Exposome‐wide association study and predictive risk score for incident Alzheimer's disease and related dementias

Figures

• 

  FIGURE 1. Analytic workflow of the study. In Step 1, we leveraged exposomic data (193 exposures) from 499,992 participants free of ADRD at baseline. Exposures were categorized into social, physical, lifestyle, ecosystems, and chemical domains. During a median follow‐up of 15.33 years, 8881 participants were diagnosed with ADRD, and 4000 participants were diagnosed with AD. In Step 2, unbiased exposome‐wide association analysis was conducted to identify ADRD (or AD)‐related exposures using Cox proportional hazard models. In Step 3, we explored the application of identified exposures predicting future ADRD (or AD) risk, investigating exposome variable interactions, and developing ERSs for population risk stratification. AD, Alzheimer's disease; ADRD, Alzheimer's disease and related dementias; ERS, exposomic risk score.

• 

  FIGURE 2. Associations between exposures and ADRD and AD risk. (A, C) Circular plots showing exposures in relation to incident ADRD (A) and AD (C). Each bar represents a significant exposure factor (p < 0.05) from Cox proportional hazards regression models. The radial distance indicates the strength of association as −log10(p value). The red dashed circle represents the Bonferroni‐corrected significance threshold (p < 2.59 × 10−4). Exposures are grouped by category and color‐coded: social, physical, lifestyle, ecosystems, and chemical. (B, D) The adjusted hazard ratios for Bonferroni‐corrected significant exposures for ADRD (B) and AD (D), in ascending order (from risk‐increasing to risk‐reducing). Dots represent hazard ratios, horizontal lines indicate corresponding 95% CIs. Hazard ratios were calculated using Cox proportional hazards regression analysis after adjustments for baseline age, sex, ethnicity, APOE status, overall health rating, and assessment center. AD, Alzheimer's disease; ADRD, Alzheimer's disease and related dementias; APOE, apolipoprotein E; CI, confidence interval.

• 

  FIGURE 3. Summary heat map for significant factors in ExWAS analysis across the full sample and subgroups. (A) ADRD. (B) AD. The color of cells indicates the effect sizes (HR). HRs were derived from Cox proportional hazards regression models adjusted for baseline age, sex, ethnicity, APOE status, overall health rating, and assessment center. Asterisks represent significant associations after Bonferroni correction (P < 2.59 × 10−4). For AD in other ethnicities, no exposures remained significant after Bonferroni correction and were therefore not displayed. AD, Alzheimer's disease; ADRD, Alzheimer's disease and related dementias; APOE, apolipoprotein E; ExWAS, exposome‐wide association study; HR, hazard ratio.

• 

  FIGURE 4. Exposome‐based predictive models for ADRD and AD risk. (A, D) ROC curves for ADRD and AD prediction models. NRI and IDI comparing models without and with exposures. (B, E) ADRD and AD prediction model interpretability through SHAP analysis. The left panel shows mean absolute SHAP values for each feature. The right panel (beeswarm plot) displays SHAP values for individual predictions. (C, F) Performance of exposome‐based predictive models in subgroups. AD, Alzheimer's disease; ADRD, Alzheimer's disease and related dementias; IDI, Integrated Discrimination Improvement; NRI, Net Reclassification Index; ROC, Receiver Operating Characteristic; SHAP, SHapley Additive exPlanations.

• 

  FIGURE 5. SHAP interaction effects for exposome‐based predictive models. (A) ADRD. (B) AD. Left panels show the top 10 exposome variable interaction pairs ranked by mean absolute SHAP interaction values. The right panels show SHAP dependence plots for top interaction pairs, revealing how feature combinations jointly influence risk predictions. AD, Alzheimer's disease; ADRD, Alzheimer's disease and related dementias; SHAP, SHapley Additive exPlanations.

• 

  FIGURE 6. Kaplan–Meier survival curves stratified by ERS quartiles. (A, C) Kaplan–Meier curves for ADRD and AD in the total population. (B, D) Subgroup‐specific survival curves stratified by sex, age, APOE status, and ethnicity. Participants were divided into ERS quartiles: Q1 (low risk), Q2 (mid‐low), Q3 (mid‐high), and Q4 (high risk). Shaded areas represent 95% CIs. AD, Alzheimer's disease, ADRD, Alzheimer's disease and related dementias; APOE, apolipoprotein E; CI, confidence interval; ERS, exposomic risk score.