[2307.12369] Early Prediction of Alzheimers Disease Leveraging Symptom Occurrences from Longitudinal Electronic Health Records of US Military Veterans

PDF from the paper entitled Early Prediction of Alzheimer’s disease, which benefits from the events of symptoms from longitudinal health records of American old military warriors, by Rumeng Li and 9 other authors

PDF view

a summary:Early prediction of Alzheimer’s (AD) is very important to intervene and timely treatment. This study aims to use automated learning methods to analyze longitudinal health records (EHRS) for patients with Alzheimer’s disease and determine signs and symptoms that can predict the beginning of the advertisement early. We used a design to control cases with the longitudinal EHRS from the Ministry of Old Warriors Affairs in the United States (VHA) from 2004 to 2021. VHA patients with AD were diagnosed after 1/1/2016 based on ICD-10-CM codes, corresponding to controls with the times, taking advantage of sex and taking advantage of inclusion with the alternative. We used an advertising -related keyboard and their accidents over time in the patient’s long eHRS as predicting AD with four automated learning models. We conducted the analyzes of the sub -group by age, sex and race/sweat, and we verified the form of the model in the “invisible” VHA group. Model discrimination, calibration, and other standards related to predictions were reported up to ten years before the ICD diagnosis. The study residents included 16701 cases and 39,097 identical controls. The average number of keywords related to AD increased (for example, “concentration”, “speaking”) per year rapidly for cases with the approaching diagnosis, from about 10 to more than 40, while it remains flat in 10 controls. Achieve the best high discrimination model (Rocauc 0.997) for predictions using data from at least ten years before ICD diagnoses. Hosmer-Lemeshow Goodness of-Falue = 0.99 is consistent with sub-groups of age, sex, race/sweat, with the exception of patients under the age of 65 (Rocau 0.746). Automated learning models can be predicted using keywords related to the disease specified from EHR notes to future advertising diagnoses, indicating its potential use to determine the risk of advertising using EHR notes, providing a reasonable price method for early examination on the adult population.