Admission-time prediction of 7-day mortality in clinician-attributed hypertensive intracerebral hemorrhage: development and external validation against the ICH score

Early mortality after intracerebral haemorrhage remains difficult to estimate at admission. We developed models for recorded mortality within 7 days after admission in two retrospective cohorts of clinician-attributed hypertensive intracerebral haemorrhage. Qingyuan records ( n = 1,433; 369 events) were used for grouped nested cross-validation, and Yunfu records ( n = 924; 182 events) for external validation. The models used 28 admission-time predictors. Predictions were averaged across 20 imputation-specific pipelines. Elastic-net logistic regression was selected from the development data. Its development and external areas under the receiver operating characteristic curve were 0.848 and 0.893, respectively. In external validation, the ICH score had an area under the curve of 0.860, giving a difference of 0.0324 (95% CI 0.0126–0.0523; DeLong P = 0.00135). The external Brier score was 0.103, with a calibration intercept of − 0.504 and slope of 1.149. At the development-selected threshold of 0.1976, sensitivity was 0.901 and specificity was 0.706. The model modestly improved discrimination over the ICH score, but overprediction, differences between centres, and unavailable treatment and outcome-ascertainment details limit clinical interpretation.