The question almost always arrives at the wrong hour. A parent with dementia is in an emergency department at three in the morning, a resident is asking whether the family wants everything done, and nobody in the room has ever said out loud what "everything" means. Since January 2016, Medicare has paid clinicians to have that conversation earlier, in daylight, on purpose: two billing codes, 99497 for the first thirty minutes of advance care planning and 99498 for each additional half hour. Ten years on, there is finally a national picture of what shows up in the claims record afterwards.

The picture comes from a retrospective cohort study published online on March 18, 2026 in Alzheimer's & Dementia: Behavior & Socioeconomics of Aging, led by Sean R. Riley with Seuli Bose-Brill and colleagues at Ohio State. The team assembled 104,366 fee-for-service Medicare decedents with Alzheimer's disease and related dementias who died between 2016 and 2022, and looked for a billed advance care planning encounter in the window from two years before death up to the last thirty days. They deliberately excluded conversations that happened only in the final month, so the finding would not simply be the sound of a death already underway. People with a billed encounter were matched ten to one against people without, with exact matching on sex, race, region and year of death.

The first number is the one families should sit with: 8,816 people, 8.4 percent of the cohort, had even one such billed conversation on record. Among those who did, 89.4 percent had exactly one; 9.0 percent had two; 1.6 percent had three or more. The rest of the study describes what a small minority experienced. In the last thirty days of life, those with any billed encounter had 12 percent lower total expenditures (ratio of expected mean cost 0.88, 95 percent confidence interval 0.85 to 0.90) and 24 percent lower odds of inpatient care (odds ratio 0.76, 0.72 to 0.79), with fewer admissions and shorter stays. Outpatient use ran in the opposite direction, with odds of 2.76 (2.63 to 2.89) and outpatient spending 41 percent higher. Odds of skilled nursing facility use were 21 percent higher, which the authors read as earlier discharge planning rather than less care. Hospice and home health care outcomes looked similar between the groups. Where more than one conversation had been billed, the association was larger: expected total expenditures at a ratio of 0.74, or 26 percent lower.

Strip out the ratios and the shape is this: money and hours moved out of the hospital and into clinics and community settings. That is not the same thing as a better death, and the study cannot claim it is. This is observational claims work. Nobody was randomised to a conversation. People whose clinicians bill an advance care planning code may already be people with engaged families, a steady primary care relationship, and settled preferences, and none of that is visible in a billing file. The authors themselves point to what claims cannot see, and the same group found a comparable pattern in a different disease: in 48,466 Medicare decedents with heart failure, published in the Journal of General Internal Medicine on May 20, 2024, billed advance care planning was associated with 19 percent lower total end-of-life spending and 2.65 times the odds of outpatient use at the end. The pattern is consistent, which is interesting, and consistent patterns in claims data are still not causes.

There is also the awkward fact that a billing code is a receipt, not a transcript. It records that at least sixteen minutes of a thirty-minute service occurred, not whether anyone explained what late-stage dementia looks like, whether a daughter understood the word "aspiration", or whether the plan was ever revisited when things changed. That gap matters most in dementia, where capacity to take part in the conversation is falling while the need for it rises. A separate national cohort study of 959,405 Medicare beneficiaries newly diagnosed with mild cognitive impairment or dementia in 2018, published in the Journal of the American Geriatrics Society on March 17, 2026, found that receiving an annual wellness visit was associated with earlier and greater advance care planning initiation, most of it billed on the same day as the wellness visit, with uptake still modest overall. The door exists. Most people never walk through it.

For a family reading this at the kitchen table, the useful takeaway is not a target, a code to request, or a decision to make. It is that the conversation is a covered, repeatable service rather than a favour to ask for, and that the one signal in this data that grew stronger with repetition was having the conversation more than once. What that conversation should contain, and when, belongs to the person living with the diagnosis and the clinician who knows them.