SHM Converge Scientific Abstract ePoster Gallery-Machine Learning-Enhanced Early Warning Scores for Predicting Deterioration in Hospitalized Oncology Patients: A Systematic Review for Hospital Medicine Practice

Machine Learning-Enhanced Early Warning Scores for Predicting Deterioration in Hospitalized Oncology Patients: A Systematic Review for Hospital Medicine Practice

Back to course

Pdf Summary

This systematic review compared Early Warning Scores (EWS) and machine-learning (ML) models for predicting clinical deterioration in hospitalized adult oncology patients. The included studies focused on hematologic malignancies, solid tumors, and mixed oncology wards, and assessed outcomes such as unplanned ICU transfer, in-hospital cardiac arrest, rapid response activation, and in-hospital mortality. Because the studies were highly heterogeneous in outcomes, predictors, and modeling methods, no meta-analysis was performed.<br /><br />Nine studies published between 2012 and 2025 were included: five evaluated traditional EWS and four evaluated ML models. Traditional scores such as MEWS, NEWS, SEWS, and MEWS-SF showed modest discrimination overall, with AUROCs ranging from 0.55 to 0.83. Performance was often lower in hematologic malignancy populations, where patients may deteriorate without obvious vital-sign abnormalities. Enhanced EWS that added markers such as SpO/FiO or lactate performed better, reaching an AUROC of 0.87.<br /><br />ML approaches, including neural networks, gradient boosting, and deep-learning models, showed the best performance, with AUROCs ranging from 0.89 to 0.95. These models likely benefit from incorporating multidimensional clinical data and temporal trends. However, most studies relied on internal validation, and only two included external validation cohorts.<br /><br />Overall, the review suggests that standard EWS may underestimate deterioration risk in oncology inpatients, while ML models offer promising improvement in early detection. Still, broader clinical use will require stronger external validation, better interpretability, and integration into electronic health record workflows.

Asset Subtitle

Mansha Gupta

Meta Tag

Author List Mansha Gupta, Naman Suroya, Shrenil Kavathia, Swathi Cherukuri

Category Research

Concept Early Warning Score

Concept Clinical Deterioration

Concept Hospitalized Cancer Patient

Concept External Validation

Concept Machine Learning Model

Distinguished Non-Finalist

Presenter Organization Midwestern University

Presenting Author Mansha Gupta

Track Technology in Hospital Medicine

Keywords

early warning scores

machine learning models

oncology patients

clinical deterioration

hematologic malignancies

solid tumors

unplanned ICU transfer

in-hospital mortality

AUROC

external validation

Early Warning Score

Clinical Deterioration

Hospitalized Cancer Patient

External Validation

Machine Learning Model