OPTIMAL HARMONIZATION OF COGNITIVE MEASURES ENHANCES THE DETECTION OF GENETIC EFFECTS ON ALZHEIMER’S DISEASE PROGRESSION: A COMPARISON OF SIX STATISTICAL APPROACHES IN 1796 MCI PATIENTS FROM THREE COHORTS

Abstract

Research on genetic variants modulating disease progression in pre-dementia stages of Alzheimer's disease (AD) requires collaborative work with several cohorts. Because of different assessment protocols, there is a need to devise reliable harmonization methods. Therefore, we compared six methods with regard to their ability to detect the impact of AD-associated genes on disease progression in patients with Mild Cognitive Impairment (MCI). We analyzed data of 1796 MCI patients from two German and one Spanish cohort. We used Cox-Regression analysis to explore conversion to AD dementia and linear mixed models (LMM) to analyze cognitive decline over up to 9 years. For the LMM, longitudinal cognitive assessments were harmonized using five methods: a) common tests across cohorts (i.e. MMSE); b) standardized composites of (partly cohort-specific) tests assessing memory and executive function; c) standardized composites adjusted for demographic variables; d) multivariate latent process LMMs that removed methodological differences between composites; e) moderated non-linear factor analysis (MNLFA) that summarized composites into a demography-adjusted latent factor. Sensitivity of each method was evaluated using APOE4 and 39 single nucleotide polymorphisms (SNPs) previously associated with AD. In addition, we examined the effect of a genetic risk score (GRS) including only genome-wide significant loci (without APOE) identified by the largest genome-wide association study from the IGAP consortium. Associations were analyzed in each cohort separately and then combined in a meta-analysis. The MNLFA was most sensitive concerning the detection of the APOE4 effect on disease progression (table 1). Only the MMSE revealed a nominally significant effect of the GRS. Significant effects of SNPs surviving correction for multiple testing were solely discovered using MNLFA (in clusterin (CLU)) and the MMSE (in CLU and TREML2). Harmonization by using the MMSE as a common measure across cohorts provided the highest sensitivity for genetic effects on disease progression in MCI. The MNLFA was the most effective method using cohort-specific neuropsychological tests and might be preferable in the absence of a common measure. Both approaches outperformed Cox-regression analyses of conversion to dementia. Our results may generalize to other fields of epidemiological research but still need confirmation in simulation studies.