Computer-Aided Detection in Mammography: Downstream Effect on Diagnostic Testing, Ductal Carcinoma In Situ Treatment, and Costs

Since 2001, Medicare has reimbursed computer-aided detection (CAD) during screening mammography. The CAD software tool is used by radiologists to identify lesions suggestive of malignant disease. Research suggests that CAD use increases the rate of false-positive findings of screening mammography and the detection of ductal carcinoma in situ (DCIS).¹ Increased DCIS detection could lead to overdiagnosis of breast cancer, particularly among older women at risk for competing causes of death. We estimated the fraction of diagnostic tests, DCIS treatments, and costs attributable to CAD dissemination within the Medicare population, among whom the risk for overdiagnosis may be elevated.

Methods

This study was approved by the institutional review board of the University of California, Davis. Informed consent was waived. Using Surveillance, Epidemiology, and End Results Medicare-linked data from January 1, 2001, through December 31, 2009, we identified screening mammograms performed on female Medicare enrollees aged 67 to 89 years,² classified mammograms by CAD use, and computed the annual prevalence of CAD use. Using annual prevalences and CAD-associated incident rate ratios from a Medicare cohort study,¹ we estimated annual attributable fractions for diagnostic mammography, ultrasonography of the breast, biopsy of the breast, and DCIS diagnoses. We also computed the attributable fraction assuming 100% CAD prevalence because nearly all US mammography units are now digital,³ and digital units typically have integrated CAD.

Extrapolating to the entire US fee-for-service Medicare population, we used incidence rate differences to estimate the number of women who underwent diagnostic testing for breast cancer or who were diagnosed as having DCIS on account of CAD use each year. We used 2013 mean Medicare reimbursement rates and published diagnostic and treatment costs (adjusted to 2013 US dollars) to estimate annual Medicare costs attributable to CAD.⁴

Results

From January 1, 2001, through December 31, 2009, annual CAD prevalence among Medicare screening mammograms increased from 3.5% to 79.7%. By 2009, 18.2% of diagnostic mammograms, 5.3% of breast ultrasonograms, 7.4% of breast biopsies, and 11.9% of DCIS diagnoses were attributable to CAD (Table 1). From 2001 through 2009, 4612 additional fee-for-service Medicare enrollees were treated for DCIS on account of CAD. If CAD were applied to all Medicare screening mammograms, 14.5% of all DCIS diagnoses would be attributable to CAD, and 1118 additional Medicare enrollees would undergo DCIS treatment each year.

Table 1.

Diagnostic Tests and DCIS Treatments Attributable to CAD Among Medicare Enrollees, 2001–2009

Outcome	Incidence Rate Without CAD per 1000 Mammograms, %	CAD-Associated IRRa	Attributable Fraction, %				2001–2009
			2001 (CAD 3.5% Prevalent)	2005 (CAD 49.8% Prevalent)	2009 (CAD 79.7% Prevalent)	CAD 100% Prevalent	Expected No. Undergoing Testing or Treatment Without CAD	Additional Medicare Enrollees Undergoing Testing or Treatment Because of CAD
Diagnostic tests
Diagnostic mammography	58.4	1.28	1.0	12.2	18.2	21.9	2 028 526	265 432
Breast ultrasonography	38.0	1.07	0.2	3.4	5.3	6.5	1 319 932	43 476
Breast biopsy	13.3	1.10	0.3	4.7	7.4	9.1	461 976	21 248
DCIS diagnosis	1.7	1.17	0.6	7.8	11.9	14.5	56 233b	4612b

Abbreviations: CAD, computer-aided detection; DCIS, ductal carcinoma in situ; IRR, incidence rate ratio.

^aAdjusted for age, race/ethnicity, rural vs urban residence, median income of elderly householders in the same zip code, time since prior mammography, use of CAD on previous examination, presence of stable and unstable comorbidities, digital (vs radiographic) mammography, year of examination, and Surveillance, Epidemiology and End Results (SEER) region. Analysis also accounts for clustering of mammograms within providers and for the sampling design.

^bIn analyses of SEER Medicare-linked data, 97.3% of Medicare enrolees diagnosed as having DCIS are treated with surgery or radiotherapy.

From 2001 through 2009, Medicare total costs for CAD use totaled $278 564 950, including $163 443 470 for supplemental fees, $53 812 033 for downstream diagnostic and interventional procedures, and $61 690 112 for DCIS treatments. If CAD were used on all Medicare screening mammograms, CAD-associated Medicare costs would exceed $67 million annually (Table 2).

Table 2.

Medicare Costs Attributable to CAD Use

Costs by Service	Cost, $
	2001–2009, 47.2% Overall Prevalence	2009, 79.7% CAD Prevalence	Annual Attributable Costs Assuming 100% CAD Prevalence
Fees for CAD use during screening	163 443 470	31 575 860	39 618 400
Diagnostic testing
Diagnostic mammography	24 457 336	4 723 180	5 928 396
Breast ultrasonography	4 347 600	839 600	1 053 800
Breast biopsy	24 626 432	4 755 377	5 969 304
Treatment of DCIS	61 690 112	11 918 016	14 954 368
Total CAD-related cost	278 564 950	53 812 033	67 524 268

Abbreviations: CAD, computer-aided detection; DCIS, ductal carcinoma in situ.

Discussion

Among Medicare enrollees undergoing screening mammography in 2009, approximately 1 in 6 diagnostic mammograms,1 in 14 breast biopsies, and 1 in 9 DCIS diagnoses were attributable to CAD. From 2001 through 2009, CAD use cost Medicare more than $278 million, most of which accrued in the latter years as CAD prevalence approached 80%. The annual Medicare costs of CAD use would exceed $67 million if CAD were used on all mammograms, representing an approximately 7% increase in the approximate $1 billion annual Medicare costs for breast cancer screening.⁵

The long-term implications of increased DCIS detection in the Medicare population are uncertain. On one hand, the intent of screening is to detect breast cancers earlier when treatments can be curative and less morbid. However, many DCIS lesions may be overdiagnosed, particularly in an older population.

Our cost analysis takes the perspective of the Medicare program; fiscal effects of CAD across the entire screening population are likely to be much higher.⁶ CAD-associated incident rate ratios may reflect the clinical effect of CAD use early in its dissemination ¹; longer-term effects may differ.

Because of broad dissemination in the United States, CAD likely accounts for a substantial fraction of diagnostic breast imaging, breast biopsies, and DCIS diagnoses among the Medicare population, with high resultant costs.