Decision Curve Analysis for Evaluating Machine Learning Models for Next-Day Transfer Out of ICU. Pozo, M., Pape, A., Locke, B., & Pettine, W. W. April, 2026. ISSN: 3067-2007 Pages: 2026.04.19.26351213![Decision Curve Analysis for Evaluating Machine Learning Models for Next-Day Transfer Out of ICU [link]](https://bibbase.org/img/filetypes/link.svg "link")
Paper doi abstract bibtex Timely identification of intensive care unit (ICU) patients likely to exit the unit can support anticipatory workflows such as chart review, eligibility screening, and patient outreach prior to transfer. Most ICU discharge prediction studies report discrimination and calibration, but these metrics do not quantify the decision consequences of acting on predictions. Using adult ICU admissions from MIMIC-IV, we represented each ICU stay as a sequence of daily clinical summaries and trained logistic regression, random forest, and XGBoost models to predict next day ICU transfer. Models achieved ROC AUC of 0.80 to 0.84 with differing calibration. We evaluated decision utility using decision curve analysis (DCA), where positive predictions trigger proactive review. Across thresholds, model guided strategies outperformed review-all, review-none, and a simple clinical rule. To translate net benefit into implementable operations, we modeled a clinical trial recruitment workflow with an 8 hour daily time constraint, incorporating chart review and consent effort. At a feasible operating threshold (0.23), the model flagged 23 charts/day and yielded 1.23 enrollments/day under conservative eligibility and consent assumptions. These results demonstrate that DCA provides a transparent framework for determining when ICU transfer predictions are worth using and how thresholds should be selected to align with real world workflow constraints.
@misc{pozo_decision_2026,
title = {Decision {Curve} {Analysis} for {Evaluating} {Machine} {Learning} {Models} for {Next}-{Day} {Transfer} {Out} of {ICU}},
copyright = {© 2026, Posted by openRxiv. This pre-print is available under a Creative Commons License (Attribution 4.0 International), CC BY 4.0, as described at http://creativecommons.org/licenses/by/4.0/},
url = {https://www.medrxiv.org/content/10.64898/2026.04.19.26351213v1},
doi = {10.64898/2026.04.19.26351213},
abstract = {Timely identification of intensive care unit (ICU) patients likely to exit the unit can support anticipatory workflows such as chart review, eligibility screening, and patient outreach prior to transfer. Most ICU discharge prediction studies report discrimination and calibration, but these metrics do not quantify the decision consequences of acting on predictions. Using adult ICU admissions from MIMIC-IV, we represented each ICU stay as a sequence of daily clinical summaries and trained logistic regression, random forest, and XGBoost models to predict next day ICU transfer. Models achieved ROC AUC of 0.80 to 0.84 with differing calibration. We evaluated decision utility using decision curve analysis (DCA), where positive predictions trigger proactive review. Across thresholds, model guided strategies outperformed review-all, review-none, and a simple clinical rule. To translate net benefit into implementable operations, we modeled a clinical trial recruitment workflow with an 8 hour daily time constraint, incorporating chart review and consent effort. At a feasible operating threshold (0.23), the model flagged 23 charts/day and yielded 1.23 enrollments/day under conservative eligibility and consent assumptions. These results demonstrate that DCA provides a transparent framework for determining when ICU transfer predictions are worth using and how thresholds should be selected to align with real world workflow constraints.},
language = {en},
urldate = {2026-04-21},
publisher = {medRxiv},
author = {Pozo, Margaret and Pape, Abigail and Locke, Brian and Pettine, Warren Woodrich},
month = apr,
year = {2026},
note = {ISSN: 3067-2007
Pages: 2026.04.19.26351213},
}
Downloads: 0
{"_id":"x6goquyEicGapYduS","bibbaseid":"pozo-pape-locke-pettine-decisioncurveanalysisforevaluatingmachinelearningmodelsfornextdaytransferoutoficu-2026","author_short":["Pozo, M.","Pape, A.","Locke, B.","Pettine, W. W."],"bibdata":{"bibtype":"misc","type":"misc","title":"Decision Curve Analysis for Evaluating Machine Learning Models for Next-Day Transfer Out of ICU","copyright":"© 2026, Posted by openRxiv. This pre-print is available under a Creative Commons License (Attribution 4.0 International), CC BY 4.0, as described at http://creativecommons.org/licenses/by/4.0/","url":"https://www.medrxiv.org/content/10.64898/2026.04.19.26351213v1","doi":"10.64898/2026.04.19.26351213","abstract":"Timely identification of intensive care unit (ICU) patients likely to exit the unit can support anticipatory workflows such as chart review, eligibility screening, and patient outreach prior to transfer. Most ICU discharge prediction studies report discrimination and calibration, but these metrics do not quantify the decision consequences of acting on predictions. Using adult ICU admissions from MIMIC-IV, we represented each ICU stay as a sequence of daily clinical summaries and trained logistic regression, random forest, and XGBoost models to predict next day ICU transfer. Models achieved ROC AUC of 0.80 to 0.84 with differing calibration. We evaluated decision utility using decision curve analysis (DCA), where positive predictions trigger proactive review. Across thresholds, model guided strategies outperformed review-all, review-none, and a simple clinical rule. To translate net benefit into implementable operations, we modeled a clinical trial recruitment workflow with an 8 hour daily time constraint, incorporating chart review and consent effort. At a feasible operating threshold (0.23), the model flagged 23 charts/day and yielded 1.23 enrollments/day under conservative eligibility and consent assumptions. These results demonstrate that DCA provides a transparent framework for determining when ICU transfer predictions are worth using and how thresholds should be selected to align with real world workflow constraints.","language":"en","urldate":"2026-04-21","publisher":"medRxiv","author":[{"propositions":[],"lastnames":["Pozo"],"firstnames":["Margaret"],"suffixes":[]},{"propositions":[],"lastnames":["Pape"],"firstnames":["Abigail"],"suffixes":[]},{"propositions":[],"lastnames":["Locke"],"firstnames":["Brian"],"suffixes":[]},{"propositions":[],"lastnames":["Pettine"],"firstnames":["Warren","Woodrich"],"suffixes":[]}],"month":"April","year":"2026","note":"ISSN: 3067-2007 Pages: 2026.04.19.26351213","bibtex":"@misc{pozo_decision_2026,\n\ttitle = {Decision {Curve} {Analysis} for {Evaluating} {Machine} {Learning} {Models} for {Next}-{Day} {Transfer} {Out} of {ICU}},\n\tcopyright = {© 2026, Posted by openRxiv. This pre-print is available under a Creative Commons License (Attribution 4.0 International), CC BY 4.0, as described at http://creativecommons.org/licenses/by/4.0/},\n\turl = {https://www.medrxiv.org/content/10.64898/2026.04.19.26351213v1},\n\tdoi = {10.64898/2026.04.19.26351213},\n\tabstract = {Timely identification of intensive care unit (ICU) patients likely to exit the unit can support anticipatory workflows such as chart review, eligibility screening, and patient outreach prior to transfer. Most ICU discharge prediction studies report discrimination and calibration, but these metrics do not quantify the decision consequences of acting on predictions. Using adult ICU admissions from MIMIC-IV, we represented each ICU stay as a sequence of daily clinical summaries and trained logistic regression, random forest, and XGBoost models to predict next day ICU transfer. Models achieved ROC AUC of 0.80 to 0.84 with differing calibration. We evaluated decision utility using decision curve analysis (DCA), where positive predictions trigger proactive review. Across thresholds, model guided strategies outperformed review-all, review-none, and a simple clinical rule. To translate net benefit into implementable operations, we modeled a clinical trial recruitment workflow with an 8 hour daily time constraint, incorporating chart review and consent effort. At a feasible operating threshold (0.23), the model flagged 23 charts/day and yielded 1.23 enrollments/day under conservative eligibility and consent assumptions. These results demonstrate that DCA provides a transparent framework for determining when ICU transfer predictions are worth using and how thresholds should be selected to align with real world workflow constraints.},\n\tlanguage = {en},\n\turldate = {2026-04-21},\n\tpublisher = {medRxiv},\n\tauthor = {Pozo, Margaret and Pape, Abigail and Locke, Brian and Pettine, Warren Woodrich},\n\tmonth = apr,\n\tyear = {2026},\n\tnote = {ISSN: 3067-2007\nPages: 2026.04.19.26351213},\n}\n\n","author_short":["Pozo, M.","Pape, A.","Locke, B.","Pettine, W. W."],"key":"pozo_decision_2026","id":"pozo_decision_2026","bibbaseid":"pozo-pape-locke-pettine-decisioncurveanalysisforevaluatingmachinelearningmodelsfornextdaytransferoutoficu-2026","role":"author","urls":{"Paper":"https://www.medrxiv.org/content/10.64898/2026.04.19.26351213v1"},"metadata":{"authorlinks":{}}},"bibtype":"misc","biburl":"https://api.zotero.org/users/1528358/collections/DH4R6ZWA/items?key=L8lztpUgybScRoWSW76OhIE0&format=bibtex&limit=100","dataSources":["hoqThwEepw4WmWudJ","zYjFbyFmZWKpRCD4j"],"keywords":[],"search_terms":["decision","curve","analysis","evaluating","machine","learning","models","next","day","transfer","out","icu","pozo","pape","locke","pettine"],"title":"Decision Curve Analysis for Evaluating Machine Learning Models for Next-Day Transfer Out of ICU","year":2026}