Order, time of mammogram reading does not affect cancer detection rate

The efficacy and accuracy of mammography readers did not diminish over time on task, according to the results of a randomized trial.

Also, fewer false-positive recalls occurred among patients whose slides were read near the end of a batch, results showed.

“Psychologists have been investigating a phenomenon of a drop in performance with time on a task called ‘the vigilance decrement’ since World War II,” Sian Taylor-Phillips, PhD, assistant professor of population evidence and technologies at University of Warwick Medical School, said in a press release. “In those days, radar operators searched for enemy aircrafts and submarines, which appeared as little dots of light on a radar screen. People thought that the ability to spot the dots might go down after too much time spent on the task.”

In the United Kingdom, mammography slides are evaluated in batches by two independent film readers. Some studies have observed a decline in detection rates and increase in recalls related to time spent interpreting mammography slides.

Taylor-Phillips and colleagues conducted a multicenter, double blind clinical trial to examine whether a vigilance decrement occurred among qualified screening mammography readers.

A total of 360 readers (radiologists, n = 186; radiography advanced practitioners, n = 143; breast clinicians, n = 31) from 46 specialized breast centers in England participated in the study.

Paired readers in the control group reviewed each batch of digital mammograms in the same order, whereas readers in the intervention group reviewed mammograms in the opposite order of one another.

Cancer detection rate served as the primary outcome measure. Secondary outcome measures included recall rates and rates of disagreement between readers.

The study included data from 1,194,147 women (mean age, 59.3 ± 7.49 years) who underwent screening mammograms between December 2012 and November 2014 (intervention group, n = 596,642; control group, n = 597,505).

The images were interpreted in 37,688 batches, with a median batch size of 35 mammograms (interquartile range [IQR], 16-46). Each reader interpreted a median of 176 batches (IQR, 96-278).

Screening mammography detected 10,484 (0.88%) cases of breast cancer after completion of all diagnostic tests.

The intervention group had a cancer detection rate of 0.88% (n = 5,272), compared with 0.87% (n = 5,212) in the control group (difference, 0.01% points; 95% CI, –0.02 to –0.04). The intervention group did not affect the cancer detection rate (OR = 1.01; 95% CI, 0.97-1.06).

The intervention group also had no effect in the subgroups of younger age, first and last five cases in the batch, the first batch of the day for both readers, or batches examined second in the day or later by both readers.

The cancer detection rate for batches read first in each workday by both readers was 0.83% (n = 580) in the intervention group and 0.88% (n = 623) in the control group (difference, –0.05%; 95% CI, –0.15 to 0.04).

The detection rate for second or subsequent batches was 0.85% (n = 2,472) in the intervention group and 0.85% (n = 2,473) in the control group (difference, 0.002%; 95% CI, –0.045 to 0.5).

The intervention group did not significantly improve recall rate (4.14% vs. 4.17%; difference, –0.03%; 95% CI, –0.1 to 0.04) or disagreement rate (3.43% vs. 3.48%; difference, –0.05%; 95% CI, –0.11 to 0.02).

An exploratory post-hoc analysis showed the cancer detection rate for individual readers did not change based on time spent working on a task, with nearly identical odds shown for cancer detection between the first and fortieth case (OR = 0.987; 95% CI, 0.92-1.04). Further, the recall rate for individual readers decreased with time spent on task (OR = 0.83; 95% CI, 0.87-0.91), and no difference occurred when readers were blinded to one another’s decision (0.88% vs. 0.87%; difference, 0.01%; 95% CI, –0.05 to 0.07).

The researchers acknowledged study limitations, including their inability to control for reading conditions. The study also did not attempt to blind second readers at centers where this was not already common practice.

“We found no reduction in performance of vigilance decrement at all,” Taylor-Phillips said. “In fact, we found the opposite of what we were expecting — breast screening readers seemed to get ‘into the zone’ and their performance improved with time on task.”

The trial’s design may aid and influence evidence-based medical research in the future, Elizabeth S. Burnside, MD, MPH, MS, of University of Wisconsin School of Medicine and Public Health, Edward A. Sickles, MD, of University of California at San Francisco, and Stephen W. Duffy, PhD, of University College London, wrote in an accompanying editorial.

“Perhaps the most important message conveyed by this study is the potential afforded by combining standardization and advanced information systems to use pragmatic study designs to answer important questions applicable to the usual care setting,” Burnside and colleagues wrote. “This opportunity promises to effectively narrow the gap between randomized trials and evidence-based interventions so that clinical advances can rapidly influence practice.” – by Cameron Kelsall

Disclosure: Taylor-Phillips reports postdoctoral fellowship funding from the U.K. National Institute of Health Research and grant support from the U.K. National Screening Committee for research related to this study. Please see the full study for a list of all other researchers’ relevant financial disclosures. Burnside, Sickles and Duffy report no relevant financial disclosures.