Development and validation of an interpretable clinical scoring model to monitor the progression of preclinical Alzheimer's disease.

BACKGROUND: Monitoring the progression of preclinical Alzheimer's disease (AD) is challenging due to the absence of obvious cognitive impairment, yet it's crucial for determining the optimal timing for disease-modifying treatments. METHODS: This prognostic study used data from the Anti-Amyloid Treatment in Asymptomatic Alzheimer’s Disease (A4) study. The development cohort included 412 participants from the placebo arm, and external validation was performed using 342 participants from the treatment arm. All participants had baseline brain amyloid-beta (Aβ) positron emission tomography (PET) standardized uptake value ratios (SUVR) below 1.4. Using the AutoScore machine learning framework, we developed two interpretable models: the AutoScore Amyloid-Beta (ASAB) model to predict Aβ accumulation (> 1.4 SUVR) and the AutoScore Phosphorylated Tau (ASPT) model to predict plasma phosphorylated tau-217 (pTau-217) elevation (> 0.3 pg/mL) over 4.5 years. Model performance was assessed using the area under the receiver operating characteristic curve (AUC-ROC) with 95% confidence intervals (CIs). RESULTS: The ASAB model achieved an AUC-ROC of 0.87 (95% CI, 0.79–0.95), and the ASPT model achieved 0.86 (95% CI, 0.78–0.94). Modified models excluding baseline Aβ SUVR or pTau-217 values maintained a strong performance (AUC-ROC 0.76–0.82). External validation demonstrated a robust performance, with an AUC-ROC of 0.83 (95% CI, 0.78–0.87) for ASAB and 0.84 (95% CI, 0.80–0.89) for ASPT. Key predictors included baseline biomarker levels, cholesterol, platelet count, alanine transaminase, and creatine kinase. CONCLUSIONS: The AutoScore-based models accurately predict longitudinal Aβ and pTau-217 levels using readily available clinical and laboratory data. These interpretable tools could help clinicians and researchers monitor preclinical AD progression and identify optimal windows for intervention.