Falls and Hospital Return: What the Record Can Tell You
A foundation model for aging reads the records a plan or health system already holds and reports where falls and hospital returns are likely to concentrate. This post is about what that can and cannot tell you, outcome by outcome.
Last updated: October 2026
A foundation model for aging reads the records a plan or health system already holds and reports where falls and hospital returns are likely to concentrate. This post is about what that can and cannot tell you, outcome by outcome.
What can a foundation model for aging tell you about falls and hospital return from the record?
It can tell you which cohorts in your population carry elevated expected risk of a fall or a 90-day hospital return, as an aggregate report drawn from coded conditions, encounters, medications and assessments you already hold. It cannot tell you which member will fall, cannot see anything that is not in the record, and does not replace a clinician's assessment.
Scienza states evidence per outcome, and the two outcomes in this post have different standing. Five outcomes passed a pre-registered sealed evaluation: dementia, stroke and TIA, Parkinson's, cognitive transition and sustained cognitive decline. Falls was not one of the five; it replicated on a second, independent look. Hospital return is a trained head without a sealed-test result.
That is the honest ranking, and it is the ranking you should hold the model to. Falls has evidence from a second look. Hospital return has training evidence and nothing more yet. The pre-registered validation manuscript has been submitted to JMIR AI.
All of it is hosted by Scienza, and all of it returns aggregates for review.
Why these two events cost what they cost
A fall at home is not itself penalized. The hip fracture it becomes is a different matter. Since January 2026 the TEAM model puts surgical hip and femur fracture treatment on the hospital's book for 30 days, so a community fall that becomes a fracture is an episode the hospital owns. In Star Ratings, the Health Outcomes Survey asks whether the plan reduced the member's risk of falling.
Hospitals are penalized for returns within 30 days under the CMS readmission penalties. ACOs carry the same return against the year's result. Most of the risk sits in the days after discharge, when follow-up lags.
Both events have a quiet period before them. That period is what the record can and cannot show.
What the record can show: the falls head
The foundation model for aging reads what you already hold. For falls, it returns an aggregate report across your cohort. Members are never named.
The result replicated on a second look, meaning it held on an independent set after the analysis plan was fixed. That is stronger than a single training result and weaker than the five pre-registered outcomes. Use it to decide where a falls program should concentrate, not to tell a family their mother will fall.
The practical check is simple. Put the cohort report next to the members your care managers already flag for fall risk. Where they overlap, the model confirms the team. Where the model points and the team was not looking, that is the cohort worth a call.
What the record can show, with less behind it: hospital return
The hospital-return head reports expected 90-day return by cohort. It runs on your cohort now. It has been trained and has no sealed-test result, and Scienza says so on every page where it appears.
What that means in practice: treat the report as a reason to look, and compare it against what your transitions team already knows. If the cohort report matches the discharges your nurses already worry about, you have learned the model agrees with them. If it does not, you have learned where to look next. Either is useful. Neither is a validated claim.
What the record cannot show, and what a call adds
The record stops at the last encounter. It does not know that a member has stopped going out, that the stairs got harder last week, or that the new medication makes her dizzy in the morning. None of that is in a claim.
It is in a conversation. Gia calls the members the report points to, by phone or video, in 44 languages, with retries. A 40-second speech check-in measures 2,500 speech biomarkers and adds how the member sounds today. Results reach the clinician in under two minutes as a ranked worklist.
Each result arrives with Proactive Decision Actions. Options, never orders. The clinician decides who gets a home safety visit, a medication review or a nurse call after discharge, and the decision writes back to the EHR. The AI co-clinician is clinical decision support, not a diagnosis. The report says where to look; the call says what is happening; the clinician says what to do.
Where to start
Pick two cohorts: members discharged in the last 30 days, and members over 75 with a fall or a fracture in the last two years. Run both heads on your own records, read the aggregate report with your clinical lead, and have Gia call the top of each list.
See how ACOs and health systems use Gia for the days after discharge, then book a working session.
Sources & References
- CMS, Transforming Episode Accountability Model (TEAM): https://www.cms.gov/priorities/innovation/innovation-models/team-model
- CMS, Hospital Readmissions Reduction Program (HRRP): https://www.cms.gov/medicare/payment/prospective-payment-systems/acute-inpatient-pps/hospital-readmissions-reduction-program-hrrp
David Kaiser is the Founder and CEO of Scienza Health, the company behind Gia, an AI co-clinician that health plans, ACOs, health systems and senior-care operators use to check in with older adults between visits.
This content is intended for informational purposes and does not constitute medical advice. Citing research does not mean this page or the product has been independently peer reviewed.