Skip to main navigation Skip to main content

CEEM : Clinical and Experimental Emergency Medicine

OPEN ACCESS
ABOUT
BROWSE ARTICLES
FOR CONTRIBUTORS

Articles

Editorial
AI & Digital Health | Critical Care

Artificial intelligence in sepsis care and research: aligning clinical questions with model selection and human oversight


Published online: August 26, 2026

1Department of Emergency Medicine, Korea University Ansan Hospital, Ansan, Korea

2Core Research and Development Center, Korea University Ansan Hospital, Ansan, Korea

Correspondence to: Sejoong Ahn (sejoongahn@naver.com)

Sukyo Lee and Sumin Jung contributed equally to this study as co-first authors.

• Received: July 23, 2026   • Revised: August 12, 2026   • Accepted: August 12, 2026

Copyright © 2026 The Korean Society of Emergency Medicine

This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0/).

  • 407 Views
  • 8 Download
Sepsis is life-threatening organ dysfunction caused by a dysregulated host response to infection [1,2]. Because of its high incidence and mortality, it remains a major global health burden [1–4]. Sepsis care depends on timely recognition, risk stratification, and early treatment decisions, making it a compelling target for artificial intelligence (AI).
At the same time, sepsis is time-sensitive, heterogeneous, and data-rich, which complicates AI development and implementation [5,6]. Patients vary in infection source, host response, comorbidities, organ dysfunction, and treatment trajectory. Relevant data include structured electronic health record (EHR) variables, clinical text, imaging, physiological signals, biomarkers, and molecular profiles.
From a modeling perspective, these data are often longitudinal, irregularly sampled, and incomplete. Missing values may reflect clinical workflow rather than random data loss, and definitions, time zero, and outcome labels may differ across studies. Models may also learn institution-specific documentation practices or treatment patterns. Temporal leakage can occur when information recorded after the intended prediction time is inadvertently included. Many of these challenges are not unique to AI. Missing data, heterogeneous cohort and outcome definitions, censoring, confounding by indication, and temporal leakage have long been recognized in sepsis research. AI can amplify or obscure these problems through high-dimensional feature construction, automated imputation or data augmentation, repeated model tuning, and complex representations, which may make underlying bias harder to detect. Generative AI adds more specific risks, including hallucination, retrieval failure, and unverified content. Evaluation of AI in sepsis therefore requires distinguishing risks that are inherited from conventional sepsis research, amplified by AI methods, or newly introduced by AI systems.
AI in sepsis should not be framed simply as a competition among algorithms. Its clinical value depends on alignment among the clinical question, data modality, temporal context, model objective, model family, explainability strategy, and human-in-the-loop implementation. A model designed for screening at emergency department triage differs fundamentally from one intended for prognosis, treatment and decision support, phenotyping, or source-linked documentation. The same algorithm, performance metric, or explanation strategy cannot be assumed to suit all of these tasks. This editorial provides a sepsis-focused synthesis of how clinical questions, model selection, evaluation, and human oversight should be aligned in clinical care and research.
When AI is developed, evaluated, or used in sepsis, the intended clinical or research task should be defined explicitly. AI has been applied to screening and early detection [7–9], prognostic prediction [10,11], treatment and decision support [12,13], phenotyping [14–16], and documentation or information synthesis [17–19]. Table 1 summarizes differences in data structure, target outputs, model families, and key cautions across these applications.
These applications should not be collapsed into a single prediction problem. Screening and prognostic models estimate current or future risk, whereas treatment-related tasks must distinguish risk prediction from expected treatment benefit. Phenotyping aims to identify heterogeneity rather than predict outcomes. Generative AI and retrieval-augmented generation (RAG) systems primarily support information synthesis and communication, rather than autonomous clinical decision-making. These distinctions shape the choice of data, evaluation methods, and clinical validation requirements.
No single model family is optimal for every sepsis AI task. Table 2 provides a concise overview of model families relevant to sepsis.
Conventional statistical models remain useful when transparency, interpretability, and ease of implementation are priorities [20]. Tree-based machine learning models may perform well with structured EHR data and nonlinear associations [21].
Time-to-event and dynamic prediction models account for event timing, censoring, and updated patient information, and they can provide horizon-specific risk estimates [22]. Temporal deep learning models may capture complex physiological trajectories [23]. Although irregular sampling and missingness are not specific to AI, flexible temporal models may incorporate workflow-related sampling patterns into learned representations, making these biases harder to recognize. Model complexity, domain shift, inference latency, and computational burden should therefore be considered when these approaches are selected.
The increasing use of unstructured and multimodal data also affects model selection. Clinical language models can process clinical text [24], and image and signal encoders can process imaging, electrocardiography, and physiological waveforms [25,26]. Multimodal fusion models can integrate EHR variables, text, imaging, biosignals, biomarkers, and omics-derived data [27]. Future sepsis AI research may increasingly incorporate host response biomarkers and high-dimensional omics, including transcriptomics, proteomics, metabolomics, and single-cell profiles. These data may further support prognosis, phenotyping, treatment response assessment, and trial enrichment.
Generative AI and RAG-based systems may support information extraction, summarization, explanation, documentation, and draft recommendations [28]. These uses should be distinguished from autonomous clinical decision-making. In high-risk conditions such as sepsis, generated text should be grounded in retrieved sources, local protocols, and patient-specific data, and it should remain subject to clinician review.
Causal inference and treatment effect models are needed when the question is whether an intervention is likely to improve outcomes, rather than simply whether a patient is at high risk [29]. In sepsis, this distinction is critical for fluids, vasopressors, corticosteroids, antibiotics, source control, and organ support. Confounding by indication, treatment timing, immortal time bias, and unmeasured confounding are not AI-specific problems; AI-based treatment-effect models inherit the same causal-identification requirements as conventional causal analyses.
Phenotyping and representation learning models may characterize sepsis heterogeneity and support trial enrichment or hypothesis generation [30]. Cluster instability and limited reproducibility are not unique to AI. However, high-dimensional representation learning may make cluster stability and interpretability more difficult to assess.
Newer approaches, including foundation models, self-supervised learning, and multimodal learning, are promising in selected settings but require careful validation before clinical use [27,31–33].
Model selection should also account for practical constraints, including performance, calibration, computational burden, inference time, and workflow integration.
Core principles of model evaluation, including discrimination, calibration, threshold-dependent performance, subgroup performance, and external or temporal validation, are not specific to AI. They remain essential because flexible, high-dimensional AI models can show strong apparent performance while learning workflow-specific patterns. The area under the receiver operating characteristic curve and the area under the precision-recall curve are useful summary measures, particularly in imbalanced datasets, but they should not be interpreted in isolation [33]. Clinical utility, uncertainty, and transportability should also be assessed.
Explainability raises a separate set of challenges. Depending on the model and task, useful methods may include coefficients, risk curves, Shapley Additive Explanations (SHAP) values, saliency maps, temporal attribution, modality ablation, and source-linked evidence [34,35]. In complex AI models, however, post hoc explanations may appear plausible without accurately reflecting how the model produced its output. They should not be interpreted as evidence of causality or clinical correctness [35]. A plausible feature attribution, attention map, or generated explanation can still be misleading if it is unstable, poorly grounded, or inconsistent with the clinical context. For generative AI, verification also requires confirming that outputs are grounded in retrieved sources and patient-specific data.
AI should not replace clinical judgment in sepsis [33,36–39]. Well-designed AI systems should instead help clinicians recognize risk earlier, track changing risk over time, synthesize complex multimodal data, communicate uncertainty, and support timely but accountable decisions. Table 3 summarizes representative AI roles and corresponding human-in-the-loop processes in sepsis care.
Human involvement is particularly important because an AI output may not capture bedside findings, treatment response, contraindications, goals of care, or other contextual factors that influence clinical decisions. Clinicians should therefore be able to review the relevant input data, supporting evidence, and uncertainty, and to modify or override an output when it does not fit the clinical context. This safeguard is especially important for treatment and decision support, where a plausible recommendation does not necessarily mean that the intervention is appropriate for an individual patient. Similarly, phenotype-informed outputs should be interpreted in relation to bedside findings and linked to a clinically meaningful action. After deployment, override patterns and clinician feedback may also help identify calibration drift, workflow mismatch, inappropriate thresholds, or safety concerns.
Human oversight is also important throughout AI research and deployment [33,36,39,40]. Table 4 summarizes key considerations across AI model development, validation, reporting, and postdeployment monitoring. In sepsis, this oversight is particularly relevant because clinical data are generated through clinical workflows. Testing frequency, monitoring intensity, treatment decisions, and documentation practices influence what is recorded in the EHR and may vary across institutions and over time. AI models may therefore learn patterns of clinical practice as well as patterns of disease. Many resulting concerns, including workflow-related missingness, temporal leakage, confounding, and limited transportability, are not unique to AI. However, flexible preprocessing, high-dimensional feature construction, imputation, and complex models may amplify or obscure these problems and make them harder to recognize. Human oversight is needed to ensure that the data, model outputs, and apparent performance remain aligned with the intended clinical or research question.
Oversight remains important after deployment because implementation can change clinical behavior and the data-generating process. Changes in case mix, documentation, treatment protocols, model use, or clinician response may alter model performance or create unintended workflow effects. Clinicians and AI researchers should therefore jointly assess clinical plausibility, actionability, performance drift, override patterns, and emerging safety concerns.
In sepsis, clinically useful AI requires alignment among the clinical question, data, model selection, evaluation, and intended workflow. Human oversight across research and clinical implementation is essential for addressing risks inherited from sepsis research as well as risks amplified or newly introduced by AI.

Conflicts of interest

The authors have no conflicts of interest to declare.

Funding

The authors received no financial support for this study.

Data availability

Data sharing is not applicable as no new data were created or analyzed in this study.

Table 1.
Clinical and research tasks, AI roles, and candidate model families in sepsis
Table 1.
Clinical or research task Typical data and timing Target output or AI role in sepsis Candidate model families Key methodological and AI-related cautions
Screening and diagnosis Demographics, comorbidities, vital signs, laboratory results, triage variables, clinical text, microbiology results, imaging, or ECG; single time point or short observation window Estimated probability of sepsis or infection, clinical alert, and supporting evidence Conventional statistical and interpretable models; tree-based structured-data machine learning models; modality-specific encoders; multimodal fusion models Clinical label and time-zero definition, temporal leakage, class imbalance, false-positive burden, and alert fatigue
Prognostic prediction Serial EHR variables, organ support, treatment response, progress or nursing notes, imaging, ECG, and physiological waveforms; longitudinal or real-time data Horizon-specific or time-to-event risk of septic shock, new or worsening organ failure, ICU transfer, short- or long-term mortality, or recovery Time-to-event and dynamic prediction models; tree-based models using trajectory features; temporal deep learning models; multimodal fusion models Irregular sampling, missingness, censoring, competing events, treatment-dependent variables, time-varying decisions, and calibration over time
Treatment and decision support Longitudinal hemodynamics, laboratory results, organ support, treatment history, clinical notes, microbiology or radiology findings, and local protocols; updated over time according to treatment exposure and response Risk-based prompts, estimated treatment effects, prioritized options, or source-linked recommendations for fluids, vasopressors, corticosteroids, antibiotics, source control, organ support, or ICU transfer Causal inference and treatment effect models for intervention-related questions; prediction models for risk estimation only; LLM/RAG-based systems for source-linked presentation Risk prediction does not establish treatment benefit; confounding by indication, immortal time bias, protocol alignment, uncertainty, and clinician verification
Phenotyping and heterogeneity EHR variables, laboratory results, biomarkers, treatment exposure, outcomes, omics-derived high-dimensional data, and optional multimodal data; baseline profiles or longitudinal cohort data Phenotype, cluster, latent representation, potential responder subgroup, or trial-eligibility assessment Phenotyping and representation learning models; multimodal fusion models; causal inference and treatment effect models for treatment effect heterogeneity Sensitivity to preprocessing and model selection, cluster instability, limited reproducibility and transportability, biological plausibility, and clinical actionability
Documentation and communication Structured EHR data combined with clinical notes, microbiology or radiology reports, retrieved guidelines, and local protocols; encounter-level, handoff-level, hospital-course, or discharge-level information Source-linked summary, handoff draft, explanation, guideline summary, or draft recommendation for clinician review Generative AI and LLM/RAG-based systems; structured-to-text systems; modality-specific encoders and multimodal fusion models when source data are diverse Hallucination, outdated or irrelevant retrieval, weak source fidelity, overconfident language, automation bias, and required clinician verification

Many cautions listed for screening, prognosis, treatment, and phenotyping reflect established methodological concerns in sepsis research that may be inherited, amplified, or obscured by AI. Hallucinations are associated with generative AI, whereas retrieval failure and weak source fidelity are particularly relevant to RAG-based systems.

AI, artificial intelligence; ECG, electrocardiography; EHR, electronic health record; ICU, intensive care unit; LLM, large language model; RAG, retrieval-augmented generation.

Table 2.
Characteristics, precautions, and explainability/verification approaches for AI model families relevant to sepsis
Table 2.
Model family Model or approach example Suitable data and task Main strength Key consideration in sepsis Explainability and verification approach
Conventional statistical and interpretable models Linear regression, logistic regression, penalized regression, GAM, EBM Tabular EHR data; screening, diagnosis, and initial risk prediction Transparent and clinically familiar; relatively easy to implement; low computational burden Limited ability to capture complex temporal, high-dimensional, or multimodal relationships; model assumptions require checking Coefficients, odds ratios, score components, nomograms, shape functions, and risk curves
Tree-based structured-data machine learning models Random forest, XGBoost, LightGBM, CatBoost Tabular EHR data; single-time-point prediction or summarized longitudinal data Captures nonlinear relationships and interactions; often provides a strong baseline for structured clinical data Less transparent; may learn institution-specific missingness or practice patterns; feature attribution should not be interpreted as evidence of causality Permutation importance, SHAP, LIME, partial dependence plots, and ICE plots
Time-to-event and dynamic prediction models Cox models, landmarking models, joint models, competing-risk models Longitudinal structured data; mortality, organ failure, ICU transfer, recovery, and other time-to-event outcomes Accounts for event timing, censoring, and updated patient information; provides horizon-specific risk estimates Requires a clear time origin, outcome definition, censoring strategy, and approach to competing events and treatment-dependent variables Survival curves, dynamic risk curves, landmark-specific estimates, and time-dependent effects
Temporal deep learning models RNN, GRU, LSTM, TCN, temporal transformer Serial EHR data and physiological waveforms; dynamic prognosis or real-time monitoring Captures complex trajectories and temporal interactions Requires relatively large datasets and computational resources; sensitive to irregular sampling, missingness, domain shift, and latency; model compression may be needed for real-time deployment Trajectory visualization, temporal attribution, occlusion analysis, saliency maps, and cautious use of attention visualization
Modality-specific encoders Clinical language models, CNN, vision transformer, pretrained or foundation-model-based encoders, one-dimensional CNN, signal transformer Clinical text, imaging, ECG, and physiological waveforms Extracts information from high-dimensional modality-specific data; pretrained or self-supervised models may reduce annotation requirements Sensitive to documentation bias, image or signal quality, artifacts, synchronization, domain shift, and computational burden Text or token attribution, class activation maps, image occlusion, waveform-segment attribution, and robustness checks
Multimodal fusion models Early fusion, late fusion, cross-attention models, multimodal transformers Combinations of EHR data, text, imaging, physiological signals, biomarkers, or omics data Integrates complementary information from multiple data sources Missing modalities, temporal alignment, heterogeneous data quality, modality dominance, model opacity, and computational burden Modality ablation, modality-specific attribution, cross-modal contribution displays, and representative case comparisons
Generative AI and RAG-based systems RAG-based LLMs, guideline-constrained LLMs, structured-to-text LLMs Structured, unstructured, or multimodal clinical data; extraction, summarization, explanation, documentation support, and draft recommendations for clinician review Supports information extraction, summarization, documentation, and source-linked communication Hallucination, retrieval failure or outdated retrieval, weak source fidelity, overconfident language, cost, and inference latency Retrieved-source display, citation and version tracking, audit trails, uncertainty display, and clinician verification
Causal inference and treatment effect models Target trial emulation, propensity score methods, marginal structural models, causal forests, uplift models, dynamic treatment regimes Longitudinal treatment and outcome data; average or individualized treatment effect estimation More appropriate than risk prediction models alone for questions about intervention benefit Confounding by indication, limited overlap, immortal time bias, unmeasured confounding, and difficulty evaluating learned treatment policies Causal diagrams, balance and overlap checks, subgroup treatment-effect plots, and sensitivity analyses
Phenotyping and representation learning models Clustering, latent class analysis, autoencoders, contrastive learning, multimodal representation learning Structured EHR, biomarkers, omics, or multimodal data; phenotype, cluster, latent representation, or responder subgroup identification Useful for studying sepsis heterogeneity, trial enrichment, and hypothesis generation Sensitive to preprocessing and model choice; clusters may be unstable, poorly reproducible, or difficult to interpret clinically Phenotype profiles, cluster heatmaps, latent-space visualization, representative cases, and stability analyses

Tabular EHR data refer to structured, row-by-column clinical data such as demographics, vital signs, laboratory results, medications, organ support, timestamps, and outcomes. Cox models include conventional Cox proportional hazards models with fixed covariates and time-dependent Cox models with covariates updated over time.

AI, artificial intelligence; GAM, generalized additive model; EBM, explainable boosting machine; EHR, electronic health record; XGBoost, Extreme Gradient Boosting; LightGBM, Light Gradient Boosting Machine; CatBoost, Categorical Boosting; SHAP, Shapley Additive Explanations; LIME, Local Interpretable Model-agnostic Explanations; ICE, individual conditional expectation; ICU, intensive care unit; RNN, recurrent neural network; GRU, gated recurrent unit; LSTM, long short-term memory; TCN, temporal convolutional network; CNN, convolutional neural network; ECG, electrocardiography; RAG, retrieval-augmented generation; LLM, large language model;

Table 3.
Human-in-the-loop design for clinical AI implementation in sepsis care
Table 3.
Clinical implementation area in sepsis AI role Human-in-the-loop design Potential risk without human oversight
Screening and diagnosis Flags possible sepsis, estimates infection probability, highlights concerning patterns, and summarizes supporting evidence The clinician reviews the relevant input data, supporting evidence, and uncertainty; confirms whether the alert is clinically meaningful; and decides whether additional evaluation or treatment is needed False-positive alerts may lead to unnecessary evaluation or treatment, whereas poorly contextualized alerts can increase alert fatigue
Prognostic prediction Tracks the risk of shock, organ failure, ICU transfer, death, or recovery and identifies worsening or improving risk patterns The clinician interprets the estimated risk in the context of bedside assessment, treatment response, comorbidities, goals of care, and information not captured by the model Risk estimates may be treated as deterministic or may influence subsequent care, potentially reinforcing self-fulfilling predictions
Treatment and decision support Suggests or prioritizes possible actions, such as fluids, vasopressor escalation, corticosteroid consideration, antibiotic optimization, source control, or ICU transfer, without autonomously determining treatment The clinician verifies the indication, contraindications, prior treatment response, bedside findings, and goals of care, and then approves, modifies, or overrides the recommendation Recommendations may overlook contraindications, prior treatment response, bedside findings, or goals of care and may be followed despite uncertain treatment benefit
LLM-assisted documentation and communication Generates source-linked chart summaries, handoff drafts, recommendation rationales, trial-screening summaries, or patient-specific guideline summaries The clinician checks the underlying source, source version, patient-specific accuracy, and final wording and remains responsible for the final documentation or communication Hallucinated, outdated, or weakly sourced information may enter clinical documentation or communication without verification
Phenotyping for clinical stratification Suggests clinically meaningful subgroups, such as high-risk organ dysfunction patterns, host response patterns, or potential treatment response profiles Clinicians review whether the suggested phenotype is consistent with bedside findings, infection source, organ dysfunction, and available biomarkers before using it to guide monitoring intensity, consultation, or trial referral Unstable or poorly interpretable phenotypes may be treated as clinically meaningful subgroups without sufficient bedside or biological support

AI, artificial intelligence; ICU, intensive care unit; LLM, large language model.

Table 4.
Key human oversight considerations in sepsis AI model development, validation, and deployment
Table 4.
Research stage Human role and key decisions Why oversight matters
Clinical or research question and intended use Define the clinical or research question, target population, care setting, prediction timing, target outcome, and intended action before model development This prevents AI-first framing before a clinically meaningful question has been defined; unclear time zero, target population, or outcome definitions reduce interpretability
Ethical, privacy, and accountability considerations Confirm IRB approval, privacy protection, deidentification, secure data storage, data access control, accountability, and audit trails Privacy breaches, re-identification risk, inappropriate data access, unclear responsibility, and lack of auditability can limit safe and trustworthy AI research
Data source, quality, and timing Define the data source, collection period, variable definitions, missing data handling, measurement units, timestamps, and data availability at the intended prediction time Clinical workflows determine what is measured and recorded; workflow-driven missingness, inconsistent timing, or postprediction information may produce misleading associations or apparent performance
Model development and performance evaluation Select the model family, preprocessing strategy, train/validation/test split, tuning approach, internal validation, and performance metrics appropriate to the intended use AUROC or AUPRC alone is insufficient; poor calibration, class imbalance, overfitting, and lack of subgroup evaluation can limit clinical usefulness; flexible preprocessing, feature construction, and model tuning may obscure bias or overfitting if their clinical and temporal implications are not reviewed
External and temporal validation Evaluate model performance across different time periods, institutions, patient groups, care settings, and practice environments when feasible Good internal performance does not guarantee transportability across institutions, populations, or practice patterns
Clinical interpretation and transparent reporting Clinicians and AI researchers jointly assess clinical plausibility, explainability, actionability, and reporting completeness Plausible explanations do not establish causality or clinical correctness; novel predictors and phenotypes require clinical interpretation and reproducibility assessment
Clinical implementation and postdeployment monitoring Define workflow integration, user interface, human review process, prospective evaluation, performance drift monitoring, update policy, and model-withdrawal criteria A technically accurate model may fail if it is not actionable, poorly integrated into workflow, or not monitored after deployment; deployment may change clinical behavior and the data-generating process; performance drift, subgroup errors, excessive alert burden, unintended workflow effects, and safety signals may therefore emerge over time

AI, artificial intelligence; IRB, institutional review board; AUROC, area under the receiver operating characteristic curve; AUPRC, area under the precision-recall curve.

  • 1. Singer M, Deutschman CS, Seymour CW, et al. The third international consensus definitions for sepsis and septic shock (Sepsis-3). JAMA 2016;315:801-10.
  • 2. Prescott HC, Antonelli M, Alhazzani W, et al. Surviving Sepsis Campaign: international guidelines for management of sepsis and septic shock 2026. Intensive Care Med 2026;52:863-936.
  • 3. GBD 2021 Global Sepsis Collaborators. Global, regional, and national sepsis incidence and mortality, 1990-2021: a systematic analysis. Lancet Glob Health 2025;13:e2013-26.
  • 4. Hwang SY, Kim T, Lim B, Ko SK, Lee SJ. Epidemiology of sepsis in emergency departments: insights from the National Emergency Department Information System (NEDIS) database in Korea, 2020-2024. Clin Exp Emerg Med 2026 Apr 8;[Epub]. https://doi.org/10.15441/ceem.26.017
  • 5. Giamarellos-Bourboulis EJ, Aschenbrenner AC, Bauer M, et al. The pathophysiology of sepsis and precision-medicine-based immunotherapy. Nat Immunol 2024;25:19-28.
  • 6. Shankar-Hari M, Ming D, Mendelson M, et al. The Lancet Commission on Sepsis: transforming sepsis care and outcomes. Lancet. 2026 Apr 22 [Epub]. https://doi.org/10.1016/S0140-6736(26)00648-3
  • 7. Adams R, Henry KE, Srdharan A, et al. Prospective, multi-site study of patient outcomes after implementation of the TREWS machine learning-based early warning system for sepsis. Nat Med 2022;28:1455-60.
  • 8. Goh KH, Wang L, Yeow AY, et al. Artificial intelligence in sepsis early prediction and diagnosis using unstructured data in healthcare. Nat Commun 2021;12:711.
  • 9. Shashikumar SP, Mohammadi S, Krishnamoorthy R, et al. Development and prospective implementation of a large language model based system for early sepsis prediction. NPJ Digit Med 2025;8:290.
  • 10. Rahman MS, Islam KR, Prithula J, et al. Machine learning-based prognostic model for 30-day mortality prediction in Sepsis-3. BMC Med Inform Decis Mak 2024;24:249.
  • 11. Shashikumar SP, Josef CS, Sharma A, Nemati S. DeepAISE: an interpretable and recurrent neural survival model for early prediction of sepsis. Artif Intell Med 2021;113:102036.
  • 12. Komorowski M, Celi LA, Badawi O, Gordon AC, Faisal AA. The Artificial Intelligence Clinician learns optimal treatment strategies for sepsis in intensive care. Nat Med 2018;24:1716-20.
  • 13. Kalimouttou A, Kennedy JN, Feng J, et al. Optimal vasopressin initiation in septic shock: the OVISS reinforcement learning study. JAMA 2025;333:1688-98.
  • 14. Seymour CW, Kennedy JN, Wang S, et al. Derivation, validation, and potential treatment implications of novel clinical phenotypes for sepsis. JAMA 2019;321:2003-17.
  • 15. Xu Z, Mao C, Su C, et al. Sepsis subphenotyping based on organ dysfunction trajectory. Crit Care 2022;26:197.
  • 16. van Amstel RB, Rademaker E, Kennedy JN, et al. Clinical subtypes in critically ill patients with sepsis: validation and parsimonious classifier model development. Crit Care 2025;29:58.
  • 17. Pak TR, Kanjilal S, McKenna CS, Hoffner-Heinike A, Rhee C, Klompas M. Syndromic analysis of sepsis cohorts using large language models. JAMA Netw Open 2025;8:e2539267.
  • 18. Zhang C, Yang H, Liu X, et al. A knowledge-enhanced platform (MetaSepsisKnowHub) for retrieval augmented generation-based sepsis heterogeneity and personalized management: development study. J Med Internet Res 2025;27:e67201.
  • 19. Boussina A, Allison C, Quintero K, et al. Medical record abstraction for quality improvement in sepsis care using artificial intelligence: a cluster randomized trial. JAMA Netw Open 2026;9:e2611885.
  • 20. Hastie T, Tibshirani R. Generalized additive models. Stat Sci 1986;1:297-310.
  • 21. Chen T, Guestrin C. XGBoost: a scalable tree boosting system. KDD '16: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining; 2016 Aug 13–17; San Francisco, CA, USA. p. 785–94.
  • 22. Van Houwelingen HC. Dynamic prediction by landmarking in event history analysis. Scand J Stat 2007;34:70-85.
  • 23. Hochreiter S, Schmidhuber J. Long short-term memory. Neural Comput 1997;9:1735-80.
  • 24. Devlin J, Chang MW, Lee K, Toutanova K. BERT: pre-training of deep bidirectional transformers for language understanding. In: Burstein J, Doran C, Solorio T, editors. Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: human language technologies, volume 1 (long and short papers); 2019 Jun 2-7; Minneapolis, MN, USA. Association for Computational Linguistics; 2019. p. 4171–86.
  • 25. Litjens G, Kooi T, Bejnordi BE, et al. A survey on deep learning in medical image analysis. Med Image Anal 2017;42:60-88.
  • 26. Faust O, Hagiwara Y, Hong TJ, Lih OS, Acharya UR. Deep learning for healthcare applications based on physiological signals: a review. Comput Methods Programs Biomed 2018;161:1-13.
  • 27. Baltrusaitis T, Ahuja C, Morency LP. Multimodal machine learning: a survey and taxonomy. IEEE Trans Pattern Anal Mach Intell 2019;41:423-43.
  • 28. Lewis P, Perez E, Piktus A, et al. Retrieval-augmented generation for knowledge-intensive NLP tasks. In: Larochelle H, Ranzato M, Hadsell R, Balcan MF, Lin H, editors. NIPS'20: Proceedings of the 34th International Conference on Neural Information Processing Systems; 2020 Dec 6-12; Vancouver, BC, Canada. Curran Associates; 2020. p. 9459–74.
  • 29. Hernán MA, Robins JM. Using big data to emulate a target trial when a randomized trial is not available. Am J Epidemiol 2016;183:758-64.
  • 30. Bengio Y, Courville A, Vincent P. Representation learning: a review and new perspectives. IEEE Trans Pattern Anal Mach Intell 2013;35:1798-828.
  • 31. Moor M, Banerjee O, Abad ZS, et al. Foundation models for generalist medical artificial intelligence. Nature 2023;616:259-65.
  • 32. Chen T, Kornblith S, Norouzi M, Hinton G. A simple framework for contrastive learning of visual representations. Proceedings of the 37th International Conference on Machine Learning; 2020 Jul 13-18. PMLR 2020;119:1597-1607.
  • 33. Lekadir K, Frangi AF, Porras AR, et al. FUTURE-AI: international consensus guideline for trustworthy and deployable artificial intelligence in healthcare. BMJ 2025;388:e081554.
  • 34. Okada Y, Ning Y, Ong ME. Explainable artificial intelligence in emergency medicine: an overview. Clin Exp Emerg Med 2023;10:354-62.
  • 35. Ghassemi M, Oakden-Rayner L, Beam AL. The false hope of current approaches to explainable artificial intelligence in health care. Lancet Digit Health 2021;3:e745-50.
  • 36. Vasey B, Nagendran M, Campbell B, et al. Reporting guideline for the early-stage clinical evaluation of decision support systems driven by artificial intelligence: DECIDE-AI. Nat Med 2022;28:924-33.
  • 37. World Health Organization (WHO). Ethics and governance of artificial intelligence for health: guidance on large multi-modal models. WHO; 2025.
  • 38. Heslin SM, Nocito R, Weingart SD. Artificial intelligence in emergency medicine critical care. Clin Exp Emerg Med 2026;13:1-4.
  • 39. Heslin SM, Singer AJ. Artificial intelligence as cognitive support in emergency medicine. Clin Exp Emerg Med 2026;13:254-7.
  • 40. Collins GS, Moons KG, Dhiman P, et al. TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ 2024;385:e078378.

Download Citation

Download a citation file in RIS format that can be imported by all major citation management software, including EndNote, ProCite, RefWorks, and Reference Manager.

Format:

Include:

Artificial intelligence in sepsis care and research: aligning clinical questions with model selection and human oversight
Download Citation

Download a citation file in RIS format that can be imported by all major citation management software, including EndNote, ProCite, RefWorks, and Reference Manager.

Format:
Include:
Artificial intelligence in sepsis care and research: aligning clinical questions with model selection and human oversight
Close
Artificial intelligence in sepsis care and research: aligning clinical questions with model selection and human oversight
Artificial intelligence in sepsis care and research: aligning clinical questions with model selection and human oversight
Clinical or research task Typical data and timing Target output or AI role in sepsis Candidate model families Key methodological and AI-related cautions
Screening and diagnosis Demographics, comorbidities, vital signs, laboratory results, triage variables, clinical text, microbiology results, imaging, or ECG; single time point or short observation window Estimated probability of sepsis or infection, clinical alert, and supporting evidence Conventional statistical and interpretable models; tree-based structured-data machine learning models; modality-specific encoders; multimodal fusion models Clinical label and time-zero definition, temporal leakage, class imbalance, false-positive burden, and alert fatigue
Prognostic prediction Serial EHR variables, organ support, treatment response, progress or nursing notes, imaging, ECG, and physiological waveforms; longitudinal or real-time data Horizon-specific or time-to-event risk of septic shock, new or worsening organ failure, ICU transfer, short- or long-term mortality, or recovery Time-to-event and dynamic prediction models; tree-based models using trajectory features; temporal deep learning models; multimodal fusion models Irregular sampling, missingness, censoring, competing events, treatment-dependent variables, time-varying decisions, and calibration over time
Treatment and decision support Longitudinal hemodynamics, laboratory results, organ support, treatment history, clinical notes, microbiology or radiology findings, and local protocols; updated over time according to treatment exposure and response Risk-based prompts, estimated treatment effects, prioritized options, or source-linked recommendations for fluids, vasopressors, corticosteroids, antibiotics, source control, organ support, or ICU transfer Causal inference and treatment effect models for intervention-related questions; prediction models for risk estimation only; LLM/RAG-based systems for source-linked presentation Risk prediction does not establish treatment benefit; confounding by indication, immortal time bias, protocol alignment, uncertainty, and clinician verification
Phenotyping and heterogeneity EHR variables, laboratory results, biomarkers, treatment exposure, outcomes, omics-derived high-dimensional data, and optional multimodal data; baseline profiles or longitudinal cohort data Phenotype, cluster, latent representation, potential responder subgroup, or trial-eligibility assessment Phenotyping and representation learning models; multimodal fusion models; causal inference and treatment effect models for treatment effect heterogeneity Sensitivity to preprocessing and model selection, cluster instability, limited reproducibility and transportability, biological plausibility, and clinical actionability
Documentation and communication Structured EHR data combined with clinical notes, microbiology or radiology reports, retrieved guidelines, and local protocols; encounter-level, handoff-level, hospital-course, or discharge-level information Source-linked summary, handoff draft, explanation, guideline summary, or draft recommendation for clinician review Generative AI and LLM/RAG-based systems; structured-to-text systems; modality-specific encoders and multimodal fusion models when source data are diverse Hallucination, outdated or irrelevant retrieval, weak source fidelity, overconfident language, automation bias, and required clinician verification
Model family Model or approach example Suitable data and task Main strength Key consideration in sepsis Explainability and verification approach
Conventional statistical and interpretable models Linear regression, logistic regression, penalized regression, GAM, EBM Tabular EHR data; screening, diagnosis, and initial risk prediction Transparent and clinically familiar; relatively easy to implement; low computational burden Limited ability to capture complex temporal, high-dimensional, or multimodal relationships; model assumptions require checking Coefficients, odds ratios, score components, nomograms, shape functions, and risk curves
Tree-based structured-data machine learning models Random forest, XGBoost, LightGBM, CatBoost Tabular EHR data; single-time-point prediction or summarized longitudinal data Captures nonlinear relationships and interactions; often provides a strong baseline for structured clinical data Less transparent; may learn institution-specific missingness or practice patterns; feature attribution should not be interpreted as evidence of causality Permutation importance, SHAP, LIME, partial dependence plots, and ICE plots
Time-to-event and dynamic prediction models Cox models, landmarking models, joint models, competing-risk models Longitudinal structured data; mortality, organ failure, ICU transfer, recovery, and other time-to-event outcomes Accounts for event timing, censoring, and updated patient information; provides horizon-specific risk estimates Requires a clear time origin, outcome definition, censoring strategy, and approach to competing events and treatment-dependent variables Survival curves, dynamic risk curves, landmark-specific estimates, and time-dependent effects
Temporal deep learning models RNN, GRU, LSTM, TCN, temporal transformer Serial EHR data and physiological waveforms; dynamic prognosis or real-time monitoring Captures complex trajectories and temporal interactions Requires relatively large datasets and computational resources; sensitive to irregular sampling, missingness, domain shift, and latency; model compression may be needed for real-time deployment Trajectory visualization, temporal attribution, occlusion analysis, saliency maps, and cautious use of attention visualization
Modality-specific encoders Clinical language models, CNN, vision transformer, pretrained or foundation-model-based encoders, one-dimensional CNN, signal transformer Clinical text, imaging, ECG, and physiological waveforms Extracts information from high-dimensional modality-specific data; pretrained or self-supervised models may reduce annotation requirements Sensitive to documentation bias, image or signal quality, artifacts, synchronization, domain shift, and computational burden Text or token attribution, class activation maps, image occlusion, waveform-segment attribution, and robustness checks
Multimodal fusion models Early fusion, late fusion, cross-attention models, multimodal transformers Combinations of EHR data, text, imaging, physiological signals, biomarkers, or omics data Integrates complementary information from multiple data sources Missing modalities, temporal alignment, heterogeneous data quality, modality dominance, model opacity, and computational burden Modality ablation, modality-specific attribution, cross-modal contribution displays, and representative case comparisons
Generative AI and RAG-based systems RAG-based LLMs, guideline-constrained LLMs, structured-to-text LLMs Structured, unstructured, or multimodal clinical data; extraction, summarization, explanation, documentation support, and draft recommendations for clinician review Supports information extraction, summarization, documentation, and source-linked communication Hallucination, retrieval failure or outdated retrieval, weak source fidelity, overconfident language, cost, and inference latency Retrieved-source display, citation and version tracking, audit trails, uncertainty display, and clinician verification
Causal inference and treatment effect models Target trial emulation, propensity score methods, marginal structural models, causal forests, uplift models, dynamic treatment regimes Longitudinal treatment and outcome data; average or individualized treatment effect estimation More appropriate than risk prediction models alone for questions about intervention benefit Confounding by indication, limited overlap, immortal time bias, unmeasured confounding, and difficulty evaluating learned treatment policies Causal diagrams, balance and overlap checks, subgroup treatment-effect plots, and sensitivity analyses
Phenotyping and representation learning models Clustering, latent class analysis, autoencoders, contrastive learning, multimodal representation learning Structured EHR, biomarkers, omics, or multimodal data; phenotype, cluster, latent representation, or responder subgroup identification Useful for studying sepsis heterogeneity, trial enrichment, and hypothesis generation Sensitive to preprocessing and model choice; clusters may be unstable, poorly reproducible, or difficult to interpret clinically Phenotype profiles, cluster heatmaps, latent-space visualization, representative cases, and stability analyses
Clinical implementation area in sepsis AI role Human-in-the-loop design Potential risk without human oversight
Screening and diagnosis Flags possible sepsis, estimates infection probability, highlights concerning patterns, and summarizes supporting evidence The clinician reviews the relevant input data, supporting evidence, and uncertainty; confirms whether the alert is clinically meaningful; and decides whether additional evaluation or treatment is needed False-positive alerts may lead to unnecessary evaluation or treatment, whereas poorly contextualized alerts can increase alert fatigue
Prognostic prediction Tracks the risk of shock, organ failure, ICU transfer, death, or recovery and identifies worsening or improving risk patterns The clinician interprets the estimated risk in the context of bedside assessment, treatment response, comorbidities, goals of care, and information not captured by the model Risk estimates may be treated as deterministic or may influence subsequent care, potentially reinforcing self-fulfilling predictions
Treatment and decision support Suggests or prioritizes possible actions, such as fluids, vasopressor escalation, corticosteroid consideration, antibiotic optimization, source control, or ICU transfer, without autonomously determining treatment The clinician verifies the indication, contraindications, prior treatment response, bedside findings, and goals of care, and then approves, modifies, or overrides the recommendation Recommendations may overlook contraindications, prior treatment response, bedside findings, or goals of care and may be followed despite uncertain treatment benefit
LLM-assisted documentation and communication Generates source-linked chart summaries, handoff drafts, recommendation rationales, trial-screening summaries, or patient-specific guideline summaries The clinician checks the underlying source, source version, patient-specific accuracy, and final wording and remains responsible for the final documentation or communication Hallucinated, outdated, or weakly sourced information may enter clinical documentation or communication without verification
Phenotyping for clinical stratification Suggests clinically meaningful subgroups, such as high-risk organ dysfunction patterns, host response patterns, or potential treatment response profiles Clinicians review whether the suggested phenotype is consistent with bedside findings, infection source, organ dysfunction, and available biomarkers before using it to guide monitoring intensity, consultation, or trial referral Unstable or poorly interpretable phenotypes may be treated as clinically meaningful subgroups without sufficient bedside or biological support
Research stage Human role and key decisions Why oversight matters
Clinical or research question and intended use Define the clinical or research question, target population, care setting, prediction timing, target outcome, and intended action before model development This prevents AI-first framing before a clinically meaningful question has been defined; unclear time zero, target population, or outcome definitions reduce interpretability
Ethical, privacy, and accountability considerations Confirm IRB approval, privacy protection, deidentification, secure data storage, data access control, accountability, and audit trails Privacy breaches, re-identification risk, inappropriate data access, unclear responsibility, and lack of auditability can limit safe and trustworthy AI research
Data source, quality, and timing Define the data source, collection period, variable definitions, missing data handling, measurement units, timestamps, and data availability at the intended prediction time Clinical workflows determine what is measured and recorded; workflow-driven missingness, inconsistent timing, or postprediction information may produce misleading associations or apparent performance
Model development and performance evaluation Select the model family, preprocessing strategy, train/validation/test split, tuning approach, internal validation, and performance metrics appropriate to the intended use AUROC or AUPRC alone is insufficient; poor calibration, class imbalance, overfitting, and lack of subgroup evaluation can limit clinical usefulness; flexible preprocessing, feature construction, and model tuning may obscure bias or overfitting if their clinical and temporal implications are not reviewed
External and temporal validation Evaluate model performance across different time periods, institutions, patient groups, care settings, and practice environments when feasible Good internal performance does not guarantee transportability across institutions, populations, or practice patterns
Clinical interpretation and transparent reporting Clinicians and AI researchers jointly assess clinical plausibility, explainability, actionability, and reporting completeness Plausible explanations do not establish causality or clinical correctness; novel predictors and phenotypes require clinical interpretation and reproducibility assessment
Clinical implementation and postdeployment monitoring Define workflow integration, user interface, human review process, prospective evaluation, performance drift monitoring, update policy, and model-withdrawal criteria A technically accurate model may fail if it is not actionable, poorly integrated into workflow, or not monitored after deployment; deployment may change clinical behavior and the data-generating process; performance drift, subgroup errors, excessive alert burden, unintended workflow effects, and safety signals may therefore emerge over time
Table 1. Clinical and research tasks, AI roles, and candidate model families in sepsis

Many cautions listed for screening, prognosis, treatment, and phenotyping reflect established methodological concerns in sepsis research that may be inherited, amplified, or obscured by AI. Hallucinations are associated with generative AI, whereas retrieval failure and weak source fidelity are particularly relevant to RAG-based systems.

AI, artificial intelligence; ECG, electrocardiography; EHR, electronic health record; ICU, intensive care unit; LLM, large language model; RAG, retrieval-augmented generation.

Table 2. Characteristics, precautions, and explainability/verification approaches for AI model families relevant to sepsis

Tabular EHR data refer to structured, row-by-column clinical data such as demographics, vital signs, laboratory results, medications, organ support, timestamps, and outcomes. Cox models include conventional Cox proportional hazards models with fixed covariates and time-dependent Cox models with covariates updated over time.

AI, artificial intelligence; GAM, generalized additive model; EBM, explainable boosting machine; EHR, electronic health record; XGBoost, Extreme Gradient Boosting; LightGBM, Light Gradient Boosting Machine; CatBoost, Categorical Boosting; SHAP, Shapley Additive Explanations; LIME, Local Interpretable Model-agnostic Explanations; ICE, individual conditional expectation; ICU, intensive care unit; RNN, recurrent neural network; GRU, gated recurrent unit; LSTM, long short-term memory; TCN, temporal convolutional network; CNN, convolutional neural network; ECG, electrocardiography; RAG, retrieval-augmented generation; LLM, large language model;

Table 3. Human-in-the-loop design for clinical AI implementation in sepsis care

AI, artificial intelligence; ICU, intensive care unit; LLM, large language model.

Table 4. Key human oversight considerations in sepsis AI model development, validation, and deployment

AI, artificial intelligence; IRB, institutional review board; AUROC, area under the receiver operating characteristic curve; AUPRC, area under the precision-recall curve.