Joshua Liu, MD

AI Delivers Rapid, Clear Interpretation of Newborn Test Results

My son was 11 days old and his blood test came back "abnormal". In the past, I would've spent six hours sick with worry waiting to hear from his doctor. Instead, AI got me the answer first - and explained it even better. Our baby was looking a bit jaundiced, so our pediatrician sent us to get his bilirubin levels tested and determine if he would need phototherapy. He told us he’d message us after he looked at the results, regardless of the outcome. We got the results in our patient portal by noon yesterday. All three lab values were flagged as “abnormal”. What did I do? Fed the results into Gemini, OpenEvidence and DoximityGPT. Here’s how they performed: → Gemini said the bilirubin levels were well below the threshold for needing phototherapy and gave the most comprehensive and consumer-friendly answer - what the test is for, what the values mean, potential causes of jaundice, suggestions on what to discuss with our doctor → DoximityGPT said the bilirubin levels were well below the threshold for needing phototherapy provided the best clinician-friendly answer (as if I were the baby’s physician) - referencing American Academic of Pediatric phototherapy guidelines, discussing additional risk factors that could influence decision for phototherapy, recommended management → OpenEvidence said there wasn’t enough information and prompted me for more (e.g. baby’s gestational age at birth) before it could give the same recommendation (although props to them for asking for more context). Triangulating the info from all three gave me a pretty good sense of what was going on. With Gemini in particular I had fun interacting with it to refresh my memory on jaundice, interpreting bilirubin levels, etc. While I still wanted confirmation from the pediatrician, I felt pretty reassured from my back and forth with these AI tools. Six hours later, I got a message from the pediatrician confirming that phototherapy wasn’t needed - but that was basically it. Unlike the AI tools, there was no explanation of the results. Nothing about how close this was to the threshold. Nothing about what to keep an eye out for. Now I was appreciative he even closed the loop and didn’t require a visit for it - I know how busy docs are, I didn’t expect an essay. Having a human pediatrician confirm things was still meaningful for us. But this is upside for AI - being able to fill in all the gaps with patient education that the physician won’t have the time for. The question is - is there some universal prompt or framework that would make AI explanations of test results actually useful? With Epic releasing its Emmi AI to help explain test results in MyChart, and Amazon releasing its Health AI agent to do the same - I guess we’ll find out. Because as much as I was able to use these AI tools to help us understand our baby’s result six hours faster and more comprehensively - I have to remind myself that my wife and I (who are both clinical) are in the minority of patients who can use AI to interpret test results reasonably well and safely.