“Where there’s variability, there’s hope for action.”
I heard this summing-up in a recent interview with Dr. John Showalter, chief product officer for Jvion, a prescriptive AI analytics company in health care. His point is that we need to compare two or more populations to find out what makes one health and another unhealthy.
Jvion spent its first several years predicting events just two to four weeks in the future. Their data analytics helped health care providers and payers plan when the emergency room would be crowded, or how many people with heart or lung disease would come in for treatment.
Today, Jvion extends its predictions far into the future, branching out into public health issues with great consequence. Already, the company has found that air quality—particularly a concern with the repeated fires in the Western U.S—and work-related stress have a greater impact on both mental and physical health than the health care field had thought
The company recently released a Behavioral Health Vulnerability Map that reports the risk in each community of hospitalizations and self-harm events related to behavioral health, down to level of a census block group. The map determines vulnerabilities by applying Jvion’s clinical AI to hundreds of publicly available data points and displaying them on maps provide by Microsoft on its Azure cloud platform. Jvion also offers a COVID-19 map.
A mission-driven company, Jvion is committed to offering these AI driven insights to the public. To allow public use of the analytical results, they have moved away from clinical data provided from clients, and are generating all their results from the combination of publicly available data and insights from AI.
What does this data consist of? Certainly, the standard public health data released by the CDC. But also environmental data such as air and water quality. Also census data.
Dr. Showalter told me that environmental data was not consistently and universally collected. So some interpolation is required to generate results for some counties based on data reported from neighboring counties.
Despite the political tensions around the 2020 census, Dr. Showalter says it has a lot of new, more precise data. The census asked very specific questions on topics of interest to health care officials, such as how a person gets to work.
Using supervised and unsupervised machine learning, Jvion so far has been able to analyze more than 10% of the U.S. population. The factors they identify can, in some cases, account for more than 98% of the variation in health care between two areas.
For instance, drilling down into different communities, they attributed a high rate of behavioral problems in one county to the presence of an army base, where binge drinking and mental health problems are prevalent. In another case, the high rate was traced to a community that arrived from Vietnam after the war and that resists getting help from medical professionals for behavioral problems. In a third case, the culprit was low English proficiency among the population, a problem correlated in the research literature with lack of access to mental health treatment.
They also introduced new levels of detail into the discussion, such as the different types of particulates in the air and their different impacts on health.
As their organization evolves, Jvion hopes to form partnerships with public health officials, advocates for policy change, and others who can use the data they’re generating. Statistics still speak loudly, even in highly politicized environments, and it would be valuable for social policy advocates to know better what health conditions they’re dealing with.
