[A]t least one in every three high spenders in a given year will be a high spender in any of the next five years. (Here, high spender is defined as in the top 10% of the annual spending distribution.) I don’t know what your prior is, but this is a much higher level of persistence than I expected.
At the end of the first year, only 472 of the original 1,682 (28 percent) remained super-utilizers. [These are patients who had 3+ hospitalizations or had a serious mental [SMI] health diagnosis and 2+ hospitalizations in a 12 month period.] At the end of the second year, just 240 (14 percent) remained, of whom only 93 (6 percent of the original 1,682) met the criteria in all twenty-four months. […] Because super-utilizer status was defined by recent use, baseline spending for this cohort was quite high, at $113,522 per capita. Per person spending for this cohort in subsequent years was much lower, falling almost 60 percent after two years.
For many reasons, these are not contradictory findings. First off, they’re computing different statistics (returning to the high spending group in any of the next several years vs. continuously remaining in the super-utilization group).
Second, a 60% drop in spending among super-utilizers, from $113,522, is still over $45,000 in spending. The top 10% of the spending distribution is over $30,000. So, though few super-utilizers may remain in that group, on average their spending remains high enough to be considered “high spenders.”
Third, super-utilization is driven by admission count, not spending. Clearly one can be a higher spender and not a super-utilizer (e.g., because of high outpatient and/or drug costs, or because of high hospital spending in one versus two or three admissions in a 12 month period). In principle, one can be a super-utilizer and not a high spender, though it’d take several very cheap hospitalizations to get there, which is probably unlikely. But, the point stands, these are different populations.
Fourth, I wrote about a study of millions of individuals covered by over 100 medium and large employers. Johnson et al.’s data are from Denver Health, “an integrated safety-net health system and the largest provider in Colorado of services to people in the state with Medicaid or no insurance.” About 40% of individuals in this study population were on Medicaid; about 30% were homeless; over 25% were uninsured. Clearly these are not statistics one would expect of those employed by medium and large employers; they are statistics one would expect of a population less likely to remain engaged with the health system. These are different populations.
Fifth, Johnson et al.’s analysis did not include utilization (and its cost) that occurred outside Denver Health. I believe drug spending, in particular, was missing. They admit that it’s possible they underestimated costs.
For all that, there’s nothing wrong with a super-utilizer focus. It’s likely that super-utilizers, as defined, are a subset of high spenders and a population one might care to design interventions for. Understanding their patterns of use and costs is a worthwhile activity. But we should be careful not to over-generalize. Super-utilizers may be different than high spenders, so spending persistence might be different as well.