Risk prediction models for mortality in patients with multimorbidity: a systematic review and meta-analysis.

Background: Multimorbidity is a major aging and public health problem that has a significant burden on a global scale. The number of risk prediction models for mortality in patients with multimorbidity is increasing; however, the quality and applicability of these prediction models in clinical practice and future research remain uncertain. Objective: To systematically review published studies on risk prediction models for mortality in patients with multimorbidity. Methods: The Wanfang, China National Knowledge Infrastructure, China Science and Technology Journal (VIP), PubMed, SinoMed, Cochrane Library, Web of Science, Embase, and Cumulative Index to Nursing and Allied Health Literature databases were searched from inception until May 30, 2024. Two independent reviewers performed study selection, data extraction, and quality assessment. The Prediction Model Risk of Bias Assessment Tool (PROBAST) checklist was utilized to assess the risk of bias and applicability. Results: Overall, 18 studies with 21 prediction models were included in this review. Logistic regression was used for model development in 12 studies, Cox regression in four, a parametric Weibull regression in one, and machine learning in one study. The incidence of mortality in patients with multimorbidity ranged from 7.6-50.0%. The most frequently used predictors were age and body mass index. The reported area under the receiver operating characteristic curve (AUC) and C-index values ranged from 0.700-0.907. Three studies were rated as having a low risk of bias, 11 as high, and four as unclear, primarily owing to poor reporting of the analysis domain. The pooled AUC value of the seven validated models was 0.81, with a 95% confidence interval ranging from 0.77-0.86, signifying a fair level of discrimination. Conclusion: The included studies revealed a degree of discriminatory ability in predicting mortality in patients with multimorbidity; however, they all demonstrated significant risks of bias based on the PROBAST checklist assessment. Future researchers should prioritize the development of new models that incorporate rigorous study designs and multicenter external validation, which may improve the precision of risk predictions and help the development of global strategies for this significant public health problem. Registration: The study protocol was registered in PROSPERO (registration number: CRD42024543170). Systematic review registration: https://www.crd.york.ac.uk/PROSPERO/recorddashboard, PROSPERO CRD42024543170.