BRENNEN

Industries / Health / Healthcare / Home health

Documentation and billing

Recording what was assessed and done at each visit, in the form payers require, so the agency gets paid and the payer can check the claim.

7

Milestones achieved, from 1999 to 2020.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1999

    One assessment for everyone

    On 25 January 1999 the US Health Care Financing Administration, now CMS, required Medicare home health agencies to complete OASIS (Outcome and Assessment Information Set), a fixed set of questions about the patient's condition and abilities, as part of each patient's assessment, and to report the answers electronically to a national database.

    govinfo.gov

    Credit

    Health Care Financing Administration

  2. 2000

    Paying per episode

    From 1 October 2000, Medicare stopped repaying home health agencies their reasonable costs and paid a set amount per 60-day episode of care, adjusted for each patient's needs using selected OASIS answers.

    govinfo.gov

    Credit

    Health Care Financing Administration, under the Balanced Budget Act of 1997

  3. 2012

    Billing outliers mapped

    The OIG screened 2010 Medicare claims against six measures of unusual billing and found that about one in four home health agencies exceeded the threshold on at least one, mostly in Texas, Florida, California and Michigan. Medicare had also paid about $5 million for claims that overlapped hospital or nursing home stays or came after the patient had died.

    Lessonsimple rules run over claims data show where to look, though passing a threshold does not prove fraud.

    oig.hhs.gov

    Credit

    HHS-OIG

  4. 2013

    New agencies barred

    CMS used new powers under the Affordable Care Act, in its own words "for the first time", to stop new home health agencies enrolling in the Miami and Chicago areas for six months from 30 July 2013. Agencies already enrolled could keep billing.

    Lessonwhen fraud clusters in places, a payer can control who may enter, not only check claims.

    cms.gov

    Credit

    CMS

  5. 2017

    Germany's new needs test

    Germany's Second Care Strengthening Act brought in a new definition of who needs care and a new assessment tool from 2017, to give equal access to long-term care benefits whether a person's impairment is physical, mental or psychological, with particular attention to dementia. The contribution rate rose by 0.2 percentage points, about €5 billion a year.

    bundesgesundheitsministerium.de

    Credit

    Federal Ministry of Health

  6. 2019

    Review before payment

    CMS's Review Choice Demonstration began in Illinois in June 2019 and spread to Ohio, Texas, Florida, North Carolina and Oklahoma. Agencies choose to have claims checked before submission (pre-claim review) or after payment. In fiscal year 2024 reviewers received 1,791,226 pre-claim requests and approved 97% of those completed; California joins from 18 January 2027.

    cms.govcms.gov

    Credit

    CMS; Palmetto GBA as reviewer

  7. 2020

    Paying by patient type

    From 1 January 2020 CMS's Patient-Driven Groupings Model pays for 30-day periods instead of 60-day episodes. Each period falls into one of 432 groups built from where the patient came from, timing, clinical group, function and other illnesses, so payment depends more on the patient than on the number of therapy visits.

    cms.gov

    Credit

    CMS

Watch for

  1. AI notes during visits

    VNS Health in New York says its home care clinicians use the Nestmed app on a tablet during visits. The app listens and drafts structured notes for the clinician to review, and VNS says it cuts documentation time "significantly", without a figure or method.

    Achieved whenan outside evaluation reports minutes saved per visit and the error rate of drafted notes against clinician-checked records.

    vnshealth.org

    Claimed by

    VNS Health and Nestmed, April 2026

Open

  1. 01

    Open checker for Medicare paperwork

    A tool that reads a synthetic certification packet (a face-to-face encounter note, a homebound statement and a plan of care) and reports, step by step, whether it meets Medicare's home health documentation rules. The face-to-face encounter is the required visit with a doctor or allowed practitioner, which must fall between 90 days before and 30 days after home care starts and relate to the reason for it; homebound means leaving home takes a considerable and taxing effort. Each flag cites the step it failed in CMS's published Home Health Review Tool.

    The brief
    Credit · unclaimed
  2. 02

    Outlier screen from public data

    An open, repeatable analysis of CMS's public file of payments, visits and patients for each home health agency ("Post-Acute Care Utilization: Home Health Agency by Geography/Provider and Case-Mix Grouping"), joined to the public file of agency owners. It flags agencies far from their peers in the style of the OIG's 2012 measures, with each flag explained and framed as a statistical outlier, not an accusation.

    The brief
    Credit · unclaimed

Open 01 · the brief

Open checker for Medicare paperwork

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A tool that reads a synthetic certification packet (a face-to-face encounter note, a homebound statement and a plan of care) and reports, step by step, whether it meets Medicare's home health documentation rules. The face-to-face encounter is the required visit with a doctor or allowed practitioner, which must fall between 90 days before and 30 days after home care starts and relate to the reason for it; homebound means leaving home takes a considerable and taxing effort. Each flag cites the step it failed in CMS's published Home Health Review Tool.

Why it is still open

CMS says insufficient documentation caused 51.4% of home health improper payments in its 2024 reporting period, out of an estimated $1.1 billion. CMS publishes its review steps as a flowchart, but searches found only proprietary vendor tools and no open checker or public test set.

cms.govcms.gov

Built with what exists

hard-coded rules for dates and provider types; a language model such as Claude for the narrative tests (see the Note "Seven more flagged"), or a local model run through llama.cpp (see the Note "A first stable number for local AI"); test packets written by volunteer clinicians or generated, with no real records.

  1. First step

    Turn the Review Tool flowchart into a written rule list, and write 30 packets, 10 clean and 20 with one known defect each.

  2. Done looks like

    A public set of at least 300 synthetic packets with seeded defects, labelled by two independent reviewers. On 100 held-out packets the tool catches at least 95% of seeded defects, wrongly flags no more than 5% of clean packets, and maps every flag to a numbered Review Tool step.

Open 02 · the brief

Outlier screen from public data

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An open, repeatable analysis of CMS's public file of payments, visits and patients for each home health agency ("Post-Acute Care Utilization: Home Health Agency by Geography/Provider and Case-Mix Grouping"), joined to the public file of agency owners. It flags agencies far from their peers in the style of the OIG's 2012 measures, with each flag explained and framed as a statistical outlier, not an accusation.

Why it is still open

The OIG's 2012 screen used claims data that are not public, and CMS still checks millions of requests one by one under the Review Choice Demonstration. A limited search found no published open method that applies the OIG approach to the public files.

oig.hhs.govportal.datarescueproject.org

Built with what exists

Python or R; the public CMS files; a language model to draft plain-language explanations of each flag in batches (see the Note "A thousand questions, one pass").

  1. First step

    Download one year of the provider-level file and reproduce CMS's national totals before building any score.

  2. Done looks like

    Code and method published, plus a back-test on an earlier year showing what share of agencies later named in public OIG or Department of Justice actions fell in the top 5% of scores, against the 5% expected by chance.