As the COVID-19 pandemic has made clear, quality medical care is unevenly distributed throughout the United States. The Centers for Disease Control and Prevention notes, “People from some racial and ethnic minority groups face multiple barriers to accessing healthcare. Issues such as lack of insurance, transportation, childcare, or ability to take time off of work can make it hard to go to the doctor.”1 Cultural differences and language barriers can also play a role, contributing to a situation where care may vary widely from one Zip code to another.
Although not traditionally thought of as a frontline specialty, radiologists have an opportunity to close these care gaps. Farouk Dako, MD, MPH, assistant professor of radiology in the cardiothoracic imaging division of the Perelman School of Medicine at the University of Pennsylvania, is on a mission to do just that. Dako sees both real-time and predictive analytics as two key tools in radiology’s arsenal when confronting healthcare disparities.2 By using such tools, in combination with straightforward human interactions in underserved communities, radiologists can combine imaging diagnoses with historical and real-time data to produce helpful predictions through the use of AI algorithms — and pave the way for an enhanced quality of life for patients. During a recent interview with the Bulletin, Dako spoke about the importance of harnessing data to survey and triage patient populations and connect them with needed care.
What are data analytics and predictive analytics and how do they relate to radiology?
Data analytics is simply the process of analyzing data to obtain useful information and to guide decisions. All radiologists analyze imaging data but there’s also a lot we can do with the non-imaging data we accumulate. For instance, practices often determine their staffing levels based on imaging type and volume.
But there are applications for data analysis beyond just staffing. For instance, by putting this data into a structured format we can detect patterns that give us a fuller picture of the circumstances confronting either an individual patient or a patient population. This gets us into the territory of healthcare disparities; we can look at how different types of imaging modalities are used by different racial or ethnic groups, for example, and determine if there are inequities in utilization.In contrast to descriptive analytics, predictive analytics involves looking at the same kind of data to predict future unknowns. On the health system level, it’s within the scope of predictive analytics to say something like, “Based on a given pattern we’re seeing on Thursdays, we anticipate 20% more cases than we think one radiologist should safely read. We can then create interventions to improve patient safety.”
On a population level, we can similarly combine imaging and non-imaging data to predict health outcomes. For example, there are lung cancer risk prediction models for nodules identified on CT combining imaging features with non-imaging data typically in the EHR. There is ongoing research into AI-based risk prediction models that, for example, can use chest X-rays to predict future cardiovascular risk.
Taking these techniques to the next level, we could combine this imaging with non-imaging data to identify high-risk patients and intervene to improve their chances at living a healthy life.
“ I’ve also found that going out into the community over time and building trust is key to welcoming folks into the healthcare system.”
Do you think that AI could have negative repercussions in low-resource communities?
Underserved areas that encounter challenges in accessing AI will be at a disadvantage compared to those areas that can more readily and effectively access it. But apart from that, even in places that use AI, it can become a problem if they don’t have enough clinical supervision to ensure that the technology is being used for the correct indications. We’re only beginning to uncover these kinds of issues as we do more research and ask forward-thinking questions about the future of AI. Because of this, we need to be vigilant as we try to introduce it into different clinical scenarios.
And this technological disparity might one day exist not just within the United States, but internationally as well between wealthier and poorer countries. I’m from Nigeria, and I consult with Nigerian radiologists as part of my global health work as a program director with RAD-AID International. Consequently, I hear these radiologists’ concerns. One major question they have is this: If AI comes to Nigeria, who’s going to own it?
There aren’t policies in place right now in Nigeria to ensure that the appropriate people will control AI in the clinical setting. The concern is that because of a trend where capable physicians often leave the country to pursue a life somewhere else, individuals with inadequate medical training will begin to fill that void. This could inadvertently result in worse health outcomes for patients than in places where more stringent protocols are in place.
Missed care opportunities are often the result of suboptimal social determinants of health. Can employing predictive analytics help resolve these barriers to quality care?
Yes, predictive analytics can help minimize missed care opportunities. But one of the major hurdles to doing so can involve getting access to the relevant data. This is key, because to be able to understand which barriers to care exist so that we can surmount them, we have to gather data that indicates when patients are being scheduled, who’s not showing up, where they live, what their patient satisfaction survey results are, and so on. We need to have patient representatives involved in understanding the problems and designing solutions to ensure a patient-centered approach.
To convince gatekeepers that radiologists should have access to this kind of data, we have to go beyond simply stating that shoring up missed care opportunities is the right thing to do. We have to speak the C-suite’s language and demonstrate the financial cost of missed care opportunities. We can take the lead to show executive leadership the revenue that’s to be gained by keeping a CT scanner running at optimal levels by highlighting the number of patients showing up for their appointments.
What are some practical steps a radiology practice can take to improve the health outcomes of underserved communities?
One thing I’ve realized by going out into the community is that misunderstandings exist. For instance, some folks think that imaging care will cost too much. Others may just not know how to engage with the healthcare system. And it’s not just community members who feel this way. I’m a physician, and I engaged with the healthcare system as an adult patient for the first time recently when I got my job as an attending and I got my insurance card. I didn’t know what to do next. If this wasn’t clear to me, then how are non-physicians supposed to know where to begin? It’s a real challenge, especially if you’re experiencing other social or economic challenges that take precedent over seeking healthcare.
Going out into the community over time and building trust is key to welcoming folks into the healthcare system. Many people probably have preconceived notions about why you’re in their community in the first place, so it’s important to let them know why you care about their health. A big part of success is just showing up, and that’s something all radiologists can do.
Most importantly, we should talk to community leaders to figure out why people aren’t using the healthcare system the way we think they should. Ultimately, we may realize that the system isn’t working for them and we can begin to think about alternate approaches.