Using Digital Signals from Credit Data for Early Detection of Alzheimer's Disease and Related Dementias

点击翻译按钮获取中文摘要

Project Summary The value of early diagnosis for Alzheimer’s disease and related dementias (ADRD) is increasingly recognized. However, available diagnostic tools rely primarily on the manifestation of cognitive symptoms that interfere with everyday activities, and screening tools to support earlier identification of individuals with ADRD are lacking. Credit data represent a unique foundational data source upon which machine learning algorithms can be developed to identify individuals at risk for ADRD and facilitate earlier diagnosis. The strength of the information signal from credit data for identifying those at risk for ADRD is supported by previous research that finds, first, that significant limitations and rapid declines in financial capacity are a hallmark of early-stage disease and, second, that afflicted individuals and their families experience negative economic consequences during early-stage disease. We propose using a massive database—that we have already constructed—of credit data from Equifax which is the basis of the Federal Reserve Bank of New York’s Consumer Credit Panel (CCP), merged at the individual level using a unique common identifier (Social Security number), with Medicare enrollment and claims data. The data encompass more than 84 million person-years of data in total, with more than 1.7 million individuals who have been diagnosed with ADRD. Our specific aims are to: (1) Estimate the effects of early-stage ADRD on a wide range of financial outcomes measured in credit data, allowing for potential differences in the effects of early-stage ADRD depending on characteristics such as race/ethnicity, education, gender, and household structure; (2) Apply machine learning methods to our already- developed massive data base with merged credit (CCP) and Medicare data in order to develop algorithms that are capable of identifying individuals at risk for ADRD; and (3) Assess the robustness of the algorithm to the inclusion of newly available years of Medicare claims and enrollment data. The findings from Specific Aim 1 are important for identifying and understanding the specific financial outcomes individuals with ADRD are most susceptible to during the early stage of disease and will help inform the machine learning models in Specific Aims 2 and 3.