How a doctor with no coding background built his own bedside AI tools [PODCAST]

Listen and followApple PodcastsSpotifyYouTubeMore apps · Browse all episodes

At 2 a.m., a surgeon needed an answer fast and had no one to ask. So he built the tool himself, with no coding background. In this episode, consultant cardiac anesthesiologist Ahmed Elsonbaty explains how a single bleeding-patient case in the operating room pushed him to create bedside clinical decision-support tools now used by more than 10,000 doctors across 150 countries. This episode is based on his article “I built clinical decision-support tools at the bedside,” published on KevinMD. You will hear why he deliberately avoided generative AI so the tools give the same traceable answer every time, how he built everything on published, documented sources, and why he insists the tools help the doctor rather than replace anyone. He also walks through submitting his work for peer review at a major journal. Hear how he did it, and the exact steps he says any physician can follow to build their own.

Partner with me on the KevinMD platform. With over three million monthly readers and half a million social media followers, I give you direct access to the doctors and patients who matter most. Whether you need a sponsored article, email campaign, video interview, or a spot right here on the podcast, I offer the trusted space your brand deserves to be heard. Let’s work together to tell your story.

PARTNER WITH KEVINMD → https://kevinmd.com/influencer

SUBSCRIBE TO THE PODCAST → https://www.kevinmd.com/podcast

Transcript

Kevin Pho: Hi, and welcome to the show. Subscribe at KevinMD.com/podcast. Today, we welcome Ahmed Elsonbaty. He’s a cardiac anesthesiologist from Cairo, Egypt. Today’s KevinMD article is “I built clinical decision support tools at the bedside.” Ahmed, welcome to the show.

Ahmed Elsonbaty: Thank you very much, Kevin. I’m so happy to meet you today.

Kevin Pho: All right. So let’s briefly share your story, and then talk about why you decided to write this article on KevinMD.

Ahmed Elsonbaty: Look, it’s a short story, but it’s impressive to me. I never thought before that I would enter the software world and build AI tools, something like that. I always heard about it, but I did not think I would go inside this world until a certain moment: 2:00 a.m. in the OR with a bleeding patient, cardiac surgery, and the junior doctor with me brought me a piece of paper with the results of a bleeding test and asked me, “What can we do?”

At this point, I started thinking, “What if I was not there?” You know that in anesthesia practice, seconds may affect the patient’s life, so I have to decide in seconds. So I told myself, “I have to make something to help these junior doctors, to help doctors when there is no one else with them.” This is the spark of the idea.

Kevin Pho: All right. So before having these tools, how would you help these junior doctors before this technology was available?

Ahmed Elsonbaty: Before that, the traditional way was learning. He asks me, and sometimes I have to teach him. But the problem is that now life is faster, cases have become more complicated, occupied, busy, so there are no chances like before for reading books at night and teaching junior doctors in the morning.

So all of these things pushed me: There should be a solution for this situation. How can we make these doctors able to learn? When you are in a far town, a small town, Manila, Lagos, any countries with few anesthesia doctors who cannot help each other, what solution is there for that? This is the spark for these tools.

Kevin Pho: Just tell us what kind of options are available to you to create some of these tools.

Ahmed Elsonbaty: Look, at the beginning of the road, I think it’s very difficult. I have to learn coding. I have to learn Python. I have to have an engineer to build my ideas. But what I found, actually, is that once you want to solve your problem, you have to solve it. The way is not so hard. You just put discipline, decide what the problem is that you have and how to solve it. And I discovered that building with doctors is better than with companies or other people who did not suffer the problems as we did. So decide and start, and you will find the way.

Kevin Pho: So give us an example of how some of these tools are helping your residents. What does that look like?

Ahmed Elsonbaty: Look, Dr. Kevin, I started with a small tool, a small tool just reading the ROTEM. I don’t know if everybody knows this. The ROTEM is an apparatus made to decide the cause of bleeding in the patient. I started with this and then expanded. But I started with only one tool for ROTEM, to analyze the results and decide what you have to do with the patient.

The problem for me was that when I searched other applications, there are marvelous applications, many applications, but I wanted to make a colleague for the doctor that never leaves him at all. How can I get that? I thought that I have to adapt the plan of management according to this patient, with these comorbidities, with these investigations. A very individualized plan for the patient. I started with only a small part, and then expanded, expanded to cover all of anesthesia: how to manage any anesthesia plan, ventilation, the ROTEM, TEG, ECMO, TEE.

The other point that is very important to me is that I did not depend on generative AI. If you ask the application the same question 100 times, the same answer will come, because every piece of information in this application is traceable and depends on published papers. And the sources of the application are inside the application. You can check them properly. It is not my own data.

Kevin Pho: So you’re telling me that if someone asks this program 100 times, it has to give the same answer 100 times.

Ahmed Elsonbaty: Yes. The answers and the information, they have to be consistent.

Kevin Pho: And you’re right, sometimes with generative AI, with the hallucinations, it can give different answers for the same questions, right?

Ahmed Elsonbaty: No, no. The same answer. Yes.

Kevin Pho: So tell me, what type of program did you use to code your bedside tools in order to give the consistent answers? What did you use?

Ahmed Elsonbaty: I used the AI tool only to, what you can say, draw the skeleton. But filling the skeleton is my job: to bring sources, published sources, documented sources, very popular sources that I depend on to give the doctor the opinion. I insist that my tool does not replace the doctors. It just helps the doctor. It gives him an opinion, gives him help with the decision. It will not replace doctors.

Kevin Pho: And the knowledge base, the knowledge base that you give the AI tools, you curate the knowledge, you give it the sources, you give it the journals that it draws from.

Ahmed Elsonbaty: Yes. I just help the doctor to take the decision. I will not replace him. And I have never, ever said my tools are holy books that you have to obey. No. I just help you to take the decision in critical periods. You have a very short time, you have to make a decision. In anesthesia practice especially, more than in other branches, seconds make a difference in patients’ lives. So you need some sort of accuracy and rapid decisions.

Kevin Pho: So how difficult was it for you to create this tool? How long did it take?

Ahmed Elsonbaty: It’s a very difficult process. Actually, I took out of the application more than I put in. Every time I put data in, I tested it. I found it was not suitable, maybe confusing, maybe not trusted, so I removed it and put in other sources of data, and tested whether it was applicable, suitable for the doctor, and could easily be used during the operations.

And one important point is that this building was not done solo by me. I put buttons for feedback and suggestions in the applications. I received suggestions, and I was surprised. I received suggestions from all over the world. “Why not put this? Why not put that?” So I continuously built and improved the programs and added more and more. So I consider that it is not unique to me. It is by the help of all over the world. Actually, the applications are now used in more than 150 countries by more than 10,000 doctors.

Kevin Pho: So did you have any training before you created the tools when it comes to programming, AI training? What kind of training did you have before creating this tool?

Ahmed Elsonbaty: No. I depended on my own. Just reading the sources, open sources, how to start, how to write prompts, something like that. So, Kevin, look, I heard an old saying before: “Once you want to do something, you can do it and do what you want.” So I decided.

Even my family suffered with me. I left them for long times to learn, to build, to remove, to rebuild. There were many nights I spent, and every now and then, while walking, I would think, “What if I put this, remove that, improve this?” This is a very hard job, but you can do it. It is not a miracle.

Kevin Pho: Now, for those who aren’t familiar in the operating room and they’re not familiar with ROTEM, just give us a 20-second primer on exactly what that is and how your tool can improve patient care. Take us into the operating room.

Ahmed Elsonbaty: As I told you before, there are many applications. I did not invent something new. But all the applications before consider that if you find something like this, you have to do this. The new thing in this application is that I specified the management according to this patient. You have an old patient with chronic renal disease, previous surgery, multiple surgeries, now diabetic on insulin treatment. So this patient has special circumstances, and you have to deal with this during your surgery. And then you come to the next point, the adjustment of the doses of the drugs. Nobody will build AI for doctors like doctors.

Kevin Pho: So how long did it take you before you could trust the results of your clinical app and you could make clinical decisions based on the information from the app? How long did it take you to trust that information?

Ahmed Elsonbaty: Look, the first application was CCA-Pro. It’s concerned with coagulation, ROTEM, TEG, ABG, ventilation, ECMO. After I finished it, I didn’t trust myself. I didn’t know if it was good or not. So I submitted it to the biggest journal in our field, the Journal of Cardiothoracic and Vascular Anesthesia. It’s the oldest journal in our field, 40, more than 40 years.

And I told myself I would test myself, whether it was a good thing or not, and I was surprised. I’m the solo author, and I received an email from Joel Kaplan. For several years, Joel Kaplan was considered a hero for us. His name was on the spine of the textbooks we studied as students. So when I received this email, I considered it a medal on my chest. They accepted the paper, and they considered it good work. This pushed me toward the big project, Anesthesia AI, to collect all of anesthesia in one application, to help more doctors all over the world.

Kevin Pho: And for those interested in creating their own clinical app using AI, what kind of advice do you have for other physicians?

Ahmed Elsonbaty: I advise them to identify the problem meticulously. What is the problem you want to solve? Search for other applications, what they said about this problem. Search for the sources you will depend on in building your application, and start to build it with good prompting, using good AI for coding.

Trial and error. Success will not be gained in one step. You will fail, you will fail, you will fail. Then ask a small circle of colleagues what their opinion is. Then start to push it in front of the other doctors, continuously receiving feedback from them, and improve it step by step until you succeed. And this was a very, very crucial point. You have to depend on well-known sources. Don’t leave the AI to answer the questions of the doctor; make your application deterministic and traceable. And I think it should be peer-reviewed in one of the specific journals to gain the trust of the doctors.

Kevin Pho: We’re talking to Ahmed Elsonbaty. He’s a cardiac anesthesiologist from Egypt. Today’s KevinMD article is “I built clinical decision support tools at the bedside.” Ahmed, let’s end with your take-home messages that you want to leave with the KevinMD audience.

Ahmed Elsonbaty: First, thank you, Kevin, for this good chance to meet you. For doctors, just start. Search for the problems around you. Build your tools. It’s not a miracle, not secrets. Don’t wait for engineering or Silicon Valley to help you. You can do it. Be meticulous. Consider that everything you write or put in your application may affect a patient’s life on the other side of the world. So be meticulous, repeat. Fail, repeat, until you succeed.

Kevin Pho: Ahmed, thank you so much for sharing your perspective and insight. Thanks again for coming on the show.

Ahmed Elsonbaty: Thank you very much, Kevin. Thank you.

Listen and followApple PodcastsSpotifyYouTubeMore apps · Browse all episodes