Skip to main content
Scienza Health
Gia Health Analytics

Your next fall and next readmission are already in your data.

Scienza's foundation model for aging reads it first. Inside your own cloud project.

Run it on your cohortWatch the two-minute overview
Google Cloud MarketplaceAWS Marketplace
  • 469,000people in the sealed test
  • 5 of 5pre-registered outcomes passed
  • 4 to 8×share of events in the top five percent by risk
  • 0records that leave your project
Illustrative shape. Sample data.
The cost you already carry

Four events. One history predicts them all.

Hospitals and health systems

A hip fracture you own for thirty days

The fall that precedes it is the first sign anyone acts on.

ACOs and plans

A return to hospital inside the window

CMS penalizes excess readmissions. The admission itself lands on the plan or the ACO.

Medicare Advantage

A dementia diagnosis you must now document

Once a member is identified, the plan owns the encounter and the care plan.

Everyone who carries the member

A stroke and the months of rehab after it

One event, then a year of cost.

The model

Seven outcomes. One foundation model for aging.

Five outcomes pre-registered and passed. Falls replicated on a second look. Hospital return trained. All seven run on your cohort today; self-serve modules release in November 2026.

  • Dementia1 year
    Released

    Aggregate dementia report. Available now on the listing.

  • Falls1 year
    Replicated

    Available now. Self-serve module releasing November 2026.

  • Hospital return30 days
    Trained

    Available now. Self-serve module releasing November 2026.

  • Stroke / TIA180 days
    Passed sealed test

    Available now. Self-serve module releasing November 2026.

  • Parkinson's1 year
    Passed sealed test

    Available now. Self-serve module releasing November 2026.

  • Cognitive transition90 days
    Passed sealed test

    Available now. Self-serve module releasing November 2026.

  • Sustained cognitive decline1 year
    Passed sealed test

    Available now. Self-serve module releasing November 2026.

Released: self-serve module on the marketplace listing. Passed sealed test: a pre-registered outcome met its threshold on the sealed set. Replicated: the result held on a second, independent look. Trained: the head exists and runs on your cohort as a custom analysis.

How it was tested

Pre-registered. Sealed. Looked at twice.

  1. 1

    The rules were locked first.

    Outcomes, horizons, cohorts and pass thresholds written down before any result was seen.

  2. 2

    Then a sealed set of about 469,000 people.

    Never used in training. Opened for evaluation only.

  3. 3

    Two registered looks.

    Each declared in advance. No fishing between them.

  4. 4

    Passed on all five. Replicated on falls.

    The top five percent by risk carry four to eight times their share of events.

Sealed-test results, as reported in the manuscript

  • Cognitive transition (90 d)0.8830.880 to 0.886
  • Sustained cognitive decline (1 y)0.8660.863 to 0.868
  • Dementia (1 y)0.8280.825 to 0.831
  • Parkinson's, medication-naive (1 y)0.8060.788 to 0.822
  • Stroke / TIA (180 d)0.7820.776 to 0.788
  • Falls, replication (1 y)0.848second look

AUROC with 95% confidence intervals, resident-level bootstrap. Calibration held on four outcomes (slopes 0.96 to 1.01).

Pre-registered validation manuscript submitted to JMIR AI. The validation summary is available now.

Request the validation summary
Inside your project

Your data never leaves.

Deploys from the listing

Into a project or account you control. Billed on your existing cloud agreement.

Reads what you already hold

Diagnoses, encounters, medications, assessments. No new data collection to start.

Returns aggregates for review

Expected incidence by cohort, site and period. Members in the highest-risk band are counted, never named.

Aggregate dementia report · 12-month horizonSample data

Expected incidence by cohort

  • Site A · 1,240 members6.1%
  • Site B · 980 members4.8%
  • Site C · 1,610 members7.3%
212Members in highest-risk bandCounted, not named
486Expected events next 12 months

Aggregate only. Nothing leaves your project.

Thirty days. Your own data. Our fee waived. Then decide.

We run the dementia report on your cohort inside your project and walk your clinical and analytics leads through the result. Cloud usage bills on your existing agreement.

Start the thirty-day evaluationWatch the overview

Questions

Where does the model run?

Inside your own cloud project, deployed from the Google Cloud Marketplace or AWS Marketplace listing. Your data does not leave your environment and Scienza does not receive member-level records.

What does the released module produce?

An aggregate dementia report for your population: expected one-year dementia incidence by cohort, site and time period. Members in the highest-risk band are counted, never named.

How was the model tested?

The analysis plan was pre-registered. The model was then evaluated on a sealed set of about 469,000 people never used in training, across two registered looks. It passed on all five pre-registered outcomes and replicated on falls.

Can we run falls or hospital return now?

Yes. Both run on your cohort now. Falls has replicated on a second look; hospital return is trained. Self-serve modules for both release in November 2026.

Is this a diagnosis?

No. Gia Health Analytics is population analytics for clinical and operational review. Every output is an option for your clinicians to review.

Part of Gia by Scienza. See the Gia care platform for the speech check-in and clinician workflows.