A new machine learning model could help public health officials anticipate the next crisis

A new machine learning model could help public health officials anticipate the next crisis

Spread the love

Diagnosing and containing an outbreak, or the health effects of a disruptive event like a natural disaster, can be a daunting task. A study published Friday by New York University suggests that a new machine learning model could improve the ability of health officials to respond to future pandemics and other public health crises.

The research was done in partnership with Carnegie Mellon University and the New York City Department of Health and Mental Hygiene.

Marketplace’s Kimberly Adams speaks with Daniel Neill, a computer science professor at NYU and director of its Machine Learning for Good Laboratory, which published the study. It explains how this machine learning model works. The following is an edited transcript of their conversation.

Daniel Neil: Our approach uses textual data from emergency room visits. So, in particular, the main reason why the patient came to the emergency room. And that textual data contains much richer data than just “a person has flu-like symptoms.” We might know exactly what kind of symptoms they have or what they’ve been exposed to, and so by detecting patterns in that textual data, we can surface new outbreaks, things that public health wasn’t already looking for as well as new outbreaks. other types of events.

Kimberly Adams: How could this algorithm be deployed in a health service to possibly identify new or unidentified outbreaks?

Neile: The hope is that public health services would actually apply this kind of approach on a day-to-day basis, where each day the algorithm would bring up clusters in the past, say, 24 hours of data that public health could review and, if necessary , to respond to. It can also help public health deal with all the myriad of issues that they have to deal with on a daily basis, which could be a cluster of cases coming from smoke inhalation, or there’s some kind of chemical exposure , or we are seeing a new cluster of drug overdoses from a new synthetic drug. So, again, the goal is to give them day-to-day knowledge of everything happening in their jurisdiction.

Adam: So maybe you could spot, I don’t know, an outbreak of something like Legionnaires’ disease sooner than you otherwise would?

Neile: Yes, it’s true. It’s a fine example of something with rare symptoms. And you can also imagine if something comes along with new symptoms, things that we’ve never seen before, like people’s noses turning blue and falling off. Now, it shouldn’t take very many cases of something like this for us to realize that we have something new and different for public health to deal with. But the irony is that typical disease surveillance systems will just map them to your existing syndrome categories and essentially miss the fact that there’s actually something new out there. So we provide a safety net to detect all these types of events that other systems might miss.

Adam: What if there is a data gap or no one is talking about their symptoms?

Neile: It’s true. It is absolutely a limitation of the system, which depends on the quality of the data, the availability of the data and the timeliness of the data. So, for example, if a jurisdiction does not get emergency department data from local hospitals in a timely manner, it will impact its entire ability to respond to all patterns of that data. Likewise, if there have been any major errors in the way the data was collected, these may propagate to what we can detect using that data. Also, you’re absolutely right, things that might not result in ER visits wouldn’t necessarily be detectable through that particular data source. There are, however, a wide variety of data sources that public health uses for outbreak detection.

Adam: One of the ways you all tested this algorithm was looking at the data that came into hospitals after Hurricane Sandy. Can you describe to me what you saw and how the algorithm responded to it?

Neile: Sure. We found a very interesting progression of clusters of cases in New York emergency departments. A day or two after Sandy passed, we saw what we expected, which was a lot of acute cases – people with leg injuries or shortness of breath, other things that are somewhat direct results of the impact of the hurricane. A few days later, we started seeing clusters of cases more related to mental health issues. So people come up with things like depression and anxiety. And then a few days after that, we saw yet another type of case. We’ve seen people come to the ER for things like dialysis or methadone maintenance. These are all things that would not typically be dealt with in a hospital emergency department. But because all the outpatient clinics were closed, people basically had to use the ER for those reasons as well. So what this really shows us is the progression of different stresses on an emergency department following a natural disaster. And I think it’s very instructive for hospital emergency personnel to know what they might need to anticipate and be prepared and have the right resources to deal with all these different types of issues.

Adam: Why is machine learning a better tool for this particular set of public health issues than how we used to?

Neile: It is by no means a task where [artificial intelligence] should replace humans. So what our system does is it makes humans, health epidemiologists, aware of events that emerge in the data that they might not otherwise see. So bringing out what’s important in all that massive, complex data that a human might care about and want to answer is really key.

Leave a Comment

Your email address will not be published.