Multitask prediction of organ dysfunction in the intensive care unit using sequential subnetwork routing
| Autor: | Alan Karthikesalingam, Eric Loreaux, Alistair Connell, Hugh Montgomery, Jessica Schrouff, Natalie Harris, Anne Mottram, Nenad Tomasev, Martin G. Seneviratne, Subhrajit Roy, Diana Mincu, Yuan Xue, Ivan Protsyuk |
|---|---|
| Rok vydání: | 2020 |
| Předmět: |
0301 basic medicine
AcademicSubjects/SCI01060 Computer science Multiple Organ Failure Multi-task learning Health Informatics Machine learning computer.software_genre Research and Applications Multitask Learning Machine Learning 03 medical and health sciences 0302 clinical medicine Deep Learning Discriminative model Intensive care Human multitasking Electronic Health Records Humans Subnetwork AcademicSubjects/MED00580 Artificial neural network business.industry Deep learning Intensive Care 030208 emergency & critical care medicine Task (computing) Intensive Care Units 030104 developmental biology Artificial intelligence Neural Networks Computer AcademicSubjects/SCI01530 business computer |
| Zdroj: | Journal of the American Medical Informatics Association : JAMIA |
| ISSN: | 1527-974X |
| Popis: | ObjectiveMultitask 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 – 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.Materials and MethodsUsing 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.ResultsSeqSNR 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.ConclusionsThe SeqSNR architecture shows superior label efficiency compared with ST and naive multitasking, suggesting utility in scenarios in which endpoint labels are difficult to ascertain. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|