Skip to content

EvidenceOSS/BRIDGE-TBI-Lite

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

1 Commit
 
 
 
 
 
 
 
 

Repository files navigation

BRIDGE-TBI-Lite — Open TBI Clinical Decision Support Rules

Status: Incubating License: Apache-2.0 Part of: Evidence Commons

"Any new TBI prediction model that cannot demonstrate calibration superiority over IMPACT Core and CRASH Basic has not earned clinical deployment."

Attribute Value
Status Incubating
Maturity Design Phase
License Apache-2.0
Part of Evidence Commons
Mission Pillar Pillar 1 (Clinical AI Evaluation & Benchmarking)

Overview

BRIDGE (Biomarker-Ready Infrastructure for Deploying Guideline-Aligned Evidence) is clinical decision support infrastructure for traumatic brain injury. BRIDGE-TBI-Lite is the planned open-source extraction containing validated TBI clinical decision rules, benchmark baselines, and evaluation tooling. The system is designed to enforce methodological discipline: any new model must demonstrate improvement over established baselines (IMPACT Core, C-statistic ~0.78-0.82 for mortality; CRASH Basic) using net reclassification improvement as the preferred comparison metric, with calibration superiority weighted above discrimination.

The parent codebase contains decision logic across the platform (evidenceos-health-platform) and research repository (evidenceos-research/evidenceos-tabular), including conformal prediction intervals for uncertainty quantification and NINDS CDE-compliant input/output schemas. No standalone extraction exists yet. This repository is intended to hold the extracted, independently usable decision rule implementations with sample size validation enforcing Riley et al. (2019/2020) criteria.

Regulatory note: BRIDGE provides clinical decision support (CDS) infrastructure. "Neurodiagnostic" describes domain positioning, not an FDA-cleared diagnostic claim.

Architecture

Component Description Platform Status
Decision Rules Executable clinical decision rules (Canadian CT Head, PECARN, CHIP, others) Designed
Baseline Models IMPACT Core + CRASH Basic reference implementations Designed
Conformal Intervals MondrianCQR uncertainty quantification from evidenceos-conformal Parent code exists
CDE Schema NINDS CDE v3.0 compliant input/output definitions from evidenceos-ontology Parent code exists
Sample Size Validator Riley pmsampsize calculation (EPV >= 10 necessary but not sufficient) Parent code exists
NRI Calculator Net reclassification improvement for model comparison Designed

Current State

What exists in the parent codebase:

  • Conformal prediction modules in evidenceos-research/evidenceos-conformal (MondrianCQR + 5 modules)
  • NINDS CDE ontology schemas (13 YAML files, 7,780 lines)
  • Tabular prediction pipeline in evidenceos-research/evidenceos-tabular
  • 12-axis multiverse engine (927 cells across 13 configs, 99.0% passing)

What does not exist yet:

  • Standalone extracted decision rule library
  • IMPACT Core and CRASH Basic reference implementations
  • NRI comparison tooling
  • INT-05 integration (BRIDGE feeding Clinical Arena for evaluation gating)
  • Test suite or CI pipeline for this repository

Extraction Plan

  1. Implement IMPACT Core and CRASH Basic as reference Python modules with published coefficients
  2. Extract conformal prediction interface from evidenceos-conformal
  3. Define NINDS CDE-compliant input schema as standalone JSON/YAML
  4. Build NRI calculator for head-to-head model comparison
  5. Integrate Riley sample size validation as a pre-analysis gate
  6. Publish v0.1.0 with baseline models, CDE schema, and comparison tooling

Ecosystem Context

graph LR
    A[evidenceos-conformal<br/>MondrianCQR] --> B[BRIDGE-TBI-Lite]
    C[evidenceos-tabular<br/>multiverse engine] --> B
    B -->|INT-05| D[Clinical Arena<br/>evaluation gate]
    B -->|INT-04| E[Evidence Capsules<br/>confidence scoring]
    style B fill:#2A9D8F,stroke:#1E3A8A,color:#fff
Loading

BRIDGE-TBI-Lite is part of the Clinic-in-a-Box product family. The full BRIDGE platform integrates with Evidence Capsules for confidence scoring (INT-04, not built) and Clinical Arena for model evaluation gating (INT-05, not built).

Canonical source: EvidenceOS/evidenceos-health-platform + evidenceos-research/evidenceos-tabular

Contributing

This project is not yet accepting contributions. The decision rule extraction and baseline model implementations are under active design. See CONTRIBUTING.md for future plans.

License

Apache-2.0 — see LICENSE for details.

About

Lightweight TBI outcome prediction SDK — conformal intervals, abstention, NINDS CDE mapping

Topics

Resources

License

Contributing

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors