Q&A: Diagnostic Robotics CEO Kira Radinsky discusses how AI can help reduce healthcare benefits costs

Tell us a little about your background before co-founding Diagnostic Robotics.

I escaped Ukraine and immigrated to Israel with my family when I was young. Since I can remember, I always wanted to be a scientist. I started my university studies at age 15 and, while working on my Ph.D., continued my research in the U.S. My research focused on analyzing all global written history using AI algorithms to predict events such as droughts, cholera outbreaks, natural disasters, and genocide.

My work was inspired by Mark Twain’s quote, “The past does not repeat itself, but it rhymes.” Although future events have unique circumstances, they typically follow familiar past patterns. Over the past several years, I devoted my life to the development of prediction techniques. The technology solution I first created inferred that Cholera outbreaks in landlocked areas are more likely to occur following storms, especially when preceded by a long drought.

I developed algorithms that analyze large-scale digital histories, social media, news media, and human web behavior and augment them with human knowledge mined from the web to provide real-time estimations of the likelihood of future events. Most recently, these algorithms have accurately predicted the first Cholera outbreak reported in Cuba in 130 years. These types of actionable predictions, which enable preventative measures, have drawn the attention of organizations such as the UN and the Gates Foundation and illustrate the vast potential for real impact on the state of humanity. I applied my approach to numerous problems, predicting disease outbreaks, mortality, and riots with 70%-90% accuracy.

I was also able to apply such methods in other various applications, including sales. I founded SalesPredict, acquired by eBay in 2016, which leveraged large-scale data mining to predict sales conversions.

In recent years, I decided to use these technologies to continue impacting healthcare. I joined Technion – Israel Institute of Technology – as a visiting professor and led a research group that processed large amounts of data obtained from both medical records and literature in a quest to create an AI system to analyze historical patterns among patients and create personalized predictions based on specific events.

Can you share the genesis story behind Diagnostic Robotics?

In 2017, Professor Moshe Shoham and I co-founded Diagnostic Robotics with a student to try to tackle an ongoing issue in healthcare, and one exacerbated by the COVID-19 pandemic – the shortage of physicians and their overburdened job responsibilities. In less than 10 years, more than 3.8B people on earth won’t have access to primary care. The current average three-hour emergency department wait in the U.S. will soon extend to eight hours (or more), as it is in China today. We decided to focus the collective intelligence of hundreds of millions of visits with primary care physicians and emergency physicians to help reduce the burden on the healthcare system. We leveraged billions of claims and historical medical records to build AI-driven triage systems to help navigate patients and create patient care journeys. When we began building the algorithms that make up our AI models, they were initially designed to help triage patients inside emergency departments and soon expanded to intake patients in oncology, surgery, and other departments. But to really address the burden on the healthcare systems one needs to prevent patient deterioration in the first place. Today, we focus on identifying and targeting patients earlier in their care journeys and matching them with relevant clinical interventions.

Many new healthcare technology solutions have emerged as a product of the COVID-19 pandemic. What makes yours stand out?

First, we feel fortunate that we were about to help many people around the world during the COVID-19 pandemic. Our remote assessment and monitoring tool helped health plans and providers reduce the number of incoming inquiries by offering people who were concerned or had questions about their potential COVID-19 symptoms with resources for self-assessment. It also helped to remotely triage symptomatic patients and help lead them to the appropriate care settings if needed. Lastly, it conducted population-level monitoring for concerning trends or hotspots of COVID-19 activity.

When we first started deploying our solutions, we knew very little about COVID-19 and its symptoms. The only known symptom at the time was a fever, based on which patients were triaged in hospitals and limited COVID tests. Our AI systems had to infer the COVID-19 symptoms through open-ended triage questionnaires in a matter of days and change national triage systems with little prior knowledge. In a week, the systems had to serve two million patients every minute. Our systems were some of the first to identify loss of smell and taste and extreme fatigue as symptoms of COVID-19 and instantaneously deploy it on a national scale to amend the patient journey and triage patients accordingly.

COVID-19 and other diseases continue to change, and so should patient journeys. We continuously train our AI models and solutions on a variety of data, including claims data, EMR records, and other datasets (such as weather patterns, etc.), and apply state-of-the-art modeling, including natural language processing algorithms, deep learning, and causal inference, to better understand the populations and offer better, timelier, and more impactful care.

These models allow us to improve patient outcomes and optimize operational strategies for health plans and providers. They are designed to reduce the number of members who experience acute problems like avoidable ED admissions and joint replacements (hip/knee/spine) as well as help members with chronic conditions, such as CHF, Diabetes, COPD, etc. that are on track for clinical deterioration.

Nearly one-third of Californians currently live in a Primary Care Health Professional Shortage Area (HPSA). What are some factors that have caused these shortages in recent years?

Though many people might say that this problem is due to the COVID-19 pandemic, I believe the issue is much deeper than that. It was announced recently that we now have a global population of 8 billion people and a U.S. population of more than 333 million. We also see California’s population rapidly growing to nearly 40 million lives. We simply don’t have enough doctors to care for our growing population and haven’t been utilizing the many technological resources that could help current physicians utilize their time. According to the AMA Masterfile, the U.S. has more than 818,000 physicians involved in direct patient care. Nearly 202,000 physicians are retired or semi-retired, and roughly 43,000 spend more time in administrative, teaching, or research roles than in patient care. Within the next five years, an estimated 35% of the physician workforce will be of retirement age.

Burnout among physicians who continue to practice is another reason for this shortage. They are severely fatigued (emotionally and physically) from the burdens of delivering equitable care and are either looking to step away or retire early. Technology is something the healthcare industry can control and better utilize to make it easier for physicians to see more patients.

How is Diagnostic Robotics helping to increase access to primary care in California and throughout the U.S.?

To tackle bandwidth issues and burnout among physicians, we have developed a solution to make their jobs more streamlined and automated. For example, physicians spend countless hours obtaining essential information from patients to get them the care they need. The task can be extremely time-consuming, leading to fewer patients receiving care. Patients may also struggle with this task because it can be overwhelming and anxiety-provoking. Our technology automates care navigation and triage by allowing physicians to administer configurable, self-service questionnaires to their patients on their own devices. It also organizes the responses to these questionnaires in a user-friendly workstream that helps providers centralize key information and aids care decision-making.

In return, providers can see more patients daily and the patients who are being seen report higher levels of satisfaction due to more time with physicians discussing treatment and developing personal relationships rather than communicating administrative information.

Another example of how we can help increase access is by navigating patients to the correct care pathways. Using 60B+ claims, 76M+ EMR records, 4M+ ER visits, and other datasets, our AI models can predict which treatments and care plans will have the greatest impact on patients’ future health and match them with care journeys that make sense. If someone needs a specialist, the solution can recommend the patient see the correct type of doctor and see their primary care doctor later, or vice versa.

We hope this will decrease the wait time for primary care appointments as patients in some areas are currently waiting weeks or months before they can see a doctor.

Healthcare benefits costs are on the rise due to inflation. What are some ways that health plans can utilize AI to help to offset that?

Our AI-powered care management solution provides health plans with a method to utilize their member data to understand how to proactively identify healthy members who are at risk of avoidable health incidents and target them to get them the care they need before they begin to deteriorate. This knowledge helps health plans reduce the number of members with unnecessary costly incidents (such as an avoidable ED visit) and encourages members to engage with their primary care physicians or other, less costly, approaches.

This same applies to providers – if their care managers can seek out patients before they get sick and set them up with a plan to reduce the likelihood of such incidents, everyone involved, including their patients, is able to reduce the cost of care. We can help reduce care costs by improving efficiencies in care management outreach. Instead of targeting members who may not be able to benefit from specific interventions or may be too far along in their existing care journey, our AI allows them to reach members who are more impactable – therefore reducing the time and money spent on monthly outreach.

What are other healthcare challenges that Diagnostic Robotics can help to solve (in California and throughout the country)?

I believe AI will become the standard in patient care in the next ten years. The number of hours people wait inside clinics will continue to grow longer as our population ages, and the physician shortage will take time to overcome.

There are many uses for AI that we already know about, and I am optimistic that we will continue to discover more. Some other things our AI models could help with in the future include improving clinical trials (predicting which patients will be the best candidates by identifying subpopulations of patients who can benefit from a medication), identifying yet-to-be-discovered drug side effects by mining the web, and understanding possible complications in surgery by using historical surgical outcome data.