Re: Association of Volume and Volume-Independent Factors With Accuracy in Screening

Studies such as the recent one by Beam et al. (1) that purport to evaluate the reading skills of radiologists inter-preting screening mammograms only point out the pitfalls and complexities of these analyses. The only true test of a radiologist’s skill in interpreting screen-ing mammograms would be to perform a study that accurately reflected the screening setting. Because this would involve the readers interpreting the same tens of thousands of mammograms, an accurate study is not feasible. Unfortu-nately, the study design used by the au-thors introduces a number of important and undocumented biases that make the conclusions highly suspect. Although the authors suggest that the initial cases were chosen by “random sampling, ” the ultimate case distribution was highly se-lected. There were 64 (43%) cancers among 148 women. As the authors ac-knowledge, this is far more cancers than a single radiologist would ever expect to see in a year let alone among a group of 148 women. This is almost 100 times more cancers than would be seen in a group of 1000 women undergoing screening. The only instructions that the radiologists were given were that the cases “did not have the mixture of mam-mograms expected from a typical screen-ing population. ” This “slight understate-ment ” of course could be interpreted in many ways. The authors should have queried the radiologists to determine what they thought was the prior prob-ability of cancer in the group because this would influence how each inter-preted the studies. I suspect that few ra-diologists expected that almost half the women had cancer. An important question that is not an-swered is the mix of cancers that were included. Other studies have included a large number of cancers that are border-line in appearance, and I suspect that the same was true for this study. This will also skew the results. An experienced radiologist would be more likely than