LAUSR

BACKGROUND Numerous risk prediction models use indicators of health to predict mortality in old age. The added value to mortality predictions based on demographic variables is unknown. OBJECTIVE To evaluate the accuracy of health indicators in predicting all-cause mortality among individuals aged 50+ using area under receiver operating characteristic curve (AUC). Specifically, to assess the added value of health indicators relative to demographic variables. METHODS We conducted a systematic review following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses. There were no restrictions on study designs, follow-up duration, language, or publication dates. We also examined the quality of studies using the Checklist for Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies. RESULTS Out of 804 studies investigating all-cause mortality in older persons, 16 studies were eligible. In community-dwelling populations, the accuracy of demographic variables and health indicators combined ranged from AUC 0.71 to 0.82, indicating modest ability to predict mortality. Age contributed the most to mortality prediction (AUC 0.65 to 0.78) and compared to age and sex, the added values of genetics, physiology, functioning, mood, cognition, nutritional status, subjective health, disease, frailty, and lifestyle ranged from AUC 0.01 to 0.10. The lack of validation samples made it difficult to assess their true added value. Findings were similar in institutionalized populations. Heterogeneity of the studies prevented us from performing a meta-analysis. CONCLUSION Age and sex contributed the most to mortality predictions in old age while the added value of health indicators is likely to be limited.