Cost-Benefit Analysis of Biopsy Methods for Suspicious Mammographic Lesions FREE

IT IS ESTIMATED that there were 184 200 new cases of breast cancer diagnosed in the year 2000.¹ Mammography has significantly improved our ability to detect early breast cancer, which is an important step toward reducing breast cancer–related mortality. A 30% reduction in mortality has been found when asymptomatic women receive mammographic screening.^{2 - 3} Two important developments have facilitated the evaluation of mammographic lesions: the development of the Breast Imaging Reporting Data System (BIRADS)⁴ by the American College of Radiology and image-guided core-needle biopsy for nonpalpable breast lesions.

The BIRADS system was developed to standardize mammographic findings and ascribe a level of suspicion of malignancy to each finding (1 = normal mammogram, 2 = benign finding, 3 = probably benign finding, 4 = suspicious abnormality, 5 = highly suggestive of malignancy). The risk of malignancy for each category of mammographic abnormality has been well established such that BIRADS 1 through 3 lesions are known to have a low risk of malignancy (<2%) and do not warrant biopsy.

Parker et al^{5 - 7} were among the first to provide a standardized approach to stereotactic breast biopsies and to establish this technique as a reliable and effective alternative to surgical breast biopsy. Advocates of stereotactic core biopsy (SCB) consider it the diagnostic procedure of choice in evaluating nonpalpable breast lesions.^{5 ,8 - 10} Potential advantages of SCB over surgical biopsy include minimal morbidity (eg, pain, breast scarring), limited recovery time, less time to diagnosis, ability to distinguish benign from malignant lesions in previously radiated breast tissue, and reduced cost compared with needle-localized biopsy (NLB). It also allows a surgeon's efforts to be directed toward the treatment of malignancy.

Growing attention has been focused on the cost savings of SCB over NLB.^{11 - 14} Hospitals, patients, and third-party payers are increasingly interested in finding ways to reduce health care costs. However, SCB only proves to be a cost-effective means of evaluating nonpalpable breast lesions if a surgical biopsy is avoided. If the mammographic lesion is sufficiently suspicious to require surgical therapy to treat a breast tumor, then SCB adds unnecessarily to the total cost of diagnosis and surgical treatment of breast cancer. Cost-effective care must be determined by total surgical treatment costs, not just the cost of diagnosis. Consequently, it is less clear which mammographic abnormalities should be evaluated by SCB vs NLB to achieve the maximum total cost reduction when both diagnosis and surgical treatment of breast cancer are considered.

The time and expense of conducting prospective randomized clinical trials have led physicians to turn to statistical modeling to assist them in clinical decision making. Models are frequently used to estimate the efficacy or cost-effectiveness of interventions when data from controlled clinical trials is lacking.¹⁵ A decision tree,^{15 - 16} such as the model used in this study, is a commonly used model that presents a clinical problem as a sequence of decisions and consequences. The likelihood of a given clinical outcome is determined by the probability of a given outcome and then summing across all possible outcomes for each clinical end point. The cost of each end point can then be compared to produce a cost-effectiveness analysis. With this background in mind, we conducted a cost-effectiveness analysis of different biopsy techniques in the evaluation of suspicious mammographic abnormalities.

We sought to determine which mammographic abnormalities are most cost-effectively evaluated with SCB compared with NLB. We took advantage of the BIRADS lexicon to simulate 3 common clinical scenarios: evaluation of lesions highly suggestive of malignancy (BIRADS 5), lesions suspicious for malignancy (BIRADS 4), and lesions suspicious for ductal carcinoma in situ (DCIS). In addition, we examined the patient-related cost of these 2 biopsy techniques as determined by the mean number of procedures required to diagnose and treat, if necessary, a given mammographic lesion.

A computer-generated mathematical model was created to determine the difference in cost and mean number of procedures required to evaluate mammographic abnormalities and treat (if cancer is present) lesions that have been evaluated first by SCB vs NLB. The model assumes equal access to SCB and NLB. Figure 1 illustrates the 2 models for SCB and NLB. The model incorporates initial modality of diagnosis with potential surgical treatment if breast cancer is diagnosed.

The model was used to simulate 3 common clinical scenarios: (1) evaluation of a mammographic lesion highly suggestive of malignancy (BIRADS 5); (2) evaluation of a suspicious mammographic abnormality (BIRADS 4); and (3) evaluation of a mammographic abnormality highly suspicious for DCIS alone (eg, microcalcification without a mass). Four input variables were included in the cost analysis: (1) risk of malignancy based on the mammographic findings as determined by the BIRADS score; (2) risk of DCIS; (3) use of breast conservation therapy (BCT); and (4) rate of positive margins following the initial surgical procedure. The risks of malignancy based on the BIRADS scores were taken from Liberman et al¹⁷ and Orel et al,¹⁸ who reported that the risk of carcinoma for a BIRADS 4 lesion was 30% to 34% and the risk for a BIRADS 5 lesion was 81% to 97%. Although early reports suggested differing diagnostic accuracy between the 11-gauge and the 14-gauge core biopsy needle, a recent report by Velanovich et al¹⁹ found that the sensitivity and specificity was similar for 11- and 14-gauge SCB as well as for NLB. Based on these reports, the risk of malignancy for BIRADS 4 lesions was assumed to be 34% and the risk of malignancy for BIRADS 5 lesions was assumed to be 90%.

Ductal carcinoma in situ has reportedly been identified in 90% of mammographic abnormalities with a characteristic appearance²⁰ ; consequently, this value was used in the current computer model. When all mammographic abnormalities are considered, however, DCIS has been found in 21% to 39% of these lesions.^{18 ,21 - 24} Consequently, a median value of 31% was chosen and held constant when making the calculations for the models simulating evaluation of BIRADS 4 and BIRADS 5 lesions.

The use of BCT at our institution is 80% and this value was held constant for all calculations. Use of BCT varies markedly in the literature from a low of 9%²⁵ to a high of 69%.²⁶ This variability reflects several factors, including the years during which the study was conducted, the age of the study patients, the geographic location of the study, and the socioeconomic status of the patients.^{27 - 29} In contrast, eligibility for BCT approaches 90% in women who have mammographically detected breast cancer.²⁹

The rate of positive margins following the initial surgical procedure was estimated at 10%, reflecting our institutional experience. The frequency of margin positivity following NLB varies in the literature from 5% to 90%.^{9 ,26 ,30 - 32} The variability of these values reflects different definitions of margin positivity (ie, microscopic vs macroscopic, distance of tumor cells from specimen margin) and is significantly affected by the volume of tissue removed at the time of biopsy.^{14 ,30} However, positive surgical margins following SCB are found to occur in 0% to 29% of biopsy specimens.^{9 ,14 ,26 ,30 ,33} While some authors have reported lower rates following SCB compared with NLB,^{9 ,14 ,33} others have found that margin positivity rates are similar between the 2 diagnostic methods.³⁴ Therefore, for the purposes of our clinical decision model, we estimated that the rate of positive margins is 10% for both diagnostic modalities, although this assumption may not be applicable to all practice environments.

To estimate procedure costs regardless of payer source or hospital charges, total National Physician Fee Schedule Relative Value Units (RVUs) for the year 2000 were used for cost-based analysis.³⁵ Table 1 lists the current procedural terminology codes and associated RVUs for the procedures used in the model. Total treatment costs were determined by summation of procedural RVUs accrued during procession down each of the clinical decision arms multiplied by the statistical likelihood of a patient choosing/undergoing that treatment pathway. Anesthetic and operating room costs, etc, were excluded from the analysis in an effort to keep the model manageable as well as avoid institutional differences in secondary costs. The total number of procedures required to evaluate and treat (if cancer was present) a lesion was also determined. Only open surgical procedures were included in the total number of procedures; this included the initial NLB as well as any operations needed to treat a breast tumor.

The cost-benefit of evaluating mammographic lesions with SCB vs NLB differs according to the suspicion for malignancy as determined by the BIRADS classification. A cost savings was found when using SCB in the evaluation of BIRADS 4 lesions but not BIRADS 5 lesions or lesions suspicious for DCIS. The patient-related cost of evaluating mammographic lesions was assessed by calculating the mean number of procedures required to evaluate and potentially treat a mammographic abnormality. Using NLB as the initial biopsy method for BIRADS 4 lesions and, to a lesser extent, for BIRADS 5 lesions, requires a patient to undergo more procedures than would be required if SCB was the initial diagnostic modality. A summary of the results is presented in Table 2.

In the first clinical scenario, we addressed the financial and patient-related costs of evaluating a mammographic lesion highly suggestive of malignancy (BIRADS 5). The total RVUs required to diagnose and treat (if necessary) a BIRADS 5 lesion did not differ significantly between SCB and NLB (15.54 vs 15.47 RVUs, respectively). A modest increase in the number of procedures needed to diagnose and potentially treat a BIRADS 5 lesions was found when NLB was used instead of SCB as the initial diagnostic method (1.65 vs 0.97, respectively).

In the second clinical scenario, we used the computer model to evaluate a lesion suspicious for malignancy (BIRADS 4). More than twice as many RVUs were required to assess BIRADS 4 lesions with NLB compared with SCB (15.66 vs 7.65, respectively). The mean number of procedures required to evaluate and potentially treat a BIRADS 4 lesion was also markedly increased when NLB rather than SCB was used as the initial biopsy method (1.25 vs 0.37, respectively).

In the final scenario, we determined the total costs and mean number of procedures needed to evaluate lesions suspicious for DCIS with SCB compared with NLB. Assessment of a lesion suspicious for DCIS with SCB would require RVUs approximately equivalent to the number of procedures needed to evaluate the same lesion with NLB initially (11.49 vs 10.17, respectively). The number of procedures required to evaluate and treat lesions highly suspicious for DCIS did not differ between NLB and SCB (1.17 vs 0.97, respectively).

The introduction of SCB has significantly affected the way nonpalpable mammographic breast lesions are evaluated. The accuracy of SCB has been well documented, making it a feasible alternative to NLB, which has served as the gold standard. Advocates of this technique also point to its advantages over NLB, namely, lower cost, less recovery time, the absence of a scar, reduced number of surgical procedures required to diagnose and treat breast cancer, and less architectural distortion, which can confound interpretation of future mammograms.

In particular, much attention has focused on the cost savings associated with SCB over NLB for the evaluation of mammographic abnormalities. Lindfors and Rosenquist¹¹ were among the first to study the cost-effectiveness of SCB. They found that the marginal cost per year of lives saved by mammographic screening can be reduced by a maximum of 23% if SCB is used instead of NLB. Several other authors have reported savings of approximately $650 to almost $2000 when SCB is used in lieu of NLB.^{12 - 14 ,30} The fact that SCB costs less to perform than NLB is clear. What is less clear, however, is which mammographic lesions should be evaluated by SCB to achieve these cost-savings. Is the cost-benefit of SCB over NLB realized for all mammographic abnormalities equally?

We found that SCB provided a financial cost advantage over NLB only in the evaluation of BIRADS 4 lesions but not BIRADS 5 lesions or mammographic abnormalities suspicious for DCIS. Using a computer-generated mathematical model, we demonstrated that twice as many RVUs are needed to initially evaluate a BIRADS 4 lesion with NLB compared with SCB. Other investigators have shown a differential cost savings of SCB over NLB based on the degree of suspicion for malignancy on mammogram results. Lee et al¹² found greater cost savings when SCB was used to evaluate lesions with a lower, rather than higher, suspicion for malignancy. Additionally, they found that cost savings were least when SCB was used to assess a highly suspicious calcification—an abnormality most consistent with DCIS. Yim et al⁹ reported a cost-benefit of SCB in evaluating indeterminate lesions (BIRADS 3 and 4). They argued that the frequency of benign diagnoses in this group provides the greatest overall cost savings when evaluation of all mammographic lesions (BIRADS 2-5) is considered. Furthermore, for those BIRADS 3 and 4 lesions that proved to be malignant, the cost savings of SCB over NLB was maintained since SCB eliminated positive margins that required reexcision and allowed for a single definitive surgical procedure.

Our demonstration of a financial cost-benefit for evaluating BIRADS 4 lesions with SCB reflects the values used in the computer model, namely, the risk of malignancy for a given mammographic abnormality, the prevalence of DCIS among all mammographic lesions, the use of BCT, and the rate of margin positivity following the initial surgical procedure. While the literature is fairly consistent with regard to the accuracy of identifying mammographic lesions suspicious for malignancy and the prevalence of DCIS among all mammographic lesions, there is less consensus regarding the use of BCT and the frequency of positive margins following NLB. Therefore, the results from our computer model are only applicable to clinical situations that reflect our rate of BCT and incidence of margin positivity.

To test the sensitivity of our clinical decision model, the rates of BCT use and margin positivity were changed to determine the applicability of our results and subsequent conclusions. First, to determine the effect of a lower rate of BCT use on the RVUs needed to evaluate various mammographic abnormalities, we calculated the total RVUs accrued when the rate of BCT was 58%.³⁶ The increase in total RVUs produced by evaluating BIRADS 4 lesions with NLB vs SCB persisted despite a decrease in BCT (16.02 vs 7.82, respectively). The total RVUs required to evaluate either a BIRADS 5 lesion or a lesion highly suspicious for DCIS with SCB vs NLB did not differ significantly when BCT use was set at 58%. Therefore, there is a cost savings of SCB over NLB for BIRADS 4 lesions even when BCT is only used approximately half of the time.

Second, to determine the effect of a higher rate of margin positivity following NLB on the RVUs needed to evaluate various mammographic lesions, a margin positivity rate of 55% was input into the computer model (reflecting the median of 5 values available in the literature).^{9 ,26 ,30 - 32} No difference in the number of RVUs required to evaluate a BIRADS 5 lesion or a lesion highly suspicious for DCIS was found if these lesions were initially evaluated with SCB or NLB. However, diagnosis of a BIRADS 4 lesion with NLB initially would require nearly twice as many RVUs compared with the number required with SCB (16.37 vs 9.05, respectively). As expected, when positive surgical margins are present half of the time following NLB, the cost savings of evaluating BIRADS 4 lesions with SCB initially persists. The current model is potentially limited by the assumption that reexcision rates are similar for lesions initially evaluated with NLB vs SCB. If reexcision rates are higher following NLB compared with SCB, additional costs would be incurred, increasing the cost-benefit of SCB over NLB. Furthermore, under these conditions, SCB would likely afford a cost savings over NLB for BIRADS 5 lesions and lesions suspicious for DCIS (in addition to BIRADS 4 lesions).

Closely related to the economic cost of evaluating mammographic abnormalities are the patient-related costs. We used the number of surgical procedures required to diagnose and potentially treat a mammographic lesion as an index of patient-related cost. The patient-related cost-benefit for SCB was inversely related to the risk of malignancy; the greatest benefit was found when the chance of malignancy was lowest. The reduction in the number of surgical procedures is important to patients who desire a biopsy method that minimizes discomfort and recovery time and maximizes diagnostic accuracy and cosmesis. Proponents of SCB have aptly demonstrated that SCB successfully meets these patient-related needs, and does so in a cost-effective manner. Kaufman et al³³ studied 113 nonpalpable breast tumors, 47 diagnosed by NLB and 66 diagnosed by SCB. An average of 1.8 operations were performed in the group diagnosed by NLB vs 1.2 operations performed in the group diagnosed by SCB. Lind et al¹⁴ studied the financial effect of using SCB vs NLB in the diagnosis of mammographically detected cancer in patients who underwent BCT. A $1300 reduction in per-patient cost was achieved when the diagnosis of breast cancer was made using SCB. The authors attribute these cost savings to the greater percentage of SCB patients (89%) who had a single surgical procedure compared with NLB patients (39%).

Stereotactic core biopsy is a safe, accurate, and cost-effective alternative to NLB for the diagnosis of mammographic abnormalities. Although SCB can never completely eliminate the need for NLB, our computer-based mathematical model provides some insight into clinical scenarios in which SCB can be used as a cost-effective substitute for a NLB. The results of our cost-benefit analysis should be interpreted within the context of the assumptions included in the model. Consequently, any conclusions drawn should consider the effect that a lower rate of BCT use or a higher rate of margin positivity may have on the final cost-benefit analysis. Our analysis suggests a cost savings for evaluating intermediate risk lesions (BIRADS 4) with SCB vs NLB in settings where SCB is available and the diagnostic accuracy of the BIRADS scoring parallels that in the published literature. Conversely, our results do not support the preferential use of SCB or NLB for the evaluation of lesions highly suggestive of malignancy (BIRADS 5) or lesions suspicious for DCIS since they can be evaluated by either SCB or NLB without a significant effect on total cost.

Presented in part at the 72nd Annual Meeting of the Pacific Coast Surgical Association, Banff, Canada, February 19, 2001.

Corresponding author: Richard Bold, MD, Division of Surgical Oncology, Suite 3010, UC Davis Cancer Center, 4501 X St, Sacramento, CA 95817 (e-mail: richard.bold@ucdmc.ucdavis.edu).

John T. Vetto, MD, Portland, Ore: In this study the authors used a clinical outcome model to compare the cost-effectiveness of needle-localized biopsy with stereotactic core biopsy using an algorithm for each procedure and applying the algorithm to 3 mammographic scenarios: BIRADS 4, BIRADS 5, and highly suspicious for DCIS. Their input variables included the risk of malignancy, the risk of DCIS, the utilization of breast preservation therapy, and the rate of positive margins.

In regards to the clinical scenarios chosen, the variability and the ability of radiologists to read a lesion as purely DCIS, the variability among screening populations of pure DCIS (the authors use a number of 31%; we would put the number higher), an increasing use of the term microinvasion, coupled with the possible utility of sentinel node lymph node biopsy for DCIS with or without microinvasion, as I have already mentioned, has been promoted by Dr Cox, all lead me to my first question: Why did the authors feel it necessary to examine lesions thought to be DCIS as a separate category?

The authors are to be congratulated for their use of the measurable unit, RVUs, to define the protean concept of cost. However, their calculations did not take into account anesthesia and recovery room costs. It appears instead that they consider these costs as being equal between a needle localized biopsy and a needle-localized lumpectomy and other definitive procedures. My partner, Dr Pommier, myself, and others at OHSU (Oregon Health Sciences University) have described a method of needle localized biopsy, which has allowed us to perform 70% of these procedures under local in the clinics, increasing the cost-benefit of needle localization biopsy at our institution. Am I correct that the authors assumed all of these costs to be equal, and do they indeed perform both needle localized biopsy and needle-localized lumpectomy in the OR under general anesthesia?

The authors are also to be congratulated for reporting results that disprove their hypothesis, the commonly held belief that stereotactic core biopsy is more cost-effective than needle localization biopsy. They found this to be the case only for BIRADS 4 lesions. As Dr Fahy stressed, the benefits of a reduced need for open biopsy in the BIRADS 4 population by the use of stereotactic core biopsy include cosmetic, psychologic, and patient comfort benefits, as well as cost savings.

Because stereotactic core biopsy, with the exception of mammotome, usually does not normalize the mammogram, does the cost of following the residual lesion and a potential further procedure negate some of the cost-benefit of stereotactic core for BIRADS 4 lesions? Further, as has already pointed out by Dr Howisey, 20% to 30% of lesions diagnosed as DCIS on stereotactic core biopsy are found to contain invasive cancer on excision and indeterminate lesions such as LCIS and atypia also require needle localization biopsy for definitive diagnosis. Both of these possibilities add cost to your management but are not on your algorithms. Could the authors comment?

The authors' finding that there is no cost advantage for stereotactic core vs needle localization for BIRADS 5 lesions begs my last questions. Which do they recommend we use? What is the role, if any, of frozen section on a needle localization for a BIRADS 5 lesion? And, are the authors recommending that we follow BIRADS 5 lesions negative on stereotactic core?

At our institution we are working toward a single treatment algorithm (slide) for breast preservation therapy based on the following assumptions: First, that stereotactic core biopsy and needle localization biopsy are not mutually exclusive. Second, that a preoperative diagnosis of malignancy by stereotactic core biopsy increases the rate of negative margins on subsequent needle-localized lumpectomy. Third, that frozen section of nonpalpable lesions is often inaccurate. And, finally, that avoiding previous open excision has been shown to increase the accuracy of sentinel node biopsy. In this algorithm, patients with benign stereotactic core biopsies are spared needle localization biopsy as we heard from Dr Howisey's paper, while those with malignancy on stereotactic core biopsy will undergo image-guided sentinel node biopsy injection during wire placement. As a caveat, this algorithm should currently be reserved for patients on sentinel lymph node protocols until completion of the ACSOG and NSABP studies, but I show it because I believe this may be what we are moving toward in terms of a future standard of care.

I. Benjamin Paz, MD, Duarte, Calif: The modeling techniques to calculate costs and estimate costs are very valuable in analysis and future studies. The problem is that the assumptions sometimes are not correct, and I think assuming that 10% of positive margins for all the surgeries is not accurate. As it has been demonstrated in prior publications, the incidence of positive margins when you don't have a prior cancer diagnosis approximates about 30% when it is done by breast surgeons vs 5% to 6% when you have a positive diagnosis prior to the surgery. So this is something that will contribute to decrease the costs of core biopsy.

The second issue is the issue of intangible benefits. The intangible benefit, for example, of time to diagnosis. Core biopsies as has been demonstrated can give you a diagnosis in half of the time. The mean time to diagnosis from core is about 3 to 4 days vs surgical biopsies from 7 to 8 days as has been demonstrated by multiple authors. So I would like the presenter to comment on those 2 issues.

Quan-Yang Duh, MD, San Francisco, Calif: I am not a breast surgeon, but I do have a question in terms of decision analysis. One of the benefits of using decision analysis is to clarify or resolve controversies by sensitivity analysis. For example, the authors can repeat the analysis by the positive margin rate, the rate of breast conservation, or the cost of the different procedures. Did you perform any sensitivity analysis to see whether the outcome is sensitive to minor changes in any of these variables?

Dr Goodnight: Dr Vetto asked about DCIS. I want to emphasize that we pulled out that category of mammograms that were thought highly suspicious for DCIS, and actually in the literature and in our experience, 90% of those patients do actually have DCIS. So that was the reason for culling out that group. It would have an impact similar to a BIRADS 5. Dr Vetto asked if we considered anesthesia and recovery costs. Actually we did not, and when we do a needle-localized biopsy, we do have those costs of an anesthesiologist or a nurse anesthetist standing by and the recovery costs. We did look at this informally and we do not think it would alter the conclusions of the paper.

Dr Vetto also asked about follow-up of patients after SCB and the necessity to follow those patients and would that impact the cost-effectiveness and therefore the algorithm and therefore the conclusion. As Dr Howisey pointed out, the SCB becomes more and more accurate and therefore the cost of follow-up of these patients is decreasing and is therefore negligible. Dr Vetto gave air to this. We did disprove our hypothesis and actually we did start out quite frankly thinking that SCB is cheaper, and therefore one would demonstrate with the model the cost-effectiveness of SCB. Having done that (disproved the hypothesis), I would still have to say for reasons that Dr Paz and others brought up, and for many other reasons, I prefer the SCB. It keeps my OR schedule free for other things. I think the radiologists have certainly said to me, you have to be careful what you wish for. They are now overwhelmed with work which they have difficulty dealing with, but nevertheless, the issue of the paper was cost-effectiveness and the conclusions were drawn thereby. But as we partially pointed out, and Dr Paz and others stated, there are advantages to SCB, which are efficiency and patient-related factors.

All of us are working toward a single diagnostic algorithm. I have no disagreement with the algorithm Dr Vetto posed. I think he has one approach. We perhaps have a slightly different approach. But smoothness and efficiency of the diagnostic algorithm is critical.

This is the question of what happens if you alter the rate of positive margins. Would that affect the algorithm and the conclusions? If you look at a composite of the literature, positive margin rate for a lumpectomy is about 55%, which we have inserted here in the algorithm. If you recall Dr Fahy's slides, once again for a BIRADS 5 lesion, there is no difference in the overall RVUs for BIRADS 5, even if you alter the rate of margin positivity. Similarly for BIRADS 4, there is the improved cost-effectiveness for the SCB similar to the data that we posed in the main presentation. Once again, for DCIS there is no difference in cost-effectiveness as measured by RVUs. So we considered this—sensitivity analysis, as described by Dr Duh—and it did not alter the conclusion.