The results of a large randomized trial conducted in the UK reported that a single reader using CAD technology would perform as well as two readers in detecting breast cancer in women aged between 50 and 70 . Based on these data, this analysis examined the cost-effectiveness of using single reading with CAD compared with double reading in the NHSBSP. As with previous cost effectiveness analyses this study suggests that CAD might be a cost-increasing intervention compared with double reading in all sizes of screening units because the savings arising from the shorter CAD reading time will be offset by the cost of staff training, the cost of purchasing, upgrading and maintaining the CAD equipment, and in particular the increased cost of assessment. [5, 30, 31]. It has been estimated that introducing single reading with CAD would produce an additional cost of £227, £253 and £590 per 1,000 women screened in high, average and low volume units respectively.
The results of the one way sensitivity analyses show that the baseline incremental cost of single reading with CAD versus double reading was highly sensitive to model parameters assumed for all sizes of units. Assessment cost, reading time per case, reader qualification and the difference in recall rates were found to have the highest impact on the incremental cost of CAD. As expected, a longer reading time per case would increase the potential saving arising from the introduction of CAD and CAD would be cost effective in all types of units. Similarly, the higher the reader qualification (and thus the cost of reading per minute) the higher are the savings arising from CAD. It has yet to be determined, if CAD were introduced routinely, who will be allowed to read mammograms (only radiologists or radiologists and radiographers). Sensitivity analyses also suggests that if the difference in recall rate is as low as 0.3%, then CAD would be cost effective in both high and average volume units . Probabilistic sensitivity analysis confirms the robustness of the study findings. When all parameters are varied, the probability of CAD being cost effective is 8% in high volume units, 7% in average volume units and only a 4% chance of being cost saving in low volume units.
If cost-effectiveness were to be measured in terms of cost per recall averted, double reading would dominate CAD in all the types of units (being less costly and more effective).
The present study improves the existing evidence on the cost effectiveness of CAD versus double reading in routine breast screening in UK. The effectiveness data used to populate the model were obtained from a large prospective randomised trial evaluating CAD versus double reading in UK breast screening units.
In December 2007, the UK Government's Cancer Reform Strategy made the commitment to replace film mammography with digital mammography . Digital mammography has been shown to be a cost effective alternative to film mammography in UK and will progressively replace film screen mammography in the next few years . This study evaluates, for the first time in the UK, the cost effectiveness of single reading with CAD against double reading assuming that screening units use digital mammography systems. As noted by Lindfors et al. , the cost effectiveness of CAD is heavily dependent of the cost of CAD technology and implementation of CAD in a digital mammography setting should be a much less labour intensive process and likely to increase its cost effectiveness.
In interpreting the results and findings, several limitations and assumptions should be considered. The study assumes that the effectiveness of CAD using a digital mammography system is the same as observed in the CADET II trial where film screen mammography was used. To date, there have been no prospective studies conducted on the effectiveness of CAD using a digital mammography system. Further research is needed to establish whether, following the introduction of digital mammography, single reading with CAD and double reading will continue to show similar effectiveness in cancer detection rates. Another limitation of this study is that the reading time is based on an examination of four digital images. When a woman is screened for the second time, the reading time per case may be longer if prior mammograms have to be viewed for comparison, thus the potential savings in reading cost by using single reading with CAD may have been underestimated [32, 33]. However, the availability of previous round mammograms could make it easier for the reader, in the case of CAD, and readers, in the case of double reading, to decide and thus reduce average reading time. As a consequence, the potential savings in reading cost resulting from the introduction of CAD in routine screening are difficult to estimate. Although the unit cost of an assessment visit used in the present study has been estimated using recent data from NHSBSP it is subject to great uncertainty. Procedures for diagnosing breast cancer (and thus the costs of assessment) vary between regions. For example in Northern Ireland where there is still a high proportion of cancer diagnosed by cytology only (4%), the average cost of an assessment visit is likely to be lower than the national average . In addition the costs of assessment are sensitive to procedural innovations. This analysis, for instance, has not taken account of the use of Vacuum Assisted Biopsy (VAB) devices. There is increasing evidence that VAB is a powerful tool to diagnose microcalcifications, to improve the diagnostic rate and to reduce under-staging. However, due to both the high cost and the high non operative diagnosis rate achieved by other sampling methods, the use of VAB has been limited in the NHSBSP . If VAB were to be used in routine clinical care the cost of assessment and thus the incremental cost of CAD versus double reading would increase significantly.
In addition, the study does not distinguish between the reading time of radiologist and non-radiologist readers. Although no significant difference in reading time between these professional groups has been reported were there to be one the potential saving arising from CAD would have been underestimated . The estimated price of purchasing and upgrading the CAD systems might be conservative. If CAD is introduced into routine practice in the screening programme the price of a CAD system and upgrading is likely to decrease due to bulk purchases and price negotiation with commercial suppliers. In addition, the number of CAD systems and workstations that screening centres would purchase is difficult to estimate since this is dependent on both screening volume and the number of readers.
There are a number of clinical issues associated with the use of CAD that should be addressed within the limitations and assumptions of this analysis.
It has been assumed that the lesion types recalled by single reading with CAD are similar to those with double reading. In CADET II, for cancer cases, there was no significant difference in recall rate between calcifications and masses .
However, there is no information on the suspected lesion type that prompted recall of non cancer cases (false positives), by either reading regimen. Thus assessment costs have been assumed to be comparable for both arms of the study. However, there is evidence that CAD has a greater sensitivity with respect to detection of microcalcification, particularly when used with digital mammography, that could result in higher recall rates and associated assessment costs . Thus assessment costs for single reading with CAD may be underestimated. Recall after screening mammography has been associated with increased anxiety and stress that could also result in increased healthcare costs and contribute to an underestimate of the true costs of recall and assessment in the current analysis [37, 38].
In addition, no account has been taken of any costs arising from the need to involve another reader to arbitrate discordant cases. In view of the high false marker rate with CAD, arbitrating all cases where there is discordance between the single reader and CAD would be equivalent to double reading with CAD and would not be cost effective . If the NHSBSP introduced single reading with CAD, appropriate training to familiarise readers with the performance of CAD systems (should minimise the number of cases that would require arbitration by a third reader [40, 41].