Predictive Modeling of Long-Term CPAP Non-Adherence in OSA patients from Post-Initiation Treatment Telemonitoring Data

Benedetta Giachetti, Simone Costantini, Elena Mugellini, Arnaud Prigent
Proceedings of the Fourth Swiss AI Days, PMLR 309:27-37, 2026.

Abstract

Long-term adherence to continuous positive airway pressure (CPAP) therapy remains a major challenge in the management of obstructive sleep apnea, despite its well-established clinical benefits. The objective of this study is to predict long-term CPAP non-adherence one year after a baseline month that does not correspond to treatment initiation but can occur within a time window ranging from the fourth to the twenty-fourth month of therapy. We propose and compare a machine learning (ML) pipeline and a deep learning (DL) approach that leverage daily CPAP telemonitoring time-series and electronic health record (EHR) variables. Both pipelines are evaluated on a held-out test set: the ML model achieved a macro F1-score of 0.83, while the DL model achieved 0.81, indicating comparable and robust predictive performance. These results suggest that CPAP usage patterns observed during an intermediate treatment phase remain highly informative for identifying patients at risk of future non-adherence and could support targeted long-term telemonitoring strategies, serving as a data-driven second opinion to assist clinicians in decision-making.

Cite this Paper

BibTeX
@InProceedings{pmlr-v309-giachetti26a, title = {Predictive Modeling of Long-Term CPAP Non-Adherence in OSA patients from Post-Initiation Treatment Telemonitoring Data}, author = {Giachetti, Benedetta and Costantini, Simone and Mugellini, Elena and Prigent, Arnaud}, booktitle = {Proceedings of the Fourth Swiss AI Days}, pages = {27--37}, year = {2026}, editor = {Kucharavy, Andrei and Delgado, Pamela and Schürch Todeschini, Valérie and Rumley, Sébastien}, volume = {309}, series = {Proceedings of Machine Learning Research}, month = {23--25 Mar}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v309/main/assets/giachetti26a/giachetti26a.pdf}, url = {https://proceedings.mlr.press/v309/giachetti26a.html}, abstract = {Long-term adherence to continuous positive airway pressure (CPAP) therapy remains a major challenge in the management of obstructive sleep apnea, despite its well-established clinical benefits. The objective of this study is to predict long-term CPAP non-adherence one year after a baseline month that does not correspond to treatment initiation but can occur within a time window ranging from the fourth to the twenty-fourth month of therapy. We propose and compare a machine learning (ML) pipeline and a deep learning (DL) approach that leverage daily CPAP telemonitoring time-series and electronic health record (EHR) variables. Both pipelines are evaluated on a held-out test set: the ML model achieved a macro F1-score of 0.83, while the DL model achieved 0.81, indicating comparable and robust predictive performance. These results suggest that CPAP usage patterns observed during an intermediate treatment phase remain highly informative for identifying patients at risk of future non-adherence and could support targeted long-term telemonitoring strategies, serving as a data-driven second opinion to assist clinicians in decision-making.} }
Endnote
%0 Conference Paper %T Predictive Modeling of Long-Term CPAP Non-Adherence in OSA patients from Post-Initiation Treatment Telemonitoring Data %A Benedetta Giachetti %A Simone Costantini %A Elena Mugellini %A Arnaud Prigent %B Proceedings of the Fourth Swiss AI Days %C Proceedings of Machine Learning Research %D 2026 %E Andrei Kucharavy %E Pamela Delgado %E Valérie Schürch Todeschini %E Sébastien Rumley %F pmlr-v309-giachetti26a %I PMLR %P 27--37 %U https://proceedings.mlr.press/v309/giachetti26a.html %V 309 %X Long-term adherence to continuous positive airway pressure (CPAP) therapy remains a major challenge in the management of obstructive sleep apnea, despite its well-established clinical benefits. The objective of this study is to predict long-term CPAP non-adherence one year after a baseline month that does not correspond to treatment initiation but can occur within a time window ranging from the fourth to the twenty-fourth month of therapy. We propose and compare a machine learning (ML) pipeline and a deep learning (DL) approach that leverage daily CPAP telemonitoring time-series and electronic health record (EHR) variables. Both pipelines are evaluated on a held-out test set: the ML model achieved a macro F1-score of 0.83, while the DL model achieved 0.81, indicating comparable and robust predictive performance. These results suggest that CPAP usage patterns observed during an intermediate treatment phase remain highly informative for identifying patients at risk of future non-adherence and could support targeted long-term telemonitoring strategies, serving as a data-driven second opinion to assist clinicians in decision-making.
APA
Giachetti, B., Costantini, S., Mugellini, E. & Prigent, A.. (2026). Predictive Modeling of Long-Term CPAP Non-Adherence in OSA patients from Post-Initiation Treatment Telemonitoring Data. Proceedings of the Fourth Swiss AI Days, in Proceedings of Machine Learning Research 309:27-37 Available from https://proceedings.mlr.press/v309/giachetti26a.html.

Related Material