Abstract
Abstract Background Timely detection and diagnosis of early‐stage Alzheimer’s disease (AD) become increasingly relevant with the introduction of new therapies. Electronic health records (EHR) data present an opportunity to identify cases with elevated risk of undiagnosed cognitive impairment and triage them for further evaluation. Such triage tools are critically important because population‐based screening of asymptomatic individuals is currently not recommended, but spontaneous detection rates are low. We previously developed a risk prediction algorithm to identify likely cases aged ≥50 years based on claims data. Method We will build on this work and develop an algorithm based on three data sources: claims, EHR data and the National Alzheimer’s Coordinating Centers (NACC) data. The claims database study, already completed, included newly‐diagnosed mild cognitive impairment (MCI) individuals and non‐MCI individual matched on age, gender and geographic region. Using data 2 years prior to diagnosis, medical conditions associated with MCI were used to build a predictive model. The EHR study will aim to augment the model with clinical data such as body mass index, smoking status, vital signs and laboratory results. The NACC study, including ∼5,400 MCI and ∼15,800 cognitively normal (CN) individuals, will be used to validate the algorithm. Result The claims database study identified 25 medical conditions that predict MCI risk (area under the curve, AUC = 0.78). The top five predictors for MCI were depression, stroke/transient ischemic attack, obstructive sleep apnea, hyperlipidemia, and hypertension. The predictive performance was better in the 50‐64 year than in the older age groups. Analyses of the EHR and NACC data are ongoing. Conclusion Data routinely collected in claims and EHR may be accurate enough to triage individuals with potential MCI for cognitive assessment in primary care.