Background Pragmatic integrated trials use routine data and systems to automate participant selection, randomisation, outcome measurements and/or other elements to deliver large trials at low cost. Here we present an example of a large pragmatic integrated trial in breast cancer screening, and discuss the situations in which these trials are appropriate and acceptable.

Methods The intervention was a simple change to the mammography test for breast screening. Interpreting whether screening mammogram show cancer is a difficult repetitive task that can result in missed cancers and false-positive recalls, and some studies have indicated that missed cancers may increase with time on task (the vigilance decrement). In the UK two readers independently evaluate each batch of mammograms to search for signs of cancer. The intervention was to change the order in which batches of mammograms were presented for interpretation, to reduce the effects of the vigilance decrement.

This was evaluated using a multicentre, double-blind, cluster randomised clinical trial at 46 breast screening centres in England for 1 year. Three hundred sixty readers participated. The primary outcome was cancer detection rate; secondary outcomes were rates of recall and disagreements between readers.

Results 1 194 147 women who had screening mammograms were randomised (596 642 in the intervention group; 597 505 in the control group), and 10 484 cases (0.88%) of breast cancer were detected. There was no significant difference in cancer detection rate with 5272 cancers (0.88%) detected in the intervention group vs 5212 cancers (0.87%) detected in the control group (difference, 0.01% points; 95% CI, −0.02% to 0.04% points). There was also no difference in recall rate, with 24 681 [4.14%] in intervention and 24 894 [4.17%] in the control group (difference, −0.03% points; 95% CI, −0.10% to 0.04% points). Patterns of cancer detection and recall with time since a break indicated that performance did not decline with time on task as predicted by the vigilance decrement theory. In fact, positive predictive value increased with time on task.

Discussion This pragmatic integrated trial in over 1 million women cost less than £300 k, and demonstrates that in certain circumstances this study design is appropriate. Considerations when planning a pragmatic integrated trial include whether consent is required at the individual or institutional level, whether the relevant outcomes are available in routine data, and the cost of the intervention.