The Largest Benefit for the Most People, an Interview with Dr. Nabile Safdar

What is your research background and your role here at Emory, specifically for AI.Health?

I'm a radiologist and a researcher. I've been studying the intersection between AI and imaging for several years, back when AI was called computer vision! I have an interest in clinical decision support, responsible AI, quality improvement and digital transformation. In my role in the Department of Radiology, my research has been to help create an environment where models, tools and algorithms that are developed in biomedical informatics or biomedical engineering find pathways into clinical spaces. 

How do you incorporate AI into the projects that you help spearhead?

A lot of our projects in the Department of Radiology are about taking the AI that we've deployed clinically and trying to figure out how it's doing from a performance perspective. We are looking to make sure that we have paired the right AI tool in the right clinical space. We are also checking that the appropriate guardrails and guidelines are in place to make sure that we're doing things safely. 

A recent project that I collaborated on involved using AI tools to detect intracranial hemorrhage. We have tools that detect blood clots in the lung or pulmonary embolism. A team and I are also looking closely at generative AI tools that make the job of reporting for radiologists easier. We are looking at AI tools that can summarize their reports or create impressions more easily. Then we have plans to expand those tools going forward. 

Within your role as Chief Artificial Intelligence Officer, what are you looking for in an AI program that would be beneficial to both clinicians and patients? What sways your decisions? 

There are literally dozens of AI efforts going on all around the system at once, and what we're trying to do is match those up to the needs of the healthcare system, providers, patients and our staff. We really want to make sure that we're spending our time on those solutions that have the most bang for the buck. One of our successes has been our AmbientAI Scribe program led by Rachel Silverman. The program has really saved doctors' time because it's listening to clinic visits and automatically creating notes, which helps them reclaim their evening time. We're expanding it to the emergency department and for nurses in hospitals, which has been huge!

We're looking closely at the suite of AI tools that our electronic medical record offers and seeing which ones need to go first. For example, there is one tool that should be deployed soon that actually reads reports and says, “Oh, this was a recommendation,” then helps us keep track of those recommendations. So if a patient was recommended to get a follow-up CT scan because there was a lung nodule, it will read it, keep track and remind the provider to order additional testing. For patient-facing systems, there has been an effort to use voice AI agents to call them after they've been discharged from the hospital to remind them to pick up their medication or walk them through measuring their own blood pressure for those who are high-risk. That's been an amazing and very interesting effort. 

A third area of focus involves helping the healthcare system operate more efficiently. These would be things like billing, revenue cycle, access and scheduling. We don’t want to ignore our research community, as they are interested in a couple of things. One is making sure that they have the necessary resources when it comes to computing and accessing the data cloud to develop new AI tools. In my role as Chief AI Officer, I'm trying to maintain several commitments – kind of acting like air traffic control for hundreds of efforts and making sure that we’re focusing on the ones that bring the most benefit to the most people. It's not really about the tools; it's about the benefit to people. It's about the outcomes for humans. 

What misconceptions do you believe people have regarding incorporating AI into the sphere of healthcare? How do you work within your role as Chief Information Officer to dispel them?

I'd say one of the major misconceptions is that AI transformation is all about technology. It is equal parts readiness of the teams and the humans who are using that technology.

To use it, they've got to be upskilled and develop their AI literacy and sometimes work on their workflows because this technology fits into their existing workflows. But sometimes it demands that we work differently. Like with any new technology, there are limitations where it can make mistakes, where clinicians will need to challenge its decisions. There is a general rule that in most healthcare uses of AI there’s a human in the loop. Very rarely is the AI going to make a clinical decision for a patient without a nurse, doctor or pharmacist reviewing it. 

What we're planning on doing as we roll out new AI tools is try to make sure there's good training and education for the people who are using it. We are planning on doing more AI literacy for everybody, where we explain how to use the tool, what to expect and what pitfalls to avoid.

 One other misconception that I'll mention is that it's always safe to use every AI tool. If you're dealing with patient information, it is not considered appropriate or safe to dump that patient information into ChatGPT and ask it questions for a couple of reasons, but one of them is patient privacy and the security of their data. If I were to input a patient's name and their information, I'm essentially giving it to a company when I do that. So, one of the things that we've done is we've secured an AI tool that is for use by folks in healthcare.

[AI.Health received a sneak peek at a tool that Dr. Safdar and his team have created to address this concern. The platform conjoins various AI tools, such as OpenAI, Gemini and Anthropic, to assist healthcare providers with next-step recommendations, record keeping, data sorting and more. The platform has not yet been rolled out to the Emory University community, but we are waiting in excitement!]  

What is a research topic that you are currently working on, and what is it that excites you about it?

Currently, I have been working with a team of researchers here at Emory to write about the concerns of AI. We’re writing about the environmental impact because that is one of the major concerns about the rapid adoption of AI. We are asking about the amount of water usage,  how much electricity it uses and what is the land use impact on communities that are near these data centers. That's been a really eye-opening experience for me. I think that's an area that is important to acknowledge. We have to be thoughtful about the impact that the use of AI has on our communities and on the de-skilling of our students or professionals.