TY - JOUR
T1 - Predicting all-cause mortality with machine learning among Brazilians aged 50 and over
T2 - results from The Brazilian Longitudinal Study of Ageing (ELSI-Brazil)
AU - Delpino, Felipe Mendes
AU - Chiavegatto Filho, Alexandre Dias Porto
AU - Torres, Juliana Lustosa
AU - Bof de Andrade, Fabíola
AU - Lima-Costa, Maria Fernanda
AU - Nunes, Bruno Pereira
N1 - F.M.D. received a postdoctoral fellowship from the National Council for Scientific and Technological Development. B.P.N. received a Research Productivity Fellow (process number: 308772/2022-9) - Level 2 from the National Council for Scientific and Technological Development. ELSI-Brazil was supported by the Brazilian Ministry of Health: DECIT/SCTIE \u2013 Department of Science and Technology from the Secretariat of Science, Technology and Strategic Inputs (Grants: 404965/2012-1 and TED 28/2017); COPID/DECIV/SAPS \u2013 Health Coordination of the Older Person in Primary Care, Department of Life Course from the Secretariat of Primary Health Care (Grants: 20836, 22566, 23700, 25560, 25552, and 27510). This study was also supported by the Conselho Nacional de Desenvolvimento Cientifico e Tecnologico \u2013 CNPq\u201D \u201CNational Council for Scientific and Technological Development \u2013 CNPq and Funda\u00E7\u00E3o de Amparo \u00E0 Pesquisa do Estado do Rio Grande do Sul - FAPERGS.
PY - 2025/3/28
Y1 - 2025/3/28
N2 - We aimed to develop a machine-learning model to predict all-cause mortality among Brazilians aged 50 and over, incorporating demographic, health, and lifestyle characteristics as predictors. We analyzed data from the Brazilian Longitudinal Study of Aging (ELSI-Brazil), waves 1 and 2 (2015–2021), a nationally representative sample from 70 municipalities across Brazil’s five regions. Nine algorithms, including Random Forest, Gradient Boosting, XGBOOST, and Logistic Regression, were tested on 9412 participants (54.6% female), with 970 deaths recorded over approximately five years. Using 59 predictor variables, we assessed performance with metrics like AUC, accuracy, precision, and F1-Score. Random Forest excelled with an AUC of 0.92 (95% CI: 0.90–0.94). SHAP analysis highlighted age, sex, BMI, medication use, and physical activity as top predictors. Integrating these models into healthcare systems can improve policy planning and enable targeted interventions, ultimately fostering better health outcomes for aging populations.
AB - We aimed to develop a machine-learning model to predict all-cause mortality among Brazilians aged 50 and over, incorporating demographic, health, and lifestyle characteristics as predictors. We analyzed data from the Brazilian Longitudinal Study of Aging (ELSI-Brazil), waves 1 and 2 (2015–2021), a nationally representative sample from 70 municipalities across Brazil’s five regions. Nine algorithms, including Random Forest, Gradient Boosting, XGBOOST, and Logistic Regression, were tested on 9412 participants (54.6% female), with 970 deaths recorded over approximately five years. Using 59 predictor variables, we assessed performance with metrics like AUC, accuracy, precision, and F1-Score. Random Forest excelled with an AUC of 0.92 (95% CI: 0.90–0.94). SHAP analysis highlighted age, sex, BMI, medication use, and physical activity as top predictors. Integrating these models into healthcare systems can improve policy planning and enable targeted interventions, ultimately fostering better health outcomes for aging populations.
UR - http://www.scopus.com/inward/record.url?scp=105001427257&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=105001427257&partnerID=8YFLogxK
U2 - 10.1038/s41514-025-00210-7
DO - 10.1038/s41514-025-00210-7
M3 - Article
C2 - 40155388
AN - SCOPUS:105001427257
SN - 2056-3973
VL - 11
JO - npj Aging
JF - npj Aging
IS - 1
M1 - 22
ER -