Circadian Rhythm Disruption and In-Hospital Mortality in ICU Patients with Obesity

Objectives Circadian rhythms are frequently disrupted in ICU patients, and obesity may exacerbate this misalignment, worsening outcomes. This study evaluates associations between circadian rhythm parameters of vital signs and in-hospital mortality, assesses whether obesity modifies these relationships, and determines if integrating circadian parameters with obesity status improves predictive model performance. Methods Retrospective observational cohort study using MIMIC-IV database (2008–2019). Included 14 064 adult patients with ≥ 24-hour ICU stays after exclusions for short stays (n = 33 107), incomplete/implausible vital signs (n = 26 010), DNR/DNI orders, or multiple admissions (latest kept). Circadian parameters (MESOR, amplitude, acrophase) derived for heart rate (HR), body temperature (BT), and mean arterial pressure (MAP) using cosinor analysis from first 24-hour vital signs. Associations assessed using logistic regression, overall and BMI-stratified (normal < 25, overweight 25–29.9, class I 30–34.9, class II 35–39.9, class III ≥ 40 kg/m ² ). Machine learning models (logistic regression, random forest, SVM, XGBoost) trained on 70/30 split and 10-fold cross-validation. Results Of 14 064 patients, 1 938 (13.8%) died in-hospital. HR MESOR (OR 1.027, 95% CI 1.024–1.030, p < 0.001), BT amplitude (OR 1.092, 95% CI 1.059–1.127, p < 0.001), MAP amplitude (OR 1.001, 95% CI 1.001–1.002, p < 0.001), class III obesity (OR 1.626, 95% CI 1.404–1.883, p < 0.001) increased mortality risk; MAP MESOR protective (OR 0.998, 95% CI 0.998–0.999, p < 0.001). HR MESOR was significant across BMI groups (ORs 1.021–1.040, p < 0.01 to < 0.001); BT acrophase in class III (OR 1.219, 95% CI 1.041–1.428, p = 0.014). The proposed random forest model achieved the best predictive performance (AUC 0.764; accuracy 85.8%; recall 88.2%; precision 87.1%), outperforming the baseline models. Conclusions Circadian rhythm parameters were significantly associated with in-hospital mortality in ICU patients, with effects modified by obesity status, and their inclusion significantly improved predictive performance.