Statistical Analyses Analyses were registered at As Predicted https://osf.io/sh27y/. Statistical power analysis (Faul et al., 2007) indicated that to detect a small effect size (f=.010, which is roughly equivalent to d=0.20) for a group x time interaction, a final n of 198 participants (n=99 per arm) would achieve 95% power with an alpha of .05. Power analyses further indicated that to examine the moderating effects of MCI alpha .05 to detect medium effects (f=0.25, roughly equivalent to d=.50) for a significant group-by-time by moderator three-way interaction. We conducted Chi-square and one-way ANOVA analyses to compare the randomized groups at baseline on age, race, education, sex, MoCA, GDS, hearing and vision. Variables with marginally significant (p<.1) differences were included as covariates in subsequent analyses. Data reduction by principal component factor analysis was used to derive composite outcomes of CAP, cognition, and everyday function for analyses. Mixed effects models were conducted in SAS (SAS Institute, Cary, NC) procedure MIXED to examine change in performance from baseline to post-test by randomized group. Mixed effects models are an advanced form of repeated measures ANOVA that account for random effects (i.e., variance in individual baseline values and change in scores) and thus tend to provide more accurate estimates. We report effects for baseline scores (intercept), the overall rate of change (time), the estimate for the difference between baseline values for the two groups, and the group-by-time interaction, which is indicative of changes in performance attributable to intervention. Primary analyses included randomized participants and analyses were repeated among those who were adherent (i.e., completed >15 sessions). Models were also conducted including MCI, group-by-MCI, time-by-MCI, and group-by-time-by-MCI interactions to examine if intervention effects varied by cognitive status. Results Recruitment and enrollment began in January 2018 and continued through December 2020, with data collection completed in April 2021. Seventy-six percent of randomized participants completed post-test. Nineteen participants were excluded from analyses due to a head injury (n=1), undergoing anesthesia (n=10), or hospitalization (n=8) before post-test. Please see Figure 1 for details. The randomized groups did not differ in sex, χ2(1)=.787, p=.375, ethnicity χ2(1)=.167, p=.683, or race, χ2(1)=.045, p=.832. The Music Reading Assessment scores did not differ between the two groups, t(266)=-.958, p=.338. The randomized groups did not significantly differ in age, F(1,266)=.293, p=.589, education, F(1,266)=.369, p=.544, MoCA, F(1,266)=.427, p=.514, or GDS, F(1,266)=.943, p=.332. However, there were marginally significant group differences for PTA in the left ear, F(1,266)=3.31, p=.070 and statistically significant group differences for both PTA in the right ear, F(1,266)=6.69, p=.010, and visual acuity, F(1,266)=8.99, p=.003. The music listening group tended to have worse hearing, while the piano training group tended to have worse visual acuity. See Table 1. The groups also differed significantly in their expectations regarding potential benefits of their randomized group, F(1,214)=15.07, p<.001. The piano training group had an average rating of 5.55 while the music listening group had an average rating of 4.99, indicating the piano training condition had higher expectations about the potential effects of their randomized condition. The piano training group also rated their condition as more challenging with 92% agreeing that the intervention was challenging as compared to 86.3% of the music listening group indicating their condition was challenging (p<.001). Principal component analysis with varimax rotation was conducted to reduce the number of variables for analyses. Please see Supplemental Table A for details. The first factor reflected cognitive performance speed and included the Trail Making Test, Digit Coding, and the TEA visual elevator and phone search subtests. The second factor included all three verbal fluency subtests. The third factor reflected CAP and included DDT, DSI, TCS, and WIN. The fourth factor included two indices from the TEA phone search while counting and visual elevator accuracy. The fifth composite included ATTR across- and within- channel performance. Baseline composites of cognitive performance, verbal fluency, ATTR, CAP, and everyday function were created by averaging z scores after the reverse-scaling of items with negative factor loadings. Post-test scores were standardized by baseline means and SDs. Timed IADL performance did not significantly load on any factor and was thus examined separately. In the primary analyses, we examined whether performance changed differentially by random assignment across composite outcomes from baseline to post-test as indicated by a significant group-by-time interaction. Covariates included hearing PTA, visual acuity, and expectations. The group-by-time interactions were not statistically significant for any of the outcomes (ps>.194). See Table 2. The results did not change when we included only those who were adherent (ps>.180). See Supplementary Table B. Piano training did not significantly enhance CAP, cognition, or everyday function as compared to music listening. We further examined if intervention effects varied by MCI status (Supplementary Table B). There were no significant group-by-time-by-MCI status interactions indicating those with and without MCI did not show differential benefit (ps>.245). Thus, MCI did not significantly moderate the effects of piano training. Effect sizes for piano training as compared to music listening are shown in supplemental materials. Improvements were not evident in the primary analyses, adherent analyses, or the subsample without MCI on measures of CAP, cognition, or everyday function. Participants with MCI randomized to piano training showed potential small effect sizes for improvement relative to music listening on Trails A, Digit Symbol Coding, and one subtest of the TEA (ds>=0.25). Raincloud plots of individual assessment and composite variables are included in the supplemental materials. Results indicated significant effects of time for ATTR in all models and significant effects of time for CAP in the primary and adherent analyses indicating improved performance from pre- to post- training. Significant effects of time were found for cognition and TEA in the primary and adherent analyses, indicating a tendency for decline across time. With respect to covariate effects, CAP, ATTR, and cognitive performance were significantly impacted by hearing. Similarly, Timed IADL and cognitive performance varied significantly by visual acuity. Those with poorer hearing or vision tended to perform worse on these outcomes. Interestingly, expectations about the assigned condition were significantly related to composite outcomes of CAP and cognition. However, having greater expectations that the randomized condition would positively affect abilities was associated with poorer performance. See Table 2. Additional sensitivity analyses examined if there were differential effects by sex. No significant group-time-sex interactions were evident (ps>.119), and the pattern of results was the same when analyses were stratified by sex.