🚀 Exciting News! 🚀 One of the prevailing challenges in healthcare analytics when developing risk-predictive models is determining whether or not to include or replace race, and other genetic markers, with social determinants of health (SDOH). For context: existing risk scores for health conditions like kidney disease, such as the estimated Glomerular Filtration Rate (eGFR), have been found to disproportionately disadvantage Black communities instead of benefitting them. Black patients with similar creatinine levels to their white counterparts are often times scored as having “healthier” kidneys, sometimes leading to under-treatment by health professionals. To help tackle this issue in healthcare analytics, this past spring semester, I worked on a machine learning project for my senior thesis focusing on how we may enhance existing similar predictive metrics and risk scores for cardiovascular health outcomes using social determinants of health! 📊 Project Overview: I developed two distinct logistic regression models using NHANES data to forecast cardiovascular health outcomes: 1) Base Model: Incorporates all current variables used in predicting the 10-year risk score for atherosclerotic cardiovascular disease (ASCVD), such as blood pressure, smoking status, and cholesterol levels, along with demographic factors like age, sex, and race (binary Black vs. white) to predict cardiovascular health outcomes. The existing ASCVD risk score uses pooled cohort equations and logistic regression models to estimate an individual's 10-year risk for ASCVD. I also calculated the true ASCVD risk scores for each patient in my dataset to observe differences in predictor variables across risk scores. 2) Enhanced Model: Adds crucial SDOH factors such as household income, poverty ratio, nutritional intake, and food security data, as well as a further stratified race variable, offering deeper insight on how these variables may also impact cardiovascular health outcomes. 🔍 Key Findings: - The traditional risk-score variables showed slightly higher accuracy in predicting outcomes. - Significant variations in risk scores across different racial groups suggest the need for a more comprehensive and nuanced race variable in future predictive models. - There was a notable link between poverty ratios and cardiovascular events, particularly among individuals living in poverty and negative cardiovascular events, underscoring the importance of socioeconomic factors in health outcomes. I invite you to dive into the details and explore the models on my GitHub. Let’s discuss how these new insights can pave the way for more inclusive, accurate and robust healthcare analytic/predictive tools. Feedback, thoughts, and collaborations are welcome! 🔗 Explore further project details on Github here! https://lnkd.in/gt-ijkVJ #HealthcareAnalytics #MachineLearning #DataScience #CardiovascularHealth #SocialDeterminantsOfHealth
Humaira Ahmed
1mo
So proud of you Noreen!
Swiss Army Knife Full Stack Software Engineer, Python/Django instructor, avid listener & learner
1moI love that this is on your github page. Well-written and easy to understand. This will enhance your resume