AI-assisted modeling of attentional control for intervention engagement
Digital cognitive interventions can improve cognition in older adults, but individual differences in transfer to executive function (EF) remain substantial and are difficult to anticipate before treatment completion. We developed a compact Facial Signature Score (FSS) from passive webcam recordings collected during therapeutic cognitive challenges in BREATHE (NCT04522791; development, N=87) and evaluated it in PSOPT (NCT06005038; external validation, N=39), an independent trial using a different intervention format. The FSS discriminated EF responders under nested cross-validation in BREATHE (AUROC=0.77) and generalized to PSOPT without feature reselection or reweighting (AUROC=0.78). The FSS was associated with continuous EF change in BREATHE (Spearman {rho}=0.49, p<0.001) and after outcome-independent recalibration in PSOPT ({rho}=0.32, p=0.045). Predictive performance was retained when the signature was computed from the first 25% and 50% of intervention sessions, and the FSS added value beyond baseline characteristics and conventional behavioral measures. Passive facial dynamics during therapeutic cognitive challenge may serve as a scalable intervention-process marker for early assessment of cognitive response among older adults.