European guidelines on breast cancer screening and diagnosis

The GDG suggests by consensus that double reading (with consensus or arbitration) over single reading be used to diagnose breast cancer in mammography screening.

Only one study with digital mammography was included in the evidence, which limited the GDG to be able to issue a strong recommendation.

The GDG notes that in the context of double reading with consensus or arbitration, no differences were observed in accuracy when arbitration or consensus or both were used to reconcile differences in interpretation between mammography readers.

• In settings with many low-volume mammography readers, the balance of benefits and harms may be even greater. The GDG refers readers to the PICO Question 7: ‘What is the optimal annual interpretive volume for radiologists reading screening mammograms?’ in the CCIB report, addressed by the QASDG regarding the experience level of mammography readers.
• In some settings, capacity (human resources of mammography readers) should be scaled up to implement double readings. In settings where double readings are already in practice, the GDG suggests continued use of double readings with consensus or arbitration.
• The GDG notes that a consideration that can favour double reading with consensus or arbitration is in those settings with many low volume mammography readers; the desirable effects of double reading with consensus or arbitration were found to be greater with less undesirable effects in these settings as compared to high volume mammography reader settings.

The GDG suggests reporting the proportion of double reading with consensus or arbitration of mammograms that occur in practice. The GDG refers this suggestion to the QASDG for consideration.

• The GDG suggests further research examining the cost-effectiveness of double vs single reading of digital mammography in different settings. Cost-effectiveness data was only identified for Spain.
• The GDG suggests new research using observational studies comparing double reading with consensus or arbitration with single reading in the context of digital mammography. Additional research could also be performed to assess accuracy within the context of double readings assessing a single reader vs with the addition of a second reader, which is performed in practice.
• The GDG suggests the use of formal radiologist blinding in research to improve the quality of evidence on double vs single readings.
• The GDG notes that newer screening strategies such as digital breast tomosynthesis (DBT) or automatic computer assisted detection (CAD) was excluded from the analysis of this question with double vs single mammography. Future research could assess the impact of double reading using CAD and/or DBT systems.

The right balance between sensitivity (true positives) and specificity (false positives) was found within the selected range of readings per reader.

• This optimal range should be implemented as part of screening programmes that use double reading (which is a conditional recommendation previously issued), as double reading of mammograms will improve the overall quality of the reading.
• Efforts to provide support with centralised reading may help implementing this recommendation.
• The number of readings should be averages over longer periods and use appropriate sampling for this measurement.
• The GDG recognised that local circumstances of the individual's performance will affect the quality of readings as much as the number of readings.
• The suggested range applies to mammography screening, not tomosynthesis.

Programmes should continue to monitor the number of readings by readers and the quality indicators of the reading process.

The GDG suggested the following research priorities:

• More published evidence is required which could also come from good monitoring data.
• Further research on elements that may influence results such as: when to read during the day, how many mammograms to read per day, time needed and variation of time for readings between readers, relationship between breast cancer detection and false positives, comparison between those reading only screening mammograms and those doing screening and clinical/diagnostic mammography.
• More research on types of training for mammography readers.
• Research into the role of artificial intelligence in double reading.

The GDG felt that no training was not appropriate but there was uncertainty about several factors that influenced the direction and strength of the recommendation. There was also uncertainty about what exactly the intervention entails.

This recommendation was discussed as being applicable to all parts of the breast cancer screening programme, although what processes the screening programme covers was not exactly defined.

Another reason for a conditional recommendation was due to the uncertainty about the evidence and the costs required for the relevant human resources (if they are not available), as well as not knowing exactly what training to provide, which may prevent programmes from proceeding. In addition, there is uncertainty on the type of information to be presented to women, depending on their sociocultural background.

This applies equally to the complete pathway of mammography screening (including further assessment when it is part of the screening programme).

There should be a periodic monitoring of the interventions using satisfaction surveys of participants.

The GDG agreed that better knowledge about what women actually want to know about will inform the interventions and the type of training, if any.