Interoperability 2.0 – Creating a Holistic Model of Care Part 2
AUTHORS: Eric Crawford, Director of Product Management, and Adam Bell, Senior Director, Advisory Services Department
Part 2: Contextualized Population Profiles
The first step in creating a holistic model of care is to begin with the end in mind. This includes bringing together multiple data sources from across the entire community of care to comprehensively understand the underlying dynamics of the population.
Next, we apply technology and data science to mesh data together and create insight. This is where the magic happens. Opportunities to impact the care delivery for populations surface as we peel back layers in the data.
Methodology Matters MORE GRAPHICS:
Following a robust methodology ensures that the insights are actionable and applicable. In other words, we want to have valid correlations between social determinants of health and actual clinical outcomes. We used the following steps to guide the process:
- Select a representative outcome variable
- Define and normalize measures
- Correlate variance in outcomes with variance in SDoH factors
- Choose the most significant factors
- Calculate a representative “burden” score
Our team focused on diabetes prevalence as the primary outcome variable. As expected, we found a high degree of variance across the United States. We created a graph data model to link that variation to selected public data sets representing social determinants of health across 5 categories. When necessary, we used population count as a normalizer for factors that crossed geographic boundaries. We created a variety of measures including:
- Proportion of population living in a food desert
- Hospital quality of care
- Percentage of population living in poverty
- Housing affordability
Our team leveraged a multiple linear regression model to isolate the most significant SDoH factors in diabetes prevalence. From there, we standardized each factor and calculated a combined “burden score” of the SDoH factors. A higher burden score indicates there are more negative SDoH factors present in the patient’s zip code, county, or state.