- Email: [email protected]
Factors Predicting Successful Needle-Localized Breast Biopsy
Original Investigations
Factors Predicting Successful Needle-localized Breast Biopsy1 Page E. Abrahamson, MSPH, Larry A. Dunlap, MHS, PA-C, M. Ahinee Amamoo, MS, Michael J. Schell, PhD M. Patricia Braeuning, MD, Etta D. Pisano, MD
Rationale and Objectives. The purpose of this study was to identify factors that predict successful removal of nonpalpable breast lesions with mammography-guided needle-localized breast biopsy. Materials and Methods. Of the 455 consecutive patients referred for needle-localized breast biopsy of one or more nonpalpable breast lesions between January 1990 and December 1994, 272 (59.8%) had sufficiently complete data to be included in this study. Medical charts, pathology laboratory reports, wire-placement mammograms, and radiographs of specimens from each patient were retrospectively reviewed to evaluate the effect of the following factors on the success of the procedure: distance from the lesion to the localizing wire, breast density, breast size, specimen volume, and lesion volume. All radiographs were independently evaluated by two radiologists who are experts in breast imaging. Results. Needle-localized breast biopsy was successful in 254 (93.3%) of 272 lesions. Placement of the localization wire within 5 mm of the breast lesion was a significant predictor of successful lesion removal (P ⫽ .007). Results from logistic regression analysis showed that needle-localized breast biopsy failure was associated with increased wire distance (P ⫽ .0006), decreased breast size (P ⫽ .02), and decreased specimen volume (P ⫽ .03). Conclusion. Needle localization wires should be placed within 5 mm of mammographically visible lesions to increase the probability of successful lesion excision. Key Word. Breast, biopsy. ©
AUR, 2003
Needle-localized breast biopsy (NLBB) is a widely used method for removal and diagnosis of nonpalpable, mammographically detected breast lesions. The failure rate for NLBB is reportedly low, ranging from 0% (1,2) to 18% (3), and the failure rate for NLBB with nonstereotactic needle insertion is even lower, ranging from 0% (4) to 6% (5). Low failure rates make it difficult to determine the causes of unsuccessful removal of breast lesions with NLBB. Acad Radiol 2003; 10:601– 606 1
From the Departments of Epidemiology (P.E.A.) and Biostatistics (M.J.S.), Lineberger Comprehensive Cancer Center (M.A.A., M.J.S.), and the Department of Radiology, School of Medicine, (E.D.P.), University of North Carolina, Chapel Hill, NC 27599; the Department of Surgery, Duke University, Durham, NC (L.A.D.); and Christ Hospital, Cincinnati, Ohio (M.P.B.). Received January 6, 2003; revision requested February 6; revision received and accepted February 13. Address correspondence to P.E.A.
©
AUR, 2003
Only one previous study (6), to our knowledge, has analyzed potential causes of NLBB failure and identified a number of factors that predict success, including accuracy of needle placement. Investigators in that study reported that the needle localization wire had to pierce the lesion or be placed less than 1 mm from the lesion for positioning to be a statistically significant predictor of success. However, standard guidelines indicate that the wire must be placed within 5 mm (7). The purpose of the current study was to identify factors related to the lesion and the procedure that predict successful removal of nonpalpable breast lesions with NLBB. MATERIALS AND METHODS Between January 1990 and December 1994, 455 consecutive NLBB procedures were performed on nonpal-
601
ABRAHAMSON ET AL
pable breast lesions in 407 patients at University of North Carolina Hospitals, Chapel Hill. Patients were included in this study if their records contained the following materials: wire-placement mammograms with two views, radiograph of the specimen, and medical records corresponding to the NLBB procedure. In accordance with these criteria, 272 lesions (59.8%) from 239 patients were included in this study. In all patients, preoperative localization involved the mammographically guided insertion of a nonstereotactic needle parallel to the chest wall. The method of wire placement is described in detail elsewhere (8). Fenestrated compression paddles were used in 180 (66%) of the 272 localizations, and fenestrated paddles with holes were used in the remaining 92 (34%). All localizations were performed with Kopans wires (Cook, Bloomington, Ind). Postlocalization wire-placement mammograms were obtained, providing true lateral and craniocaudal views. In most cases, biopsies to remove the lesion and localization wire were performed with local anesthesia in an outpatient setting by one of 15 surgeons at University of North Carolina Hospitals. Radiographs of all specimens were obtained and reviewed by one of six radiologists as part of routine patient care. If the lesion was not present in the specimen, this information was communicated immediately to the operating surgeon, who removed more tissue if necessary. Both wire-placement mammograms were reviewed by two radiologists (E.D.P., M.P.B.) to document the proximity of the wire to the lesion, the breast density, the lesion volume and type, and the position of the lesion within the breast. A four-point system was used for grading breast density, based on an abbreviation of terminology from the Breast Imaging Reporting and Data System, or BI-RADS. This system, which has been in use at our institution since 1989 and which resembles the Wolfe classification system, required both participating radiologists to classify breast tissue as dense, mostly dense, mostly fatty, or fatty. Each lesion was categorized as a noncalcified mass, calcified mass, calcification, area of asymmetric density, or architectural distortion. Its position in the breast was noted as anterior, posterior, axillary, or central and was further specified by an imaginary line drawn from the lesion to the nipple and “read” as the hour hand on a clock (1–12 o’clock). A research assistant measured the breast from the craniocaudal view only. On the basis of these measurements, breast size then was estimated by using a general formula for calculating the
602
Academic Radiology, Vol 10, No 6, June 2003
area inside an ellipse. The volume of the specimen was obtained from the pathology report. To determine the distance between the wire and the lesion, the two radiologists’ readings were averaged for each of the two mammographic views; the larger of the two averages was then used as the final value. Similarly, each radiologist measured the dimensions of the lesion, and the averages were used to calculate lesion volume. The two radiologists’ readings also were averaged for breast density. Medical charts were reviewed for age, race, and occurrence of procedural complications. Radiographs of the specimens were reviewed by the same two radiologists to verify partial or full removal of the lesion and to determine whether localization had been successful. If the radiograph indicated that no part of the lesion had been excised, the localization was deemed a failure. When the two radiologists’ readings did not agree, the radiographs were reassessed to reach a consensus. The factors chosen for consideration in the analysis included distance from the lesion to the localizing wire, breast density, breast size, specimen volume, and lesion volume, because these are the clinical factors thought to be associated with the successful removal of nonpalpable breast lesions by means of NLBB. Age, lesion type, and lesion position in the breast were not considered in the analysis because they are not thought to be independent predictors of successful NLBB. The continuous variables in the study were categorized as had been done in previous studies, when possible (6). However, when variables are categorized, there tends to be an underfitting of the data. Therefore, the Wilcoxon rank sum test, which takes advantage of the continuous nature of the data, was used to determine the significance of the variables. Transformations of independent variables were performed to approximate Gaussian distribution of the data. Specifically, a log transformation of the lesion volume and square root transformations of the wire distance and the specimen volume were performed. A logistic regression model with a backward elimination procedure and a significance-level-to-stay criterion of P ⬍ .05 was used to determine which factors predict successful removal of nonpalpable breast lesions with NLBB. A separate analysis based on a scoring system was performed to classify the risk profile related to NLBB failure. First, an isotonic regression analysis (9) was used to determine whether categorical splitting of the variables found to be significant in the logistic regression analysis could lead to a better model. Backward elimination was then performed by using an approach described by Schell
Academic Radiology, Vol 10, No 6, June 2003
and Singh (10), with the isotonic cut-points being used to classify variables into three groups according to the associated risk of NLBB failure (low, moderate, or high). Next, a point system was developed in which each risk level was assigned a point value: two points for high risk, one for moderate risk, and zero for low risk. The total risk score for each patient was computed by summing the risk points associated with each variable. One risk point was assigned when the variable value was missing.
RESULTS A total of 272 NLBBs were performed in 239 patients. Multiple NLBBs performed in the same patient during a single visit included two concurrent ipsilateral NLBBs in 10 patients, two concurrent bilateral NLBB in seven patients, and four concurrent NLBBs in one patient (three in one breast and one in the opposite breast). Thirteen patients had two or more NLBBs performed at separate visits within the study period. Of the 18 patients in whom more than one NLBB was performed at the same visit, one patient who underwent two concurrent bilateral NLBBs experienced a failure. Tables 1 and 2 show the number of NLBB successes and failures in relation to lesion and patient characteristics. NLBB failed in 18 of the 272 lesions (6.6%); this failure rate is consistent with those found in previous studies. Patient age ranged from 30 to 86 years, with a mean of 57 years. Most lesions were characterized as calcification only (49%), mass only (42%), or mass with calcification (8%). Less common findings included architectural distortion and asymmetric breast tissue. Lesions were most commonly positioned in the central (49%) or posterior (38%) area of the breast and less frequently in the anterior (13%) or axillary (⬍1%) part. Nearly half (49%) of the lesions were in breasts characterized as mostly fatty. In the univariate analysis, the distance of the wire from the lesion was the strongest predictor of successful NLBB. When we evaluated the proximity of the wire placement to the breast lesion, 5 mm was the maximum distance required for a statistically significant prediction of NLBB success. As demonstrated by the results shown in Table 3, NLBB failed about 3% of the time when the localization wire was placed less than 1 mm from the lesion and about 5% of the time when it was placed within 1–3 mm. The failure rate, however, increased dramatically with greater distance. When the distance be-
SUCCESSFUL NEEDLE-LOCALIZED BREAST BIOPSY
tween the wire and the lesion was 6 –10 mm, the failure rate increased to 25%. In addition, the results of 2 tests for general association revealed that NLBB failure rates did not differ significantly by supervising surgeon or radiologist, type of localization paddle, patient race, number of localizations per visit, or presence of procedural complications such as infection, hematoma, eccymosis, and edema (data not shown). For each of the categorical variables recorded by both radiologists (lesion type, lesion location, and breast density), there were no significant differences in relation to NLBB success. Specimen volume measurements were not indicated in the pathology reports for 48 lesions; NLBB failed in only one of these lesions. No failures occurred in the 10 lesions for which breast size was not indicated in the record. Results from the backward-elimination logistic regression analysis are reported in Table 4. NLBB failure was associated with increased wire distance (P ⫽ .0006), decreased specimen volume (P ⫽ .03), and decreased breast size (P ⫽ .02). The 2 statistic for the final model was 18.2. No relationship was observed between success and either lesion volume or breast density. The isotonic regression exploratory analysis indicated cut-points for each of the three variables found to be statistically significant in the logistic regression analysis (wire distance, breast size, and specimen volume). Risk scores ranged from 0 to 6, with a median score of 2. Table 5 displays the cut-points found with isotonic regression and the number of risk points assigned to each level. For example, if a patient has a breast size of 150.3 cm2, a specimen volume of 23 cm3, and a wire distance of 17 mm, then she would have a risk score of 3, obtained by summing one, zero, and two risk points, respectively, for each variable. Table 6 displays the risk scores, which were combined into three groups (0 –2, 3– 4, 5– 6) that were assessed for model fit with logistic regression. The 2 statistic for the logistic model with the grouped risk scores was 36.5. This twofold increase in the 2 value over that obtained with the original logistic regression model indicates that the relationship between the significant variables and NLBB failure is probably not linear in the logit. The probability of NLBB failure was significantly greater for patients having more than one risk factor in the moderateor high-risk category. NLBB failed in only 1% of patients with risk scores of 0 –2 and 12% of those with risk scores of 3 or 4. For patients with risk scores of 5 or 6, the failure rate increased to 83%.
603
ABRAHAMSON ET AL
Academic Radiology, Vol 10, No 6, June 2003
Table 1 NLBB Successes or Failures in Relation to Lesion and Patient Characteristics: Categorical Data NLBB Outcome Factor Wire distance from lesion (mm) ⱕ5 ⬎5 Lesion volume (cm3) ⬍0.5 0.5–2.0 ⬎2.0 Breast size (cm2) ⱕ100 101–150 ⬎150 No data Specimen volume (cm3) ⱕ10 ⬎10 No data Breast density Dense Mostly dense Mostly fatty Fatty Age at NLBB (y) ⬍50 50–65 ⬎65 Lesion type Mass Mass with calcification Calcification Other§ Lesion position in breast Anterior Posterior Axillary Central Lesion position relative to clock 1–3 4–6 7–9 10–12
Success (n ⫽ 254)
Failure (n ⫽ 18)
Total
238 (95) 16 (76)
13 (5) 5 (24)
251 21
88 (91) 81 (95) 85 (94)
9 (9) 4 (5) 5 (6)
97 85 90
72 (90) 79 (93) 93 (96) 10
8 (10) 6 (7) 4 (4) 0
80 85 97 10
P Value* .007
.41
.31
.88 27 (93) 180 (92) 47
2 (7) 15 (8) 1
29 195 48
40 (91) 53 (95) 123 (92) 38 (97)
4 (9) 3 (5) 10 (8) 1 (3)
44 56 133 39
.62
.21 78 (90) 101 (94) 75 (96)
9 (10) 6 (6) 3 (4)
87 107 78
110 (95) 21 (100) 121 (92) 2 (67)
6 (5) 0 (0) 11 (8) 1 (33)
116 21 132 3
.33†‡
.23†㛳 36 (100) 94 (92) 2 (100) 122 (92)
0 (0) 8 (8) 0 (0) 10 (8)
36 102 2 132 .33†
67 (93) 34 (100) 46 (90) 107 (93)
5 (7) 0 (0) 5 (10) 8 (7)
72 34 51 115
Note.—Percentages are in parentheses. *Except where otherwise indicated, P values were determined with the 2 test. †These P values were determined by means of the Fisher exact test. ‡Data in “Other” were excluded from analysis. §This category included asymmetric breast tissue and architectural distortion. 㛳Data in “Axillary” were excluded from analysis.
604
SUCCESSFUL NEEDLE-LOCALIZED BREAST BIOPSY
Academic Radiology, Vol 10, No 6, June 2003
Table 2 NLBB Successes or Failures in Relation to Lesion and Patient Characteristics: Continuous Data Factor
Mean
Median
P Value*
Wire distance from lesion (mm) Lesion volume (cm3) Breast size (cm2) Specimen volume (cm3) Breast density† Age at NLBB (y)
2.0 0.55 136.8 55.2 2.6 56.9
0.5 0.90 126.3 30.0 3.0 57.0
.0003 .10 .23 .24 .73 .23
*P values were determined by means of Wilcoxon rank sum test with normal scores. †Breast density was scored as follows: dense ⫽ 1, mostly dense ⫽ 2, mostly fatty ⫽ 3, and fatty ⫽ 4.
Table 3 NLBB Failure Rates by Distance between Localization Wire and Breast Lesion Distance (mm)
No. of Lesions*
No. of Failures†
⬍1 1–3 4–5 6–10 ⬎10
142 82 27 12 9
4 (3) 4 (5) 5 (18) 3 (25) 2 (22)
*n ⫽ 272. †Percentages are in parentheses.
Table 4 Results of Logistic Regression Analysis of NLBB Success in Relation to Lesion and Patient Characteristics Factor
P Value
Wire distance Lesion volume Specimen volume Breast size Breast density
.0006 .17 .03 .02 .30
DISCUSSION Using a standard analytic approach based on logistic regression, we identified three key factors that were significantly related to NLBB failure: increased distance between the localization wire and lesion, decreased specimen volume, and decreased breast size. The risk point system is a practical method for identifying patients with lesions at high risk of localization failure. While the probability of NLBB failure was relatively low for patients
Table 5 Variable Cut Points Found by Isotonic Regression and the Number of Risk Points Assigned to Each Level Cut Points from Isotonic Regression for Each Factor
Risk Points
Specimen volume (cm3) ⬍24 24–57.9 ⱖ58 Wire distance (mm) 0–2 3–9 10–25 Breast size (cm2) ⬍100.2 100.5–204.9 ⱖ205
0 1 2 0 1 2 2 1 0
Table 6 Distribution of NLBB Failures by Risk Score Risk Score
No. of NLBB Failures
0 1 2 3 4 5 6
0/12 (0) 0/62 (0) 2/97 (2) 7/75 (9) 4/20 (20) 4/5 (80) 1/1 (100)
Note.—Percentages are in parentheses. The numbers of NLBB failures for each risk group are as follows: 0 –2 risk points, two of 171 (1%); 3– 4 risk points, 11 of 95 (12%); and 5– 6 risk points, five of six (83%).
with just one moderate- or high-risk factor, the probability increased substantially for those with two or three. Precise placement of the localization wire appears to be a significant predictor of NLBB success in nonpalpable breast lesions. To our knowledge, this is the first study to report data supporting the current teaching that the localization wire should be placed within 5 mm of the lesion to enhance the probability of successful lesion removal. While Jackman and Marzoni (6) hypothesized that a distance of 5 mm might be adequate, their study, in which they used the same method that we used in our study to determine wire distance, revealed that only penetration of the lesion by the needle (or needle placement ⬍1 mm from the lesion) was a significant predictor of NLBB success. Perfection is not always possible, and our results support the teaching that placement within 5 mm of the
605
ABRAHAMSON ET AL
lesion is sufficient for a high probability of successful lesion removal. It should be noted, however, that the proportion of failures in this study was lower when the wire was placed closer than 5 mm (Table 3). Like Jackman and Marzoni (6), we observed an association between NLBB failure and decreased specimen volume despite the fact that data on specimen volume were missing for 48 patients (including one in whom NLBB failed). While greater specimen volume would seem to be a more logical predictor of NLBB success, studies reporting failure rates by specimen volume have not consistently shown this association (3,11,12). The significant association between small breast size and NLBB failure was unexpected. We hypothesized initially that this association may be due to higher density in smaller breasts, but there was no association between density and NLBB failure, and smaller breasts were not more likely than larger breasts to be dense rather than fatty. The high proportion of fatty breasts seen in this study may be due to a disproportionate percentage of obese women within our hospital’s patient population. Another explanation may be that surgeons remove smaller specimens in patients with small breasts to preserve breast tissue, which may make failure to remove the entire lesion more likely. The only factor that significantly affected risk and that is under the control of the radiologist is the location of the wire relative to the lesion. The results of this study suggest that the radiologist performing NLBB must place the wire within 5 mm of the lesion to increase the probability of success, and therefore, that the radiologist is justified in repeating the wire placement to ensure positioning within 5 mm of the lesion. This is especially important in women who have small breasts or in women who have requested a minimal amount of tissue removal for cosmetic reasons. Unlike Jackman and Marzoni (6), we did not find NLBB failure to be associated with lesion type, lesion volume, or the number of lesions per breast. We measured lesion volume in three dimensions, however, rather than using diameter alone, and the present study included fewer patients who underwent NLBB of two lesions in one breast during a single visit. There is no obvious explanation for the inconsistent results for lesion type, since these variables were presumably evaluated similarly in the two studies, which included nearly equal numbers of patients within a comparable age range. Although the failure rate of 6.6% is within the range reported in the literature, it is higher than the 2.5% failure rate reported by Jackman and Marzoni (6). One explanation for our higher failure rate is that our hospital is an
606
Academic Radiology, Vol 10, No 6, June 2003
academic center where trainees work, whereas Jackman and Marzoni work in private practice. As a referral center, our hospital may also have more difficult cases or more patients with concurrent illness. Our study was somewhat limited by the number of missing records, but we do not believe that this limitation biased our results, since there is no reason to suspect that the factors related to missing records vary between successful NLBB and unsuccessful NLBB. Another limitation is that lesion outcome was not recorded. Thus, it is not possible to study NLBB failure in relation to whether the lesion was malignant or benign. Because of the growing use of mammographic screening, increasing numbers of women with nonpalpable breast lesions will be referred for NLBB, although percutaneous needle biopsy may obviate NLBB in most women with benign lesions (13). A better understanding of the causes of NLBB failure may help improve the accuracy of the procedure, prevent malignant lesions from being missed, and reduce the need for repeated excisions. Further studies of NLBB failure could elucidate additional characteristics of lesions at risk for unsuccessful removal and offer insight into the inconsistencies between our results and those previously published (6). REFERENCES 1. Allen MJ, Thompson WD, Stuart RC, et al. Management of non-palpable breast lesions detected mammographically. Br J Surg 1994; 81:543–545. 2. Proudfoot RW, Mattingly SS, Stelling CB, Fine JG. Nonpalpable breast lesions: wire localization and excisional biopsy. Am Surg 1986; 52:117– 122. 3. Norton LW, Zeligman BE, Pearlman NW. Accuracy and cost of needle localization and excisional biopsy. Arch Surg 1988; 123:947–950. 4. Roses DF, Mitnick J, Harris MN, et al. The risk of carcinoma in wire localization biopsies for mammographically detected clustered micro calcifications. Surgery 1991; 110:877– 886. 5. Rissanen TJ, Makarainen HP, Mattila SI, et al. Wire localized biopsy of breast lesions: a review of 425 cases found in screening or clinical mammography. Clin Radiol 1993; 47:14 –22. 6. Jackman RJ, Marzoni FA Jr. Needle-localized breast biopsy: why do we fail? Radiology 1997; 204:677– 684. 7. Kopans D. Breast imaging. Philadelphia, Pa: Lippincott-Raven, 1998. 8. Kopans DB, Swann CA. Preoperative imaging-guided needle placement and localization of clinically occult breast lesions. AJR Am J Roentgenol 1989; 152:1–9. 9. Robertson T, Wright FT, Dykstra RL. Order-restricted statistical inference. New York, NY: Wiley, 1988. 10. Schell MJ, Singh B. The reduced monotonic regression method. J Am Stat Assoc 1997; 92:128 –135. 11. Gallagher WJ, Cardenosa G, Rubens JR, McCarthy KA, Kopans DB. Minimal-volume excision of nonpalpable breast lesions. AJR Am J Roentgenol 1989; 153:957–961. 12. Tinnemans JG, Wobbes T, Hendriks JH, van der Sluis RF, Lubbers EJ, de Boer HH. Localization and excision of nonpalpable breast lesions. A surgical evaluation of three methods. Arch Surg 1987; 122:802– 806. 13. Parker SH, Burbank F. A practical approach to minimally invasive breast biopsy. Radiology 1996; 200:11–20.
Factors Predicting Successful Needle-localized Breast Biopsy1 Page E. Abrahamson, MSPH, Larry A. Dunlap, MHS, PA-C, M. Ahinee Amamoo, MS, Michael J. Schell, PhD M. Patricia Braeuning, MD, Etta D. Pisano, MD
Rationale and Objectives. The purpose of this study was to identify factors that predict successful removal of nonpalpable breast lesions with mammography-guided needle-localized breast biopsy. Materials and Methods. Of the 455 consecutive patients referred for needle-localized breast biopsy of one or more nonpalpable breast lesions between January 1990 and December 1994, 272 (59.8%) had sufficiently complete data to be included in this study. Medical charts, pathology laboratory reports, wire-placement mammograms, and radiographs of specimens from each patient were retrospectively reviewed to evaluate the effect of the following factors on the success of the procedure: distance from the lesion to the localizing wire, breast density, breast size, specimen volume, and lesion volume. All radiographs were independently evaluated by two radiologists who are experts in breast imaging. Results. Needle-localized breast biopsy was successful in 254 (93.3%) of 272 lesions. Placement of the localization wire within 5 mm of the breast lesion was a significant predictor of successful lesion removal (P ⫽ .007). Results from logistic regression analysis showed that needle-localized breast biopsy failure was associated with increased wire distance (P ⫽ .0006), decreased breast size (P ⫽ .02), and decreased specimen volume (P ⫽ .03). Conclusion. Needle localization wires should be placed within 5 mm of mammographically visible lesions to increase the probability of successful lesion excision. Key Word. Breast, biopsy. ©
AUR, 2003
Needle-localized breast biopsy (NLBB) is a widely used method for removal and diagnosis of nonpalpable, mammographically detected breast lesions. The failure rate for NLBB is reportedly low, ranging from 0% (1,2) to 18% (3), and the failure rate for NLBB with nonstereotactic needle insertion is even lower, ranging from 0% (4) to 6% (5). Low failure rates make it difficult to determine the causes of unsuccessful removal of breast lesions with NLBB. Acad Radiol 2003; 10:601– 606 1
From the Departments of Epidemiology (P.E.A.) and Biostatistics (M.J.S.), Lineberger Comprehensive Cancer Center (M.A.A., M.J.S.), and the Department of Radiology, School of Medicine, (E.D.P.), University of North Carolina, Chapel Hill, NC 27599; the Department of Surgery, Duke University, Durham, NC (L.A.D.); and Christ Hospital, Cincinnati, Ohio (M.P.B.). Received January 6, 2003; revision requested February 6; revision received and accepted February 13. Address correspondence to P.E.A.
©
AUR, 2003
Only one previous study (6), to our knowledge, has analyzed potential causes of NLBB failure and identified a number of factors that predict success, including accuracy of needle placement. Investigators in that study reported that the needle localization wire had to pierce the lesion or be placed less than 1 mm from the lesion for positioning to be a statistically significant predictor of success. However, standard guidelines indicate that the wire must be placed within 5 mm (7). The purpose of the current study was to identify factors related to the lesion and the procedure that predict successful removal of nonpalpable breast lesions with NLBB. MATERIALS AND METHODS Between January 1990 and December 1994, 455 consecutive NLBB procedures were performed on nonpal-
601
ABRAHAMSON ET AL
pable breast lesions in 407 patients at University of North Carolina Hospitals, Chapel Hill. Patients were included in this study if their records contained the following materials: wire-placement mammograms with two views, radiograph of the specimen, and medical records corresponding to the NLBB procedure. In accordance with these criteria, 272 lesions (59.8%) from 239 patients were included in this study. In all patients, preoperative localization involved the mammographically guided insertion of a nonstereotactic needle parallel to the chest wall. The method of wire placement is described in detail elsewhere (8). Fenestrated compression paddles were used in 180 (66%) of the 272 localizations, and fenestrated paddles with holes were used in the remaining 92 (34%). All localizations were performed with Kopans wires (Cook, Bloomington, Ind). Postlocalization wire-placement mammograms were obtained, providing true lateral and craniocaudal views. In most cases, biopsies to remove the lesion and localization wire were performed with local anesthesia in an outpatient setting by one of 15 surgeons at University of North Carolina Hospitals. Radiographs of all specimens were obtained and reviewed by one of six radiologists as part of routine patient care. If the lesion was not present in the specimen, this information was communicated immediately to the operating surgeon, who removed more tissue if necessary. Both wire-placement mammograms were reviewed by two radiologists (E.D.P., M.P.B.) to document the proximity of the wire to the lesion, the breast density, the lesion volume and type, and the position of the lesion within the breast. A four-point system was used for grading breast density, based on an abbreviation of terminology from the Breast Imaging Reporting and Data System, or BI-RADS. This system, which has been in use at our institution since 1989 and which resembles the Wolfe classification system, required both participating radiologists to classify breast tissue as dense, mostly dense, mostly fatty, or fatty. Each lesion was categorized as a noncalcified mass, calcified mass, calcification, area of asymmetric density, or architectural distortion. Its position in the breast was noted as anterior, posterior, axillary, or central and was further specified by an imaginary line drawn from the lesion to the nipple and “read” as the hour hand on a clock (1–12 o’clock). A research assistant measured the breast from the craniocaudal view only. On the basis of these measurements, breast size then was estimated by using a general formula for calculating the
602
Academic Radiology, Vol 10, No 6, June 2003
area inside an ellipse. The volume of the specimen was obtained from the pathology report. To determine the distance between the wire and the lesion, the two radiologists’ readings were averaged for each of the two mammographic views; the larger of the two averages was then used as the final value. Similarly, each radiologist measured the dimensions of the lesion, and the averages were used to calculate lesion volume. The two radiologists’ readings also were averaged for breast density. Medical charts were reviewed for age, race, and occurrence of procedural complications. Radiographs of the specimens were reviewed by the same two radiologists to verify partial or full removal of the lesion and to determine whether localization had been successful. If the radiograph indicated that no part of the lesion had been excised, the localization was deemed a failure. When the two radiologists’ readings did not agree, the radiographs were reassessed to reach a consensus. The factors chosen for consideration in the analysis included distance from the lesion to the localizing wire, breast density, breast size, specimen volume, and lesion volume, because these are the clinical factors thought to be associated with the successful removal of nonpalpable breast lesions by means of NLBB. Age, lesion type, and lesion position in the breast were not considered in the analysis because they are not thought to be independent predictors of successful NLBB. The continuous variables in the study were categorized as had been done in previous studies, when possible (6). However, when variables are categorized, there tends to be an underfitting of the data. Therefore, the Wilcoxon rank sum test, which takes advantage of the continuous nature of the data, was used to determine the significance of the variables. Transformations of independent variables were performed to approximate Gaussian distribution of the data. Specifically, a log transformation of the lesion volume and square root transformations of the wire distance and the specimen volume were performed. A logistic regression model with a backward elimination procedure and a significance-level-to-stay criterion of P ⬍ .05 was used to determine which factors predict successful removal of nonpalpable breast lesions with NLBB. A separate analysis based on a scoring system was performed to classify the risk profile related to NLBB failure. First, an isotonic regression analysis (9) was used to determine whether categorical splitting of the variables found to be significant in the logistic regression analysis could lead to a better model. Backward elimination was then performed by using an approach described by Schell
Academic Radiology, Vol 10, No 6, June 2003
and Singh (10), with the isotonic cut-points being used to classify variables into three groups according to the associated risk of NLBB failure (low, moderate, or high). Next, a point system was developed in which each risk level was assigned a point value: two points for high risk, one for moderate risk, and zero for low risk. The total risk score for each patient was computed by summing the risk points associated with each variable. One risk point was assigned when the variable value was missing.
RESULTS A total of 272 NLBBs were performed in 239 patients. Multiple NLBBs performed in the same patient during a single visit included two concurrent ipsilateral NLBBs in 10 patients, two concurrent bilateral NLBB in seven patients, and four concurrent NLBBs in one patient (three in one breast and one in the opposite breast). Thirteen patients had two or more NLBBs performed at separate visits within the study period. Of the 18 patients in whom more than one NLBB was performed at the same visit, one patient who underwent two concurrent bilateral NLBBs experienced a failure. Tables 1 and 2 show the number of NLBB successes and failures in relation to lesion and patient characteristics. NLBB failed in 18 of the 272 lesions (6.6%); this failure rate is consistent with those found in previous studies. Patient age ranged from 30 to 86 years, with a mean of 57 years. Most lesions were characterized as calcification only (49%), mass only (42%), or mass with calcification (8%). Less common findings included architectural distortion and asymmetric breast tissue. Lesions were most commonly positioned in the central (49%) or posterior (38%) area of the breast and less frequently in the anterior (13%) or axillary (⬍1%) part. Nearly half (49%) of the lesions were in breasts characterized as mostly fatty. In the univariate analysis, the distance of the wire from the lesion was the strongest predictor of successful NLBB. When we evaluated the proximity of the wire placement to the breast lesion, 5 mm was the maximum distance required for a statistically significant prediction of NLBB success. As demonstrated by the results shown in Table 3, NLBB failed about 3% of the time when the localization wire was placed less than 1 mm from the lesion and about 5% of the time when it was placed within 1–3 mm. The failure rate, however, increased dramatically with greater distance. When the distance be-
SUCCESSFUL NEEDLE-LOCALIZED BREAST BIOPSY
tween the wire and the lesion was 6 –10 mm, the failure rate increased to 25%. In addition, the results of 2 tests for general association revealed that NLBB failure rates did not differ significantly by supervising surgeon or radiologist, type of localization paddle, patient race, number of localizations per visit, or presence of procedural complications such as infection, hematoma, eccymosis, and edema (data not shown). For each of the categorical variables recorded by both radiologists (lesion type, lesion location, and breast density), there were no significant differences in relation to NLBB success. Specimen volume measurements were not indicated in the pathology reports for 48 lesions; NLBB failed in only one of these lesions. No failures occurred in the 10 lesions for which breast size was not indicated in the record. Results from the backward-elimination logistic regression analysis are reported in Table 4. NLBB failure was associated with increased wire distance (P ⫽ .0006), decreased specimen volume (P ⫽ .03), and decreased breast size (P ⫽ .02). The 2 statistic for the final model was 18.2. No relationship was observed between success and either lesion volume or breast density. The isotonic regression exploratory analysis indicated cut-points for each of the three variables found to be statistically significant in the logistic regression analysis (wire distance, breast size, and specimen volume). Risk scores ranged from 0 to 6, with a median score of 2. Table 5 displays the cut-points found with isotonic regression and the number of risk points assigned to each level. For example, if a patient has a breast size of 150.3 cm2, a specimen volume of 23 cm3, and a wire distance of 17 mm, then she would have a risk score of 3, obtained by summing one, zero, and two risk points, respectively, for each variable. Table 6 displays the risk scores, which were combined into three groups (0 –2, 3– 4, 5– 6) that were assessed for model fit with logistic regression. The 2 statistic for the logistic model with the grouped risk scores was 36.5. This twofold increase in the 2 value over that obtained with the original logistic regression model indicates that the relationship between the significant variables and NLBB failure is probably not linear in the logit. The probability of NLBB failure was significantly greater for patients having more than one risk factor in the moderateor high-risk category. NLBB failed in only 1% of patients with risk scores of 0 –2 and 12% of those with risk scores of 3 or 4. For patients with risk scores of 5 or 6, the failure rate increased to 83%.
603
ABRAHAMSON ET AL
Academic Radiology, Vol 10, No 6, June 2003
Table 1 NLBB Successes or Failures in Relation to Lesion and Patient Characteristics: Categorical Data NLBB Outcome Factor Wire distance from lesion (mm) ⱕ5 ⬎5 Lesion volume (cm3) ⬍0.5 0.5–2.0 ⬎2.0 Breast size (cm2) ⱕ100 101–150 ⬎150 No data Specimen volume (cm3) ⱕ10 ⬎10 No data Breast density Dense Mostly dense Mostly fatty Fatty Age at NLBB (y) ⬍50 50–65 ⬎65 Lesion type Mass Mass with calcification Calcification Other§ Lesion position in breast Anterior Posterior Axillary Central Lesion position relative to clock 1–3 4–6 7–9 10–12
Success (n ⫽ 254)
Failure (n ⫽ 18)
Total
238 (95) 16 (76)
13 (5) 5 (24)
251 21
88 (91) 81 (95) 85 (94)
9 (9) 4 (5) 5 (6)
97 85 90
72 (90) 79 (93) 93 (96) 10
8 (10) 6 (7) 4 (4) 0
80 85 97 10
P Value* .007
.41
.31
.88 27 (93) 180 (92) 47
2 (7) 15 (8) 1
29 195 48
40 (91) 53 (95) 123 (92) 38 (97)
4 (9) 3 (5) 10 (8) 1 (3)
44 56 133 39
.62
.21 78 (90) 101 (94) 75 (96)
9 (10) 6 (6) 3 (4)
87 107 78
110 (95) 21 (100) 121 (92) 2 (67)
6 (5) 0 (0) 11 (8) 1 (33)
116 21 132 3
.33†‡
.23†㛳 36 (100) 94 (92) 2 (100) 122 (92)
0 (0) 8 (8) 0 (0) 10 (8)
36 102 2 132 .33†
67 (93) 34 (100) 46 (90) 107 (93)
5 (7) 0 (0) 5 (10) 8 (7)
72 34 51 115
Note.—Percentages are in parentheses. *Except where otherwise indicated, P values were determined with the 2 test. †These P values were determined by means of the Fisher exact test. ‡Data in “Other” were excluded from analysis. §This category included asymmetric breast tissue and architectural distortion. 㛳Data in “Axillary” were excluded from analysis.
604
SUCCESSFUL NEEDLE-LOCALIZED BREAST BIOPSY
Academic Radiology, Vol 10, No 6, June 2003
Table 2 NLBB Successes or Failures in Relation to Lesion and Patient Characteristics: Continuous Data Factor
Mean
Median
P Value*
Wire distance from lesion (mm) Lesion volume (cm3) Breast size (cm2) Specimen volume (cm3) Breast density† Age at NLBB (y)
2.0 0.55 136.8 55.2 2.6 56.9
0.5 0.90 126.3 30.0 3.0 57.0
.0003 .10 .23 .24 .73 .23
*P values were determined by means of Wilcoxon rank sum test with normal scores. †Breast density was scored as follows: dense ⫽ 1, mostly dense ⫽ 2, mostly fatty ⫽ 3, and fatty ⫽ 4.
Table 3 NLBB Failure Rates by Distance between Localization Wire and Breast Lesion Distance (mm)
No. of Lesions*
No. of Failures†
⬍1 1–3 4–5 6–10 ⬎10
142 82 27 12 9
4 (3) 4 (5) 5 (18) 3 (25) 2 (22)
*n ⫽ 272. †Percentages are in parentheses.
Table 4 Results of Logistic Regression Analysis of NLBB Success in Relation to Lesion and Patient Characteristics Factor
P Value
Wire distance Lesion volume Specimen volume Breast size Breast density
.0006 .17 .03 .02 .30
DISCUSSION Using a standard analytic approach based on logistic regression, we identified three key factors that were significantly related to NLBB failure: increased distance between the localization wire and lesion, decreased specimen volume, and decreased breast size. The risk point system is a practical method for identifying patients with lesions at high risk of localization failure. While the probability of NLBB failure was relatively low for patients
Table 5 Variable Cut Points Found by Isotonic Regression and the Number of Risk Points Assigned to Each Level Cut Points from Isotonic Regression for Each Factor
Risk Points
Specimen volume (cm3) ⬍24 24–57.9 ⱖ58 Wire distance (mm) 0–2 3–9 10–25 Breast size (cm2) ⬍100.2 100.5–204.9 ⱖ205
0 1 2 0 1 2 2 1 0
Table 6 Distribution of NLBB Failures by Risk Score Risk Score
No. of NLBB Failures
0 1 2 3 4 5 6
0/12 (0) 0/62 (0) 2/97 (2) 7/75 (9) 4/20 (20) 4/5 (80) 1/1 (100)
Note.—Percentages are in parentheses. The numbers of NLBB failures for each risk group are as follows: 0 –2 risk points, two of 171 (1%); 3– 4 risk points, 11 of 95 (12%); and 5– 6 risk points, five of six (83%).
with just one moderate- or high-risk factor, the probability increased substantially for those with two or three. Precise placement of the localization wire appears to be a significant predictor of NLBB success in nonpalpable breast lesions. To our knowledge, this is the first study to report data supporting the current teaching that the localization wire should be placed within 5 mm of the lesion to enhance the probability of successful lesion removal. While Jackman and Marzoni (6) hypothesized that a distance of 5 mm might be adequate, their study, in which they used the same method that we used in our study to determine wire distance, revealed that only penetration of the lesion by the needle (or needle placement ⬍1 mm from the lesion) was a significant predictor of NLBB success. Perfection is not always possible, and our results support the teaching that placement within 5 mm of the
605
ABRAHAMSON ET AL
lesion is sufficient for a high probability of successful lesion removal. It should be noted, however, that the proportion of failures in this study was lower when the wire was placed closer than 5 mm (Table 3). Like Jackman and Marzoni (6), we observed an association between NLBB failure and decreased specimen volume despite the fact that data on specimen volume were missing for 48 patients (including one in whom NLBB failed). While greater specimen volume would seem to be a more logical predictor of NLBB success, studies reporting failure rates by specimen volume have not consistently shown this association (3,11,12). The significant association between small breast size and NLBB failure was unexpected. We hypothesized initially that this association may be due to higher density in smaller breasts, but there was no association between density and NLBB failure, and smaller breasts were not more likely than larger breasts to be dense rather than fatty. The high proportion of fatty breasts seen in this study may be due to a disproportionate percentage of obese women within our hospital’s patient population. Another explanation may be that surgeons remove smaller specimens in patients with small breasts to preserve breast tissue, which may make failure to remove the entire lesion more likely. The only factor that significantly affected risk and that is under the control of the radiologist is the location of the wire relative to the lesion. The results of this study suggest that the radiologist performing NLBB must place the wire within 5 mm of the lesion to increase the probability of success, and therefore, that the radiologist is justified in repeating the wire placement to ensure positioning within 5 mm of the lesion. This is especially important in women who have small breasts or in women who have requested a minimal amount of tissue removal for cosmetic reasons. Unlike Jackman and Marzoni (6), we did not find NLBB failure to be associated with lesion type, lesion volume, or the number of lesions per breast. We measured lesion volume in three dimensions, however, rather than using diameter alone, and the present study included fewer patients who underwent NLBB of two lesions in one breast during a single visit. There is no obvious explanation for the inconsistent results for lesion type, since these variables were presumably evaluated similarly in the two studies, which included nearly equal numbers of patients within a comparable age range. Although the failure rate of 6.6% is within the range reported in the literature, it is higher than the 2.5% failure rate reported by Jackman and Marzoni (6). One explanation for our higher failure rate is that our hospital is an
606
Academic Radiology, Vol 10, No 6, June 2003
academic center where trainees work, whereas Jackman and Marzoni work in private practice. As a referral center, our hospital may also have more difficult cases or more patients with concurrent illness. Our study was somewhat limited by the number of missing records, but we do not believe that this limitation biased our results, since there is no reason to suspect that the factors related to missing records vary between successful NLBB and unsuccessful NLBB. Another limitation is that lesion outcome was not recorded. Thus, it is not possible to study NLBB failure in relation to whether the lesion was malignant or benign. Because of the growing use of mammographic screening, increasing numbers of women with nonpalpable breast lesions will be referred for NLBB, although percutaneous needle biopsy may obviate NLBB in most women with benign lesions (13). A better understanding of the causes of NLBB failure may help improve the accuracy of the procedure, prevent malignant lesions from being missed, and reduce the need for repeated excisions. Further studies of NLBB failure could elucidate additional characteristics of lesions at risk for unsuccessful removal and offer insight into the inconsistencies between our results and those previously published (6). REFERENCES 1. Allen MJ, Thompson WD, Stuart RC, et al. Management of non-palpable breast lesions detected mammographically. Br J Surg 1994; 81:543–545. 2. Proudfoot RW, Mattingly SS, Stelling CB, Fine JG. Nonpalpable breast lesions: wire localization and excisional biopsy. Am Surg 1986; 52:117– 122. 3. Norton LW, Zeligman BE, Pearlman NW. Accuracy and cost of needle localization and excisional biopsy. Arch Surg 1988; 123:947–950. 4. Roses DF, Mitnick J, Harris MN, et al. The risk of carcinoma in wire localization biopsies for mammographically detected clustered micro calcifications. Surgery 1991; 110:877– 886. 5. Rissanen TJ, Makarainen HP, Mattila SI, et al. Wire localized biopsy of breast lesions: a review of 425 cases found in screening or clinical mammography. Clin Radiol 1993; 47:14 –22. 6. Jackman RJ, Marzoni FA Jr. Needle-localized breast biopsy: why do we fail? Radiology 1997; 204:677– 684. 7. Kopans D. Breast imaging. Philadelphia, Pa: Lippincott-Raven, 1998. 8. Kopans DB, Swann CA. Preoperative imaging-guided needle placement and localization of clinically occult breast lesions. AJR Am J Roentgenol 1989; 152:1–9. 9. Robertson T, Wright FT, Dykstra RL. Order-restricted statistical inference. New York, NY: Wiley, 1988. 10. Schell MJ, Singh B. The reduced monotonic regression method. J Am Stat Assoc 1997; 92:128 –135. 11. Gallagher WJ, Cardenosa G, Rubens JR, McCarthy KA, Kopans DB. Minimal-volume excision of nonpalpable breast lesions. AJR Am J Roentgenol 1989; 153:957–961. 12. Tinnemans JG, Wobbes T, Hendriks JH, van der Sluis RF, Lubbers EJ, de Boer HH. Localization and excision of nonpalpable breast lesions. A surgical evaluation of three methods. Arch Surg 1987; 122:802– 806. 13. Parker SH, Burbank F. A practical approach to minimally invasive breast biopsy. Radiology 1996; 200:11–20.