Prediction models for overall survival and all-cause mortality risk in older adults with cancer: a systematic review.

Abstract

Mortality risk prediction models can support decision making in older adults with cancer; however, existing models are associated with a high risk of bias. This systematic review assessed published prediction models for overall and all-cause mortality in adults with cancer aged 65 years or older. We searched for publications in Ovid Embase, Ovid Medline, Cochrane CENTRAL, and EBSCO CINAHL on Nov 25, 2022, and updated the search on Feb 24, 2024. We included 250 studies, of which 182 (72·8%) reported both model development and internal validation. 176 (70·4%) of 250 models predicted overall survival; 40 (16·0%) models focused on lung cancer and 30 (12·0%) models on colorectal cancer. 43 (17·2%) models were specifically developed for older adults; 138 (55·2%) models did not incorporate geriatric variables such as comorbidities, nutrition, and cognition. Risk of bias was high in all models, largely owing to inappropriate handling of continuous predictors, univariable selection of predictors, and inadequate control for overfitting. These limitations preclude clinical use. Future models predicting overall and all-cause mortality in older adults with cancer should adhere to existing methodological guidelines and incorporate geriatric domains.