Understanding the social and environmental underpinnings of health and disease is imperative to reduce known health care disparities and manage existing disease burden within the Type 2 Diabetic population. While the importance of factors such as diet or access to health services and medications are well recognized broadly in the literature, explorations with respect to quantifying the exact degree and contours of diabetes specific Social Determinants of Health (SDOH) remains nascent. Understanding how the factors that make up SDOH impact such as time to diagnosis, time to initial treatment, persistency to prescribed treatment among other factors, is paramount for long-term improvement of health outcomes.

To address this gap in the knowledge base, Symphony Health partnered with HealthWise Data to evaluate how SDOH factors combined with medical, and pharmacy claims-based features may impact the persistency of diabetic patients being actively treated. 

A key drivers analysis was conducted to identify statistically significant features that were either barriers or influencers of patient persistency.

The foundation of the data used is Symphony Health’s Integrated Dataverse (IDV®), linked to and combined with HealthWise Data’s HealthWise360 Universe, using Symphony Health’s proprietary Synoma® tokenisation. 

Supervised machine learning was utilised to identify key features within Symphony’s IDV® and HealthWise Data classifying patients as more or less likely to be persistent three (3) months after starting a new treatment. 

The most influential features identified via the machine learning models (i.e., XGBoost and Logistic Regression) were used to develop qualitative profiles of patients more likely to be persistent beyond 3 months, and less likely to be persistent at or less than three (3) months.

Many social and environmental factors, including food insecurity, healthy lifestyles, and preventive services were observed to directly impact a patient’s likelihood of remaining persistent or discontinuing treatment. Features derived from clinical / healthcare utilisation data highlight profiles of patients that tend to have higher levels of disease burden, morbidly obese with concomitant medications versus those without. These study results demonstrate the interrelatedness of clinical / healthcare claims data and SDOH data and how, when integrated, a better understanding of patient population dynamics can be developed. What factors can be impacted through intervention and those that cannot be overcome.