Data is (still) Good...

Harnessing the power of AI, ethically, is essential to improving our healthcare system.

?Data’s gotten a bad rap lately. Or more precisely, consumer data. And oftentimes with good reason. Advocates, lawmakers, and individual consumers have become increasingly concerned about the amount of data that’s being collected about them and how it’s being used. This friction has led to a well-publicized feud between Apple and Facebook (and their iconic CEOs, Tim Cook and Mark Zuckerberg) about data portability across apps, a flurry of fines around the world, and even an executive order from President Biden.

?Less talked about these days is data for good. A call to action as much as a movement, data for good aims to address global humanitarian issues around poverty, human rights, and the environment. Global health and healthcare is certainly on that list, and it’s what we’re focused on at Change Healthcare.

?

With Great Data Comes Great Responsibility

?In the age of digital healthcare, there is more data about patients than ever before. The question is: How is this data properly used, if at all? With access to patient information—even if it is fully HIPAA-compliant, de-identified data—comes great responsibility. It’s our job as technology practitioners to use technology data and analytics to improve healthcare outcomes and economics while upholding privacy, confidentiality, and the security of data at all costs to ensure we act responsibly.

?I define data responsibility in two ways: The first is the data version of the Hippocratic oath. Principally, this means protecting individual health data, creating platforms that guarantee regulatory compliance in real time, and ensuring ethical use. While these capabilities in this area are a point of pride and differentiation, they’re just one part of our broader responsibilities. Just as “First, do no harm” is a low bar for doctors, we have to protect the data of those we serve.

?Because we have HIPAA-compliant, de-identified patient-level data, we can provide instant scale to researchers who are, for example, studying the variations of care based on ethnicity, race, socioeconomic status, or geography. These social determinants of health (SDoH) can be difficult to measure for individual institutions because of a lack of scale. Any single provider, institution, or health system will never have enough data, with enough heterogeneity, to provide a holistic view into populations. But with more data points available, we can provide instant scale to allow researchers to home in on exactly the answers they seek.

?Critically, we are also pulling in nonclinical data. Clinical data is, by definition, self-selecting; we’ll always have more information from people who have higher utilization rates, and the way they interact with providers can skew our view as well. Often, these data tell us the symptom but not the cause. For example, if a certain population group in a specific location has relatively high ER-utilization rates, it doesn’t tell us why that is. Data on socioeconomic factors, health and financial literacy, and other nonclinical inputs can. After all, 80% of SDoH come from our lives outside the doctor’s office. So, we need a window to see the world outside the clinical setting.

?We’re looking for ways our data and AI can help researchers better understand how SDoH affects outcomes, and therefore, how we might address these inequities in healthcare. That’s a broader responsibility that goes far beyond simply protecting privacy. It’s a responsibility to use our technologies to improve our healthcare system as a whole and patient care specifically.

?

Addressing Healthcare Inequities with Data and AI

?An example of using data for good is the work that Michael Pencina and Laine Thomas are engaged in at Duke University School of Medicine. They’ve been studying the dramatic differences in COVID-19 acuity and quality of care experienced by varied demographic cohorts. With Change Healthcare data, they didn’t have to manually gather data sets from multiple sources. Rather, they used our data and built models on top of it. It saved them months. Amid a public health crisis, this time savings is vital. Drs. Pencina and Thomas were also able to benchmark their own clinical data from the Duke Health System and benchmark it against national numbers. This allows providers an instant, broad view of the efficacy of their treatments. While continuing to study COVID-19 outcomes by population segments, they’re also planning further studies on subjects—such as maternal morbidity—where variations on outcomes are known but poorly understood.

?

?AI Isn’t Bottom-Up, and It’s Only as Good as Your Data

?AI models are only as powerful as the data they depend on. The old saying, “garbage in, garbage out” applies here. AI will never be bottom-up. Providing AI tools to individual providers—or even large health systems—will never provide enough data, or the right data, to allow AI to do what it does best. The sample sizes are just too small.

?One way to think about this is comparing AI with trillions of data points to an individual doctor. When presented with a case, they’re basically doing the same thing: parsing the data available to them to come up with a recommended solution. For doctors, they’re mining their education, their experience, the experience of the colleagues they’re consulting with, and whatever clinical information they have access to. Imagine being able to do that at scale, analyzing millions of actual medical records in seconds. AI simply supercharges this process for the doctor; it’s just analyzing a much, much larger data set. Exponentially larger, in fact, and constantly updated. This empowers the doctor and ultimately benefits the patient.

?It’s the same for a hospital or a health system. AI can certainly be helpful in analyzing the data they have on hand, but that information is a fraction of a fraction of all of the clinical data out there and does not include the important SDoH data that can be culled from nonclinical data.

?

Better Data, Better Healthcare, Better Outcomes

?By applying AI to huge data sets responsibly, we have the chance to transform our healthcare system. And it isn’t just with the SDoH data that I’ve been discussing in this post but also in medical imaging, at the point of care, with revenue cycle management to lower costs, and on and on. In our current environment, individuals are rightfully concerned about how their data is collected and used. But collected, managed, and leveraged responsibly, data—when coupled with AI—can have a transformative, positive effect on healthcare. This, in turn, helps individuals, providing better healthcare outcomes for every patient.