Development and validation of clinically deployable ischemic stroke risk prediction model incorporating cardiometabolic and geriatric comorbidities in Ugandan older adults

Abstract

Ischemic stroke(IS) is major cause of disability and death among older adults, yet geriatric-specific risk prediction tools are limited in low-resource settings. This study identified cardiometabolic and geriatric risk factors and developed a nomogram to estimate individualized IS risk. Overall, 210 participants(age, 74.6±9.1yrs; 61.4% women; 60.5% with caregiver) receiving care in public and private geriatric facilities in Uganda from July 30, 2023, to July 30, 2025 were retrospectively enrolled. The primary outcome was incident IS. Univariable and multivariable logistic regression identified factors associated with IS risk. A predictive model was developed, internally validated using bootstrap resampling, and evaluated for discrimination and calibration. Prevalent conditions included hypertension(41.0%), type-2-diabetes mellitus(T2DM) (26.2%), Coronary Heart disease(CHD)(24.8%), and malnutrition(25.2%). During follow-up, IS occurred in 32(15%) of the participants. IS was independently associated with physiotherapy (OR 8.49; 95%CI 1.874-18.475; P=0.006), hypertension(OR 6.52; 95%CI 1.402-14.333; P=0.017), and CHD(OR 3.70; 95%CI 2.435-19.604; P<0.001). Dementia(OR 1.65; 95%CI 0.275-9.906; P=0.182) and gender(OR 3.48; 95%CI 0.846-14.340; P=0.084) were retained for model development using liberal threshold(P<0.200) to avoid excluding potentially important confounders. The model demonstrated excellent calibration, with bias-corrected curves near ideal and mean absolute error 0.025. Among older adults in Uganda, we found that HTN, CHD, and physiotherapy were associated with IS risk. A five-variable nomogram incorporating HTN, CHD, physiotherapy, gender, and dementia demonstrated good calibration and may provide a practical tool for individualized risk stratification in resource-limited settings. External validation in larger, diverse populations with comprehensive assessment of relevant variables is warranted to validate these findings and confirm the model’s performance and generalizability.