Developing a Risk Prediction Model for Heart Failure (HF) Incidence in Patients with Diabetes Mellitus Using AI & Ml Techniques
Author(s)
Verma V1, Sharma S2, Markan R2, Dawar V2, Bhargava S3, Brooks L4, Musle A2, Kumar S2, Nayyar A2
1Optum, Gurgaon, HR, India, 2Optum, Gurugram, HR, India, 3Optum Tech, Eden Prarie, MN, USA, 4Optum, Basking Ridge, NJ, USA
Presentation Documents
OBJECTIVES:
This model aims at predicting incidence of HF in patients with Diabetes, based on multiple risk factors. Early identification will enable healthcare providers in early intervention and delaying disease progression.METHODS:
Based on ICD-9 and ICD-10 codes, 165,963 diabetics, aged 18 years and above were identified using Optum’s Market Clarity Database. Patients with continuous eligibility for 5 years (2015 to 2019), having at least one yearly claim for Diabetes were included in the analysis. Of the overall cohort, 53,534 diabetic patients were found to have at least one inpatient OR 2 outpatient claims (>=60 days apart) for HF between January 2019 to December 2019. Patients having claim for HF in the pre index period (5 years) were excluded from the analysis. A multivariate analysis was conducted to develop a prediction model by incorporating 36 variables like demographics, comorbidities and Sign and Symptoms during pre-index period using feature selection techniques. Training and evaluation of Logistic Regression, XGBoost and Random Forest Classifier were executed. The models were trained and tested using 80:20 ratio of total subjects.RESULTS:
The AUROC is 0.828 for Logistic Regression. The model identified HF and non-HF patients with 68% and 78% precision respectively. Odds ratio (OR) indicates higher probability of having HF in Diabetics with risk factors like Atrial fibrillation (OR: 2.55), Renal failure (OR: 2.77), Ventricular hypertrophy (OR: 2.06), COPD (1.86), Myocardial infraction (2.51), Respiratory (1.31) and Musculoskeletal symptoms (1.24). Further analysis would be performed by including more features and multiple iterations on various ML models to enhance HF prediction.CONCLUSIONS:
The predictive model worked effectively in identifying risk factors for HF in patients with diabetes. It would improve outcomes by providing physicians with tools and data to identify patients risk factors. Payers can utilize this model in optimizing their pricing strategy for patients at risk of developing HF.Conference/Value in Health Info
2023-05, ISPOR 2023, Boston, MA, USA
Value in Health, Volume 26, Issue 6, S2 (June 2023)
Code
RWD4
Topic
Clinical Outcomes, Real World Data & Information Systems, Study Approaches
Topic Subcategory
Clinician Reported Outcomes, Electronic Medical & Health Records, Reproducibility & Replicability
Disease
Cardiovascular Disorders (including MI, Stroke, Circulatory)