Clinical Manifestations

Traditional cognitive testing requires trained staff, in-person visits, and suffers from between-visit variability, contributing to underdiagnosis of Mild Cognitive Impairment (MCI). Storyteller, a self-administered remote story-recall task, enables low-burden repeat assessments in ADNI4. We examined whether its longitudinal trajectories improve discrimination between MCI and Cognitively Normal (CN) participants.

ADNI4 participants with baseline clinical diagnosis of CN (n = 76) or MCI (n = 21) who completed Storyteller at baseline and 6 months were included. Mixed linear regression was used to examine effects of demographic variables (age, sex, education level), baseline diagnosis, and time on Storyteller G-match scores. Logistic regression models with 10-fold cross-validation compared classification performance of baseline G-match, change in G-match over a 6-month period, and both variables together, using Receiver Operating Characteristic (ROC) curve Area Under the Curve (AUC) and 95% confidence intervals.

Age (p = 0.837), sex (p = 0.708), and education (p = 0.440) did not differ between groups, but the MCI group had lower MMSE (p = 0.001), G-match at baseline (p <0.001) and 6 months (p <0.001); 8 (10.5%) CN participants had CDR-G 0.5 (Table 1). The mixed model showed significant effects of sex (male: -6.133±2.253, p = 0.006), education years (1.221±0.471, p = 0.010), baseline MCI (-9.437±2.566, p <0.001), timepoint (6 months: 3.622±0.992, p <0.001), and MCI*timepoint interaction (MCI declined more: -8.997±2.13, p <0.001) (Figure 1). For MCI classification, demographic baseline of sex and education level had an AUC of 0.507 [0.323-0.691] (p = 0.932), while baseline G-match AUC=0.754 [0.592-0.915] (p = 0.006) and 6-month change AUC=0.712 [0.538, 0.886] (p = 0.022) each outperformed random; combining both improved AUC to 0.800 [0.620-0.980] (p = 0.004), and further including demographics had an AUC of 0.817 [0.659-0.975] (p = 0.001) (Figure 2).

Storyteller detected greater decline in MCI than CN, improving classification when longitudinal measures were included. Remote, self-administered follow-up provided additional diagnostic power after only six months.