Immune-inflammatory biomarkers for in-hospital mortality risk stratification in older patients with sepsis: a multicentre leave-one-hospital-out validation study

Abstract

Older adults with sepsis have high mortality, but whether circulating immune-inflammatory biomarkers improve prediction beyond routine clinical information across hospitals is uncertain. We analysed 816 patients aged ≥ 60 years from a prospectively assembled multicentre cohort at four tertiary hospitals; 225 (27.6%) died in hospital. Five prespecified models were evaluated using leave-one-hospital-out internal–external validation. Routine-clinical extreme gradient boosting achieved an area under the receiver operating characteristic curve (AUROC) of 0.688, an area under the precision–recall curve (AUPRC) of 0.453, and a Brier score of 0.186. Adding 19 immune-inflammatory biomarkers yielded an AUROC of 0.684, an AUPRC of 0.468, and a Brier score of 0.184. Compared with the routine-clinical model, differences were small and imprecise (ΔAUROC, − 0.003; ΔAUPRC, 0.015; Brier-score improvement, 0.002), and decision-curve benefit was inconsistent. The immune-enhanced model was poorly calibrated (slope, 0.751; intercept, − 0.189). Findings were similar in patients aged ≥ 65 years, but the incremental signal was not maintained after excluding SCHX, the smallest and highest-mortality centre. These data do not support clinical use of the full biomarker panel for mortality prediction.