Abstract
OBJECTIVE: To develop and validate a mortality-risk prediction model for patients with advanced splenomegaly-subtype schistosomiasis, enabling accurate prognosis assessment and informed resource allocation. DESIGN: Retrospective, population-based cohort study using clinical data from a single-centre registry. SETTING: Endemic regions of Hubei Province, China. PARTICIPANTS: The study includes 628 patients with advanced splenomegaly-subtype schistosomiasis from the Hubei Provincial Advanced Schistosomiasis Registry between September 2014 and January 2015. The splenomegaly subtype is defined as splenomegaly extending below the umbilical line or with a transverse diameter exceeding the mid-abdominal line. We divided the study population into two cohorts. The derivation cohort included 452 patients selected from several counties within the registry. These patients had a confirmed diagnosis of advanced splenomegaly-subtype schistosomiasis. Only those with complete data were retained. The external validation cohort comprised 176 patients from geographically distinct counties in the same registry, and the same inclusion and exclusion criteria were applied. 10-fold cross-validation was employed to evaluate the model's generalisation on the derivation cohort. OUTCOME MEASURES: 6-year all-cause mortality was the outcome measure. Baseline variables, included age, serum aspartate aminotransferase, albumin, splenectomy history and frequency of ascites ≥5 episodes were analysed using Cox proportional hazards regression. Model performance was assessed via C-statistics, net reclassification improvement (NRI) and integrated discrimination improvement (IDI). RESULTS: Five predictors were integrated into the mortality risk model. The C-statistic for predicting 6-year mortality was 0.79 (95% CI 0.74 to 0.83) in the derivation cohort, with robust validation in the external validation cohort (0.78, 95% CI 0.70 to 0.86). Simplified models using subsets of predictors showed slight reductions in discrimination (NRI and IDI). Patients with a frequency of ascites ≥5 episodes had significantly lower survival rates and should be given special attention in clinical practice. CONCLUSIONS: The validated prediction model identifies high-risk patients using accessible clinical variables. The approach may optimise prognostication and prioritisation of healthcare resources for advanced schistosomiasis. The model's performance remains to be confirmed in a prospectively enrolled cohort.