Age‐related relative comorbidity burden of mild cognitive impairment: A US database study

Le résumé fourni par la source

Abstract Background As novel therapies emerge, timely detection and diagnosis of early Alzheimer’s disease (AD) become increasingly relevant. While primary care would be the most suitable setting for early detection, diagnosis rates of mild cognitive impairment (MCI) are low and systematic screening of asymptomatic individuals is currently not recommended. We investigated whether a predictive algorithm based on established AD risk factors can help to identify individuals at elevated risk of having undiagnosed MCI, in particular in younger age cohorts. Method Our retrospective study used MarketScan insurance claims data to identify individuals aged ≥50 years with MCI (but without dementia) and assemble controls without MCI/dementia diagnoses matched on age, sex, and geographic region. Medical conditions previously identified as AD risk factors in the literature were considered MCI risk factors if their prevalence in MCI individuals was statistically significantly higher than in controls. Logistic regression and extreme gradient boosting (XGBoost; a machine learning method) were applied to discriminate MCI from control individuals based on area under the curve (AUC). Analyses were conducted for the overall cohort and by age group (50‐64, 65‐79, and ≥80 years). Result Twenty‐five medical conditions had statistically higher frequency in the MCI cohort (n = 5,185) compared to controls (n = 15,555). The top MCI risk factors (with odds ratio for the 50‐64, 65‐79, and ≥80 years subgroups, respectively) were: depression (4.4, 3.1, 1.9); stroke/transient ischemic attack, TIA (6.4, 3.0, 2.1); obstructive sleep apnea, OSA (3.6, 2.5, 1.7); hyperlipidemia (1.7, 1.5, 1.4); and hypertension (1.6, 1.5, 1.5) (all p‐values <0.0001). The logistic regression models for identifying MCI individuals reached AUCs of 0.75, 0.69, and 0.66, for the 3 age groups, respectively. The XGBoost approach reached AUC = 0.94 (training set) and AUC = 0.78 (test set). Each approach was most accurate when discriminating MCI for the youngest age group. Conclusion Our two modeling approaches were able to predict MCI with acceptable accuracy. The predictions were more accurate in the 50‐64 age group than in the older groups, suggesting that the use of prediction models for triage in primary care will play a greater role in younger age groups.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.