Rationale and objectives: To determine the incidence of radiology resident preliminary interpretation errors for plain film, body computed tomography, and neuroradiology (neuro)computed tomographic examinations read on call.
Materials and methods: We retrospectively reviewed the data in a prospectively acquired resident quality assurance (QA) database dating between January 2000 and March 2007. The database comprises all imaging studies initially interpreted by an on-call resident and later reviewed by a board-certified attending radiologist who determined the level of discrepancy between the two interpretations according to a graded scale from 0 (no discrepancy) to 3 (major discrepancy). We reviewed the data with respect to resident training level, imaging modality, and variance level. Statistical analysis was performed with chi2 test, alpha=0.05. We compared our results with other published series studying resident and attending accuracy.
Results: A total of 141,381 cases were entered into the database during the review period. Of all examinations, 95.7% had zero variance, 3.3% minor variance, and 1.0% major variance. There was a slight, statistically significant increase in overall accuracy with increased resident year from 95.4% of examinations read by first-year residents (R1s) to 96.1% by fourth-year resident (R4s) (P<.0001). Overall percentages of exams with major discrepancies were 1.0% for R1s, 1.1% for second-year residents, 1.0% for third-year residents, and 0.98% for R4s.
Conclusions: The majority of preliminary resident interpretations are highly accurate. The incidence of major discrepancies is extremely low and similar, even with R1s, to that of attending radiologists published in other studies. A slight, statistically significant decrease in the error rate is detectable as residents gain experience throughout the 4 years of residency.