Industries / Health / Healthcare / Hospice
Admission
Deciding who can receive hospice care, referring them, and the person's own choice to enroll.
7
Milestones achieved, from 1983 to 2023.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1983
An entry test for hospice
When the US Congress added hospice to Medicare in 1983, it set the terms most US hospice care still follows. A doctor must judge that the person has six months or less to live if the illness runs its normal course, and the person agrees to give up Medicare-paid treatment aimed at the terminal illness in exchange for care focused on comfort. Medicare still pays for care unrelated to that illness.
Credit
US Congress; Centers for Medicare & Medicaid Services (CMS)
1996
A scale for daily ability
The Palliative Performance Scale (PPS) rates a person's physical state as a percentage, adapted from the older Karnofsky scale used in cancer care. Victoria Hospice in British Columbia, Canada, used it to profile admissions: 83% of 213 people admitted to its hospice unit scored 20% to 50%, and those admitted at 10% lived 1.88 days on average. The hospice still publishes the scale, now in a second version.
Credit
F. Anderson, G. M. Downing, J. Hill, L. Casorso and N. Lerch, Victoria Hospice (Journal of Palliative Care, 1996)
1999
Entry checklists fail a test
Researchers tested criteria based on National Hospice Organization guidelines on 2,607 seriously ill people with lung, heart or liver disease who left five US hospitals alive. Of the 923 people the broadest criteria called eligible, 70% lived longer than six months. The narrowest criteria picked out only 19 people, and 53% of them lived longer too.
LessonFor heart, lung and liver disease, no checklist then in use could pick out the people with six months or less to live.
Credit
E. Fox, J. Lynn and colleagues, for the SUPPORT investigators (JAMA, 1999)
2000
Doctors guess too long
At five outpatient hospice programs in Chicago, 343 doctors estimated how long 468 patients would live at the time of referral. Median survival was 24 days. Only 20% of the estimates were within a third of the actual survival, 63% were too optimistic, and on average doctors overestimated survival by a factor of 5.3.
LessonThe longer a doctor had known the patient, the less accurate the estimate, so a second opinion from someone less close may help.
Credit
N. A. Christakis and E. B. Lamont (BMJ, 2000)
2004
A register in every practice
Keri Thomas, a family doctor in Yorkshire, England, developed the Gold Standards Framework from 1998 to help general practices find people in their last year of life and plan their care. By the framework's account, from 2004 every UK general practice kept a palliative care register and held team meetings about the people on it. Its prompt, the "surprise question" ("Would you be surprised if this person died within the next year?"), was pooled in a 2024 review of 56 groups covering 68,829 patients: it flagged 69% of those who went on to die, and 40% of the people it flagged died.
LessonA register only helps the people put on it: England's practice registers held 0.56% of patients in 2023/24, by the House of Commons Health and Social Care Committee's count.
goldstandardsframework.org.ukpmc.ncbi.nlm.nih.govpulsetoday.co.uk
Credit
Keri Thomas, Gold Standards Framework; A. Gupta, S. Straw and colleagues (BMJ Supportive & Palliative Care, 2024)
2021
Hospice without giving up treatment
The Medicare Care Choices Model, run from 2016 to 2021, let people who qualified for hospice get supportive care from a hospice while still receiving treatment for their illness. Among enrollees who died, 83% went on to use the full hospice benefit, against 65% of a matched comparison group, and they spent 42 days in hospice against 19. Their net Medicare spending was $7,604 (13%) lower. Only 7,263 people enrolled, fewer than 1% of those eligible in the hospices' areas.
LessonLow payments and hard recruitment wore the test down: of 141 hospices accepted, only 44 took part for all six years.
Credit
CMS Innovation Center; evaluation by an independent contractor
2023
Nudges, but no hospice change
At nine cancer clinics of a large academic health system in Pennsylvania, a machine-learning model flagged patients with at least a 10% predicted risk of dying within 180 days, and their clinicians got weekly lists, comparisons with their peers and text reminders. Serious illness conversations, planned talks about prognosis, goals and wishes, rose from 1.3% to 4.6% of all visits. Over 40 weeks and 20,506 patients, cancer drug treatment near death fell, but there was no effect on hospice enrollment or length of stay.
LessonMore conversations about dying did not, by themselves, bring people into hospice sooner.
Credit
C. R. Manz, R. B. Parikh and colleagues (JAMA Oncology, 2020 and 2023)
Watch for
Software flags hospice eligibility
Axxess, which sells home health and hospice software, says its Care Predictor, an AI tool, flags home health patients who may be eligible for hospice or palliative care "much earlier". It gives no accuracy figures.
Achieved whena study by someone outside Axxess reports how often the flagged patients were in fact eligible and whether they reached hospice sooner.
Claimed by
Axxess (Jaime Carlson, senior vice president), May 2025
England's identification target
The UK government aims, by March 2029, to identify 10% more people nearing the end of life earlier and to cut their hospital admissions and stays by 10%, under a Modern Service Framework for palliative and end-of-life care due in autumn 2026.
Achieved whenNHS data show 10% more people identified, for example on practice palliative care registers, by March 2029 against the 2023/24 level.
Claimed by
Department of Health and Social Care, June 2026
Open
01
A plain election form, tested
A plain-language version of the US hospice election statement, the form a person signs to enter hospice under Medicare, and of its addendum listing what the hospice will not cover. Test it against CMS's model text with readability scores and a short quiz on what the reader has agreed to, taken by adult volunteers who are not patients and are not choosing care. Publish both versions, the quiz and the results.
The briefCredit · unclaimed02
Time in care, worldwide
An open, updatable dataset of how long people receive hospice or palliative care before they die, by country, setting and illness, taken from the reports that registries and regulators already publish, such as MedPAC in the US, the Palliative Care Outcomes Collaboration (PCOC) in Australia, Sweden's palliative care register and UK hospice studies. Each figure carries its source, year and exact definition, and the dataset is refreshed when a new report comes out.
The briefCredit · unclaimed
Open 01 · the brief
A plain election form, tested
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A plain-language version of the US hospice election statement, the form a person signs to enter hospice under Medicare, and of its addendum listing what the hospice will not cover. Test it against CMS's model text with readability scores and a short quiz on what the reader has agreed to, taken by adult volunteers who are not patients and are not choosing care. Publish both versions, the quiz and the results.
Why it is still open
The HHS Office of Inspector General (OIG) found that election statements in more than a third of sampled 2012 inpatient stays lacked required information, often failing to say that the person gives up Medicare coverage of treatment for the terminal illness, or that hospice care is palliative rather than curative. Kehl and McCarty measured reading levels in hospice papers for families in 2012 (mean grade levels of 9 to 11), but searches found no test of whether readers understand the election form itself. CMS's FY 2027 rule requires the addendum for everyone who elects hospice, so more people will read it.
Built with what exists
CMS's published model election statement and addendum; standard readability formulas (Flesch-Kincaid, SMOG); a language model such as Claude to draft plain versions for people to edit; an online survey panel of adult volunteers; approval from a university ethics committee.
First step
Write the 10 quiz questions from the items the OIG found missing, and try them on 20 volunteers with the CMS text.
Done looks like
At least 300 volunteers randomly given the CMS text or the plain text, with the share answering each of 10 quiz questions correctly published for both. The plain version reads at or below a US grade 6 level and scores higher on the quiz by a statistically significant margin.
Open 02 · the brief
Time in care, worldwide
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open, updatable dataset of how long people receive hospice or palliative care before they die, by country, setting and illness, taken from the reports that registries and regulators already publish, such as MedPAC in the US, the Palliative Care Outcomes Collaboration (PCOC) in Australia, Sweden's palliative care register and UK hospice studies. Each figure carries its source, year and exact definition, and the dataset is refreshed when a new report comes out.
Why it is still open
The largest comparison, Jordan and colleagues' 2020 review of 169 studies from 23 countries, is a one-off in which US studies made up most of the data. The authors reported inconsistent terms and methods and almost no data from low- and middle-income countries, and called for consistent reporting so that services can be compared. Searches found no updated or open version.
Built with what exists
public reports in PDF; pdftotext; a small, low-cost model for repeated extraction, such as Muse Ember (see the Note "The jobs agents repeat"), with every figure checked by a person against its page; a public spreadsheet and code repository.
First step
Enter the figures for the US, Australia, Sweden and England from their latest public reports, with their definitions side by side.
Done looks like
At least 15 countries, each with figures for at least two years, every figure traced to a page. A second person rechecking 50 figures at random finds no more than one error.