case_id stringclasses 10 values | indication stringclasses 6 values | phase stringclasses 2 values | site_profile stringclasses 10 values | early_site_metrics stringclasses 10 values | signal_summary stringclasses 10 values | risk_assessment_draft stringclasses 1 value | gold_risk_level stringclasses 2 values | gold_failure_mode stringclasses 10 values | gold_correct_action stringclasses 10 values |
|---|---|---|---|---|---|---|---|---|---|
CSQDD-001 | Oncology | Phase II | New site to indication; low trial experience | Screen fail 62%; consent-to-randomization median 28d; query rate high | High friction pipeline | Low risk | high | screen_fail_spike | Flag high risk. Add prescreen workflow, retrain site, add backup sites, extend timeline |
CSQDD-002 | Immunology | Phase III | High volume site; competing trial ongoing | Enroll rate 0.3/mo vs planned 1.0; early dropouts 12% | Competitive displacement | Low risk | high | competitive_displacement | Flag high risk. Rebalance site mix, add non competing regions, adjust targets |
CSQDD-003 | Neuro | Phase II | Urban site; long travel catchment | Missed visit rate 18%; reschedule rate 22% | Visit burden signal | Low risk | medium | visit_adherence_collapse | Flag medium risk. Add home support, reduce visit load, transport support |
CSQDD-004 | Metabolic | Phase III | Primary care referral model | Referral conversion 6% vs expected 18%; prescreen backlog | Referral gap | Low risk | high | referral_pipeline_gap | Flag high risk. Build referral network, simplify prescreen, add outreach staff |
CSQDD-005 | Infectious | Phase II | Hospital site; limited after-hours staffing | Off-hours screens missed 40%; consent delays | Operational window failure | Low risk | medium | operational_window_loss | Flag medium risk. Add 24/7 screening coverage, streamline consent, extend window if safe |
CSQDD-006 | Oncology | Phase III | Biomarker dependent; pathology turnaround slow | Biomarker TAT 21d; tissue inadequate 30% | Biomarker ops drift | Low risk | high | biomarker_ops_delay | Flag high risk. Add central lab capacity, pre-screen registry, broaden tissue options |
CSQDD-007 | Cardiology | Phase III | High enrolling site but data quality weak | SDV findings high; query rate 3x; protocol deviations 9% | Data integrity risk | Low risk | high | data_quality_drift | Flag high risk. Increase monitoring, retrain staff, tighten deviation controls |
CSQDD-008 | Immunology | Phase II | Experienced site; tight eligibility | Screen fail 48%; slow enroll | Eligibility mismatch | Low risk | high | eligibility_mismatch | Flag high risk. Adjust criteria or add earlier line sites; update forecasts |
CSQDD-009 | Neuro | Phase III | MRI heavy protocol | MRI scheduling delay 35d; missed scans 14% | Capacity constraint | Low risk | high | capacity_constraint | Flag high risk. Secure imaging slots, add MRI-ready sites, extend timeline |
CSQDD-010 | Metabolic | Phase II | Run-in intensive diet stabilization | Run-in completion 62%; early withdrawal 20% | Run-in attrition | Low risk | medium | run_in_attrition | Flag medium risk. Shorten run-in, add coaching support, reforecast enrollment |
Clinical Site Quality Drift Detection v0.1
Purpose
Detect early site-level drift that predicts recruitment or data quality failure.
Model task
Return one JSON object
- risk_level
low, medium, high - failure_mode
one allowed label - correct_action
one short paragraph
Scoring
0 to 100
- risk accuracy 30
- failure mode accuracy 35
- action similarity 25
- format pass 10
Run
python scorer.py --predictions predictions.jsonl --test_csv data/test.csv
- Downloads last month
- 316