Skip to content
    Real-World Studies

    Machine Learning Models Predict Persistent Fever in Hospitalized Cancer Patients

    Pucher et alNPJ digital medicine

    Machine Learning Models Predict Persistent Fever in Hospitalized Cancer Patients
    Persistent fever following the initiation of intravenous antibiotics is common among hospitalized patients with cancer, frequently triggering antimicrobial escalation and additional diagnostic imaging. To assist during clinical reassessment, researchers developed and externally validated machine learning models aimed at predicting fever persistence 48 to 72 hours after antibiotic initiation using electronic health record data from two academic hospitals. The cohort comprised adult oncology patients presenting with fever within 24 to 48 hours of initiating intravenous antibiotic treatment.
    The predictive algorithms incorporated patient demographics, comorbidities, laboratory values, vital signs, and longitudinal temperature patterns. The best-performing model achieved an AUROC of 0.82 (95% CI, 0.79–0.84) on internal validation and 0.78 (95% CI, 0.77–0.79) during external validation, with temperature-related parameters emerging as the dominant predictors. In addition, decision curve analysis demonstrated positive net clinical benefit across a broad range of evaluated thresholds.
    A retrospective assessment indicated that approximately one-quarter of antibiotic escalations and chest CT scans were performed in patients who were correctly predicted not to experience persistent fever. These findings indicate that short-term fever persistence can be predicted using routinely collected clinical data; however, the authors specify that the model should function strictly in an adjunctive capacity rather than for standalone rule-out decisions, highlighting the need for prospective investigations to evaluate clinical safety, provider adoption, and real-world outcomes.

    Key Takeaways

    • The best-performing machine learning model predicted persistent fever at 48 to 72 hours with an AUROC of 0.82 (95% CI, 0.79–0.84) internally and 0.78 (95% CI, 0.77–0.79) externally, with temperature-related variables serving as the dominant predictors.
    • Retrospective assessment showed that approximately one-quarter of antibiotic escalations and chest CT scans occurred in patients correctly predicted not to remain febrile.
    • Clinical implementation should be restricted to an adjunctive role rather than standalone rule-out decision-making pending prospective trials evaluating safety, clinician uptake, and real-world impact.
    Read the original sourceNPJ digital medicine