Like 80% of Americans, you likely have. 

That’s where Ada Health comes in. It’s a mobile app that helps people quickly diagnose their symptoms by assessing their ailments against its database of 3,600 conditions. And it’s using AI to do it.

Since launching in 2016, Ada’s app now covers 99.5% of all medically diagnosable conditions. It looks at more than 10,000 symptoms and risk factors across medical disciplines, including mental health, pregnancy, rare diseases, and pediatrics, linked to more than 31,000 different ICD-10 codes, a system that codes diagnosis and procedures for insurance claims processing. The app has more than 13 million users around the world. 

“To sift through all the possibilities and information that turns up online leads to a great deal of confusion and anxiety,” said Dr. Claire Novorol, who is the cofounder and chief medical officer of Ada Health. “Our AI-based system assessment and acuity-specific care navigation helps eliminate inefficiencies in the patient journey and accelerates time to diagnosis, treatment, and care.”

Novorol founded Ada Health with Daniel Nathrath, who now serves as chief executive officer, and Martin Hirsch, Ph.D., a former medical researcher who was looking to apply his probabilistic reasoning approaches to clinical diagnosis. Novorol said that they felt there was a more personalized, granular, and reliable way to self-assess symptoms than an online search, and that became the Ada app.

The app is powered by a probabilistic reasoning system using Bayesian methods, one of the core concepts for computer science and AI. Its probabilistic reasoning takes into account an existing evidence base. Novorol explained that the most frequently entered symptoms are fairly common: cough, headache, abdominal pain, tiredness, and nausea. But what sets Ada apart is the product’s ability to differentiate among the common causes and the less common ones to identify potentially rare conditions. 

“It’s the long tail of many thousands of atypical symptoms and presentations,” she said. “At the same time, it’s delivering safe and appropriate results, consistently. It’s not sending every patient to emergency care—that’s costly and creates an overburdened system.”

Ada is a “white box system,” which is a form of explainable AI, Novorol said. For any given suggestion, there is transparency in how the technology came to that result. It identifies its medical reasoning and logic. “This level of rigorous validation is essential in our field,” she said. “We take safety very seriously.”

For clinicians, Ada has become an essential tool in striving for patient-first care. Novorol explained that Ada reduces the strain on overworked clinicians by helping patients find the right care more quickly; on the medical provider side, it automatically imports electronic health records, saving time-consuming work manually documenting medical records. In the EU, Ada’s app is technically considered a Class IIa medical device under the Medical Device Regulation terms. It’s one of the few AI-based symptom assessment platforms to be included under such designation. 

Ada is exemplary of the move by health care companies to adopt AI to achieve the “quadruple aim,” explained by the National Institutes of Health as improving population health, improving the patient’s care experience, and enhancing the caregiving experience. According to a 2023 report by Accenture, a human-plus-technology model is the answer, creating omnichannel connections with patients and lessening the burden on clinicians. It estimates 70% of health care workers’ tasks could be augmented or automated by technology like AI.

“In health care, AI extends far beyond research and data processing,” Novorol said. “We’re seeing AI’s transformative impact in diagnosis, medical imaging interpretation, drug discovery, remote monitoring and robotics, and candidate identification. AI-driven solutions are a crucial cornerstone of current and emerging health-tech initiatives.”

For Novorol, evolving the Ada product is about testing new types of AI. Currently, they are experimenting with large language models (LLMs) and how they could complement Ada. She explained that LLMs are an exciting horizon in the health-tech industry. Traditionally, LLMs have had clear limitations that make them challenging to employ in a clinical setting, namely, they are “black box systems” and opaque in their decision-making process, Novorol said. But now there may be ways to use their approach in responsible ways.

“We work closely with clinicians at multiple health systems around the world, and what we hear is that doctors are very open to the benefits of AI,” Novorol said. “However, transparency and explainability are particularly important in a clinical setting. They want to ensure the highest levels of safety and medical quality when deploying AI technologies.”