Related Experiment Video
Updated: Jun 16, 2026
Enzymatic Synthesis of Epoxidized Metabolites of Docosahexaenoic, Eicosapentaenoic, and Arachidonic Acids
Published on: June 28, 2019
EHR-Driven Delivery of EPA Assessments Avoids Cherry-Picking Bias
Phillip D Jenkins1, Shelby Willis1, Julie Doberne1
1Surgical Data and Decision Sciences Lab, Department of Surgery, Oregon Health & Science University, Portland, Oregon.
Background:
Accurate entrustment measurement is critical for the overall validity of the national EPA initiatives and competency-based education (CBE) advancement. Yet when EPA assessments are initiated manually, residents and faculty often choose to assess cases where there is favorable performance. This cherry-picking bias threatens to distort the veracity of the EPA assessment cohort, blunt meaningful feedback, and halt educational progress. EHR-triggered assessments can mitigate cherry-picking bias.
Methods:
We analyzed 6890 EPA assessments across 10 general surgery programs (August 2023-October 2025). We compared entrustment scoring for manual versus EHR-triggered workflows. For 2 programs (A and B) that implemented EHR data integration during the study, we performed a time-based analysis to measure assessment changes from manual preintegration to automated postintegration workflows.
Results:
EHR-trigger implementation caused an immediate dip in entrustment levels, followed by a longer, variable increase as resident learning benefited from more consistent feedback (Program A: SD 0.94 to 1.00, variance ratio 1.13, Levene p = 0.57; Program B: SD 0.60 to 0.85, variance ratio 2.01, Levene p = 0.10). Automated triggering increased score heterogeneity, consistent with capture of a wider performance spectrum. Across all programs, mean assessment scores declined from manual to automated EHR-triggered workflows (2.72 ± 0.91, n = 2,141 vs 2.65 ± 0.94, n = 4,749; p = 0.005). Individual programs showed variability (Program G: Δ = -0.40, p < 0.001; E: Δ = -0.19, p = 0.003; F: Δ = -0.21, p = 0.001; B: Δ = +0.16, p = 0.009), but the overall mean decline remained -0.11. Triggering significantly changed entrustment distributions in 4 programs (χ² = 9.26-31.73, p < 0.05) and significantly decreased skewness toward more balanced measurement across entrustment levels.
Conclusions:
Automated EHR-triggered EPA assessments mitigated cherry-picking and enabled more comprehensive and realistic entrustment scoring throughout the procedure learning process.
