Practice Effects in Mild Cognitive Impairment Increase Reversion Rates and Delay Detection of New Impairments

Abstract

Objective: Cognitive practice effects (PEs) can delay detection of progression from cognitively unimpaired to mild cognitive impairment (MCI). They also reduce diagnostic accuracy as suggested by biomarker positivity data. Even among those who decline, PEs can mask steeper declines by inflating cognitive scores. Within MCI samples, PEs may increase reversion rates and thus impede detection of further impairment. Within an MCI sample at baseline, we evaluated how PEs impact prevalence, reversion rates, and dementia progression after 1 year.

Methods: We examined 329 baseline Alzheimer's Disease Neuroimaging Initiative MCI participants (mean age = 73.1; SD = 7.4). We identified test-naïve participants who were demographically matched to returnees at their 1-year follow-up. Since the only major difference between groups was that one completed testing once and the other twice, comparison of scores in each group yielded PEs. PEs were subtracted from each test to yield PE-adjusted scores. Biomarkers included cerebrospinal fluid phosphorylated tau and amyloid beta. Cox proportional models predicted time until first dementia diagnosis using PE-unadjusted and PE-adjusted diagnoses.

Results: Accounting for PEs increased MCI prevalence at follow-up by 9.2% (272 vs. 249 MCI), and reduced reversion to normal by 28.8% (57 vs. 80 reverters). PEs also increased stability of single-domain MCI by 12.0% (164 vs. 147). Compared to PE-unadjusted diagnoses, use of PE-adjusted follow-up diagnoses led to a twofold increase in hazard ratios for incident dementia. We classified individuals as false reverters if they reverted to cognitively unimpaired status based on PE-unadjusted scores, but remained classified as MCI cases after accounting for PEs. When amyloid and tau positivity were examined together, 72.2% of these false reverters were positive for at least one biomarker.

Interpretation: Even when PEs are small, they can meaningfully change whether some individuals with MCI retain the diagnosis at a 1-year follow-up. Accounting for PEs resulted in increased MCI prevalence and altered stability/reversion rates. This improved diagnostic accuracy also increased the dementia-predicting ability of MCI diagnoses.

Keywords: Alzheimer’s disease; biomarkers; cognitive aging; dementia progression; mild cognitive impairment; practice effects.

FIGURE 1

Full Cox proportional models for time until first dementia diagnosis by PE-unadjusted and PE-adjusted 12-month diagnoses. Cox proportional hazard models compared progression to dementia between those who were classified with mild cognitive impairment at follow-up (Stable MCI) and those who reverted to cognitively normal (Reverters). Models used classifications (Stable MCI vs. Reverter) as the independent variable of interest; months from baseline until first dementia diagnosis as the dependent variable; and all variable data (16 – months from baseline). Covariates were age and education, fixed at the average level within the sample (age: 73.1 years; education: 16.4 years). The left graph bases diagnoses on the PE-unadjusted 12-month data; the right graph uses diagnoses based on the PE-adjusted 12-month data. Each model presents a hazard ratio (HR; [CI]) that indicates how much more likely the Stable MCI group was to convert to dementia compared to the Reverters. Wald tests and likelihood-ratio tests (LRT) are also included with associated p-values to denote the significance of the HR. The Y-axis of each model provides the survival probability and the X-axis of each model provides the time frame until dementia conversion.

FIGURE 2

Full Cox proportional models for time until first dementia diagnosis by PE-unadjusted and PE-adjusted 12-month diagnoses. Cox proportional hazard models compared progression to dementia between those who were classified with mild cognitive impairment at follow-up (Stable MCI) and those who reverted to cognitively normal (Reverters). All models used classifications (Stable MCI vs. Reverter) as the independent variable of interest and months from baseline until first dementia diagnosis as the dependent variable. Covariates were age and education, fixed at the average level within the sample (age: 73.1 years; education: 16.4 years). Models in the top row display results completed with PE-unadjusted scores; models in the bottom row display results completed with the PE-adjusted scores. Each row designates the time frame for each model measured in months from baseline. Time frames were restricted to demonstrate how predictive the classification was for studies with various follow-up periods. As these hypothetical studies would not know if a participant converted to dementia past their follow-up period, those who converted after the endpoint of that specific model were censored (i.e., recoded as non-converters). Each model presents a hazard ratio (HR; [CI]) that indicates how much more likely the Stable MCI group was to convert to dementia compared to the Reverters. Wald tests and likelihood-ratio tests (LRT) are also included with associated p-values to denote the significance of the HR. The Y-axis of each of the 6 models provides the survival probability and the X-axis of each model provides the time frame until dementia conversion.