Factors Associated with False Positive Results on Screening Mammography in a Population of Predominantly Hispanic Women

Screening mammography has been an accepted method of breast cancer screening in the United States for decades. Universal screening mammography results in increased cancer detection rate and a reduction in breast cancer mortality compared with clinical breast exams alone (1). Among women ages 50 to 74 years, biennial screening mammography confers a 25.8% relative risk reduction in breast cancer mortality (2). However, guidelines for screening mammography in the United States remain controversial, with differing recommendations from major medical organizations. In 2013, the American Cancer Society (ACS) recommended that women begin screening annually at age 45, with the transition to biennial screening at age 55; in addition, women ages 40 to 45 should be given the option to start screening annually (3). In contrast, the United States Preventive Services Task Force (USPSTF) recommends that average-risk women start screening biennially at age 50 (4).

Much of the debate about screening mammography arises from an increasing awareness of the risks of false positive results, defined as recall breast imaging or biopsy not yielding breast cancer. Of particular interest are the psychological effects of false positive mammography, with studies showing consistently higher scores of psychological distress in women with false positive results, persisting up to 12 months (5), and even similar levels of distress in women with false positive results as those with breast cancer diagnoses (6). In addition, false positive results incur considerable financial consequences, with an estimated cost in the United States of approximately $2.8 billion between 2012 and 2013 and an average $200 out-of-pocket cost to patients (7).

In the United States, the likelihood of a false positive screening mammogram is substantial. For annual mammography, the probability of a false positive recall mammogram was 16.3% for a woman's first screening mammogram and 9.6% for subsequent mammogram, and the probabilities of a false positive biopsy at these screenings were 2.5% and 1.0%, respectively (8). The cumulative probability of false positive studies over 10 years for annual screening ranges from 43.1% to 61.3% (8–11). There are several variables that alter such probabilities, however. Biennial screening decreases these cumulative probabilities to 29% to 42% (8, 10), contributing to the recent ACS and USPTSF recommendations to extend screening intervals to biennially. In addition, younger age (particularly age 40–45), dense breasts, and breast cancer risk factors increase the probability of a false positive result (12).

There is a dearth of information, however, regarding false positive mammography in minorities, particularly for the Hispanic population. Adherence to screening mammography programs is a particular concern in this population, especially after a false positive mammogram. In one study, Hispanic women were less likely to continue with screening mammography after experiencing a false positive result (13). Rates of false positive mammograms are higher at facilities serving “vulnerable” populations (i.e., lower educational attainment, racial/ethnic minorities, and limited income), perhaps reflecting the concern that these patients might be less likely to follow-up after an abnormal study and are at greater risk of diagnosis with cancer at presentation (14). There is a need, therefore, to more closely examine the rates of and contributions to false positive mammography in these populations to tailor screening recommendations to their unique needs.

In this study, we aim to further examine the relationships between mammographic breast density, breast cancer risk status, age, and frequency of screening to false positive results on screening mammograms in a predominantly Hispanic population.

We conducted a retrospective cohort study called the Know Your Risk: Assessment at Screening (KYRAS) for breast cancer study. Women were approached for enrollment during routine screening mammography at the Avon Breast Imaging Center at Columbia University Medical Center (CUMC) in New York, New York. They completed a questionnaire on sociodemographic characteristics and breast cancer risk factors and gave consent to access their electronic health records (EHR) for research purposes. The inclusion criteria for this analysis include: (i) women, age ≥18 years, (ii) English or Spanish-speaking, (iii) no previous diagnosis of breast cancer, (iv) providing permission to access EHR information. The study was approved by the institutional review board at CUMC.

Participants completed a one-time comprehensive survey on demographic characteristics and breast cancer risk factors, including family history of breast and ovarian cancer. We collected information on age (dichotomized as less than 50 years and 50 or older), highest educational level [less than 8 years of education, 8–11 years (without graduating high school)], high school graduation or General Equivalency Diploma, vocational or technical school or military training, some college or university classes but no degree, associate's or bachelor's degree, graduate degree, post-graduate degree, or professional degree), race [white, black, Asian (Chinese, Japanese, Filipino, Hawaiian, Other Pacific Islander, Other Asian-American), Native American or Alaskan Native, other], ethnicity (Hispanic, not Hispanic), and Jewish ancestry (Ashkenazi, Sephardi, both). We also calculated body mass index (BMI) using the subject's self-reported height and weight, categorizing BMI as underweight (<18.5 kg/m²), normal weight (18.5–24.9 kg/m²), overweight (25.0–29.9 kg/m²), or obese (≥30.0 kg/m².

We collected information on additional breast cancer risk factors, including first-degree family history of breast cancer, age of menarche, first live birth (if applicable), and history of breast biopsy (including number of biopsies and presence of atypia) to calculate 5-year and lifetime absolute risk of invasive breast cancer according to the Gail breast cancer risk assessment tool (BCRAT; ref. 15). We were unable to calculate breast cancer risk in women <35 years of age or those with a prior history of breast cancer, ductal or lobular carcinoma in situ, as the Gail model does not produce a reliable risk calculation for these women. In addition, we used a modified Six-Point Scale family history screener to determine eligibility for genetic testing for breast cancer susceptibility genes (16). This model accounts for first and second-degree family history of breast and ovarian cancer, age at diagnosis (before or after age 50), bilateral breast cancer, male breast cancer, and Ashkenazi Jewish descent (16). We dichotomized risk status as high risk or low/average risk for analysis. High-risk status for breast cancer was defined as having a greater than 1.66% 5-year risk or 20% lifetime risk of invasive breast cancer based upon the Gail model, or a score of 6 or greater on the modified Six-Point Scale, which indicates eligibility for BRCA genetic testing.

We also collected information on the women's breast cancer risk perceptions and their breast cancer screening behaviors. We asked about the participant's perceived risk of getting breast cancer (very low, moderately low, neither low nor high, moderately high, very high) and perceived risk compared to other women (much lower, about the same, much higher; ref. 17). Finally, we assessed self-reported mammography screening frequency (yearly, every 1–2 years, every 2–3 years, other) to determine screening behaviors.

We collected the participants' mammography and biopsy results from the EHR. From the radiology reports, we retrieved information on breast density, frequency of screening, number of mammograms, follow-up time since first mammogram, and recall imaging dating back to 1989.

Mammographic breast density was assessed qualitatively by radiologists using a standardized Breast Imaging Reporting and Data System (BIRADS) classification system: (i) “mostly fatty” (<25% density) or “scattered areas of fibroglandular density” (25%–50%), defined as low breast density, and (ii) “heterogeneously dense” (51%–75%) or “extremely dense” (>75%), defined as high breast density. For analysis, we used the breast density category for the woman's latest mammogram.

Frequency of screening was determined in the EHR by calculating the median number of days between mammograms. Frequency was categorized as yearly, biennial, or every 3+ years. If the median screening interval was between 274 days (9 months) and 548 days (18 months), then it was coded as yearly screening; a median interval between 548 days (18 months) and 913 days (30 months) was coded as biennial screening (12, 18). If individuals had more than 913 days between their mammograms, then they were coded as screening every 3+ years, which is considered as non-compliant according to both the ACS and USPSTF screening guidelines (3, 4). If the interval between breast imaging was less than 274 days (9 months), then it was coded as recall breast imaging. If these data were not available, we used the woman's self-reported response in the KYRAS questionnaire, which was also categorized as yearly, biennial, or every 3+ years. If the woman was having her first screening mammogram at the time of enrollment, screening frequency was not documented.

All of the breast pathology reports were reviewed by a medically trained coder (JEM) and were categorized as: (i) non-proliferative breast disease, (ii) proliferative breast disease without atypia, (iii) atypical hyperplasia, (iv) lobular carcinoma in situ, (v) ductal carcinoma in situ, (vi) invasive breast cancer, or (vii) non-diagnostic. A false positive biopsy was defined as any breast biopsy not yielding breast cancer.

The primary outcome of interest was a false positive result on screening mammography, defined as recall breast imaging or biopsy not resulting in breast cancer. Recall imaging was defined as receiving a BIRADS score of 0, 3, 4, or 5 on the screening mammogram without resulting in invasive breast cancer or ductal carcinoma in situ. Category 0 indicates a need for additional imaging evaluation, Category 3 recommends a follow-up time of 6 months, Category 4 indicates a suspicious abnormality, and Category 5 is highly suggestive of malignancy (19). False positive biopsies were defined as any breast biopsy that did not result in invasive breast cancer or ductal carcinoma in situ. This is consistent with the definitions of false positive results on screening mammography used in prior literature (20).

Descriptive statistics were generated for baseline variables of interest. We conducted univariate analyses using the χ² test to examine the relationship of each of the exposures of interest (age, race/ethnicity, education, breast cancer risk status, breast density, and frequency of screening mammograms) with the outcome of interest (ever had recall breast imaging or biopsy) to determine significant relationships (P < 0.1). After identifying the significant predictors, we used multivariable logistic regression models to examine the magnitude and direction of association. We also controlled for race/ethnicity, BMI, and total years of follow-up since first documented mammogram in the EHR. The covariates were selected based upon the 10% rule and biological significance. If the odds ratio (OR) between each variable of interest and the outcome of interest changed by more than 10% after including the covariate into the univariate model, it was considered to be a potential confounder, and thus included in the final multivariable model. All analyses were conducted using SAS version 9.3 (Cary, NC).

From November 2014 to October 2015 (Fig. 1), of 13,023 women who presented for mammography screening at CUMC, 2,757 (21.2%) women were approached for enrollment into the KYRAS study and 396 (14.3%) refused participation. Of the 2,361 enrolled participants, 2,335 (98.9%) had complete questionnaire data. For our analysis, we excluded those who had a previous diagnosis of breast cancer (N = 42), refused access to their electronic health record (EHR) for research purposes (N = 102), or were missing EHR information (N = 172). Our final sample consisted of 2,019 women.

The baseline characteristics by race/ethnicity of the study population are described in Table 1. Based upon data collected from the EHR, the study population was similar in demographic and clinical characteristics to the remaining screening cohort not enrolled in the KYRAS study (Supplementary Table S1). The median age of the participants was 59 years (range, 29–91). Seventy-six percent of the study population was Hispanic, 10% non-Hispanic white, 10% non-Hispanic black, and 4% other. A large percentage of our population was identified as meeting high-risk criteria for breast cancer either by the Gail model (17.7%), by their eligibility for BRCA genetic testing (11.7%), or by either criteria (24.7%). Approximately 18% of the Hispanic women were considered high risk for breast cancer, compared to 68% of non-Hispanic white women, 31% non-Hispanic black women, and 29% other. Twenty-four percent of the non-Hispanic white women were of Ashkenazi Jewish descent, and a large proportion of these (51%) met family history criteria for BRCA genetic testing.

We were able to obtain screening frequency data from the EHR for 84% of our population. In these women, self-reported screening frequency showed a statistically significant correlation (P = 0.0339) with EHR screening frequencies using the Kendall's tau-b correlation method. Therefore, for the other 16% of women who were missing EHR screening frequencies, we assumed that their self-reported screening frequency was equivalent to their actual screening frequency. Among participants, 69% screened yearly, 17% screened biennially, 10% screened every 3+ years, and 4% of women were screening for the first time. Despite having lower breast cancer risk, Hispanic women presented for screening mammography more frequently, with 72% screening yearly compared to 61%, 64%, and 52% screening yearly among non-Hispanic whites, non-Hispanic blacks, and other women, respectively.

With a median follow-up of 8.9 years (range, 0–26) since their first documented mammogram in the EHR, the women had a median of 7 mammograms (range, 1–27). About 53% of the women in our analysis had at least one false positive result on their screening mammograms (Table 2): 52% of them had at least one recall breast imaging and 12% of them had at least one biopsy that did not result in breast cancer. For the women who had a false positive biopsy, 55.0% had nonproliferative breast lesions, 27.5% had proliferative disease without atypia, 7.2% had atypical hyperplasia, 6.8% had nondiagnostic findings, and 3.5% had other benign findings. In addition, 4 women had newly diagnosed screen detected breast cancer on the mammogram conducted between November 2014 to October 2015, including 2 invasive breast cancers and 2 cases of ductal carcinoma in situ.

In univariate analysis, age, frequency of screening, and breast density were significantly associated with ever having a false positive result (Table 3). After controlling for total years of follow up, age, race/ethnicity, BMI, breast density, and breast cancer risk status, frequency of mammographic screening and breast density were significantly associated with having a false positive result on a screening mammogram. Those who screened annually had 2.18 times the odds of having a false positive result compared with those who screened biennially [OR, 2.18; 95% confidence interval (CI), 1.70–2.80]. Having dense breasts was associated with a greater than 60% increased risk of having a false positive result compared to women with less dense breasts (OR, 1.64; 95% CI, 1.32–2.04). We also found that for every year of screening follow-up there was an 8% increase in the likelihood of having a false positive result (OR, 1.08; 95% CI, 1.07–1.10). Older women (50 years or older) tended to have a lower risk for having a false positive result compared with women aged less than 50 years, after controlling for known confounders (OR, 0.80; 95% CI, 0.62–1.02). However, this association was not significant.

We compared women who screened annually versus biennially by sociodemographic and clinical factors, and examined the relationship between age, race/ethnicity, breast cancer risk status, education, and perceived breast cancer risk with screening frequency. In the univariate analysis, we found that none of these factors was significantly associated with screening frequency (Table 4). In the multivariable logistic regression model, we found that only race was significantly associated with screening behaviors, after controlling for the covariates. Hispanic women were more likely to screen annually compared to non-Hispanic white women (OR, 1.92; 95% CI, 1.17–3.15). In fact, all minority populations were more likely to screen annually compared with non-Hispanic white women (Table 4).

We identified specific factors associated with having a false positive result on a screening mammogram and described the magnitude of these relationships. Fifty-three percent of the women in our study received a false positive result over a median follow-up time of 8.9 years. This finding is in agreement with recent models that predict that 61.3% of women who screen annually will have a false positive result over 10 years (8). Through the combination of self-reported information and data extracted from the EHR, we found that more frequent mammography screening and higher breast density were significantly associated with having a false positive result on a screening mammogram, after controlling for BMI, years of follow-up, and race/ethnicity. Women who screened annually were 2.18 times more likely to have a false positive result compared with those who screened biennially, and those who had denser breasts were 1.64 times more likely to have a false positive result compared with women with less dense breasts.

Our finding that more frequent screening is associated with increased false positive mammography results is consistent with those of similar studies, which showed lower cumulative rates of false positives in women over a 10-year period with biennial versus annual screening (8, 10, 18, 21). These studies have driven the recent changes in USPSTF and ACS guidelines for screening mammography in average-risk women, and our results further reinforce this.

Although prior studies investigating the role of age in false positive mammography have shown the highest rate of false positive results among younger women ages 40 to 49, with rates decreasing with increasing age, our multivariable analysis did not show that age was a significant predictor of false positives in our study population (12). Younger women have denser breasts, and other studies have similarly demonstrated that after controlling for breast density, age was no longer significantly associated with having a false positive result (22, 23). Our univariate analysis supports this, as age was initially significantly associated with having a false positive mammogram, but after controlling for breast density and screening duration, this relationship was attenuated.

In our subgroup analysis, we found that Hispanic women tended to be at lower breast cancer risk compared with non-Hispanic white women. However, Hispanic women were actually screening more frequently than non-Hispanic white women (71.9% of Hispanic vs. 60.8% of non-Hispanic white women screening annually). In light of the concerns regarding false positives, these women may actually be exposing themselves to unnecessary harms. Previous studies examining mammography trends in Hispanic women have primarily been focused on increasing uptake and reducing barriers to access breast cancer screening (24–28). Consequently, much of the work has been to remove these barriers and improve screening adherence in this population (29). Although these efforts have done much to increase screening uptake, we must also consider the potential harms of over-screening in this population, including decreased willingness to continue with future screening mammography after false positive results (13). Future efforts should therefore be focused on education of both patients and providers in risk-stratification of Hispanic women to inform appropriate screening intervals in this population.

There are several limitations to our study. First, this is a retrospective study and we did not assess changes in screening patterns over time as recommendations have been updated. However, the majority of women self-reported screening annually, which was confirmed in their medical record. Second, we used the Gail breast cancer risk model to calculate risk in our participants, which was developed in non-Hispanic white women and has not been well-validated in Hispanic populations (15). Third, we recruited participants from a population actively engaged in screening mammography at an academic medical center. Therefore, there may be selection bias for women who were more likely to be compliant with screening practices and have access to the medical system. Finally, we conducted a single institution study at an urban academic medical center and our results may not be generalizable to screening populations in other geographic regions.

Our study has several strengths, however. One is that we explored breast density specifically as a variable of interest, which could be a useful marker to inform women about how they should approach their own screening choices (30, 31). Another unique aspect of our study is our predominantly Hispanic population, a population that has been underrepresented in studies of screening mammography and breast cancer risk assessment. Additional strengths include our large sample size and use of a comprehensive questionnaire that thoroughly assessed breast cancer risk and genetic testing eligibility in individuals.

In summary, our study identified frequency of screening mammography and breast density as variables influencing rates of false positive results in a predominantly Hispanic population in New York City, both reinforcing the findings of prior studies and contributing new data for an under-studied population. Hispanic women were more likely to undergo annual screening mammography despite lower breast cancer risk compared with non-Hispanic white women. Futures studies should aim to identify interventions to inform providers and patients on breast cancer risk and appropriate mammography screening intervals, to reduce the rates of false positive mammography results and thus minimize the potential harms of screening.

No potential conflicts of interest were disclosed.

Conception and design: W. Ueng, R. Kukafka, K.D. Crew

Development of methodology: W. Ueng, R. Kukafka, K.D. Crew

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.S. Trivedi, A. Vanegas, J. Vargas, R. Kukafka, K.D. Crew, R. Sandoval

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): W. Ueng, M.S. Trivedi, H.S. Yi, K.D. Crew

Writing, review, and/or revision of the manuscript: J.E. McGuinness, W. Ueng, M.S. Trivedi, R. Kukafka, K.D. Crew

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.E. McGuinness, M.S. Trivedi, R. David, A. Vanegas, J. Vargas, K.D. Crew

Study supervision: R. David, A. Vanegas, R. Kukafka, K.D. Crew

The study was supported by the National Cancer Institute (grant R01 CA177995-01A1) and the Avon Foundation for Women.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.