Development and validation of a Cog-Free risk predicting tool for dementia in a community setting.

Background: Dementia poses a growing global public health burden, particularly in low- and middle-income countries where cognitive screening coverage remains limited. Current risk estimation tools often depend on cognitive testing or biomarkers, restricting their applicability in community and primary care settings. Aims: To establish and validate a data-driven analytical framework for developing a cognitive-testing-free dementia risk estimation tool (Cog-Free) using routinely collected health examination data. Methods: For this prospective cohort study, we developed Cog-Free, an internet-based dementia risk estimation tool, using 38 Least Absolute Shrinkage and Selection Operator-selected risk-associated variables and the optimal machine-learning algorithm (logistic regression). The optimal algorithm was internally validated with bootstrap resampling and externally tested in the Chinese Longitudinal Healthy Longevity Survey cohort. The tool was trained and internally validated in 2962 dementia-free adults aged ≥ 65 years (2018-2024), and its performance was compared with three established cognitive-testing-free tools. Results: Cog-Free achieved the highest area under the receiver operating characteristics curve in the internal validation set (0.86 [95% confidence interval (CI) 0.82-0.89]) with an accuracy of 0.81 (95% CI 0.78-0.83), sensitivity 0.78 (95% CI 0.68-0.85) and specificity 0.81 (95% CI 0.78-0.84), significantly outperforming three existing tools (DeLong's test, Conclusions: Cog-Free provides a data-driven, cognitive-testing-free and easily accessible approach for early dementia risk screening using routine health data. Its performance and web-based design suggest potential utility as a pre-screening and risk stratification tool within community health systems, including settings with limited access to cognitive testing.