COVID-19 has accelerated the adoption of AI in healthcare. AI based tools and solutions can work quickly, be deployed at scale, and respond to the dynamic nature of the crisis. Use-cases span all facets of responding to the pandemic, from diagnosis and triage, to treatment and combating new transmission.
A wide range of players—including startups, established companies, universities, and more—are bringing their capabilities and perspectives to the table. Startups like Current Health, a UK-based remote-monitoring company supporting Mayo Clinic and Baptist Health with their COVID-19 response, are benefitting the industry’s rapid digital adoption, and scaling quickly to meet demand. The CEO of Current Health told the WSJ in March that it could double its workforce to accommodate this increased interest. On the other end of the spectrum, tech giants such as Microsoft, Google, Apple, Amazon, and Facebook are involved in initiatives related to remote communication between patient and clinician, contact tracing, and drug development, among others. With their wide reach, capabilities, and financial resources, these players are in a unique position to have an impact within the US’s fragmented healthcare system and capitalize on a growing demand for consumer tools that put patients in control of their care. Universities are playing an important role too. Penn Medicine for instance designed a COVID-19 chatbot in partnership with Google that risk-stratifies users to facilitate patient triage.
These developments are likely to impact the future of AI in healthcare. COVID-19 has spurred unprecedented data-sharing and collaboration efforts. C3.ai, for example, has created a data lake accessible through APIs that contains a diverse and growing number of COVID-19-related datasets such as genomic sequences of virus samples and image data. These initiatives have potential to establish new norms and infrastructures that support future AI development. The current focus on addressing privacy and security concerns, particularly in relation to contact tracing, is also likely to further AI adoption by strengthening the case for universal privacy standards. Such standards increase consumer and healthcare entity comfort with third-party data sharing, supporting digital health efforts such as remote patient monitoring with wearables and AI-driven therapy chatbots.
For details on the framework we used for our use-case analysis click here.
The COVID-19 AI use-case landscape
The healthcare industry is quickly adopting AI tools to address the COVID-19 pandemic across all domains and functions. Use-cases that just weeks ago were in the design phase are now in the test or spread phases as organizations scramble to bring solutions to market. This brings with it huge opportunities but also creates some risks.
Consumer wearables present an opportunity to help the public assess their risk for COVID-19. The smart ring company Oura, for example, is partnering with UCSF on a study to determine whether its product can detect COVID-19 before symptoms appear. Chatbots can also help screen patients and direct them to the appropriate resources and care settings. This use-case has entered the spread phase with health systems, governments, and tech companies offering the service. Penn Medicine, for example, designed a COVID-19 chatbot in partnership with Google to facilitate patient triage.
AI companies and academics are developing AI-based tools for COVID-19 detection that can supplement PCR testing. These include image-based diagnostics as well as applications using markers such as voice, the sound of a patient’s cough, or more standard metrics like temperature, heart rate, and oxygen saturation. For example, Thirona, a med tech company in the Netherlands, created an AI algorithm that can detect COVID-19 from chest X-rays at levels comparable to or better than human readers . Such tools have potential to facilitate remote testing and expand access.
AI tools are helping scientists accelerate the COVID-19 drug development process and improve understanding of the disease. Google’s DeepMind is predicting protein structures associated with SARS-CoV-2 to inform drug and vaccine development. Other researchers are using ML and natural language processing (NLP) to mine existing COVID-19-related papers for answers to key questions about the disease and identify promising therapies from the body of existing drugs. These tools can visualize connections between diseases, biological processes, and drugs that target those processes. Once researchers identify target therapies, clinical trial sponsors are running ML-driven adaptive trials that test multiple interventions as part of a single study.
Once providers identify COVID-19-positive patients, machine learning (ML) algorithms are being used to quantify disease progress based on imaging and predict which patients will require closer monitoring or transfer to an ICU. Lower-risk patients can return home with remote monitoring tools that alert providers of likely adverse events. Remote monitoring tools are also used in the hospital to minimize clinician contact with patients. This use-case is seeing rapid adoption.
In our previous paper, we note that the barriers for AI to break into the actual delivery of care are high. Although still in clinical trials, technologies like Admetsys’ closed-loop insulin/glucose administration system are generating excitement as hospitals take steps to protect frontline staff.
On the administrative side of care delivery, ML helps hospitals forecast care needs to address resource shortages and optimize capacity by prioritizing patients for ICU stepdown and discharge. Tools for these use-cases have passed initial development and are being tested by innovative organizations or scaled across the industry.
Promote public health
Population health tools employing ML can help governments and healthcare organizations identify high risk populations and prioritize intervention resources such as testing and face mask distribution. Tools to predict infection hotspots where strict stay-at-home orders are needed, facilitate contact tracing, and monitor social distancing adherence will become increasingly essential as states continue to relax social distancing guidelines. While contact tracing and social distancing adherence solutions have potential to facilitate a relaxing of containment measures, adoption will depend on developers’ abilities to address privacy considerations and public acceptance of increased surveillance.
AI tools can also be used to monitor overall sentiment of the pandemic. Researchers from Penn’s Center for Digital Heath, for example, created an NLP-based tool to track self-reported COVID-19 symptoms and trends in language about stress, anxiety, and overall sentiment.
Notably, the contribution of AI to modeling COVID-19 spread and deaths has been limited. Commonly cited forecasts, such as the COVID-19 model from IHME, rely on standard epidemiological models. Two AI-based models did, however, predict the outbreak before the WHO’s announcement. These include BlueDot’s AI algorithm and Boston Children’s Hospital’s Health Map, though human interpretation was required to understand the magnitude of the results.
Notably, there are a wide range of players bringing capabilities and perspectives to the table during this crisis. These organizations fall into four broad categories:
Small & mid-sized vendors
COVID-19 presents an opportunity for many small and mid-sized health tech companies. Multiple vendors with relevant capabilities are offering free tools during the crisis to cement their place in the market. Qventus, an AI-based resource and patient flow optimization vendor, has made their localized COVID-19 demand model and scenario planner widely available. They have also rolled out additional capacity optimization products. In the care delivery domain, AgileMD is offering their adaptive clinical decision support pathways to health systems and Zebra Medical Imaging hopes to offer their CT-based diagnostic algorithm to hard-hit hospitals at no cost.
Other vendors have realized an expanded customer base. Buoy Health, for example, built a COVID-19 symptom-checking tool that serves as a triage system for the state of Massachusetts. In the area of remote monitoring, previously fledgling companies have scaled quickly to help providers combat COVID-19. The CEO of Current Health, a UK-based startup, told the WSJ in March that it could double its workforce to accommodate the increased demand. The company is currently supporting Mayo Clinic and Baptist Health with their COVID-19 response
University-based collaborations focused on AI applications have proliferated. Examples focused on diagnosis and screening include Cough for the Cure, an MIT and Stanford effort to develop a COVID-19 diagnostic based on a patient’s cough, and a project within Carnegie Mellon’s CyLab working on a remote fever-scanning technology. In an effort to support care decisions, researchers at NYU designed a ML algorithm to predict which COVID-19 patients will develop severe cases. Furthermore, University of Oxford’s Vaccine Group developed a vaccine candidate currently in human trials and Safepaths, a project out of MIT, is working to create a privacy-focused contact tracing platform.
Established healthcare data & workflow companies
Healthcare data and workflow companies are adapting existing solutions and creating new tools to address the crisis. GE Healthcare, for example, set up an AI-driven command center in Oregon with 64 hospital partners representing 90% of beds in the state. GE’s algorithms process about two million data points a day from the hospitals’ data streams to predict bottlenecks and optimize resource use across the state. IBM Watson Health is also applying its command center tools to help manage patient flow during the crisis. Additionally, they are offering Watson Assistant, a conversational AI platform, pretrained with COVID-19 questions.
Epic is another active player in the COVID-19 AI application ecosystem. Their deterioration index, an algorithm that uses patient vital sign, lab, and nurse assessment data to assign a risk score, is being applied by many hospitals to assess COVID-19 patients . Epic is also working with Cleveland Clinic on home monitoring for confirmed and suspected COVID-19 cases to alert clinicians of new or worsening symptoms. At the time of writing, the tool has been implemented but does not yet include predictive monitoring.
Big tech companies
Tech giants such as Microsoft, Google, Apple, Amazon and Facebook are also stepping up to address COVID-19. Amazon is donating $5M-worth of Alexa devices to hospitals to facilitate remote communication between a patient and their clinician and Microsoft, Verily, and Apple have all developed screening and triage chatbots. Microsoft’s service is currently used by the CDC and large health systems such as Providence St. Joseph and Novant. Through an unprecedented partnership, Google and Apple released APIs that allow authorities to build contact tracing apps that access Android and iOS bluetooth systems. These bluetooth-based apps will not collect location data; instead, they will record proximity to other users. While the companies also say they plan to build bluetooth-based contact tracing functionality into their underlying platforms, the development timeline is unclear.
Outside of contact tracing, Google’s DeepMind is predicting protein structures associated with COVID-19. Microsoft is leveraging its Azure platform to do the same through its collaboration with ImmunityBio. Facebook AI is also offering its expertise to help with the crisis by partnering with academic researchers on various COVID-related initiatives. For example, it is partnering with NYU Langone’s Predictive Analytics Unit and Department of Radiology to build machine learning-based hospital-specific forecasts for COVID-19. These forecasts aim to help organizations optimize resource allocation and workflows.
COVID-19 will forever change healthcare—and the role of AI in the industry is no exception. Here we highlight a few ways the current landscape is likely to impact its future. While these predictions are optimistic, it is important to note that any shortcuts taken now due to the urgency of the situation, such as noisy data or rushed implementations, could harm future adoption if the now-expanded user base loses trust in the technology.
Today’s data-sharing and collaboration efforts have potential to establish new norms and infrastructures that support future development.
The COVID-19 pandemic has resulted in unprecedented information-sharing and collaboration. C3.ai, for example, has created a data lake accessible through APIs that contains a diverse and growing number of COVID-19-related datasets such as genomic sequences of virus samples and image data. Data-sharing partnerships among hospitals are also working to break down long-standing barriers to interoperability. In this environment, developers are putting goals of profitability to the side and creating open-source tools.
While not all collaborations will continue post-Covid, the ones that do will lay the foundation for an infrastructure supporting increased AI deployment. This infrastructure will likely include more prominent health information exchanges and increased efforts on the part of healthcare data companies such as Epic to further data interoperability.
The current focus on addressing privacy and security concerns may lead to established standards that further AI adoption.
Disease surveillance and contact tracing are essential to safely reopening the U.S. economy, but privacy and security concerns related to these efforts are acute. Through their partnership, Google and Apple are addressing privacy concerns with decentralized models that allow users to decide when their information is shared. Some states, however, have raised concerns that Google and Apple’s privacy settings prevent them from collecting enough information to make contact tracing efforts effective. Even so, a recent poll from the Washington Post and University of Maryland shows many Americans lack trust in big tech to keep their data anonymous.
This philosophical misalignment on privacy issues between government agencies and big tech highlights a need for federal data privacy standards. Such standards could further contact tracing in three ways:
- By increasing big tech’s willingness to adjust product settings to work better for government agencies,
- By promoting consumer comfort with contact tracing tools from big tech and public health agencies, and
- By opening the door for new players to develop innovative products and gain users.
To date, attempts to pass a federal consumer privacy law have been unsuccessful. The urgency of the COVID-19 pandemic, however, appears to be pushing efforts to establish universal privacy standards forward.
Big tech will play an even greater role in healthcare and support further interest in consumer-driven care.
As noted earlier, big tech companies have launched multiple AI initiatives to support COVID-19-related efforts. Due to the fragmented nature of the US healthcare system, the large reach, capabilities, and financial resources of these players put them in a unique position to have an impact. For example, Apple and Google together reach about 80% of US adults through their iOS and Android smartphone platforms.
Furthermore, adoption of digital health tools that put patients in control of their care are growing faster than ever before. As consumer companies steeped in technology capabilities with growing panels of healthcare experts, tech giants are in an advantageous position to capitalize on this demand. With new tools to support consumer-driven and home-based care, patient interest in self-directed, convenient care will only increase.
In a previous post, we predicted an increase in partnerships between mid-sized healthcare services players and these tech giants. We expect these collaborations, as well as biopharma-related partnerships, to accelerate in the current environment.
Click here to return to the COVID-19 AI use-case landscape.
Acknowledgement, Dr. Marc Herant, Managing Partner, Recon Strategy for co-developing the framework
 The AI system, CAD4COVID-Xray, performed similarly to six readers at low and intermediate sensitivity operating points and outperformed human readers at high sensitivity points. Study published in the Journal of Radiology of May 8, 2020.