Can AI Speed Up Vaccine Distribution?

When it comes to COVID-19, it sometimes seems that every silver lining has a cloud. We were barely done celebrating the scientific miracle that brought us multiple rapidly developed, highly effective vaccines when we had to confront the unpleasant truth that we lacked the smart, efficient infrastructure needed to get those shots into people’s arms. Even now, only a small percentage of the U.S. population has received a shot — and while the Biden administration is stepping on the gas, it’s becoming clear that we need a new toolkit to help us accelerate our efforts.

One potentially promising approach is to use AI systems to steer our vaccination programs.Companies such as IBM are now using machine learning technologies to help healthcare systems understand their patient demographics, identify high-risk populations, and factor in vaccine-availability modeling to figure out smarter ways to distribute available doses. AI tools are also being used to help streamline the supply chain and speed up production and transportation of doses.

Not foolproof

Still, such technologies aren’t foolproof. Stanford Medical Center — an institution that has more access than most to experts in AI systems — drew criticism late last year after it emerged that its distribution algorithm had assigned vaccines to just seven out of more than 1,300 frontline doctors. “Our algorithm, that the ethicists, infectious disease experts worked on for weeks … clearly didn’t work right,” one administrator admitted. 

Such episodes play into longstanding concerns about AI in healthcare, with human experts fearing that they will need to surrender agency to algorithms that may or may not be correctly programmed. In fact, some have argued that using complex criteria of any kind to determine who should get vaccinated is self-defeating, and that we should establish simple age-based rules to enable healthcare systems to move quickly and get more shots into people’s arms. 

Others point to the fragmentation of the U.S. healthcare system — and healthcare data ecosystem — as a sign that AI tools will be hard to roll out effectively. Israel’s world-leading vaccination program, for instance, is driven in part by a powerful electronic record system that gives healthcare providers streamlined access to unified medical data for all the country’s citizens. Having universally accessible and consistently formatted data makes it far easier to use AI to identify high-risk patients. but such systems are necessarily harder to scale up in the United States, where every hospital has its own system of medical records. 

A simpler approach

Despite these challenges, I’m less bearish than many on the notion of using AI to support complex healthcare decision-making because its ability to support vaccine distribution isn’t limited to such use cases. In fact, AI is perhaps most powerful not when it’s being used to decide who gets a shot, but rather when it’s being used to take care of other, more mundane parts of the vaccine distribution process.

Consider this: in the United Kingdom, thousands of patients last year used an AI-powered app to make appointments to get a flu shot. The AI tool, accessed through a simple conversational “chat-bot” interface, didn’t determine who was eligible for vaccination, but it did help people to learn their own vaccination status, figure out locations and times, and get basic information about the flu vaccine. It also triaged patients with more complex questions or concerns to their local doctor’s office so they could get help from a human nurse or administrator.

A similar approach could be used to manage COVID-19 vaccination appointments here in the United States. That would ease the strain on government planners and healthcare workers who are currently flooded with inquiries — and for patients, it would offer a big upgrade from the clunky birthday-invitation apps currently being used to manage vaccination bookings. Conversational AI can answer questions, offer information, and provide reassurance as well as simply booking appointments, making them the perfect tool for helping patients to navigate these chaotic, confusing times. 

In the COVID era, similar tools can also be used to support telemedicine and remote health monitoring, ensuring that patients don’t pack into overburdened clinics and emergency rooms in search of reassurance or information that could be delivered remotely and, in many cases, automatically. Reducing foot traffic through our hospitals and clinics gives vaccination workers a little bit more breathing room to do their jobs and for overworked healthcare professionals those incremental gains can make a big difference..

The next steps

So should we abandon distribution algorithms and simply focus on streamlining healthcare operations? Not at all. In the COVID era, we’re facing a huge number of complex and interconnected problems — some epidemiological, some medical, some logistical, some operational — and we should use data and AI-powered analytics to make the best decisions we possibly can. Inevitably, there will be stumbles along the way, but the power of modern AI platforms to tease out useful signals from large bodies of data is game-changing for many healthcare applications, and we should leverage this important technology in any way we can.

But along the way, it’s important to remember that when we talk about AI in healthcare, we aren’t only talking about computers making big decisions about who gets vaccinated and who doesn’t. We’re also talking about quieter, less dramatic applications that can incrementally ease the strain on our healthcare workers, eliminate or reduce logistical bottlenecks, and help keep our hospitals and clinics working efficiently while we work to get everyone vaccinated. It’s by automating and streamlining our healthcare systems and vaccination processes, as much as by using algorithms to decide where vaccines go, that we’ll ultimately win the war against COVID-19. 

Adam Odessky is the chief executive officer and co-founder at Sensely in San Francisco.