Prereading of Screening Mammograms by Technologists May Not Improve Interpretation

Prereading of screening mammograms for breast cancer by radiologic technologists may not benefit interpreting radiologists, a new study suggests.

The study, published in Radiology, included 109,596 women who underwent screening mammograms for breast cancer from September 2017 to May 2019. It was designed to evaluate the radiologists’ early screening outcome measures when blinded to warning signals from technologists during interpretation of mammograms compared with nonblinded. The warning signals are alerts to abnormalities suspicious for cancer observed by the technologist.

“Radiologists should be aware of the possible impact of (additional) information during the reading of images,” noted Tanya Geertse, who is affiliated with the Dutch Expert Centre for Screening in the Netherlands. “This information can also make them more insecure. Therefore, it does not always have to yield a positive effect.”

Among the participants, 53,291 were included in a blinded group and 56,305 were in the nonblinded group. Screenings were assessed by 41 technologists, with a median of 14.6 years of experience, and 14 radiologists, with a median of 12 years of experience.

Those in the blinded group had lower overall recall rates (2.1% vs 2.4%, P = .001) and higher positive predictive values of recall (30.6% vs 26.2%, P = .02) than those in the nonblinded group, the study said. No difference in cancer detection rate was reported (6.5 per 1,000 screening examinations in the blinded group versus 6.4 per 1,000 screening examinations in the nonblinded group, respectively; P = .75).

“I was a bit surprised by the result,” Geertse said. “We already knew that the technologists are sometimes too sensitive and give relatively many warnings. Some radiologists also complained about this.”

Prereading was introduced in the Dutch screening program in 2003, with the decision to present warning signals to the radiologist upon opening the screening. The program was intended to increase motivation of technologists and decrease unnecessary recall and improve cancer detection.

“When implementing such tools (for example computer-aided detection systems) in practice, careful evaluation is needed to decide whether these markers should be presented during or after reading images,” Geertse said.

While the study showed that presenting technologists’ warning signals to radiologists during their initial interpretation were not effective, more consideration should be given to the timing of the information. The study authors suggested that warning signals could be presented immediately after interpretation to allow radiologists an opportunity to change their recall decisions.

“The procedure in the Dutch screening program has been adjusted in the meantime,” Geertse said. “The radiologists first read the images, and immediately after interpretation (if applicable), the warning signal is shown. In that case, the radiologist has the option to change his or her recall decision. Research is needed to show whether this procedure produces better results.”

Careful consideration also is needed to decide when to present markers from artificial intelligence (AI) and new computer-aided detection (CAD) systems.

“In both its objectives for improved accuracy and its process for identifying and alerting the radiologist to areas of potential suspicious imaging findings, the Dutch prereading strategy is analogous to CAD and newer AI-driven adjunct mammographic screening technologies,” Solveig Hofvind, Ph.D., and Christopher I. Lee, M.D., M.S., wrote in an associated editorial.

“All of these warning systems are meant to help streamline the work of interpreting radiologists by helping them focus on the most suspicious area of a mammogram. However, as we have learned from the study by Geertse et al and seminal studies on the accuracy of traditional CAD for mammography, these systems may not be benefiting interpreting radiologists or the women undergoing routine screening.”