Data Science and SDoH Data Fuels COVID-19 Research
Historically, medical research and analysis has depended on clinical trial data, which lacks important information related to lifestyle, health literacy, and other social determinants of health (SDoH). Including SDoH insights in data analysis can provide a more detailed view of individuals’ real-world experiences.
In this case study, learn how researchers at the Duke University School of Medicine are using Data Science as a Service (DSaaS) to explore the efficacy of various COVID-19 treatment regimens and identify socioeconomic disparities in care.
Rich healthcare data from 140 million patients gives Duke’s researchers a granular view of multiple data points, including the medications used to treat acute patients, length of inpatient stays, and time from admission to treatment. These insights help researchers uncover opportunities to improve care, particularly for at-risk populations.
Duke University School of Medicine—one of the nation’s leading institutions for medical education, clinical care, and biomedical research. Duke Med is the third-ranked medical school for research nationally.1
Duke University School of Medicine researchers sought to explore the efficacy of various COVID-19 treatment regimens during the pandemic. This vital research project intends to provide statistical evidence to guide care recommendations. With the right data, it would also allow the researchers to benchmark Duke University Medical Center’s outcomes not only internally but also to thousands of COVID-19 cases nationally. The opportunity is to provide insight on best practices with a particular focus on socioeconomically driven disparities in care.
Historically, research and analysis have depended on clinical trials. While valuable, these trials are often relatively small, especially when examining specific subgroups. Additionally, this data–while providing important information gleaned in the clinical setting– frequently lacks important information related to lifestyle, health literacy, and other social determinants of health (SDoH) which can dramatically impact health outcomes. The result is an incomplete, limited view of what treatments are working and for whom.
"Clinical trials consistently under-enroll diverse populations, resulting in relatively small sample sizes for important subgroups,” explains Michael Pencina, Ph.D., vice dean for data science and information technology at Duke University School of Medicine and director of Duke AI Health. “We need to understand who the therapies work for and in what setting. Clinical trial data is a key starting point but is just not big enough if we want to examine treatments in diverse subgroups and understand what works best for whom.”
Laine Thomas, Ph.D., associate professor of biostatistics and bioinformatics at the Duke Clinical Research Institute, added: “We’re interested in whether people that are in a high-resourced healthcare environment–people that are in the high end of the economic distribution–may have access to better, quicker healthcare. You can hypothesize that, at the high end of the income distribution, there’s generally more access to healthcare. And we can then investigate whether that changes the experience–what condition a person is in when they present with COVID, what hospital they present to– that would be a marker for a lot of differences in the community.”
Duke researchers are using Change Healthcare’s Data Science as a Service (DSaaS) to access voluminous, deidentified claims data culled from millions of healthcare transactions. Beyond the clinical data available in DSaaS, it also includes deep, rich, SDoH insights on race and ethnicity, economic stability, and health literacy, which dramatically expanded Duke’s ability to determine how well various treatments were working and whether there was variation in efficacy among different groups of patients. Change Healthcare’s DSaaS is provided in a secure, hosted environment, ensuring regulatory compliance and ethical use at every step in the process. This saves vital time by eliminating the need for separate, timeconsuming compliance reviews.
Leveraging healthcare data from 140 million patients, Change Healthcare’s DSaaS gave Dr. Thomas and her team a granular view of multiple data points, including the medications being used to treat acute COVID-19 patients, length of inpatient stays, time from admission to treatment, and more. The data provided also allowed researchers to adjust their models in context. That is, they had access to detailed (anonymized) histories, allowing for further refinement of their findings.
In addition, SDoH data was layered into the analysis, providing a detailed view on individuals’ realworld experiences. Duke investigators found that treatments and outcomes varied not just by basic demographic factors such as age and gender but also by economic stability, race, and ethnicity. These insights allow researchers to identify disparities in care and treatments and to uncover opportunities to improve care, particularly for at-risk populations.
Gaining a snapshot of the effects of SDoH on treatment of a specific illness, such as COVID-19, is powerful, but it only hints at the remarkable value that can come from tracking utilization and acuity trends over time. DSaaS has the potential to allow researchers–whatever their focus–to unlock key insights that inspire meaningful innovation that improves patient care.
Looking ahead, the Duke researchers plan to use this data to investigate how COVID-19 affects various subgroups, which could help identify interventions that aren’t obvious on the surface.
Furthermore, these same data sets can be used to perform modeling on other public health challenges. Thomas used the example of maternal health and morbidity; it’s known that different populations have different levels of maternal morbidity, but to truly address health disparities, a more granular view may provide new insights. Using DSaaS and Change Healthcare’s SDoH data will also allow researchers to measure their own institution’s performance against national benchmarks.
Pencina pointed out that DSaaS will greatly accelerate their ability to run new studies and models. Rather than the painstaking process of gathering multiple data sets from multiple locations, then running compliance on each, DSaaS allows researchers to access the data needed from one source and then build new models. This could translate into significant savings; Thomas says processes that traditionally have taken 24 months may now be completed in as few as six months.
Pencina and Thomas expect that using data within DSaaS will translate into better insights, increased innovation, and improved outcomes.
1 U.S. News & World Report, usnews.com/best-graduate-schools/top-medical-schools/research-rankings