MRI‐based measure of dementia risk derived using machine learning is strongly associated with measures of longevity and biological age

Abstract

AbstractBackgroundMachine learning and artificial intelligence methods have been applied to brain images to estimate measures of brain aging. However, only rarely have these measures been examined in the context of biologic age. Here, we investigated associations of an MRI‐based measure of dementia risk the Alzheimer’s disease pattern similarity (AD‐PS) scores with measures of longevity and biological age.MethodParticipants were those from visit 5 of the Atherosclerosis Risk in Community Study with cognitive status adjudication and MRI available. The AD‐PS score was estimated based on previously reported machine learning methods. We evaluated associations of the AD‐PS score with all‐cause mortality. Participants were stratified by AD‐PS tertiles and the analyses were adjusted for age, race, sex, hypertension and smoking. AD‐PS score was examined in association with 32 proteins reported to be associated with age. In these analyses, we used Bonferroni correction (α = 0.05,p<0.0016). Finally, associations with a deficit accumulation index(DAI) based on 38 health items was investigated. In both cases, linear regression models were adjusted for age, race and sex. Sensitivity analyses using only cognitively normal (CN) individuals were performed.ResultMortality – A total of 356 participants died within 8 years of follow‐up. The AD‐PS score was significantly associated(p<0.001) with time to all‐cause mortality. Participants in the lowest tertile had lower all‐cause mortality rate compared to those in the highest tertile (HR: 0.43;95% CI:0.31,0.60). The association remained significant when restricting the sample to only CN subjects (HR for lowest tertile:0.53; 95% CI:0.35,0.81, p = 0.0028).Age related proteins ‐ The AD‐PS scores were significantly associated (p<0.05, uncorrected) with 10 of the 32 proteins. Growth/differentiation factor 15(GDF‐15) and pleiotrophin remained significant after correction for multiple‐testing. Analysis of CN participants showed a subset of the same proteins to be significant (p<0.05,uncorrected) but only the GDF‐15 remained significant after correction for multiple‐testing.DAI ‐ A linear regression model showed a significant association between DAI and AD‐PS scores overall (coeff = 0.52; 95% CI:0.37‐0.67) and in the CN subset (coeff = 0.3,95% CI:0.13‐0.48).ConclusionWhile the AD‐PS scores were created as a measure of dementia risk, our analyses suggest that they could also be capturing brain aging.