JMIR Publications

ABSTRACT

Background:

As the number of US adults 65 years and older continues to grow, so does the need for sustainable, scalable approaches to cognitive assessment and support. Alzheimer’s disease (AD) affects over 6 million US adults, a number projected to reach 13 million by 2050, yet cognitive decline remains underdiagnosed and is often identified late in the disease course. Given older adults’ increasing access to and use of the Internet, Internet-based cognitive programs have emerged as a promising approach to cognitive assessment and monitoring. Building on previously successful approaches to cognitive enhancement program design, we developed LookAlikes, an easily accessible, interactive, and engaging cognitive assessment program that tracks multiple cognitive domains over time and provides instantaneous feedback.

Objective:

This 2-phase study aimed to (1) optimize the LookAlikes user interface and validate the accuracy of its speech-to-text and scoring systems (Phase 1) and (2) establish the feasibility and acceptability of LookAlikes among older adults with subjective memory concerns (Phase 2). We hypothesized that the program’s systems would achieve sufficient accuracy to support future longitudinal cognitive tracking and that LookAlikes would be feasible and acceptable to the target population.

Methods:

In both phases, participants (≥65 years of age) with subjective memory concerns were recruited from a convenience sample of older adults ineligible for an NIH-funded national randomized controlled trial (NCT05565833). All participants were asked to use LookAlikes daily for 12 days (<2 min/session) and complete an end-of-study online questionnaire assessing ease of use, program design, overall satisfaction, and likelihood of recommending the program, with open-ended questions to capture unanticipated feedback. In Phase 1 (N=40), we compared 480 audio files to processed speech-to-text results using word-by-word counts and calculated word error rate (WER) against a prespecified accuracy target of ≥85%. Phase 2 enrolled a separate, subsequent sample of 43 participants, and feasibility was measured by program use, duration of use, and frequency of use. Quantitative data were analyzed in SPSS Statistics (v.26); qualitative responses were analyzed using line-by-line coding and content analysis.

Results:

In Phase 1, all 40 participants completed the 12 LookAlikes activities (completion range 12-19 days) and the end-of-study questionnaire. Speech-to-text accuracy was 99.1% (WER=191/22,029) across 480 audio files, surpassing the 85% target. In Phase 2 (N=43), 97.7% of participants completed all 12 activities, with 74.4% completing them within 14 days; 1 participant withdrew. Acceptability was high: 92.7% reported playing LookAlikes every day, 75% indicated they would play for fun outside a study, and 85.3% liked the post-task scoring feedback. On a 5-point Likert scale, mean ratings for satisfaction (4.49), ease of use (4.61), and convenience (4.70) were consistently high. Participants reported low confusion (mean 1.54) and frustration (mean 1.59). The majority (87.8%) reported no privacy concerns.

Conclusions:

The LookAlikes speech-to-text and scoring systems exceeded prespecified accuracy targets, and the program was feasible and highly acceptable to older adults with subjective memory concerns. Together, these complementary findings support progression to a longitudinal study evaluating LookAlikes to measure and track cognition over time.