Data and code for: Prognostic models for spontaneous brainstem haemorrhage — a systematic review and meta-analysis of discrimination, with PROBAST appraisal

This deposit contains the complete analysis underlying a systematic review and meta-analysis of prognostic models for spontaneous brainstem haemorrhage (PROSPERO CRD420261528626; reported in accordance with PRISMA 2020).Forty-seven multivariable prognostic models from 32 studies (5,462 patients) were included. C-statistics were pooled after Fisher-z transformation using random-effects models fitted by REML, with cluster-robust variance estimation (CR2 small-sample correction) clustering by study, because several models within a study share participants. Pooled C was 0.879 (95% CI 0.831-0.915) for 30-day mortality, 0.895 (0.826-0.938) for functional outcome and 0.841 (0.797-0.877) for mortality at other timepoints. No model met PROBAST criteria for low risk of bias (24 high risk, 23 unclear; all concerns arose in the analysis domain), and certainty of evidence assessed with GRADE was very low for both main outcomes.Contents:- the full model-level extraction table (49 models, 6 sheets) with a data dictionary;- the two analysis-ready datasets (main analysis, 47 models; subgroups and sensitivity analyses, 49 models) in tab-separated format;- nine R scripts that reproduce every reported number and figure;- console logs recording all pooled estimates, subgroup and sensitivity analyses, small-study tests and prediction intervals;- the search strategies for all seven databases, the model-by-model PROBAST judgements, the record of C-statistics reconstructed from published contingency tables, the validation of the Hanley-McNeil standard-error approximation, and the completed PRISMA 2020 checklist.Analysis was performed in R 4.6.1 with metafor 5.2.1 and clubSandwich 0.7.0. Running "Rscript code/00_run_all.R" from the deposit root regenerates every reported value to three decimal places and rebuilds all figures.