There’s a business intelligence group in every health system. Its purpose is often both broad and vague, something like, “increase revenue and patient safety while complying with all regulations and standards.”
These BI groups are called upon when a health system is facing a particularly thorny challenge such as a new shared-risk payer contract in place that requires careful risk mitigation, among other possible examples.
In the business world, BI is typically done around a product: how many downloads, how many times did a user click the red button, etc. In healthcare, however, BI is typically done at the highest level only: the number of lives saved, dollars spent.
Healthcare BI teams spend time segmenting the patient population, then tracking the outcomes of these segments of patients. The problem, however, is that this top-down approach often misses the intricacies of interventions in the middle. There’s often too many confounding factors: which intervention actually worked?
This linking of patients to interventions and, finally, to outcomes, requires that BI teams spend much more time deep in the details. This linked understanding requires the same amount of rigor that AI initiatives need in order to be successful, safe and effective. And this understanding happens only through careful engineering.
At Penn Medicine, we’ve created integrated product teams around our AI applications. These product teams consist of data scientists, physicians, and software engineers, just to name a few.
This year we’ve begun to include a BI analyst as well, to help us make better design decisions such that the linking of patients to interventions and outcomes is feasible and easily reported via dashboards and reports.
In our first joint program, we deployed a machine learning application to better find incomplete patient records. Our pilot showed an immediate improvement, but those gains started to diminish after deployment.
Because of the level of engineering that went into our BI dashboard, we had insight into what was happening in those middle steps, and were able to quickly zero in on the issue and make corrections.
Business intelligence in healthcare is about making the right decisions. Data science in healthcare is about providing insights that allow for better decision making. Health systems that take advantage of the natural alignment of these two disciplines will likely see better outcomes, faster.
Mike Draugelis is chief data scientist at Penn Medicine.
