OBJECTIVES:

This study helps in predicting the incidence of AD based on multiple variables that were identified during the prodromal phase. Identification and timely intervention delays disease progression and improves the quality of life that leads to reduction in overall cost of care.

METHODS:

Based on ICD-9 and ICD-10 codes,115,652 patients aged 60 years and older, from 2019 to 2020, were identified using Optum’s Market clarity database. 28,048 patients with AD were identified in the entire sample having at least two outpatient claims (at least 30 days apart) OR one inpatient claim. Patients having claim or encounter for AD in the pre index period (3 years) were excluded from the analysis. Only patients having continuous eligibility were included in the study.

Overall, 37 potential risk predictors like demographics, comorbidities, signs, and symptoms were identified based on feature selection techniques. Training and evaluation of Logistic Regression, XGBoost, Random Forest Classifier and K-nearest Neighbor were executed. The models were trained using 80:20 ratio of total subjects.

RESULTS:

The AUROC is 0.873 for Logistic Regression. The model identified AD and non-AD patients with 77% and 84% precision respectively. Odds ratio (OR) indicates higher probability of having AD in patients with risk factors as amnesia (OR: 9.04), Mild cognitive impairment (OR: 5.88), Hypertension (OR: 1.63), Psychosis and Memory loss (OR: 1.57) and Musculoskeletal symptoms (OR: 1.10). Further, an in-depth analysis would be performed by including more signs and symptoms and comorbidities. and running multiple iterations on various ML models to enhance the AD prediction.

CONCLUSIONS:

This model can predict the AD patients during their early phase of disease and enables healthcare providers in planning the most effective course of treatment for them. This in turn saves time and money and attain better healthcare outcomes.