Read this first

Nobody who works on this site is a clinician. Nothing published here diagnoses, treats or predicts anything, and no page on this site should change what a doctor has told you or delay a call to one.

What this page does is show our working. Every figure we publish is listed below with the exact query that produced it, so that anyone who wants to can pull the same rows from the same public source and check us.

The rest of the site starts at the front page.

Where every number on this site comes from

One public dataset, four survey questions, and a small number of rules about which rows we use. All of it is written down here.

Memory Care at Home STL desk. Published September 10, 2026.

There is a great deal of confident writing about memory loss on the internet, and a lot of it rests on numbers whose origin is never stated. We would rather be checkable than confident, so this page is longer and duller than it needs to be.

The source

Every survey figure on this site comes from one place: the Centers for Disease Control and Prevention's Alzheimer's Disease and Healthy Aging Data, resource identifier hfr9-rurv on data.cdc.gov. The underlying collection instrument is the Behavioral Risk Factor Surveillance System, a telephone survey the states run with the CDC. The dataset is public domain and is attributed to the CDC Division of Population Health.

We use one class within that dataset, Cognitive Decline, which contains four questions. Nothing on this site uses any other class, and in particular it does not use the Caregiving class, which asks a different question family of a different population.

The exact row selector

For all four questions we take the rows matching every one of these conditions, and no others.

• Class = Cognitive Decline
• YearStart = 2019 and YearEnd = 2022
• StratificationCategoryID1 = AGE and StratificationID1 = 65PLUS, that is, adults aged 65 years or older
• StratificationCategoryID2 = OVERALL and StratificationID2 = OVERALL, that is, not broken out by sex or by race and ethnicity
• Data_Value is present, that is, the CDC did not suppress the estimate

The de-duplication rule

This is the part that most needs stating, because getting it wrong would silently change numbers.

The dataset files two separate families of rows under YearStart 2019. One carries YearEnd 2019 and covers thirty-seven locations. The other carries YearEnd 2022 and covers fifty-four. They are not copies of each other. For twenty-two of the jurisdictions present in both, the two families report different values for at least one of the four questions, so mixing them would produce a table that is internally inconsistent and matches no published series.

Our rule is therefore simple and absolute: we use the YearEnd 2022 family only, for every question, on every page. We chose it because it is the only family with complete coverage of the reporting jurisdictions. We never fall back to the other family to fill a gap.

For Missouri specifically, the two families agree exactly on all four questions, so the state figures quoted on this site are the same whichever family you pull. That is a coincidence of Missouri's reporting history, not a general property of the dataset, and it is why the rule is stated rather than assumed.

What we exclude, and why the count is 46

The dataset includes five roll-up rows that are not jurisdictions: Midwest, Northeast, South, West, and United States, DC and Territories. We drop all five, because ranking a region against a state is meaningless.

We also drop rows the CDC has suppressed. The dataset marks these with a footnote reading that the sample size of the denominator, or the age group for age standardisation, is under fifty, or that the relative standard error is over thirty percent. Puerto Rico is suppressed on these questions and therefore does not appear.

Finally, a jurisdiction enters our data file only if it carries a value for all four questions. This is a deliberate strictness: it means every page on this site ranks the same set of places, so a reader moving between two pages is comparing like with like, and no page can quietly drop or add a state. That leaves 46 states and jurisdictions.

What the numbers are, and are not

These are self-reports. Adults answered questions about their own memory over the telephone. Nobody in this survey was examined, tested or diagnosed by anyone, and the CDC does not present these figures as a measure of dementia prevalence. Neither do we.

Every estimate carries a confidence interval, which the CDC publishes and which we print in the last column of every table. Where two jurisdictions are a point or two apart and their intervals overlap heavily, the difference between them is not meaningful, and we say so on the pages rather than letting the ordering imply otherwise.

None of these figures describe an individual. A state-level percentage cannot indicate anything about the person you care for, and no page here attempts to.

Who wrote this and who pays for it

Pages are written by the Memory Care at Home STL desk. There is no named medical author, because there is no clinician on this site and inventing one would be worse than admitting it. We do not publish an author name we cannot stand behind.

This site is published by OwnersFirm. It carries exactly one paid placement, for New Plan Care, which is a client of OwnersFirm. That placement appears in a bordered box labelled as a featured placement, with a disclosure stating the relationship, and it is kept entirely separate from the survey data. No figure on this site was selected, ordered, filtered or framed to suit it, and no provider is ranked anywhere on this site.

Corrections

If a figure here does not match what you get from the CDC resource using the selector above, that is a defect and we want to know. The dataset itself is also revised from time to time by the CDC, and the pull date printed on every page is the date our copy was taken, not the date the CDC last updated the source.

The local copy lives in this network's repository as engine/data/cdc_cognitive_decline.json, pulled September 10, 2026.