AI4PD

← Back to CVC ProjectsHomeApproachEvidenceTeamPartners

Partner with AI4PD

Validate a Parkinson's progression model on your cohort, and keep the cross-site evidence it produces

AI4PD is building a clinician-facing Parkinson's digital twin: tools that estimate a patient's current state and forecast progression, each returned to the physician with explicit uncertainty. The modeling is tractable. The bottleneck is assembling and validating multi-institution longitudinal evidence across sites and populations, which is what we need cohort partners for. Identifiable data never leaves your institution; you receive only de-identified records, under a signed agreement, on UT Austin secure infrastructure. Your site keeps ownership, and you keep the validation evidence your cohort helps produce.

We partner with movement-disorders neurologists and cohort custodians — bajaj@cs.utexas.edu.

How engagement works

  1. Scoping call. What your cohort covers, what questions matter to your clinic, and what is feasible.
  2. Agreements. A signed Data Use Agreement, and the IRB pathway for your institution: we establish an IRB reliance agreement or defer to your IRB, as you prefer.
  3. De-identification at your site. Identifiable data stays with you. You share only de-identified records (HIPAA Safe Harbor or Expert Determination, your choice). Because no PHI is transferred, AI4PD acts as neither a Covered Entity nor a Business Associate, and no BAA is required. If a project ever needs limited identifiers, that is scoped and signed separately before anything moves.
  4. Joint harmonization. We map your source schema and build a de-identified longitudinal timeline together, with audit trails. Expect roughly a few days of your data manager's time for the initial mapping; after that, records flow from your existing systems with no new clinical workflow.
  5. Clinician review. Your neurologists review the outputs.
  6. Validation and follow-up. Cross-site results are reported back to you; the timeline and models update as new visits arrive.

Your data, your terms

You retain ownership.
Participating clinicians and sites keep ownership of their data. Participating does not transfer it.
Withdraw and delete.
You may withdraw at any time. On withdrawal we delete your data and any single-site models derived from it.
No secondary use.
We use your data only for the validation work scoped in the DUA. No transfer to third parties, and no commercial use of cohort-derived models, without your written consent.
Publication.
Publication terms and co-authorship are defined in the agreement.
Access on request.
Access to your cohort's data is restricted to a named project team and logged. The access log is available to your institution on request.

What a clinical partner contributes

No site is asked to hand over a finished, cleaned dataset. Harmonization is done jointly, and each data stream arrives on whatever clock your clinic already generates it.

  • Longitudinal clinical records — EHR, medication histories, and follow-up outcomes over time.
  • Imaging and biomarkers — Structural and functional imaging and available biomarker panels, contributed as acquired.
  • Clinical scores per visit — MDS-UPDRS and related motor and non-motor assessments on the visit cadence.
  • Sensor and wearable streams — Continuous motor data where collected.
  • DBS / LFP records — Deep-brain-stimulation programming and local-field-potential data at sites where available.

The timeline grows at the pace your clinic already works.

What a partner gets

Decision support, returned to the physician first

The core target is 12-month progression forecasting against named endpoints (MDS-UPDRS-III, time-to-motor-fluctuation), each output returned with explicit uncertainty and named evidence gaps. This is the capability your cohort would help validate; it is in development and not yet prospectively validated, and we say so on every output.

Around that target, the collaboration aims to produce: current-state estimation, differential-diagnosis and levodopa-response support, DBS programming support where applicable, and ranked therapy and monitoring plans. Every output is human-in-the-loop decision support, returned to the physician. The system never diagnoses or prescribes autonomously.

Validation evidence your cohort produced

  • Cross-site results. How the model performs on a cohort like yours, including where it fails.
  • Calibration you can check. Whether its confidence is honest: when it says it is 80% sure, is it right about 80% of the time. Methods detail (baseline comparisons and ablations) is in the validation report.

A harmonized, de-identified patient timeline

  • Schema mapping co-developed from your source systems.
  • Audit trails over the de-identified longitudinal timeline.
  • A reusable asset for your own downstream research, on the use terms above.

Governance, stated honestly

Where data lives.
De-identified data resides on UT Austin secure computing infrastructure approved for restricted data, not on general-purpose HPC or researcher laptops.
Legal instrument.
Transfer is governed by a signed Data Use Agreement covering de-identified records. We name the UT Austin IRB of record and the responsible data custodian in the agreement.
EU and other jurisdictions.
For sites subject to GDPR, such as EU collaborators, transfer is handled either through fully anonymized data outside GDPR scope or under appropriate safeguards (for example, Standard Contractual Clauses), specified per site before any data moves.
Constrained sharing.
The default is data-minimizing transfer of de-identified records to UT Austin infrastructure. For sites that require records not to leave their walls, we will scope a no-egress alternative.
Responsible use.
Outputs are decision support only and go to the physician first. External domain review, including DBS where applicable, and responsible-use documentation are part of the engagement.

Funding partners

Infrastructure partners

Oden Institute for Computational Engineering and Sciences

Shared compute and administrative support for cross-campus, restricted-data deployments.

Texas Advanced Computing Center (TACC)

High-performance GPU clusters for large-scale modeling and validation.

Learn more

How to start

For clinical-cohort, data-sharing, and validation partnerships, write to the PI directly: Prof. Chandrajit Bajaj (Principal Investigator) — bajaj@cs.utexas.edu (faculty profile). For institutional or partnership inquiries, use the Oden Institute contact portal.

To discuss philanthropic support, the UT Austin giving page is here.