All over the United States, reducing the rate of infant mortality has been an ongoing struggle. However, over the past several years in our own home state of Indiana, we’ve seen a determined effort on the part of healthcare providers to address this tragic state of affairs. The good news? This effort that has begun to pay off, as Indiana recently recorded its lowest-ever infant mortality rate.
In our work with the state and with a local Indiana healthcare system client, expanding data analysis to include social and economic data characteristics has helped advance this effort. At the recent Innovation Digi 2020 conference held November 5 in Indiana, our Data Science Team Lead Amrutha Wheeler detailed how this shift in focus and capability is helping save lives.
Moving past clinical data to real answers
“Clinical data alone is not sufficient,” Wheeler told attendees. “External data sources contain valuable information around social determinants of health—features like a mother’s medical history combined with their education level, access to resources, support system, and so on—help us better understand patient risk and predict patient outcomes.”
Until now, however, the value of these social determinants of health (SDOH) datasets hasn’t been fully explored. But with this important and ongoing work, things are coming into focus. Working to identify and integrate clinical and SDOH datasets allows for the building of more complete patient profiles, which in turn is leading to a better understanding of risk and the development of tailored treatment programs that address it.
Finding the social features that affect infant mortality
To identify the main social determinate drivers linked to infant mortality, our team closely examined clinical data and deeply explored individual patient cases to develop a hypothesis that would end up guiding our future analysis and ideation around a solution. When the team determined the 51 features most likely to influence infant mortality, they then mapped those features to available information sources and datasets.
Models were then developed to analyze individual factors, such as the number of prenatal visits and whether the mother had underlying conditions such as diabetes or hypertension. However, the results of this analysis led to further questions. For example, diabetes had a negative impact on infant mortality, which further inquiry determined was correlated with the number of prenatal visits. Within the sample, mothers with diabetes went to the doctor more, and a higher number of prenatal visits correlated with a decreased risk of infant mortality.
After isolating the clinical factors that had the most impact on infant mortality, these social determinants added another layer of insight. For example, language, accessibility to healthcare, and education level were the significant social drivers of a mother’s maternal health, which is directly related to an infant’s mortality.
Leveraging relationships to improve health outcomes
The data science capabilities were essential to the success of the project, Wheeler explained. But more so were the relationships the Resultant team has developed and cultivated across both the commercial and public sector healthcare organizations.
“We didn’t just focus on easily available SDOH datasets,” she said. “We worked with national, regional, and local partners in government, nonprofit, and commercial spaces to access data that is far more granular and lends itself to richer analysis.”
This same approach can be applied to other health concerns, “without recreating the wheel each time,” Wheeler said. “We leveraged generalizable predictive machine learning models that can be tailored to specific health outcomes, reducing time to value and increasing confidence in the outputs by clinical stakeholders.”
Find out more about our work on infant mortality here.
