AI Models Might Make American Healthcare Better – But It’s An Uphill Climb

Video caption: Although algorithmic work can fine-tune clinical responses, we’re also going to need an overhaul on how the country’s healthcare delivery system works.

For anyone who’s thinking about reforming America’s healthcare system, this video can be helpful. It doesn’t purport to answer all the questions – in fact, John Guttag specifically insists that it won’t answer the biggest question – but it does start to look at some ways that we could do better with health outcomes. And that’s important for a country that has sort of a black eye in terms of its effectiveness in this area.

In fact, when you watch the beginning of the presentation, you see some of those statistics that many of us are already familiar with: the United States spending double the global runner upcountry on healthcare per person, and achieving only relatively low life expectancy, etc. – as Guttag says, America is “not doing well on the world stage.” Even the World Health Organization and other global analysts, as you can see here, give America pretty low marks on all of these metrics.

So Guttag goes into parts of why that may be the case. The system, he explains, impedes delivery of the best care, opining: “Many of our citizens and residents don’t actually get access to this good care … predictive models can make a difference here; they really can help.”

Guttag also gives us three specific criteria – the right treatment, the right provider, and the right time. When these three converge, you get better results, in general.

He also adds another thing that’s critically important to care outcomes – an appropriate cost to the society. That’s one to throw around in a societal or political context, where AI’s applications would be, at best, complementary.

Anyway, later on, he talks about theoretical healthcare plans for John Doe and Jane Doe, and about looking at the rates of adverse outcomes to figure out what’s best for healthcare methodology. Custom or individual analysis, he stresses, is important, but how do you get there?

Much of this, he says, is just a model, and analysts should take it “to the field” to understand how it works out in practice. There’s a bit of humor where he talks about finding a magazine and picking out the best doctor. Obviously, it’s a little more complicated than that.

(image caption: don’t just use the yellow pages – or whatever the online equivalent is)

“What we really need is not a ranking of providers, but an app that matches patients and providers,” Guttag says, pointing out a lot of the limitations that, practically, hold us back from doing a deeper dive where that could be helpful.

If you want a lot more detail into the nuts and bolts of what Guttag is suggesting, watch the part where he outlines the orthopedic study and chronicles the findings in aid of evaluating outcomes.

In a nutshell, one thing you find out is that the best outcomes are ‘volume-based’ – in other words, if you find a provider that does a certain thing all day, that provider is probably going to be more proficient than someone else who is a jack of all trades or a generalist.

Guttag also talks about other fixes that result in fewer trips to the hospital per capita.

“Even though we’re trying to deliver (healthcare) to the population,” he says, “we have to make the decisions not based on averages, but on individuals. And to do that at scale, we really need to deploy AI based models.”

This is just one of many powerful presentations on AI and healthcare that came out of this week’s conference. Anyone who has skin in the game is likely to see this as relevant to the ongoing struggle to improve our healthcare models, and advance modern medicine, in that context.