Kaustav P. Shah, MD

Introduction

As AI and machine learning gain traction in healthcare, it’s crucial to understand which hospitals are adopting these tools, and which are not. In this interview, Dr. Kaustav P. Shah, fellow at the University of Pennsylvania and VA internist, shares findings from his national study on hospital AI use. He’ll present his work, “Hospital Usage of Machine Learning Tools in a Nationally Representative Survey”, at the SGIM 2025 Annual Meeting.

Q: What initially drew you to study AI and machine learning in hospitals, and why was this an important topic for you to explore?

I study innovation through technology in healthcare. I work on many projects in the AI and machine learning (ML) space including secondary data work like this project to understand trends in AI usage along with operational work at our health system implementing AI tools. This is an important study since it gives a national overview of how hospitals are using predictive machine learning tools and what types of hospitals may be early users.

Q: What do your findings reveal about disparities in AI/ML adoption across different hospital types and communities?

We found that hospitals that are rural or located in areas with higher social disadvantages were less likely to use predictive AI/ML tools. Hospitals with health system affiliation were much more likely to use AI/ML tools. This may be concerning since we know that there is often a digital divide in the adoption of new technologies in healthcare both on the patient and system level. Although, we don’t have a lot of rigorous data showing clinical benefit from AI/ML tools, as the technology improves, I think we will need to spend time and resources to make AI/ML adoption easier for lower-resourced hospitals.

Q: How does your approach differ from previous methods of tracking AI adoption, and why is that important for understanding the full picture?

The AHA survey and IT supplement is an annual survey with a high response rate across acute care hospitals in the United States. There is a lot of excitement and hype around AI in healthcare, however, it is hard to measure national usage since most of the AI/ML solutions don’t have reimbursement mechanisms that allow tracking through traditional methods like claims data. Other surveys are not as nationally representative, especially in reaching smaller and rural hospitals.  This data source is a great tool to help answer our research question on how acute care hospitals are using AI including some interesting follow up questions around whether the models are developed in-house or from vendors. There are limitations to this work as it doesn’t include ambulatory-only settings (where we see a lot of AI usage through ambient documentation).

Q: With AI tools spreading quickly, how can general internists and frontline clinicians stay informed, and cautious, about what’s truly helpful for patients?

First, I think focusing on the principles of good research design that we are all taught as general internists (understanding the intervention, sample population, control arm, inclusion and exclusion) is important when reading AI/ML research. Becoming familiar with some of the common ways that predictive machine learning tools are evaluated can be helpful since terminology is sometimes a little different than what we as clinicians are used to thinking of when we read a clinical trial for a new drug. There is not a lot of randomized trial data (although this is improving) but pushing for rigorous, randomized studies including clinically meaningful patient and clinician outcomes is important. Lastly, it is important to think about cost and implementation challenges of AI/ML tools especially with the differences in usage we saw in our study in hospitals in lower-resourced settings. 

Q: What do you hope to gain by presenting this work at SGIM’s 2025 Annual Meeting?

I am looking forward to presenting this work and hearing comments from clinicians, health system leaders, digital health vendors, and startups that will be at the 2025 SGIM annual meeting. There is a lot of excitement about AI/ML in healthcare and so I hope every clinician takes some time to think about how AI/ML may be helpful to augment their clinical practice and how we can reorient our current delivery system using these new technologies to help better serve our patients and communities.

Visit Dr. Shah’s poster at SGIM25!