SHORT-TERM PREDICTION OF CONVERSION FROM MILD COGNITIVE IMPAIRMENT TO DEMENTIA IN PATIENTS IN THE MEMENTO COHORT, 2011-2016

Abstract

Limited knowledge is available on models able to combine all potential biomarkers and evaluate their respective role on top of clinical and neuro-psychological examinations to predict conversion from mild cognitive impairment (MCI) to dementia, including of Alzheimer's disease (AD). We used data of the nation-wide Memento cohort, which consecutively recruited 2323 participants in 28 memory clinics for either isolated cognitive complaints MCI, while not demented. In participants with baseline Clinical Dementia Rating (CDR) = 0.5, a short-term clinical prediction model of conversion to dementia was built using primarily socio-demographic characteristics, personal and family medical history, neurological and physical examination (including cardiovascular risk factors), and cognitive performances assessed by an extensive neuropsychological battery. The added values of APOE ε4 carrier status and MRI biomarkers (brain parenchymal fraction, hippocampal volume, white matter lesions) on prediction abilities were then evaluated. All models were adjusted for age and sex. Overall, 1374 participants had a baseline CDR=0.5 (mean age 71 years, 60% women). During a mean 2.8 years of follow-up, 189 individuals became demented (133 Alzheimer's dementia cases) yielding an incidence of 49.5 per 1000 person-years. The best predictive clinical model (C-statistic 0.87, 95% Confidence interval (CI), 0.84 to 0.91) included: visit at clinic motivated by behavioral disorders, lower physical complaints, lower performances at cognitive tests (MMSE, gestural praxis, Copy of Rey-Osterrieth complex figure at 3 minutes, Free and cued total immediate recalls test, time to perform Trail Making Test part A), limitations at Instrumental Activities of Daily Living (IADL), higher score at short physical performance battery. On top of clinical and neuro-psychological examinations, with at least 1 APOE ε4 allele and lower hippocampal volume were associated with significant improvement of predictive performances (c-statistic 0.89, CI 0.86 to 0.92). Internal validation was performed using bootstrap (1000 iterations/samples). Restricting modelling of conversion to AD did not show difference in the nature of the predictive variables. Excellent performance was found with a model based on a set of variables measured routinely at the clinic, able to accurately predict dementia on the short term.