Inter-observer and intra-observer variability of mammogram interpretation: a field study.
To evaluate the performance of radiologists in mammographic mass screening, seven radiologists read blindly the mammograms of 45 women (two views for each breast). The films included 12 normal, 24 benign disease and 9 cancers. The readings were repeated after 2 years. As expected, variability was higher among radiologists than between the two readings of the same radiologist, but general reproducibility was moderate. Kappa values for a positive/negative classification were 0.45 at the first and 0.44 at the second reading (inter-observer comparisons). For the intra-observer comparisons, Kappa values ranged from 0.35 to 0.67 (mean 0.56). Generally, accuracy was low partly due to the difficulty of the cases. A slight increase in sensitivity was observed at the second reading. The level of agreement is a good indicator of accuracy. Proper training and standardization of criteria are essential before mass breast screening is implemented.