David Rein
Senior Fellow
NORC at the University of Chicago
September 2026
Unless noted otherwise, numbers below refer to all Medicare beneficiaries ages 65 and older with a Highly Likely or Likely dementia diagnosis in claims, which we treat as our summary case definition.
Franklin County, Ohio, home to Columbus, is one of the most typical counties in America. Ohio used to be considered a national election bellwether because the state closely mirrors national distributions of race and ethnicity, employment, education, and urban-rural populations. Franklin County, with a diagnosed prevalence of 9.4 percent, is also a bellwether for diagnosed dementia outcomes. Across diagnosed prevalence, mortality, hospitalization, emergency visits, and long-stay nursing care, Franklin County sits closer to national average values than any other of the 248 counties with at least 50,000 beneficiaries ages 65 and older.
A caveat about dementia diagnosis: The Dementia DataHub reports diagnosed dementia identified in Medicare claims. In our recent validation study, the DataHub’s case definitions identified about half of the Health and Retirement Survey respondents who’s cognitive measures indicate dementia. Sensitivity varied by geography, insurance type, and demographics. Places with high diagnosed dementia prevalence may have more disease, but the diagnostic signal also contains noise generated by local coding practices, access to care, intensity of health care use, and the cultural norms around telling a person they have dementia and writing that determination into a payment record.
Now the extremes: Among counties with at least 50,000 beneficiaries, Miami-Dade County, Florida, sits farther from the national averages than any other county (see below the map for methods). In Miami-Dade, 20.4 percent of beneficiaries carried a Highly Likely or Likely diagnosis in 2022, more than double the national rate, yet only 11.5 percent of those diagnosed died during the year, as compared to 19.0 percent nationally, and they made 995 emergency department visits per 1,000 diagnosed compared to 1,336 nationally. High prevalence combined with low mortality and low emergency use is consistent with the hypothesis of earlier diagnosis resulting in a younger and healthier diagnosed pool, and with more diagnosis in general. Our system’s public-use file data alone cannot verify this explanation, but pairing these data with other sources could.
Barnstable County, Massachusetts, is Miami-Dade’s mirror image. Miami-Dade sits far above the national figures on diagnosed prevalence and far below on mortality, hospitalization, emergency visits, and long nursing facility stays among the diagnosed. Barnstable is below the nation on prevalence and above it on all four of the others. Among counties with 50,000 or more beneficiaries, no two profiles are farther apart. The inverse of our Miami-Dade hypothesis is that in Barnstable, patients tend to be diagnosed later, less frequently, and closer to death. Again, this is a hypothesis, not a finding.
We identified these comparisons by sifting through our own public use file and asking what we missed. In this case, we compared each of the 248 counties with at least 50,000 beneficiaries to the national estimates on five measures at once: diagnosed prevalence, and among the diagnosed, mortality, hospitalization, emergency visits, and long nursing facility stays. Each gap was scaled by how much counties typically vary on that measure, and the five scaled gaps were combined into a single distance. Franklin County sits closest to the national figures across all five; Miami-Dade sits farthest, driven mostly by a diagnosed prevalence that was more than double national average coupled with lowest mortality rate among the diagnosed cases of any large county. Barnstable County, Massachusetts, is Miami-Dade’s opposite: 7.1 percent diagnosed prevalence compared to Miami-Dade’s 20.4 percent, and mortality among the diagnosed of 22.2 percent against 11.5, with hospitalization, emergency visits, and long nursing facility stays all above the national figures where Miami-Dade’s are all below. Of the 248 large counties, no two sit farther apart.
Everything in this post came from an experiment we ran on our own public use file. Surveillance systems produce far more data than any one team can analyze, I thought many interesting patterns in the data were going undetected and unexplored. Some of those patterns might tell us something about how dementia gets diagnosed and how that differs by place. To test the idea, I set up 14 inexpensive AI agents across two rounds of exploration. Each was assigned a topic area, such as health care utilization, disparities, or data anomalies, and asked to produce specific, data-driven findings. The agents reported to a more capable manager agent that got them started, fielded their questions, and decided which results were worth a second look. Most of what they produced was unremarkable or wrong. But they also surfaced findings I doubt we would have noticed on our own: a state line inside a single metropolitan area where the diagnosed prevalence rate jumps five points, a diagnosis difference that appears to vanish in the place you would least expect, and insights about variable prevalence among Medicare beneficiaries younger than 65. Over the next few months, Dementia DataHub staff are verifying these findings as time allows and writing them up. Every post in the series uses only public use file data, so that each one also illustrates what the file can and cannot tell you. I hope they spark your curiosity.
If you are so inspired, we invite you to download the data to test your own questions. Fill out the data request form to access the data yourself. If you find something interesting, let us know.
