproducible cohort selection and survival prediction with SEER cancer data. Experiments are performed for 1- and 5-year survival of breast and lung cancer with cases diagnosed between 2004 and 2009. We compare minimal data preprocessing with 1-n encoding of categorical inputs and apply logistic regression and multilayer perceptron (MLP) models. Encoding with 1-n vectors proves beneficial throughout all experiments. For lung cancer, MLP models show a slightly superior performance. Moreover, importance of input attributes is analyzed with logistic regression weights and ablation analysis for MLPs. Thu, 29 Nov 2018 00:00:00 +0000 https://proceedings.mlr.press/v85/hegselmann18a.html https://proceedings.mlr.press/v85/hegselmann18a.html Access to electronic health record (EHR) data has motivated computational advances in medical research. However, various concerns, particularly over privacy, can limit access to and collaborative use of EHR data. Sharing synthetic EHR data could mitigate risk. In this paper, we propose a new approach, medical Generative Adversarial Network (medGAN), to generate realistic synthetic patient records. Based on input real patient records, medGAN can generate high-dimensional discrete variables (e.g., binary and count features) via a combination of an autoencoder and generative adversarial networks. We also propose minibatch averaging to efficiently avoid mode collapse, and increase the learning efficiency with batch normalization and shortcut connections. To demonstrate feasibility, we showed that medGAN generates synthetic patient records that achieve comparable performance to real data on many experiments including distribution statistics, predictive modeling tasks and a medical expert review. We also empirically observe a limited privacy risk in both identity and attribute disclosure using medGAN. Thu, 29 Nov 2018 00:00:00 +0000 https://proceedings.mlr.press/v85/goodfellow18a.html https://proceedings.mlr.press/v85/goodfellow18a.html Electronic Health Records (EHRs) contain an overwhelming amount of information about each patient, making it difficult for clinicians to quickly find the most salient information. Accurate, concise summarization of relevant data can help alleviate this cognitive burden. In practice, clinical narrative notes serve this purpose during the course of care, but they are only intermittently updated and are sometimes missing information. We address this problem by learning to generate topics that should be in summaries of structured health record data at any point during a stay. We use the detailed, high-dimensional structured data to predict existing clinical note topics. Our model can generate topics based on structured health record data, even when a real note does not exist. We demonstrate that using structured data alone, we are able to generate note topics comparable to the performance of using prior notes alone. Our method is also capable of generating the first note in the stay. We demonstrate that our predicted topic distributions are meaningful using the downstream task of predicting in-hospital mortality. We show that our generated note topic vectors perform comparably or even outperform topics from the actual notes on predicting in-hospital mortality. Thu, 29 Nov 2018 00:00:00 +0000 https://proceedings.mlr.press/v85/gong18a.html https://proceedings.mlr.press/v85/gong18a.html Medical data are not missing at random. The problem is more acute when the observations are over an extended period of time; any particular variable is observed at relatively few time points. We taking missing values to be the norm, and treat “presentness” (the times of observations) as additional features to augment the values observed. A joint model over both avoids the “missing at random” assumption. We use piecewise-constant conditional intensity models (PCIMs) to build a generative model of observation times and values. We demonstrate its effectiveness in reconstruction of monitor readings of patient vitals from sparse EHR data. Thu, 29 Nov 2018 00:00:00 +0000 https://proceedings.mlr.press/v85/fauber18a.html https://proceedings.mlr.press/v85/fauber18a.html Health risks from cigarette smoking – the leading cause of preventable death in the United States – can be substantially reduced by quitting. Although most smokers are motivated to quit, the majority of quit attempts fail. A number of studies have explored the role of self-reported symptoms, physiologic measurements, and environmental context on smoking risk, but less work has focused on the temporal dynamics of smoking events, including daily patterns and related nicotine effects. In this work, we examine these dynamics and improve risk prediction by modeling smoking as a self-triggering process, in which previous smoking events modify current risk. Specifically, we fit smoking events self-reported by 42 smokers to a time-varying semi-parametric Hawkes process (TV-SPHP) developed for this purpose. Results show that the TV-SPHP achieves superior prediction performance compared to related and existing models, with incorporation of time-varying predictors having greatest benefit over longer prediction windows. Moreover, the impact function illustrates previously unknown temporal dynamics of smoking, with possible connections to nicotine metabolism to be explored in future work through a randomized study design. By more effectively predicting smoking events and exploring a self-triggering component of smoking risk, this work supports development of novel or improved cessation interventions that aim to reduce death from smoking. Thu, 29 Nov 2018 00:00:00 +0000 https://proceedings.mlr.press/v85/engelhard18a.html https://proceedings.mlr.press/v85/engelhard18a.html In this work, we present a specific case study where we aim to design effective treatment allocation strategies and validate these using a mouse model of skin cancer. Collecting data for modelling treatments effectiveness on animal models is an expensive and time consuming process. Moreover, acquiring this information during the full range of disease stages is hard to achieve with a conventional random treatment allocation procedure, as poor treatments cause deterioration of subject health. We therefore aim to design an adaptive allocation strategy to improve the efficiency of data collection by allocating more samples for exploring promising treatments. We cast this application as a contextual bandit problem and introduce a simple and practical algorithm for exploration-exploitation in this framework. The work builds on a recent class of approaches for non-contextual bandits that relies on subsampling to compare treatment options using an equivalent amount of information. On the technical side, we extend the subsampling strategy to the case of bandits with context, by applying subsampling within Gaussian Process regression. On the experimental side, preliminary results using 10 mice with skin tumours suggest that the proposed approach extends by more than 50% the subjects life duration compared with baseline strategies: no treatment, random treatment allocation, and constant chemotherapeutic agent. By slowing the tumour growth rate, the adaptive procedure gathers information about treatment effectiveness on a broader range of tumour volumes, which is crucial for eventually deriving sequential pharmacological treatment strategies for cancer. Thu, 29 Nov 2018 00:00:00 +0000 https://proceedings.mlr.press/v85/durand18a.html https://proceedings.mlr.press/v85/durand18a.html The localization of epileptic zone in pharmacoresistant focal epileptic patients is a daunting task, typically performed by medical experts through visual inspection over highly sampled neural recordings. For a finer localization of the epileptogenic areas and a deeper understanding of the pathology both the identification of pathogenical biomarkers and the automatic characterization of epileptic signals are desirable. In this work we present a data integration learning method based on multi-level representation of stereo-electroencephalography recordings and multiple kernel learning. To the best of our knowledge, this is the first attempt to tackle both aspects simultaneously, as our approach is devised to classify critical vs. non-critical recordings while detecting the most discriminative frequency bands. The learning pipeline is applied to a data set of 18 patients for a total of 2347 neural recordings analyzed by medical experts. Without any prior knowledge assumption, the data-driven method reveals the most discriminative frequency bands for the localization of epileptic areas in the high-frequency spectrum (>=80 Hz) while showing high performance metric scores (mean balanced accuracy of 0.89 +- 0.03). The promising results may represent a starting point for the automatic search of clinical biomarkers of epileptogenicity. Thu, 29 Nov 2018 00:00:00 +0000 https://proceedings.mlr.press/v85/d-amario18a.html https://proceedings.mlr.press/v85/d-amario18a.html This paper presents the results of an analytics-based study to determine key differences in junior resident level and expert level surgical skill when engaging with a neurosurgical simulator. Window-based time series discretization and sequential pattern analysis were used on positional data to identify frequent movement patterns and instrumentation techniques associated with each skill class. Cross-validation confirmed that a Bayesian classification model constructed using these patterns can be used to predict skill level with a high degree of confidence and accuracy on a small sample of neurosurgeons who engaged with the simulator. An analysis of movement speed also revealed that the junior residents exhibited a high degree of very slow and very fast movements, whereas the expert surgeons displayed a significantly more consistent technique of moderate-speed movements. Finally, the analysis was integrated within a cloud-based learning framework, helping to provide beneficial feedback on movement proficiency to resident surgeons in training. The presented work makes two key contributions to the field of machine learning in the medical field: the study 1) employs a low-level behaviour-based analysis of surgical technique, as opposed to high-level summary metrics such as blood loss and average force, and 2) avoids the use of expert information on neurosurgical skill within the AI engine, and thus employs an analysis that is entirely uninformed. Thu, 29 Nov 2018 00:00:00 +0000 https://proceedings.mlr.press/v85/buffett18a.html https://proceedings.mlr.press/v85/buffett18a.html There are established racial disparities in healthcare, including during end-of-life care, when poor communication and trust can lead to suboptimal outcomes for patients and their families. In this work, we find that racial disparities which have been reported in existing literature are also present in the MIMIC-III database. We hypothesize that one underlying cause of this disparity is due to mistrust between patient and caregivers, and we develop multiple possible trust metric proxies (using coded interpersonal variables and clinical notes) to measure this phenomenon more directly. These metrics show even stronger disparities in end-of-life care than race does, and they also tend to demonstrate statistically significant higher levels of mistrust for black patients than white ones. Finally, we demonstrate that these metrics improve performance on three clinical tasks: in-hospital mortality, discharge against medical advice (AMA) and modified care status (e.g., DNR, DNI, etc.). Thu, 29 Nov 2018 00:00:00 +0000 https://proceedings.mlr.press/v85/boag18a.html https://proceedings.mlr.press/v85/boag18a.html We present a new approach to ensemble learning for risk prognosis in heterogeneous medical populations. Our aim is to improve overall prognosis by focusing on under-represented patient subgroups with an atypical disease presentation; with current prognostic tools, these subgroups are being consistently mis-estimated. Our method proceeds sequentially by learning nonparametric survival estimators which iteratively learn to improve predictions of previously misdiagnosed patients - a process called boosting. This results in fully nonparametric survival estimates, that is, constrained neither by assumptions regarding the baseline hazard nor assumptions regarding the underlying covariate interactions - and thus differentiating our approach from existing boosting methods for survival analysis. In addition, our approach yields a measure of the relative covariate importance that accurately identifies relevant covariates within complex survival dynamics, thereby informing further medical understanding of disease interactions. We study the properties of our approach on a variety of heterogeneous medical datasets, demonstrating significant performance improvements over existing survival and ensemble methods. Thu, 29 Nov 2018 00:00:00 +0000 https://proceedings.mlr.press/v85/bellot18a.html https://proceedings.mlr.press/v85/bellot18a.html