Global performance of machine learning models to predict all-cause mortality: systematic review and meta-analysis

Q: How was the systematic review conducted?

The study followed PRISMA recommendations and used a protocol registered in PROSPERO (CRD42023476567). Searches were performed in PubMed, LILACS, Web of Science, and Scopus without restrictions on study year, country, or language.

Q: What was the overall predictive performance of the machine learning models?

Across 88 included studies, the pooled AUC was 0.831 (95% CI 0.797–0.865). The results also showed extreme heterogeneity (I2: 100%).

Q: What limitations and equity-related gaps were identified?

Most models included no social variables (89.8%), equity-oriented subgroup analyses were scarce (<10%), and external validation on other datasets was limited (8.0%). The authors note that context dependence and extreme heterogeneity require local validation before implementation, and that the limited evidence may perpetuate social disparities.

Aug 3, 2026

Upload by Aurelia

A systematic review evaluated how machine learning models predict all-cause mortality across global populations. The protocol was registered in PROSPERO and followed PRISMA guidance, using searches in PubMed, LILACS, Web of Science, and Scopus. Eighty-eight studies were synthesized with random-effects models, heterogeneity assessed by I2 and quality evaluated using TRIPOD + AI. The pooled performance showed an overall AUC of 0.831, with extreme heterogeneity and limited use of social variables, scarce equity-focused subgrouping, and few external validations. Findings indicate strong predictive capability but context dependence requiring local validation, while also highlighting potential inequity perpetuation in public health implementation.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages35

Reading Time1 h 28 min

AccessFree to download

Methods

Search strategy

Results

Study characteristics and performance

Subgroup analyses and meta-regression

Discussion

[www. nature.com/scientificreports](www. nature.com/scientificreports) OPEN Global performance of machine learning models to predict all-cause mortality: systematic review and meta-analysis Felipe Mendes Delpino1􀀍, Ludmila Pereira Pimenta2, Diego Ferreira Gonzalez3, Audêncio Victor2,4, Cinthia Fonseca Araujo5, Keisyanne De Araujo-Moura2, J. Jaime Miranda6, Sandro Rogério Rodrigues Batista7,8, Alexandre Dias Porto Chiavegatto Filho2 & Bruno Pereira Nunes1,9 We aimed to review the literature on the performance of machine learning models to predict all-cause mortality. The systematic review was protocolled in PROSPERO (CRD42023476567) following PRISMA guidelines. Searches were conducted in PubMed, LILACS, Web of Science, and Scopus databases. Studies predicting all-cause mortality using machine learning were analyzed with random-effects models, with heterogeneity assessed using I2 statistics and quality evaluated using TRIPOD + AI. The meta-analysis included 88 studies. Most of the studies were from the United States (n = 25) and China (n = 20). Overall pooledAUC was 0.831 (95% CI 0.797–0.865), with extreme heterogeneity (I2:100%) . The majority of the studies included no social variables in the models (89.8%). Subgroup analysis showed similar performance between general population studies and disease-specific populations. Models from high-income countries were similar to those from low-and middle-income countries. Meta-regression showed covariates that affected the results: algorithm, population type, study quality score, and study CI imputation. Equity-oriented sub-group analysis (< 10%) and external validation in other datasets (8.0%) were scarce. Overall, machine learning models showed high performance to predict all-cause mortality, but also highlighted equity gaps. The limitations reduce the potential of public health’s evaluation and deployment due to the risk of perpetuation of social disparities. Extreme heterogeneity indicates highly context-dependent performance requiring local validation before implementation assessment. Keywords Machine learning, Mortality, Prediction, Review, Meta-analysis Recent technological advancements due to improvements in computing power have allowed the collection of large volumes of data1,2. In the health area, these innovations present an opportunity to increase the accuracy of outcome predictions3,4, such as all-cause mortality, a complex outcome in public health approaches, especially due to the multifactorial nature of this outcome5–7. The use of predictive models for all-cause mortality could be useful to provide better care for individuals and populations, mainly to offer approaches to prevent premature mortality and events associated with poor disease management, which could lead to avoidable deaths. Machine learning models can offer new perspectives for predicting all-cause mortality8–11 considering its ability to deal with complex relationships among variables. However, the performance of these models can be influenced by the context and the representativeness of the available data. 1Postgraduate Program in Nursing, Federal University of Pelotas, Gomes Carneiro, 01, Pelotas, Rio Grande do Sul, Brazil. 2School of Public Health, University of São Paulo, São Paulo, São Paulo, Brazil. 3Faculty of Medicine, Federal University of Pelotas, Pelotas, Rio Grande do Sul, Brazil. 4Faculty of Epidemiology and Population Health, London School of Hygiene and Tropical Medicine, Keppel Street, London WC1E 7HT, UK. 5Postgraduate Program in Epidemiology, Federal University of Pelotas, Pelotas, Brazil. 6Sydney School of Public Health, Faculty of Medicine and Health, University of Sydney, Sydney, NSW 2006, Australia. 7Faculty of Medicine, Universidade Federal de Goiás, Goiânia, Brazil. 8Postgraduate Program in Medical Sciences, Faculty of Medicine, University of Brasília, Brasília, Brazil. 9Department of Health and Kinesiology, University of Illinois Urbana-Champaign, Urbana, IL, USA.

FAQs

How was the systematic review conducted?

The study followed PRISMA recommendations and used a protocol registered in PROSPERO (CRD42023476567). Searches were performed in PubMed, LILACS, Web of Science, and Scopus without restrictions on study year, country, or language.

What was the overall predictive performance of the machine learning models?

Across 88 included studies, the pooled AUC was 0.831 (95% CI 0.797–0.865). The results also showed extreme heterogeneity (I2: 100%).

What limitations and equity-related gaps were identified?

Most models included no social variables (89.8%), equity-oriented subgroup analyses were scarce (<10%), and external validation on other datasets was limited (8.0%). The authors note that context dependence and extreme heterogeneity require local validation before implementation, and that the limited evidence may perpetuate social disparities.

Aug 3, 2026

Upload by Aurelia