An HMO developing a platform for the discovery of clinical insights from medial information, Israel’s Maccabi Healthcare Services, plans to deploy an AI system that can identify people at risk of developing COVID-19 complications. The work is being done in partnership with the Kahn-Sagol-Maccabi Research and Innovation Institute and Medial EarlySign and looks at factors like preexisting conditions, the three organizations announced today. The AI could enable the HMO to fast-track patients for testing at a time when COVID-19 test kits are tough to come by. As of early this month — citing a growing shortage of reagents — the Israel Health Ministry said it wouldn’t test people for COVID-19 symptoms unless they had recently traveled.
According to Medial EarlySign CEO Dr. Jeremy Orr, the COVID-19 risk detection system shares elements with the company’s existing flu complications model, whose design was informed by an analysis of tens of millions of people treated by Maccabi and billions of lab results, structured electronic health record data, vital signs, demographics, and other data points. Following a pilot analysis of Maccabi patients’ anonymized records, it identified the top 2% of highest-risk patients (approximately 40,000 people), taking into account variables like:
- Respiratory disease such as pneumonia, bronchiolitis, and influenza
- Hospital admission history
- Weight and BMI
- Medications prescribed for respiratory illnesses or conditions, such as asthma and cough
- Heart disease
- Smoking history
- Digestive disease
- Immunosuppression therapies
When a person flagged by the system as high risk contacts a nurse or a doctor to report COVID-19-like symptoms, the system automatically notifies the medical professional of that person’s status. From there, the potentially infected person can expect priority treatment at Maccabi testing facilities and drive-in testing stations. Alternatively, they might be offered an at-home test that can be administered remotely.
Medial EarlySign says it’s in “advanced negotiations” with health systems in the U.S. that have expressed interest in incorporating the algorithm into their COVID-19 health care protocols.
Efforts to apply AI to COVID-19 patient data are underway within ICUs, too. At Stanford University, a team led by physician Ron Li is evaluating whether an algorithm trained on over 130,000 patient encounters — Deterioration Index — could accurately identify which COVID-19 patients’ condition will deteriorate. Bayesian Health, a startup spun out of Johns Hopkins University, is working on an early warning model for acute respiratory distress syndrome, a type of respiratory failure that can be caused by COVID-19. And the University of Chicago Medical Center is testing an upgrade to its AI eCart system that will monitor oxygen to signal when a COVID-19 patient’s lungs might be failing.