Misdiagnosis and How AI Can Help Patients Advocate for Better Care | Paul Battle

What we talked about.

In this episode of the Sunny Ray Show, host Sunny Ray talks with Paul Battle, who joins from Hyderabad, India, about misdiagnosis and how AI can help patients advocate for better care. Paul, a former senior executive at Lenovo with a healthcare background, explains how his late wife's misdiagnosis during her breast cancer journey pushed him to build a company focused on diagnosis rather than billing and administrative efficiency. He argues that misdiagnosis is the number three killer, cites figures on yearly deaths and disabilities in the United States, and explains why primary care doctors see only a small share of the roughly 28,000 diagnoses. He describes a hematology test in which his model matched a specialist's cases 28 for 28, and says a general chatbot would not do the same. The conversation also covers inaccurate electronic medical records, a patented patient interview, the role of family history and where someone grew up, the direct to consumer and self-insured employer business, bootstrapped funding with a current raise, and the choice to start with men's health. Paul closes by pointing listeners to a free trial.

After losing his wife to a misdiagnosed cancer, a former Lenovo executive builds AI to help patients catch what doctors miss.

The questions, and the answers.

What problem are you solving, and why did you get into this space?

It's personal. I come from a healthcare background, built a career in technology at IBM and Lenovo, and my late wife was misdiagnosed during her breast cancer journey. I wanted to marry technology and healthcare, so with my co-founder I set out to tackle misdiagnosis head on, moving from late reaction to prediction, risk assessment and early detection.

How big is misdiagnosis as a problem?

I believe it's the number three killer after cancer and heart disease, and probably higher because many deaths are never attributed to it. In the United States there are 400,000 deaths and 400,000 disabilities a year from it, and at least 15% of diagnoses are wrong. You and I will likely have one or two in our lifetimes.

What did your enterprise healthcare experience teach you, and what was missing?

At Lenovo I ran healthcare as a vertical. Organizations used technology for operational efficiency like billing and appointment setting, which AI does well. Almost nothing focused on making doctors more effective at diagnosis, because they felt they only needed administrative help. I learned to ask what people actually need, and to show them the data.

What did you believe about medicine that turned out to be wrong?

Doctors are incredibly talented, but they're human. There are 28,000 different diagnoses, and a primary care doctor may see around a thousand in an entire career, under 4%. They will make mistakes and take unnecessary steps. We're also finding that specialists make mistakes, and our model can catch those too.

What results have you seen from your model?

We worked with a top United States hematology professor who gave us cases, one he called impossible. We got his answer in 45 minutes, and across 28 cases we went 28 for 28, with the correct diagnosis ranked number one, without lab tests. We're drafting a publication. Generic LLMs might get 20 out of 28.

Why can't someone just put their symptoms into a general chatbot?

We trained and run our model against proprietary databases and data that no one else is using today. A general LLM won't match that performance, and then you're left with the cases it missed. On top of that, we have a patented patient interview and a patented way to diagnose and assess risk.

Why does the data in a patient's record matter so much?

About 40% of an electronic medical record is incorrect, whether it's in Epic, Oracle or Athena Health. A doctor's note saying someone might be pre-diabetic can end up read as diabetic. So it's not only about asking the right questions, it's making sure the answers are correct and the wrong ones don't get run against.

How is the company funded and why start with men's health?

We're completely bootstrapped, mostly by my co-founder and me with some friends and family, and we built it affordably with a development team in Brazil. We're raising 2.5 to 3 million at a 12.5 million pre-money valuation. We started with men's health because men are more likely to share details with an AI avatar than a doctor.