{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,3,27]],"date-time":"2026-03-27T20:10:41Z","timestamp":1774642241878,"version":"3.50.1"},"reference-count":47,"publisher":"Oxford University Press (OUP)","issue":"9","license":[{"start":{"date-parts":[[2021,6,21]],"date-time":"2021-06-21T00:00:00Z","timestamp":1624233600000},"content-version":"vor","delay-in-days":0,"URL":"http:\/\/creativecommons.org\/licenses\/by-nc\/4.0\/"}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"published-print":{"date-parts":[[2021,8,13]]},"abstract":"<jats:title>Abstract<\/jats:title><jats:sec><jats:title>Objective<\/jats:title><jats:p>Multitask learning (MTL) using electronic health records allows concurrent prediction of multiple endpoints. MTL has shown promise in improving model performance and training efficiency; however, it often suffers from negative transfer \u2013 impaired learning if tasks are not appropriately selected. We introduce a sequential subnetwork routing (SeqSNR) architecture that uses soft parameter sharing to find related tasks and encourage cross-learning between them.<\/jats:p><\/jats:sec><jats:sec><jats:title>Materials and Methods<\/jats:title><jats:p>Using the MIMIC-III (Medical Information Mart for Intensive Care-III) dataset, we train deep neural network models to predict the onset of 6 endpoints including specific organ dysfunctions and general clinical outcomes: acute kidney injury, continuous renal replacement therapy, mechanical ventilation, vasoactive medications, mortality, and length of stay. We compare single-task (ST) models with naive multitask and SeqSNR in terms of discriminative performance and label efficiency.<\/jats:p><\/jats:sec><jats:sec><jats:title>Results<\/jats:title><jats:p>SeqSNR showed a modest yet statistically significant performance boost across 4 of 6 tasks compared with ST and naive multitasking. When the size of the training dataset was reduced for a given task (label efficiency), SeqSNR outperformed ST for all cases showing an average area under the precision-recall curve boost of 2.1%, 2.9%, and 2.1% for tasks using 1%, 5%, and 10% of labels, respectively.<\/jats:p><\/jats:sec><jats:sec><jats:title>Conclusions<\/jats:title><jats:p>The SeqSNR architecture shows superior label efficiency compared with ST and naive multitasking, suggesting utility in scenarios in which endpoint labels are difficult to ascertain.<\/jats:p><\/jats:sec>","DOI":"10.1093\/jamia\/ocab101","type":"journal-article","created":{"date-parts":[[2021,5,14]],"date-time":"2021-05-14T19:11:00Z","timestamp":1621019460000},"page":"1936-1946","source":"Crossref","is-referenced-by-count":17,"title":["Multitask prediction of organ dysfunction in the intensive care unit using sequential subnetwork routing"],"prefix":"10.1093","volume":"28","author":[{"given":"Subhrajit","family":"Roy","sequence":"first","affiliation":[{"name":"Google Health, London, United Kingdom"}]},{"given":"Diana","family":"Mincu","sequence":"additional","affiliation":[{"name":"Google Health, London, United Kingdom"}]},{"given":"Eric","family":"Loreaux","sequence":"additional","affiliation":[{"name":"Google Health, London, United Kingdom"}]},{"given":"Anne","family":"Mottram","sequence":"additional","affiliation":[{"name":"Google Health, London, United Kingdom"}]},{"given":"Ivan","family":"Protsyuk","sequence":"additional","affiliation":[{"name":"Google Health, London, United Kingdom"}]},{"given":"Natalie","family":"Harris","sequence":"additional","affiliation":[{"name":"Google Health, London, United Kingdom"}]},{"given":"Yuan","family":"Xue","sequence":"additional","affiliation":[{"name":"Google Health, Mountain View, California, USA"}]},{"given":"Jessica","family":"Schrouff","sequence":"additional","affiliation":[{"name":"Google Health, London, United Kingdom"}]},{"given":"Hugh","family":"Montgomery","sequence":"additional","affiliation":[{"name":"Centre for Human Health and Performance, University College London, London, United Kingdom"}]},{"given":"Alistair","family":"Connell","sequence":"additional","affiliation":[{"name":"Google Health, London, United Kingdom"}]},{"given":"Nenad","family":"Tomasev","sequence":"additional","affiliation":[{"name":"DeepMind, London, United Kingdom"}]},{"given":"Alan","family":"Karthikesalingam","sequence":"additional","affiliation":[{"name":"Google Health, London, United Kingdom"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-0435-3738","authenticated-orcid":false,"given":"Martin","family":"Seneviratne","sequence":"additional","affiliation":[{"name":"Google Health, London, United Kingdom"}]}],"member":"286","published-online":{"date-parts":[[2021,6,21]]},"reference":[{"issue":"8","key":"2021081407020193400_ocab101-B1","doi-asserted-by":"crossref","first-page":"591","DOI":"10.1097\/00003246-198108000-00008","article-title":"APACHE-acute physiology and chronic health evaluation: a physiologically based classification system","volume":"9","author":"Knaus","year":"1981","journal-title":"Crit Care Med"},{"issue":"24","key":"2021081407020193400_ocab101-B2","doi-asserted-by":"crossref","first-page":"2957","DOI":"10.1001\/jama.1993.03510240069035","article-title":"A new Simplified Acute Physiology Score (SAPS II) based on a European\/North American multicenter study","volume":"270","author":"Le Gall","year":"1993","journal-title":"JAMA"},{"issue":"7","key":"2021081407020193400_ocab101-B3","doi-asserted-by":"crossref","first-page":"707","DOI":"10.1007\/BF01709751","article-title":"The SOFA (Sepsis-related Organ Failure Assessment) score to describe organ dysfunction\/failure. On behalf of the Working Group on Sepsis-Related Problems of the European Society of Intensive Care Medicine","volume":"22","author":"Vincent","year":"1996","journal-title":"Intensive Care Med"},{"issue":"8","key":"2021081407020193400_ocab101-B4","doi-asserted-by":"crossref","first-page":"1280","DOI":"10.1007\/s00134-012-2578-0","article-title":"Comparison between SAPS II and SAPS 3 in predicting hospital mortality in a cohort of 103 Italian ICUs. Is new always better?","volume":"38","author":"Poole","year":"2012","journal-title":"Intensive Care Med"},{"issue":"4","key":"2021081407020193400_ocab101-B5","doi-asserted-by":"crossref","first-page":"423.e1","DOI":"10.1016\/j.jcrc.2011.08.016","article-title":"Caution when using prognostic models: A prospective comparison of 3 recent prognostic models","volume":"27","author":"Nassar","year":"2012","journal-title":"J Crit Care"},{"issue":"1","key":"2021081407020193400_ocab101-B6","doi-asserted-by":"crossref","first-page":"1936","DOI":"10.1038\/s41598-019-38491-0","article-title":"DeepSOFA: a continuous acuity score for critically ill patients using clinically interpretable deep learning","volume":"9","author":"Shickel","year":"2019","journal-title":"Sci Rep"},{"key":"2021081407020193400_ocab101-B7","first-page":"994","author":"Johnson","year":"2017"},{"key":"2021081407020193400_ocab101-B8","doi-asserted-by":"crossref","first-page":"101820","DOI":"10.1016\/j.artmed.2020.101820","article-title":"Early detection of sepsis utilizing deep learning on electronic health record event sequences","volume":"104","author":"Lauritsen","year":"2020","journal-title":"Artif Intell Med"},{"issue":"1","key":"2021081407020193400_ocab101-B9","doi-asserted-by":"crossref","first-page":"6145","DOI":"10.1038\/s41598-019-42637-5","article-title":"Data-driven discovery of a novel sepsis pre-shock state predicts impending septic shock in the ICU","volume":"9","author":"Liu","year":"2019","journal-title":"Sci Rep"},{"issue":"7767","key":"2021081407020193400_ocab101-B10","doi-asserted-by":"crossref","first-page":"116","DOI":"10.1038\/s41586-019-1390-1","article-title":"A clinically applicable approach to continuous prediction of future acute kidney injury","volume":"572","author":"Toma\u0161ev","year":"2019","journal-title":"Nature"},{"key":"2021081407020193400_ocab101-B11","author":"Wang","year":""},{"issue":"1","key":"2021081407020193400_ocab101-B12","doi-asserted-by":"crossref","first-page":"247","DOI":"10.1186\/s13054-019-2486-6","article-title":"Multi-morbidity states associated with higher mortality rates in organ dysfunction and sepsis: A data-driven analysis in critical care","volume":"23","author":"Zador","year":"2019","journal-title":"Crit Care"},{"key":"2021081407020193400_ocab101-B13","first-page":"1552","author":"Xue","year":"2020"},{"key":"2021081407020193400_ocab101-B14","author":"Sebastian"},{"key":"2021081407020193400_ocab101-B15","author":"Yu"},{"key":"2021081407020193400_ocab101-B16","author":"Vandenhende"},{"key":"2021081407020193400_ocab101-B17","author":"Mao"},{"issue":"1","key":"2021081407020193400_ocab101-B18","doi-asserted-by":"crossref","first-page":"368","DOI":"10.1186\/s12859-017-1776-8","article-title":"A neural network multi-task learning approach to biomedical named entity recognition","volume":"18","author":"Crichton","year":"2017","journal-title":"BMC Bioinform"},{"key":"2021081407020193400_ocab101-B19","doi-asserted-by":"crossref","first-page":"141","DOI":"10.3389\/fpls.2020.00141","article-title":"Doing more with less: a multitask deep learning approach in plant phenotyping","volume":"11","author":"Dobrescu","year":"2020","journal-title":"Front Plant Sci"},{"key":"2021081407020193400_ocab101-B20","author":"Miquel","year":"2017"},{"key":"2021081407020193400_ocab101-B21","author":"Ngufor","year":"2015"},{"issue":"1","key":"2021081407020193400_ocab101-B22","doi-asserted-by":"crossref","first-page":"96","DOI":"10.1038\/s41597-019-0103-9","article-title":"Multitask learning and benchmarking with clinical time series data","volume":"6","author":"Harutyunyan","year":"2019","journal-title":"Sci Data"},{"key":"2021081407020193400_ocab101-B23","author":"McDermott"},{"key":"2021081407020193400_ocab101-B24","first-page":"802","author":"Suresh","year":"2018"},{"key":"2021081407020193400_ocab101-B25","author":"Razavian"},{"key":"2021081407020193400_ocab101-B26","first-page":"1180","author":"Wang","year":"2014"},{"key":"2021081407020193400_ocab101-B27","first-page":"301","author":"Choi","year":"2016"},{"issue":"1","key":"2021081407020193400_ocab101-B28","first-page":"216","article-title":"SNR: sub-network routing for flexible parameter sharing in multi-task learning","volume":"33","author":"Ma","year":"2019","journal-title":"Proc AAAI Conf Artif Intell"},{"issue":"1","key":"2021081407020193400_ocab101-B29","doi-asserted-by":"crossref","first-page":"160035","DOI":"10.1038\/sdata.2016.35","article-title":"MIMIC-III, a freely accessible critical care database","volume":"3","author":"Johnson","year":"2016","journal-title":"Sci Data"},{"issue":"1","key":"2021081407020193400_ocab101-B30","doi-asserted-by":"crossref","first-page":"18","DOI":"10.1038\/s41746-018-0029-1","article-title":"Scalable and accurate deep learning with electronic health records","volume":"1","author":"Rajkomar","year":"2018","journal-title":"NPJ Digit Med"},{"key":"2021081407020193400_ocab101-B31"},{"issue":"10","key":"2021081407020193400_ocab101-B32","doi-asserted-by":"crossref","first-page":"1419","DOI":"10.1093\/jamia\/ocy068","article-title":"Opportunities and challenges in developing deep learning models using electronic health records data: a systematic review","volume":"25","author":"Cao","year":"2018","journal-title":"J Am Med Inform Assoc"},{"issue":"4","key":"2021081407020193400_ocab101-B33","doi-asserted-by":"crossref","first-page":"c179","DOI":"10.1159\/000339789","article-title":"KDIGO clinical practice guidelines for acute kidney injury","volume":"120","author":"Khwaja","year":"2012","journal-title":"Nephron Clin Pract"},{"key":"2021081407020193400_ocab101-B34"},{"key":"2021081407020193400_ocab101-B35","first-page":"1","author":"Louizos"},{"issue":"8","key":"2021081407020193400_ocab101-B36","doi-asserted-by":"crossref","first-page":"1735","DOI":"10.1162\/neco.1997.9.8.1735","article-title":"Long short-term memory","volume":"9","author":"Hochreiter","year":"1997","journal-title":"Neural Comput"},{"key":"2021081407020193400_ocab101-B37","first-page":"1724","author":"Cho","year":"2014"},{"key":"2021081407020193400_ocab101-B38","first-page":"1","author":"Jasmine"},{"key":"2021081407020193400_ocab101-B39","first-page":"249","article-title":"Understanding the difficulty of training deep feedforward neural networks","volume":"9","author":"Glorot","year":"2010","journal-title":"J Mach Learn Res"},{"key":"2021081407020193400_ocab101-B40","first-page":"1","author":"Kingma","year":"2015"},{"key":"2021081407020193400_ocab101-B41","first-page":"2575","article-title":"Variational dropout and the local reparameterization trick","volume-title":"Advances in Neural Information Processing Systems 28","author":"Kingma","year":"2015"},{"key":"2021081407020193400_ocab101-B42","author":"Kendall"},{"key":"2021081407020193400_ocab101-B43","first-page":"233","author":"Davis","year":"2006"},{"key":"2021081407020193400_ocab101-B44","volume-title":"An Introduction to the Bootstrap. Number 57 in Monographs on Statistics and Applied Probability","author":"Bradley","year":"1993"},{"key":"2021081407020193400_ocab101-B45","first-page":"1","article-title":"Statistical comparisons of classifiers over multiple data sets","volume":"7","author":"Dem\u0161ar","year":"2001","journal-title":"J Mach Learn Res"},{"issue":"13","key":"2021081407020193400_ocab101-B46","doi-asserted-by":"crossref","first-page":"3521","DOI":"10.1073\/pnas.1611835114","article-title":"Overcoming catastrophic forgetting in neural networks","volume":"114","author":"Kirkpatrick","year":"2017","journal-title":"Proc Natl Acad Sci U S A"},{"key":"2021081407020193400_ocab101-B47","doi-asserted-by":"crossref","first-page":"112","DOI":"10.1016\/j.jbi.2018.04.007","article-title":"Benchmarking deep learning models on large healthcare datasets","volume":"83","author":"Purushotham","year":"2017","journal-title":"J Biomed Inform"}],"container-title":["Journal of the American Medical Informatics Association"],"original-title":[],"language":"en","link":[{"URL":"http:\/\/academic.oup.com\/jamia\/article-pdf\/28\/9\/1936\/39731941\/ocab101.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"syndication"},{"URL":"http:\/\/academic.oup.com\/jamia\/article-pdf\/28\/9\/1936\/39731941\/ocab101.pdf","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2022,12,27]],"date-time":"2022-12-27T12:12:38Z","timestamp":1672143158000},"score":1,"resource":{"primary":{"URL":"https:\/\/academic.oup.com\/jamia\/article\/28\/9\/1936\/6307184"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2021,6,21]]},"references-count":47,"journal-issue":{"issue":"9","published-online":{"date-parts":[[2021,6,21]]},"published-print":{"date-parts":[[2021,8,13]]}},"URL":"https:\/\/doi.org\/10.1093\/jamia\/ocab101","relation":{},"ISSN":["1527-974X"],"issn-type":[{"value":"1527-974X","type":"electronic"}],"subject":[],"published-other":{"date-parts":[[2021,9,1]]},"published":{"date-parts":[[2021,6,21]]}}}