Skip to content
DraconiticResearch
Search Sources API
181,713 studies31 sourcesUpdated 11 min ago
Peer-reviewedCC BYnpj Health Systems

Discordance between actual care and recommendations by a reinforcement learning model for sepsis

  • Peter C. Nauka1
  • Jason N. Kennedy2
  • Xinlei Chen3
  • Matthieu Komorowski4
  • Chung-Chou H. Chang3
  • Romain Pirracchio5
  • Lu Tang3
  • Bryan J. McVerry2
  • Derek C. Angus2
  • Milos Hauskrecht6
  • Christopher W. Seymour2
6 affiliations
  1. 1Division of Pulmonary, Allergy, Critical Care, and Sleep Medicine, Department of Medicine, University of Pittsburgh Medical Center
  2. 2Department of Critical Care Medicine, University of Pittsburgh School of Medicine
  3. 3Department of Biostatistics, University of Pittsburgh
  4. 4Department of Surgery and Cancer, Imperial College London
  5. 5Department of Anesthesia and Perioperative Care, Zuckerberg San Francisco General Hospital and Trauma Center, University of California San Francisco
  6. 6Department of Computer Science, University of Pittsburgh
Read full textPDFDOICite

Abstract

Fluid and vasopressor management of sepsis requires a complex set of sequential treatment decisions tailored to individual responses. We examined the frequency and implications of disagreement between reinforcement-learning (RL), using the AI Clinician as a test example, and actual clinical decisions. A novel discordance score was defined as the Euclidean distance between clinical team actions and RL-recommended treatments (range 0 to 5.66), across 25 treatment options of vasopressors and intravenous (IV) fluids. Among 30,678 ICU patients with sepsis cared for across 309,415 time-epochs, discordance was more common among epochs of patients with SOFA > 6 than those with SOFA < 6, and for intravenous fluid choices than for vasopressors. After multivariable adjustment, greater discordance was associated with 90-day mortality (adjusted odds ratio (OR) 1.11, 95%CI: 1.05,1.18) in the SOFA strata > =6 but not for SOFA < = 4 (adjusted odds ratio (OR): 1.02, 95%CI: 0.97,1.08).

Related

  1. Delphi consensus guidelines for fungal sepsis in non-neutropenic critically ill patients in resource-limited settings: fungal infection study forum expert recommendations
    Shivaprakash Mandya Rudramurthy, Inderpaul Singh Sehgal, Ritesh Agarwal +18
    Scientific ReportsPeer-reviewedCC BY-NC-ND 4.0Full textInfectious diseaseCrossref
  2. Targeted cystatin C testing for large negative eGFR discordance in adults with diabetes: development and external validation of a machine-learning model
    Xuanxuan Wang, Mengyu Zhang, Ke Zhang +3
    Frontiers in EndocrinologyPeer-reviewedCC BY 4.0Endocrinology & metabolismCrossref
  3. Machine learning clustering of multidimensional clinical features identifies four prognostically distinct sepsis subtypes
    Rui Zhang, Yingli Zhang, Xirou Wu +6
    Scientific ReportsPeer-reviewedCC BY-NC-ND 4.0Full textEpidemiology & public healthCrossref
  4. Reinforcement learning–coupled neural ode modelling reveals ferroptotic resolution failure as a shared dynamical axis linking periodontitis and coronary atherogenesis
    Pradeep Kumar Yadalam, Roshan Noor Mohamed, Sakeenabi Basha +2
    Scientific ReportsPeer-reviewedCC BY-NC-ND 4.0Full textGeneral medicineCrossref
  5. Reinforcement learning as a foundation for psychodynamics
    Zack Dulberg, Jonathan D. Cohen
    PsyArXivPreprintCC BY 4.0Full textNeurosciencePsychiatry & mental healthOSF

Cite

BibTeXRIS
BibTeX
@article{nauka2026discordance,
  title = {{Discordance between actual care and recommendations by a reinforcement learning model for sepsis}},
  author = {Peter C. Nauka and Jason N. Kennedy and Xinlei Chen and Matthieu Komorowski and Chung-Chou H. Chang and Romain Pirracchio and Lu Tang and Bryan J. McVerry and Derek C. Angus and Milos Hauskrecht and Christopher W. Seymour},
  journal = {npj Health Systems},
  year = {2026},
  month = oct,
  doi = {10.1038/s44401-026-00156-2},
  url = {https://doi.org/10.1038/s44401-026-00156-2},
  note = {CC BY},
}
RIS
TY  - JOUR
TI  - Discordance between actual care and recommendations by a reinforcement learning model for sepsis
AU  - Peter C. Nauka
AU  - Jason N. Kennedy
AU  - Xinlei Chen
AU  - Matthieu Komorowski
AU  - Chung-Chou H. Chang
AU  - Romain Pirracchio
AU  - Lu Tang
AU  - Bryan J. McVerry
AU  - Derek C. Angus
AU  - Milos Hauskrecht
AU  - Christopher W. Seymour
PY  - 2026
DA  - 2026/10/25
JO  - npj Health Systems
DO  - 10.1038/s44401-026-00156-2
UR  - https://doi.org/10.1038/s44401-026-00156-2
AB  - Fluid and vasopressor management of sepsis requires a complex set of sequential treatment decisions tailored to individual responses. We examined the frequency and implications of disagreement between reinforcement-learning (RL), using the AI Clinician as a test example, and actual clinical decisions. A novel discordance score was defined as the Euclidean distance between clinical team actions and RL-recommended treatments (range 0 to 5.66), across 25 treatment options of vasopressors and intravenous (IV) fluids. Among 30,678 ICU patients with sepsis cared for across 309,415 time-epochs, discordance was more common among epochs of patients with SOFA > 6 than those with SOFA < 6, and for intravenous fluid choices than for vasopressors. After multivariable adjustment, greater discordance was associated with 90-day mortality (adjusted odds ratio (OR) 1.11, 95%CI: 1.05,1.18) in the SOFA strata > =6 but not for SOFA < = 4 (adjusted odds ratio (OR): 1.02, 95%CI: 0.97,1.08).
SN  - 3005-1959
ER  -