{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/forecasting-individualized-disease","title":"Forecasting Individualized Disease Trajectories using Interpretable Deep Learning","arxiv_id":"1810.10489","date":"2018-10-24","proceeding":null,"authors":["Ahmed M. Alaa","Mihaela van der Schaar"],"abstract":"Disease progression models are instrumental in predicting individual-level\nhealth trajectories and understanding disease dynamics. Existing models are\ncapable of providing either accurate predictions of patients prognoses or\nclinically interpretable representations of disease pathophysiology, but not\nboth. In this paper, we develop the phased attentive state space (PASS) model\nof disease progression, a deep probabilistic model that captures complex\nrepresentations for disease progression while maintaining clinical\ninterpretability. Unlike Markovian state space models which assume memoryless\ndynamics, PASS uses an attention mechanism to induce \"memoryful\" state\ntransitions, whereby repeatedly updated attention weights are used to focus on\npast state realizations that best predict future states. This gives rise to\ncomplex, non-stationary state dynamics that remain interpretable through the\ngenerated attention weights, which designate the relationships between the\nrealized state variables for individual patients. PASS uses phased LSTM units\n(with time gates controlled by parametrized oscillations) to generate the\nattention weights in continuous time, which enables handling\nirregularly-sampled and potentially missing medical observations. Experiments\non data from a realworld cohort of patients show that PASS successfully\nbalances the tradeoff between accuracy and interpretability: it demonstrates\nsuperior predictive accuracy and learns insightful individual-level\nrepresentations of disease progression.","url_abs":"http://arxiv.org/abs/1810.10489v1","url_pdf":"http://arxiv.org/pdf/1810.10489v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[],"tasks":[{"task_slug":"deep-learning","task_name":"Deep Learning"},{"task_slug":"disease-prediction","task_name":"Disease Prediction"},{"task_slug":"disease-trajectory-forecasting","task_name":"Disease Trajectory Forecasting"},{"task_slug":"state-space-models","task_name":"State Space Models"}],"methods":[{"method_slug":"lstm","method_name":"LSTM"},{"method_slug":"sigmoid-activation","method_name":"Sigmoid Activation"},{"method_slug":"tanh-activation","method_name":"Tanh Activation"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/disease-trajectory-forecasting-on-uk-cf-trust","task":"Disease Trajectory Forecasting","dataset":"UK CF trust","model":"PASS","rank_in_archive_order":1,"of":3,"metrics":{"AUC (ABPA)":"0.687","AUC (Aspergillus)":"0.640","AUC (Diabetes)":"0.771","AUC (E. Coli)":"0.701","AUC (I. Obstruction)":"0.577","AUC (K. Pneumonia)":"0.718","I. Obstruction":"0.577"},"uses_additional_data":false}],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}