Standardized LDH-to-lymphocyte ratio improves early mortality prediction in severe fever with thrombocytopenia syndrome: A 15-day competing-risk bedside model

BACKGROUND: Severe fever with thrombocytopenia syndrome (SFTS) is a tick-borne viral disease associated with a high mortality risk. Early triage is critical, but risk prediction can be biased because many patients are admitted several days after symptom onset and some leave hospital early. METHODOLOGY/PRINCIPAL FINDINGS: We conducted a retrospective single-center cohort study of 459 consecutively hospitalized patients with laboratory-confirmed SFTS. The primary analysis included 392 patients with ascertained in-hospital vital status. Of these, 387 patients with complete admission predictor data formed the derivation cohort for a prespecified 15-day prediction horizon after symptom onset, selected to capture the clinically relevant early high-risk phase of SFTS. Symptom onset was treated as time zero, hospital admission as delayed entry (left truncation), and discharge alive within 15 days as a competing event. We compared the admission-based standardized lactate dehydrogenase-to-lymphocyte ratio (sLLR) with other ratio biomarkers for prediction of 15-day in-hospital death. We then developed a prespecified five-predictor bedside model including age, neurological manifestations, prothrombin time, platelet count, and sLLR. Individualized 15-day death risk was estimated as CIF@15 from cause-specific Cox models. Model performance was assessed by discrimination, calibration, clinical utility, and prediction error, with bootstrap internal validation. Among 387 patients in the derivation cohort, 67 died within 15 days. Admission sLLR showed the best discrimination for 15-day mortality (area under the curve [AUC] 0.797, 95% confidence interval [CI] 0.738-0.855). Using an ROC-derived threshold (sLLR ≥ 2.79), the 15-day cumulative incidence of in-hospital death was 48.2% versus 10.3% in the lower-sLLR group (P < 0.001). The five-predictor model improved discrimination (AUC 0.867, 95% CI 0.824-0.910) compared with the corresponding model without sLLR (P = 0.009), showed good calibration, provided higher net benefit across clinically relevant thresholds, and achieved low prediction error (Brier score at day 15: 0.097). sLLR added prognostic information beyond standard admission variables. A simple five-factor model may provide a practical tool for real-world early risk stratification of hospitalized patients with SFTS, helping clinicians identify those at increased risk of death within 15 days after symptom onset, although external validation in independent multicenter cohorts is still needed.