Sharing executable phenotype definitions beyond OHDSI network？

Researchers
1

Hi OHDSI community,

I’ve been thinking about a gap in phenotype sharing and would love your input.

The Problem I’m Seeing

When I review published studies using OMOP data, I often find:

  • Paper describes: “AKI defined as creatinine increase >0.3 mg/dL within 48 hours”
  • Actual implementation: Buried in 200 lines of SQL or Python, with implicit assumptions about baseline, measurement timing, etc.
  • Result: Two teams implementing “the same” phenotype get different cohorts

ATLAS cohort definitions work great within OHDSI, but I’ve struggled with:

  1. Sharing with non-OHDSI collaborators - JSON exports require ATLAS to interpret
  2. Reviewing published logic - Hard to diff two phenotype definitions
  3. Temporal patterns - Expressing “rising trend over 6 hours” vs “single threshold”

What I’ve Been Experimenting With

I’ve been working on an open-source project called PSDL (Patient Scenario Definition Language) - a YAML-based format for clinical scenario definitions.

Example (AKI detection):

scenario: AKI_Detection
version: "0.3.0"

audit:
  intent: "Detect early acute kidney injury"
  rationale: "KDIGO criteria for AKI staging"
  provenance: "KDIGO Clinical Practice Guideline (2012)"

signals:
  Cr:
    ref: creatinine
    concept_id: 3016723  # OMOP concept
    unit: mg/dL

trends:
  cr_delta_48h:
    expr: delta(Cr, 48h)
    description: "Creatinine change over 48 hours"

logic:
  aki_stage1:
    when: cr_delta_48h >= 0.3
    severity: medium
    description: "AKI Stage 1 - creatinine rise ≥0.3 mg/dL"

Key features:

  • Human-readable - Clinicians can review without SQL knowledge
  • OMOP-native - Uses concept_ids directly
  • Temporal operators - delta, slope, ema, min, max over time windows
  • Audit metadata - Built-in intent, rationale, provenance fields
  • Portable - Same file works across different OMOP instances with site-specific mappings

GitHub: GitHub - Chesterguan/PSDL: Patient Scenario Definition Language

Questions for the Community

  1. Is this a real problem? Or do existing tools (ATLAS, CohortDiagnostics) already solve this well enough?
  2. Temporal logic - How do you currently express patterns like “creatinine rising over 6 hours” in ATLAS? Is there a standard approach?
  3. Cross-network sharing - When collaborating with sites outside OHDSI, how do you share phenotype definitions?
  4. What’s missing? - Looking at the format above, what would make it more useful for your work?

I’m genuinely trying to understand if this fills a gap or duplicates existing efforts. Happy to contribute to OHDSI tools if there’s a better path forward.

Thanks for any feedback!

Links:

This is an open-source side project - feedback and criticism welcome!

I