Industry field guide
Specialty rehabilitation hospital assessment records for AI
A specialty hospital can test whether software compares assessments from the same episode and measurement basis.
- 2022 NAICS
- 622310
- Practice pack
- Example and answer key (JSON)
- Updated
What this guide tests
Which records belong to E, and can their counts alone demonstrate improvement under an unchanged condition?
See the specimen and fields · Work through the fictional records
Task and source context
A specialty hospital can test whether software compares assessments from the same episode and measurement basis. CMS publishes inpatient rehabilitation assessment and reporting materials. A fictional exercise can test provenance while leaving actual functional evaluations private. CMS inpatient rehabilitation reporting resources
Invent an episode and create two harmless task counts at different times. Give them different assistance categories and include one unassigned worksheet. Ask which records belong to the episode and what comparisons the evidence supports. Score task, time and stated conditions independently.
Use abstract identifiers and original measures. Keep real inpatient records, mobility videos, outcome scores and clinical recommendations outside the dataset. Have rehabilitation staff review the fictional scoring design.
Industry research
The U.S. Census Bureau’s industry definitions establish which activities these codes cover. The workflow sources in this guide describe the task or record type. Fiund’s pilot below uses fictional material to test a specific decision within that scope.
622310 · Specialty (except Psychiatric and Substance Abuse) Hospitals
This industry comprises establishments known and licensed as specialty hospitals primarily engaged in providing diagnostic and medical treatment to inpatients with a specific type of disease or medical condition (except psychiatric or substance abuse).
Build a pilot sample
Set up this pilot
Use fictional rehabilitation task counts from different time windows and assistance levels. Grade same-episode attribution and preserve conditions when discussing any comparison.
Prepare the test packets
- Start with the worked case. Use the fictional record types below as a guide to the source files you need. Give each packet a stable ID and save its source versions with it.
- Change one fact. Make one packet with enough evidence to answer the question and one where the evidence points to a different answer. Keep the other fields fixed so a reviewer can tell what changed.
- Remove the decisive evidence. Give the reviewer a third packet with a plausible near-match in its place. Have them write what can still be established and what evidence would resolve the rest.
Fields the reviewer needs
- Episode, task and date
- Defines the observation.
- Count, time window and assistance
- Preserves the measurement basis.
- Worksheet link and comparison limit
- Identifies what is missing.
Use What health care and social services teams can contribute to AI to turn the specimen into a brief with a defined task, source and review owner.
Fictional exercise
Task counts under different assistance conditions are not an unqualified improvement score
A fictional episode contains original tabletop sorting counts with abstract assistance labels. It includes no inpatient, mobility video, clinical score or functional-assessment claim.
| Record ID | Source | What it records |
|---|---|---|
| RA1 | Episode E, task T | Day 1: 4 inert objects sorted in 60 seconds, condition INDEPENDENT under the original exercise definition. |
| RA2 | Episode E follow-up | Day 3: 6 objects in 60 seconds, condition PROMPTED. Same task label but helper cue allowed. |
| RA3 | Loose worksheet | 8 objects in 60 seconds, condition INDEPENDENT, no episode or date. No reviewed E linkage. |
Question for the model. Which records belong to E, and can their counts alone demonstrate improvement under an unchanged condition?
Show the reference answer
RA1 and RA2 explicitly belong to E and share task T and a 60-second window, but their assistance conditions differ. Report four independent objects and six prompted objects rather than an unqualified two-object improvement under the same conditions. RA3 lacks episode and date, so it cannot supply a matched independent follow-up. The fixture supports provenance and condition checking, not a clinical outcome or rehabilitation judgment.
How to check the result
Ask for the answer, the record IDs that support it and any fact the evidence leaves unresolved. Have a reviewer compare the response against the reference before deciding whether the sample is ready to expand.
Errors to flag in this case
- Comparing counts while omitting a changed assistance condition.
- Assigning an unlinked worksheet to the episode because it completes the desired comparison.
Change the inputs and run it again
Add a reviewed day-4 independent E/T count in 90 seconds. Require the time-window difference to remain explicit before comparing rates or outcomes.
Score the decision and the evidence separately
Save the reviewer’s expected answer, supporting record IDs and unresolved fields in a separate answer key. Score the model on the decision, its citations and its handling of the deliberate gap. A correct answer citing a different case counts as an evidence error.
Before expanding collection, have a second reviewer run the same checks on the held-out packet. Keep related versions, views and exports of the same case together so they stay on one side of the training and evaluation split.
Check permission and scope
Use original abstract measures and inert props. Exclude inpatient records and actual outcome scores; rehabilitation staff review the fictional rubric.
Leave out patients, real charts, security controls, clinical operations and staff identifiers.
Describe an authorized archive through Fiund’s existing-data assessment, or describe a new capture capability through the collection brief. The contributor agreement covers the terms for participation.
Sources and further reading
Resources for this workflow
Includes worksheet
What health care and social services teams can contribute to AI
Separate a staff training simulation from a governed clinical or care-data project.
Includes worksheet
Build an AI evaluation sample for service handoffs
Use fictional records to test whether a handoff reached a documented state.
Company and program reviews
Use these reviews to compare sourcing, capture or evaluation approaches. Confirm the applicable program, accepted sources and terms with the provider.
Compare the approach
Protege review for AI teams and data owners
Compare the company’s healthcare data-provider pathway separately from a fictional staff workflow.
Compare the approach
Sapien review: dataset quality review and expert checks
Compare an independent quality review process for an administrative or nonclinical evaluation.
Related industry examples
Industry guide · NAICS 622110
Hospital discharge packet version and handoff data for AI
A hospital can evaluate whether software distinguishes a drafted discharge packet from the approved version and a separate receiving acknowledgment.
Industry guide · NAICS 622210
Inpatient behavioral health follow-up evidence for AI
A referral, booked appointment and completed follow-up represent different states.
Industry guide · NAICS 623110
Skilled nursing facility assessment window data for AI
Facility staff may collect observations across several shifts and assemble them into a dated assessment.