Saturday, January 19, 2019

How Skewed Is the Distribution of Medical Spending?

Following up on some topics I raised in this post.

I found this document titled The Concentration of Health Care Spending by the National Institute for Health Care Management. It gives some good figures for the distribution of actual annual health care spending. See the chart on the 3rd page. The top 1% of spenders spend ~$90,000 on average. The top 5% of spenders spend ~$40,700 on average. The top 10% of spenders spend ~$27,000 on average. The top 30% spend ~$12,000 on average. You can get cute with these numbers and infer that, for example, spenders between the 1st and 5th percentile spend about $28,000 on average. (Do a weighted-average of the 1% spending $90,000 and another 4% of individuals in the 1-5 percentile range spending $28,000 and you get the $40,700 figure.) People in the 5-10 percentile range spend on average $13,000 a year. People in the 10 to 30 percentile range, about $5,000 a year. Clearly this is a pretty skewed distribution, with the very top percentiles accounting for a large share of total spending. If someone is in the top 1% of spending year after year, health spending could easily be higher than their lifetime income.

But wait. Are people persistently in the top 1% (or top 10% or top 20%), year after year? When we talk about "the top 1% of spenders", is it the same people year after year?  Turns out the answer is "No." (Or mostly "No.") Look at the figure on page 8. It shows the probability that someone in a high-spending quantile will transfer out of that quantile within a year. If you're in the top 1% in one year, the probability is only 20% that you'll be in that quantile the next year; you have an 80% chance of moving to a lower-spending percentile. If you're in the top 5%, you have a 62% chance of moving to a lower-spending quantile. If you're in the top 10%, you still have better-than-chance odds of moving to a lower spending quantile. Clearly this demonstrates that there is some persistence in remaining in a high-spending quantile. Which is consistent with our intuition that there are unhealthy people who need ongoing, expensive care throughout their lives. If transition probabilities were completely random, people in the top 1% would only have a 1% chance of being in that group in the next year; 20% is a lot higher than 1%. But it's a lot lower than 100%.

I tried to get clever with this and come up with the transition matrix for various quantiles. I spent about an hour thinking about it and gave up. I think someone cleverer than me could come up with at least a transition matrix that is consistent with these figures; I'm thinking there is not an exact solution to that puzzle. If someone knows about a successful attempt to come up with these transition probabilities, or if someone knows what keywords to search for so I can look them up, feel free to share. Having that transition matrix would allow me to come up with expected future health care costs for various quantile ranges. (Given the transition matrix, you can multiply through to determine your odds of being in any given quantile in the next year, the year after that, the year after that, and so on. And with the average cost per quantile two paragraphs up, you can calculate their expected future costs in each year, given their current status. Use some interest rate to discount future spending, maybe add some assumptions about "absorbing states", because people eventually die, and you can come up with at least a rough-and-ready estimate for future costs. And that would tell you roughly what a health insurance policy would cost in a free market.)

I wanted to show that the expected future cost for someone in the top 1% is actually modest. Plainly it's less than "$90,000 per year, every year forever, discounted using a reasonable interest rate." Most people who transition out of the top 1% probably transition to another high-spending quantile. They may still be in, say, the top 5% or 10% in the next year. They don't transition to a perfectly random quantile, such that  they are equally likely to be at the 2nd percentile and the 99th percentile in the next year. But the odds of several very bad years in a row gets pretty small, if you simply multiply through the probabilities. Here is a darker consideration: some of those people who are in the top spending quantiles several years in a row are in the last years of their life. It's not a happy thought, but it means that there aren't that many more future years to consider, which means that future expected health costs are not quite as high as the distribution of one year's actual health cost implies. Most of the people who will be in the top 1% are probably insurable on a prospective basis, before the severity of their condition is fully realized. And anyway, a lot of this very high spending is Hail Mary medicine, enormously costly with very little chance of extending life. A sensible health insurance plan that excludes this kind of spending would be still more affordable.

I don't doubt that there are some people who are "uninsurable" in the sense that Ed Dolan means (explained in my previous post). There are probably some illnesses or injuries that are so severe and require such expensive care that even expected future health spending exceeds lifetime income. Once again, the solution to that problem is to insure health status changes rather than having insurance pay all your medical bills. The coverage trigger should be the diagnosis or the injury event, rather than the acquisition of health care years later. Your insurer at the time you were diagnosed with diabetes should eat all the future costs of health care related to treating your diabetes. If you switch insurers, that part of your future health spending should stay with the previous insurer. That way you can shop around and other health insurers don't treat you like a hot potato. (They might charge you more because your poor health is predictive of other problems not related to your diabetes, but they don't have to price or underwrite against you for spending related to your diabetes, which at that point is no longer "fortuitous." At that point, your condition a known entity.) As David Friedman once put it, the trick is to insure before the die is cast. Our stupid system insists on insuring after the dice have already come up snake-eyes.

By the way, I've read some posts from Ed Dolan's excellent blog, here. Go read some. I highly recommend it. I agree with a lot of his policy proposals and a lot of what he says about health policy. Implementing his solutions would vastly improve our health care system, and if I were in a position to deal, I'd be quite happy to accept these as "second best" solutions. I pounced on his Econtalk interview because he repeats a bad argument (in my opinion) that vastly overstates the "some people are uninsurable" problem.

No comments:

Post a Comment