Empirical Identification and Longitudinal Characterization of Mild Cognitive Impairment Subtypes using Latent Mixture Modeling
Rationale: Research in conventional mild cognitive impairment (MCI), a prodromal stage between normal aging and Alzheimer’s dementia (AD), has demonstrated neuropsychological heterogeneity using clustering techniques. The current dissertation aimed to 1) empirically establish baseline neuropsychological MCI subtypes; 2) explore longitudinal characterization of empirical subtypes using rigorous norms; and 3) examine the probability of transition between subtypes over time.
Design: Study 1 included 806 MCI participants from the Alzheimer’s Disease Neuroimaging Initiative (ADNI). Unique neuropsychological MCI subtypes and their associations with AD markers were investigated using latent profile analysis (LPA). Study 2 included 825 ADNI participants with baseline MCI that had follow-up at 12-months (n=751) and 24-months (n=639). Demographically-corrected T-scores were derived from the performance of 284 “robust” normal control participants assessed at baseline, 12-, and 24-months. Serial LPAs established neuropsychological subtypes for the MCI participants at each time point. Study 3 employed latent transition analysis to evaluate the likelihood of subtype change over time, as well as the influence of AD-risk factor covariates on transition probabilities.
Results: Study 1 produced 3-classes: mixed impairment, amnestic impairment, and cognitively normal neuropsychological subtypes. Amnestic and mixed classes had higher positivities on markers of AD than the cognitively normal class. In Study 2, 4-neuropsychological classes were separately established at baseline, 12-, and 24-months: multi-domain impairment ([MLT]), amnestic impairment (AMN), dysexecutive/below average cognition (DYS/BA), and average cognition (AVG) classes. The MLT and AMN subtypes declined over time on the majority of measures, while the AVG subtype had stable neurpsychological performance. The DYS/BA subtype demonstrated stable memory performance and improvement on language and attention/executive measures. Study 3 indicated a high probability (>86%) for participants of all subtypes to remain in their class over time. Covariates that modestly increased the likelihood of transition between classes included worse functional ability and AD-biomarker positivity.
Conclusions: This dissertation research used latent mixture models to establish analogous longitudinal neuropsychological profiles in conventionally diagnosed MCI. Results suggest that individuals are most likely to remain within their subtype across two years, including cognitively normal “false-positives.” Future studies should examine empirical MCI subtypes with the use of actuarial methods that may improve diagnostic accuracy.