PROMISE-AD: Progression-aware Multi-horizon Survival Estimation for Alzheimer's Disease Progression and Dynamic Tracking
AI predicts Alzheimer's progression with 99.7% 5-year accuracy, outperforming baselines.
Alzheimer's disease progression prediction is complicated by irregular visit schedules, censored outcomes, and the need for calibrated multi-year risk estimates. To address this, Qing Lyu and colleagues from multiple institutions introduce PROMISE-AD (Progression-aware Multi-horizon Survival Estimation for Alzheimer's Disease). The model converts each pre-index clinical visit into a token comprising standardized measurements, missingness masks, longitudinal changes, time-normalized slopes, visit timing, and non-diagnostic categorical features. A temporal Transformer then fuses global, attention-pooled, and latest-visit representations to estimate a progression score and latent discrete-time mixture hazards. Training combines survival likelihood, horizon-specific focal risk loss, progression ranking, hazard smoothness, and mixture-balance regularization, followed by isotonic calibration for 1-, 2-, 3-, and 5-year risks.
In held-out testing across three seeds, PROMISE-AD achieved an integrated Brier score (IBS) of 0.085 ± 0.012 and C-index of 0.808 ± 0.015 for cognitively normal to mild cognitive impairment conversion, the lowest IBS among compared methods. For MCI-to-AD conversion, it attained the highest C-index (0.894 ± 0.018) and near-ceiling 5-year discrimination (AUROC 0.997 ± 0.003; AUPRC 0.999 ± 0.001). Ablation studies confirmed the importance of longitudinal change features, fused temporal representations, mixture hazards, cognitive and functional measures, APOE4 status, and recent conversion-proximal visits. These results demonstrate that progression-aware survival modeling can deliver interpretable, multi-horizon risk estimates for Alzheimer's conversion, enabling personalized clinical decisions from routine tabular data.
- Achieved 0.894 C-index and 0.997 AUROC for MCI-to-AD conversion at 5 years, outperforming all baselines.
- Handles irregular visit intervals and censored data using a temporal Transformer with mixture hazards.
- Provides calibrated 1-, 2-, 3-, and 5-year risk estimates from standard clinical measurements and APOE4 status.
Why It Matters
Enables early, personalized Alzheimer's risk prediction from routine clinical data, guiding intervention timing.