Factors Causing Variability When Reading Mammograms

Abstract

Aims: This thesis investigated the impacts on radiologist performance of 1) the time of day at which an image is read, 2) how long a radiologist has been awake and the hours slept the night before and 3) the number of cases read without a break. Studies 2 and 3 also examined whether the influence of the studied factors varied according to reader experience.

Methods: Data were collected during conference workshops, and radiologist accuracy was evaluated with BreastScreen Reader Assessment Strategy test sets, each containing 60 digital mammography cases. In study 1, 197 reader assessments were used to assess specificity, lesion sensitivity and JAFROC; in study 2, 133 reader assessments were used to assess sensitivity, specificity, lesion sensitivity, JAFROC and ROC AUC; and in study 3, 10 radiologists completed a test set without breaks to assess sensitivity, specificity, lesion sensitivity and ROC AUC; interactions between the fixed-series sequence and radiologist experience were also examined.

Results: Study 1: The specificity was lower in the late morning and late afternoon than the early morning or mid-afternoon. Study 2: The lesion sensitivity of less experienced radiologists was lower for those awake less than 2 hours than those awake 8 to 10 hours and lower for those awake 4 to 6 hours than those awake 8 to 12 hours. The ROC AUC of less experienced radiologists was also lower for those with less than 6 hours of sleep than for those who slept longer. Study 3: Linear interactions were found between experience and the fixed-series sequences for sensitivity and lesion sensitivity, with experienced readers improving during the last series and less experienced readers deteriorating in performance.

Conclusions: Time of day, hours awake, hours slept and the number of cases read without a break can all impact radiologist performance, which has significant implications for workday planning and accurate reporting.