Background: This project evaluated the feasibility, acceptability, and validity of smartphone app cognitive assessments in a multisite study designed to validate blood and digital biomarkers of early-onset cognitive changes in underrepresented populations.

Method: Eighty-one participants (mean age 52 ± 6.6 years; 78% female) from a racially and ethnically diverse cohort (42% Black, 30% multiracial, 15% White, 11% Asian; 37.5% Latino/Hispanic) remotely completed a ~30-minute battery of cognitive and motor tasks on their personal smartphones. Analyses focused on the flanker task as a measure of executive function. Participants completed questionnaires on technology familiarity, user experience, and subjective cognitive decline. A subset completed blood draws (N=37) for plasma Alzheimer’s disease biomarkers (Aβ42/40), as well as vascular (total and HDL cholesterol) and metabolic (fasting glucose) labs, from which a vascular risk score was derived using published methods. Another subset (N=36) completed in-person neuropsychological testing. Descriptive statistics summarized participant characteristics and user experience. Linear regressions adjusted for age, sex, and education examined associations between flanker performance, neuropsychological measures, vascular risk, and Alzheimer’s disease biomarkers.

Results: Participants completed 100% of available smartphone assessments. Technology familiarity was high (100%), with most preferring mobile assessments over pen-and-paper testing (78%). Greater subjective cognitive concerns were associated with worse total scores on the smartphone executive functioning task (β= -0.29, p=0.005), and higher scores corresponded with better performance on gold standard measures of executive functioning (e.g. Trail Making Test Part B [β = –0.50, p = 0.001]).

Elevated vascular risk scores were associated with reduced flanker accuracy (β= -0.33, p= 0.04) and slower reaction times (β= -0.38, p=0.02). Higher fasting glucose related to lower accuracy (β= -0.44, p = 0.003) and slower reaction time (β= -0.44, p=0.002).

Lower Aβ42/40 ratios (reflecting higher amyloid burden) were associated with worse smartphone-app executive function scores, particularly among individuals with low vascular burden (N= 12, β= 0.70, p=0.025).

Conclusions: Smartphone cognitive assessments were feasible, acceptable, and sensitive to subjective cognitive impairments within a diverse midlife cohort. Validity was supported by associations with gold-standard tasks, risk factors, and biomarkers. These provide preliminary support that remote assessments can reflect biologically relevant cognitive changes in underrepresented populations.