Is It Time to Stop Paying for Computer-Aided Mammography?

Abstract

Computer-aided detection (CAD) is a technology designed to address the problem of screening mammography’s imperfect sensitivity. Now used on over 90% of US mammograms, CAD essentially acts like an automated second reader by marking potentially suspicious spots for radiologists to review before making final recommendations. Early studies suggested that CAD could increase cancer detection rates by 20%.1 But subsequent research suggested little, if any, impact of CAD on cancer detection and raised concerns that CAD may increase recall and biopsy rates.2,3 However,most clinical studies to date have assessedCAD impacts when used with film mammograms. Digital mammographyhas now largely supplanted filmmammography in the United States. When used in the context of digital mammography, does CADyield net benefits towomen?A study by Lehman et al4 in this issue of JAMA Internal Medicine addresses this important question. In an observational study of 323 973 women undergoing digital screening mammography in diverse US practices, Lehmanet al4 found thatCADusewasnot associatedwith any improvement in sensitivity, specificity, positive predictive value, cancerdetection rates, or otherproximal screeningoutcomes. Indeed, among radiologists who interpreted digital mammograms with and without CAD, sensitivity was worse with CAD, contrary to CAD’s design. While earlier evaluations suggested that community radiologists often overreacted to CAD output, leading to higher rates of diagnostic investigation,2,5 Lehmanet al4 found little, if any, impact of CAD in modern digital mammography practice. It is possible that, with years of CAD use, many radiologistshave learned that theyieldofCAD isminimal so theynow largely ignoreCADoutput. It is also conceivable that improvements indigitalmammographytechnologyhaveswampedany incremental impacts of CAD on interpretation that may have been previously detectable. This observational studymay be confoundedbyunmeasured radiologist ormammography facility factors, althoughearlier research adjusting for these factors also found no benefits of CAD.2 Like all subgroup analyses, analyses of outcomes among subsets of radiologistsmust be interpretedcautiously.Despite these limitations, this study4 is another large-sample, real-world evaluation of CAD’s interpretive outcomes suggesting that CADyields no clinically significant benefits in typical mammography practice. The field of implementation science should take interest in interventions like CAD that are widely adopted in advance of strong evidence of effectiveness.Whatmade CAD so alluring topatients,practitioners,orboth,andwhywerepayerswilling (at least initially) to finance CAD?How is it that CAD is applied on 90% of US mammograms when it yields no clear benefits to patients? Thefirstessential step forbroadCADadoptionwasUSFood andDrugAdministration (FDA)approval in 1998.BecauseCAD is a device rather than a drug, the evidence bar for FDA approvalwas comparatively low. Its approvalwasbasedonsmall studies of CAD’s “safety” and “effectiveness.” Effectiveness, for example, was demonstrated by studies in which radiologists read sets of mammograms with enriched breast cancer prevalence, suggesting that CADcouldprompt increased cancerdetection. In addition,Congressionalmembers, lobbiedby industry, pressured the FDA to approve CAD.6 Evenso,CADwas still a longshot.UseofCADrequired film mammograms to be fed into machines to digitize images for computer analysis, andCADoutputhad tobeviewedondedicateddevices separate fromactualmammograms.Mammographywas already a loss-leader formany radiology practices, yet CAD added unreimbursed technician and radiologist effort. Without reimbursement, few mammography facilities could justify the capital costs forCAD installation.At the time, establishing reimbursement fornewpreventive services, such as CAD, required Congressional amendment of the Medicare statute.While Congress had previously addedMedicare benefits for preventive services, such as prostate cancer screening, these efforts required strong Congressional sponsorship and auspicious political winds.7 CAD ended up having both. Representing Silicon Valley (home of CAD’s leading manufacturer), California CongressRelated article page 1828 Digital ScreeningMammography and Computer-Aided Detection Original Investigation Research