Development and validation of a dementia risk prediction model for low- and middle-income countries: the 10/66 study

BACKGROUND: Most people with dementia live in LMICs, underscoring the need for LMIC-specific identification of high-risk individuals. This study aimed to develop and validate a simple dementia risk prediction model for these settings. METHODS: Data from seven 10/66 Study sites were analyzed. Over 100 candidate predictors were screened based on existing models and the 2024 Lancet Commission, including LMIC-specific variables (eg, food insecurity and household assets). Predictors were selected using LASSO and modelled with the Fine-Gray method to generate a risk score. Predictive accuracy was pooled via meta-analysis. RESULTS: 11143 participants were included, among whom 1069 (9.6%) developed dementia during follow-up. A five-factor risk score comprising age, social engagement, physical activity, hypertension, and difficulty in handling money was developed. The pooled c-statistic was 0.75 (95% CI: 0.72-0.78), with good calibration across sites. Decision curve analysis showed a modest net benefit, with variation across countries. CONCLUSION: It is possible to predict incident dementia with reasonable accuracy using a simple model across different LMICs. Our findings support the use of context-specific risk assessment tools to identify individuals at elevated dementia risk in LMIC settings, which may inform resource allocation for dementia care services and public health planning.