Setting realistic expectations about what AI can and can't do today is essential to make effective use of these technologies, while allowing the health care industry to focus its time and energy on the most valuable-use cases during a time of crisis. To help health care leaders navigate the AI field in the present COVID-19 pandemic, we've broken down some emerging applications by how valuable they are in the immediate term and long term.
Where AI adds value today
Health care organizations are pursuing AI applications that help them deal with their most pressing challenges—preventing infection spread and alleviating the strain on clinicians. Most of today's valuable AI applications rely on relatively simple rules-based models, leverage data that is already available, or involve the redeployment of existing AI solutions to combat COVID-19.
1. Identifying and protecting vulnerable populations: AI can help health care organizations identify patients with the greatest risk for severe complications and take a proactive approach to prevent infection.
- Medical Home Network (MHN) is using artificial intelligence and their health risk data to identify vulnerable or socially-isolated Medicaid patients. Patients identified as being at-risk are contacted by a care manager via phone, email, or text with information on the steps they should take to avoid infection and how they can access care if needed. Targeted community outreach can help educate patients and relieve the strain on emergency departments.
- Clinical AI vendor Jvion released a publicly available COVID Community Vulnerability Map that identifies locations with populations at the greatest risk for severe outcomes if COVID-19 is contracted. Jvion used a de-identified health dataset of 30 million patients along with social determinants of health data to estimate risk levels at the census block level. Providers can use this map to prioritize limited resources and deploy preventative interventions.
2. Chatbot screening and triage for low-acuity cases: Chatbots can alleviate bottlenecks in the health system by keeping the "worried well" out of clinics and reserving limited resources for the most sick.
- OSF HealthCare recently added the patient engagement vendor Gyant's new COVID-19 Emergency Response Assistant feature to their existing chatbot "Clare." The tool uses AI to screen patients and direct them to the right care settings or resources. In the first two days after deploying the new feature, the chatbot had 14,000 interactions.
- Providence St. Joseph Health augmented their existing chatbot "Grace" with Microsofts AI capabilities for a similar purpose. Providence's new Coronavirus Assessment Tool was added to the chatbot in just three days. Microsoft's Healthcare Bot Service has also supported COVID-19 bots at Novant Health, Virginia Mason Health System, and most recently the CDC's Coronavirus Self-Checker, which launched last Thursday. Microsoft is also making a set of COVID-19 response templates built on CDC guidelines available for free to help customers deploy their own chatbots.
3. Remote patient monitoring to expand provider capacity: Overwhelmed hospitals are setting up new facilities or leveraging non-traditional spaces to deliver care. Remote patient monitoring can be used to track a growing inpatient population with limited staff resources, all while keeping providers at a safe distance to control the spread of infection.
Sheba Medical Center in Israel recently converted staff dormitories and an underground parking garage into care units for COVID-19 patients. Sheba is using AI-powered remote patient monitoring to predict complications such as respiratory failure or sepsis. A sensor that sits under a patient's mattress analyzes the patient's movement and vitals, sending warning alerts to providers if their condition deteriorates.
- Last week, the White House announced a new open dataset called CORD-19 that will contain nearly 30,000 scientific articles about the virus in a machine-readable format. Data scientists are being called to apply machine learning (ML) and NLP techniques to this open-source dataset to answer outstanding questions about the virus' genetics, incubation, treatment, symptoms, and prevention.
Where AI shows long-term promise
Looking to the future, there are other ways health care leaders can leverage AI to navigate the pandemic as it unfolds. These applications show a lot of promise, but require a greater understanding of the disease and more complete data.
1. Early detection and diagnosis: ML imaging models are widely used to identify diseases like cancer or heart conditions before a human doctor. AI can detect subtle differences in complex images, but more COVID-19 imaging data is needed to train an ML model. Furthermore, physical signs of the disease may show up in scans well after a patient's initial infection, making it unclear whether imaging is the best method for early diagnosis.
2. Predicting the exact path of the disease going forward: Although an AI algorithm developed by BlueDot accurately predicted the first outbreak before the World Health Organization's announcement, predicting the virus' spread becomes more difficult as the pandemic grows. Complete data on COVID-19 is challenging to aggregate across countries with varying data collection processes and data sharing standards. Another significant barrier is the availability of test kits. Without sufficient testing resources or consistent testing processes available globally, we lack up-to-date data on disease counts.
Your top resources for COVID-19 readiness
You're no doubt being inundated with a ton of information on how to prepare for possible patients with the 2019 coronavirus (COVID-19). To help you ensure the safety of your staff and patients, we pulled together the available resources on how to safely manage and prevent the spread of COVID-19.