Sulindac, a Nonselective NSAID, Reduces Breast Density in Postmenopausal Women with Breast Cancer Treated with Aromatase Inhibitors

Nonsteroidal anti-inflammatory drugs (NSAID) including aspirin have been extensively studied for their inhibition of cyclooxygenase 2 (COX2) and its proinflammatory/protumorigenic metabolite, prostaglandin E₂ (PGE₂; ref. 1). Early demonstration that PGE₂ increases estrogen synthesis via alternate promoter use in CYP19, the aromatase gene, provided initial linking of PGE₂ to estrogen in breast cancer risk (2–6). In 1996, PGE₂ was described as “one of the most potent factors stimulating aromatase expression” in tissues (7). Evidence that PGE₂ promoted mammary tumors in mice, along with lower breast cancer incidence among regular aspirin/NSAID users, advanced earlier interest in NSAIDs for cancer chemoprevention (8). However, success of the endocrine therapies, including the aromatase inhibitors (AI) for hormone-responsive breast cancers, and rare, but serious, toxicities of NSAIDs diminished enthusiasm for their use in prevention despite strong mechanistic evidence (9).

After two decades, unpleasant and common side effects of endocrine therapies and inactivity toward aggressive, non–hormone-responsive cancers have resulted in poor uptake for primary breast cancer prevention (10, 11). In contrast, additional evidence of NSAID benefit in the breast has renewed interest in the prevention potential of anti-inflammatory agents in high-risk individuals. Such evidence includes lower breast cancer incidence in patients at genetic risk who regularly take NSAIDs (12, 13). In addition, experimental models link inflammation in breast tissue stroma and adipose with diffuse and focal desmoplastic-like changes that have been observed early in tumorigenesis and in patients with more dense breast tissue on mammography, an independent risk factor for breast cancer (14–16). Mammographically dense breast tissue is a radiologic feature correlated with stromal features including a predominant contribution of fibrillary collagens (17). Experimentally, inflammation affects the local extracellular matrix (ECM) including promoting physical changes in collagen (e.g., stiffness) and altering the permissiveness of the tissue to tumor growth (18). Unclear is the extent to which localized tissue inflammation in the adipose or in the stroma underlie associations between percent mammographic breast density (BD) and breast cancer risk.

Despite strong preclinical data, whether or not the inhibition of PGE₂ modifies breast cancer risk through effects on breast tissue stroma and adipose is unknown. Initial efforts to examine this question by evaluating the effect of PGE₂ inhibition using celecoxib, a selective COX2 inhibitor, on tissue biomarkers of proliferation (Ki-67) were terminated due to toxicity concerns. Revisiting this, a recently completed study of women at increased risk for breast cancer treated with 400 mg celecoxib twice daily for 6 months using random periareolar fine-needle aspiration sampling of the breast support favorable changes in cytology but no change in tissue Ki-67 or estrogen receptor (ER) expression (19). Observational studies of NSAID use for effects on BD using mammographic measurements have also been largely null though for women followed prospectively, those who continued NSAID use were more likely to remain low density than those who discontinued use (20–23). Separately, Wood and colleagues reported a dose- and duration-dependent inverse relationship between BD and NSAID use (24). In both positive studies, these associations were greater in younger women who on average have higher baseline BD. In the only randomized placebo-controlled trial to assess the effect of NSAIDs on mammographic BD in postmenopausal women at increased risk for breast cancer, a single daily dose of 325 mg aspirin for 6 months showed no effect on BD or on serum estradiol, estrone, or free estradiol concentrations when compared with placebo (25, 26).

To examine the effect of NSAID use on BD, we conducted an open-label, single-arm study of sulindac, a nonselective NSAID, using a noncontrast, breast magnetic resonance imaging (MRI) method measuring percent BD (27). Sulindac was selected for its pleotropic anticancer activity attributed to both COX-dependent and COX-independent mechanisms (28). A dose of 150 mg twice daily was selected based on our prior observation that GDF15, a biomarker of sulindac's non-COX activity (29), was significantly increased in the nipple aspirate fluid of women who received 6 weeks of 150 mg twice daily sulindac compared with single daily dosing (30). We targeted postmenopausal women on AIs with an intact unaffected, contralateral breast as an “at-risk” group with homogeneously low estrogen levels and for whom NSAID may confer additional benefit on AI-associated arthralgia symptoms. Because no prior studies have assessed the effect of AI on BD by MRI, a nonrandomized control arm of postmenopausal patients on AI (observation group) was followed in parallel.

This open-label phase II study was designed to test the effect of 12 months' daily sulindac at 150 mg twice daily on MRI-based percent BD (MRPD) and to assess the feasibility of obtaining paired breast biopsies for tissue studies. Further, while most studies report no AI-specific effect on BD using mammography (31), no prior studies used quantitative fat-water decomposition MRI to measure change in BD. Thus, a nonrandomized observation arm from the same patient population as the sulindac group was included to obtain an estimate of the effect of 12 months' AI therapy on MRPD (i.e., AI-only observation group). Enrollment for both study groups targeted postmenopausal women with a history of stage 0–III hormone receptor–positive breast cancer on AI therapy. The main inclusion criteria were presence of an intact, nonirradiated, noninvolved contralateral breast and BIRADs score of 2, 3, or 4 or presence of scattered, heterogeneous or dense fibroglandular tissue on mammogram in the past 12 months, intent to remain on AI therapy for the duration of the study, and willingness to undergo breast MRI at three time points. For sulindac treatment, participants were asked to refrain from NSAID use except low-dose aspirin (≤ 81 mg/day) throughout the study. Prior to starting sulindac, participants had to demonstrate adequate renal function, normal or controlled blood pressure, and have no contraindications to NSAIDs. Further, to wash out any NSAID effects on BD, participants enrolled to sulindac were asked to stop NSAID use for 4 months prior to their baseline MRI. For sulindac, an optional core-needle biopsy procedure at baseline and after 6 months' sulindac therapy was included to explore sulindac effects on breast tissue. In recruiting eligible subjects to the sulindac study, eligible women not interested in participating or excluded during screening for any reason (e.g., NSAID intolerance) were offered participation in the observation-only study. All participants were asked to undergo three MR breast imaging studies (baseline, 6, and 12 months). No breast biopsies were obtained from the observation group.

At the initiation of the study, the prespecified enrollments were 75 and 60 subjects to sulindac and observation groups, respectively. Following slow accrual and limited resources, accrual targets were reduced in year 5 to 50 subjects initiating sulindac with the expectation that at least 40 would complete 12 months of intervention and the prespecified futility analysis was treated as the primary analysis. There was no plan to test differences between the nonrandomized groups, and no breast biopsy or adverse event (AE) data were obtained for the observation arm. All patients were enrolled at the University of Arizona Cancer Center and at the Stony Brook Cancer Center. The protocol was approved by the Institutional Review Board at each site, and all enrolled patients provided written informed consent.

The primary endpoint was relative change from baseline in MRPD at 12 months. Absolute change at 12 months and relative and absolute change at 6 months are included as additional endpoints. MRPD was obtained using a previously validated measure of the amount of fibroglandular tissue in the breast (27, 32). Briefly, the volumetric fraction of fibroglandular tissue in the breast was obtained and converted to MRPD as a validated comparator to percent mammographic density. During the trial, three MRI scanners were used for collecting MR breast images. The same scanner was used in collecting all breast images for each individual patient (Supplementary Methods). IDEAL fat‐water separation was performed for all data sets with multipeak fat spectrum and R2* correction. The fat‐water separation was performed with in‐house software written in Matlab (MathWorks) using the complex echo images to ensure long‐term stability. Automated breast segmentation was used to generate masks for the whole breast (28). To assess effects of AI and sulindac on the fat and water compartments of the breast, change over time in each fraction was explored separately.

On evidence that inflammation influences collagen physical properties in breast tissue including straightness, we explored collagen fiber length, width, and straightness in normal breast biopsy tissue and change after 6-month treatment with sulindac using second-harmonic generation (SHG) microscopy. As described (33), SHG occurs when a microscope laser field undergoes a nonlinear, second-order polarization as it passes through noncentrosymmetric ordered structures like fibrillary collagen. Briefly, 5-μm hematoxylin and eosin (H&E)–stained normal breast tissue section pairs from 32 subjects blinded to baseline or 6-month status underwent SHG microscopy. Three randomly selected regions of denser stroma (226.32 μm × 226.32 μm) were identified and imaged using a Zeiss LSM 510 META NLO Two-Photon laser scanning confocal microscope system with a 40×/1.3 oil immersion Zeiss Plan-Neofluar objective. An excitation wavelength of 890 nm, with an approximately 100-fs width pulse at an 80 MHz repetition rate, was provided by a Coherent Ti:Sapphire laser XR (Coherent Inc.). Laser scan speed was per pixel at 3.20 μs, and acquisition time of each image (averaged by eight scans) was around 60 seconds. All images were processed using the computer-assisted image feature extraction software, CT-FIRE V2.0 BETA (12/2017) supported by MATLAB. CT-FIRE is an automated tracking algorithm for collagen fiber feature extraction from the SHG image that yields descriptive statistics for length, width, and straightness. Here, straightness is defined as the ratio of the end-to-end length to the total length of the fiber (34). Feature extraction was conducted using default settings, i.e., all fibers with minimum fiber length of 30 pixels, image resolution of 300 dpi, and max fiber width of 15 pixels.

All patients who received at least one dose of sulindac were included for toxicity analysis and AE reporting. AEs were graded using CTCAE v4. Because earlier studies suggested that the cardiac toxicity of NSAIDs may be related to elevations in blood pressure, repeated blood pressure measures were obtained in both the sulindac and observation arms. Summary findings are reported.

Continuous measures were compared using Wilcoxon rank-sum tests, and categorical variables were compared using Χ² tests with exact P values based on Monte Carlo simulation. Linear mixed-effects models for longitudinal data were used to assess change in MRPD at 6 and 12 months in each of the study arms. For the primary endpoint, relative change in BD at 12 months was defined as (MRPD at 12 months – MRPD at baseline)/MRPD at baseline. To improve efficiency and power, log-transformed MRPD was used as the dependent variable in the linear mixed model so that a log fold change could be estimated through a linear combination of estimated coefficients from the fitted model. The corresponding relative change was estimated using a back-transformed log fold change minus 1. The fixed effects of the model included study arm, visit, and an interaction between arm and visit. Factors that differed between the study arms and that may influence the primary endpoint were included as covariates: body mass index (BMI) at every time point, time on AI, and study site. The dependence structure for longitudinal data from the same subject was selected using Akaike information criterion, and the final selected one was Toeplitz. Only subjects who started sulindac intervention and had baseline MRPD were included in the primary analysis (n = 48 for the sulindac arm). As a secondary endpoint, absolute change was defined as the difference between MRPD at 6 and 12 months from MRPD at baseline. Absolute change was analyzed in a separate linear mixed-effect model with adjustment for baseline MRPD. Analysis of the residual indicated that no data transformation on the absolute change was needed. Other covariates in the model were the same as those in model for log-transformed MRPD. Compound symmetry structure was used to model correlation among longitudinal measurement from the same patient. The choice of covariance structure for reporting of absolute or relative change in MRD was considered separately. All analyses for primary and secondary MRPD endpoints were repeated for sulindac patients who were adherent to the protocol (completed ≥ 80% of study agent). Pill counts were recorded for intervention adherence at each clinic visit. Thirty-nine of the 50 evaluable subjects (78%) were adherent as defined by having taken ≥80% of the planned sulindac dose for the 12-month duration.

To explore if change in MRPD over 12 months differed in our study for women at higher risk of developing breast cancer based on baseline mammographic density pattern, we first examined the relationship between baseline MRPD in our study population and mammography density categories (Supplementary Fig. S1). Women with heterogeneously or extremely dense breast patterns on mammography have a 2- to 3-fold greater risk of developing breast cancer than women with predominantly fatty or scattered fibroglandular mammography patterns (31). Based on the mean and standard deviation of MRPD in each category, we selected an MRPD cutoff point of 25% as it best separated women with a scattered fibroglandular pattern (mean baseline MRPD 16.7% ± SD 7.5) from women with heterogeneously and extremely dense patterns (mean baseline MRPD 26.1% ± SD 13.1 and 55.5% ± SD 16.9, respectively).

Sensitivity to missing data was assessed by multiple imputation for MRPD and BMI measurements using each patient's baseline age, race/ethnicity, BMI, time on AI, study site, mean arterial pressure, and prior NSAID use. Ten imputed data sets were generated, and all estimates of MRPD change were constructed using Rubin's rule (34).

The effect of sulindac on breast tissue collagen features included analyses of collagen fiber length, width and angle using linear mixed models as described for the primary analysis to estimate change in these measures. The linear correlation between the change in MRPD and the change in each measure from tissue biopsy imaging was assessed using Pearson correlation coefficient or Spearman rank correlation coefficient if more appropriate.

Between November 28, 2012 and February 19, 2018, 257 women were screened for eligibility: 58 enrolled to sulindac and 56 to observation (Fig. 1). Of the 58 consented to sulindac, 50 started sulindac after an NSAID washout. Of these 50, 48 had baseline breast MR images (one was lost in IT security change on instrument, and one was corrupted by image artifact). Of the 50 that started sulindac, 48 (96%) completed to 6 months, and 43 (86%) to 12 months. Of the 56 who consented to observation, only 46 (82%) underwent the baseline MRI procedure. Of these, 42 (91%) and 40 (87%) completed the study to 6- and 12-month MR imaging, respectively. Of the 40 who completed to 12 months, 39 had breast MR (one subject did not complete end-of-study MRI but completed end-of-study pain and quality-of-life questionnaires).

Baseline characteristics are summarized for the sulindac arm and nonrandom observation control arm (Table 1). Both sulindac and observation arms were majority non-Hispanic White of similar age and BMI. At baseline, the two study groups had similar distributions of MRPD (Table 1; Supplementary Fig. S2). Imbalances between the study arms for time on AI at baseline reflects the run-in/NSAID washout for sulindac participants. After adjusting for baseline MRPD, BMI at every time point, time on AI, and study site, the relative MRPD nonsignificantly decreased by 3.6% [95% confidence interval (CI), −8.9 to 2.0] over 12 months in the AI-only observation group (Table 2). The corresponding absolute reduction in MRPD was −0.65% (95% CI, −1.76 to 0.47; Table 3). Additional adjustments for age, MRI scanner, low-dose aspirin use, or scan date did not change the estimates (Supplementary Table S1). No significant change in either relative or absolute MRPD was observed at 6 months (Tables 2 and 3). Further, small relative and absolute decreases in MRPD over 12 months did not differ significantly by baseline BD strata of ≤ versus >25% (Tables 2 and 3).

In the sulindac group, the adjusted relative MRPD significantly decreased by 9.8% (95% CI, −14.6 to −4.7) over 12 months (Table 2). The corresponding absolute reduction in MRPD was −1.4% (95% CI, −2.51 to −0.29; Table 3). Further adjustments for MRI scanner, age, scan dates, and low-dose aspirin use did not change the estimates. At 6 months, a small 4.2% decrease in relative MRPD was observed (95% CI, −8.1 to −0.10). Absolute change in MRPD at 6 months was not significant. Larger reductions in MRPD were seen in women enrolled to sulindac with baseline MRPD > 25% (−14.6%) versus those ≤ 25% (−8.1%; Tables 2 and 3). The difference between strata was not significant (P = 0.269).

Preplanned sensitivity analyses on the main outcome of relative change in MRPD after 12 months included effects of adherence to sulindac and missing data for MRPD and BMI. Most participants who stopped study agent also withdrew early (Fig. 1). Of the 43 who completed to 12 months, 39 (91%) were adherent at >80% sulindac dose. For missing data, imputed adjusted results using a pooled estimate of change in relative MRPD for the sulindac arm for 6 and 12 months were −5.6% (95% CI, −10.2 to −0.9) and −13.3% (95% CI, −19.2 to −7.3), respectively. Results for the observation arm were unchanged from Table 3.

Because MRPD is derived from fat and water (i.e., fibroglandular) signals in the breast, we conducted exploratory analyses of the relative change in the fat and water fraction of the breast separately. Neither the water nor the fat volume changed significantly from baseline to 12 months in the AI-only observation arm (Table 4). In the sulindac arm, the estimated relative breast water volume significantly decreased by 7.6% (95% CI, −13.3 to −1.5) at 12 months with a nonsignificant increase in mean fat volume of 5.8% (95% CI, −0.4 to 12.3).

At 6 months, a total of 36 of 50 (72%) participants who started sulindac agreed to optional core-needle biopsy at baseline and 33 to repeat the procedure at 6 months. For the 36 subjects at baseline, 34 underwent a research biopsy and 2 consented to tissue biopsy from the contralateral breast during a planned surgery. At 6 months, 31 underwent research biopsy and 2 provided tissue biopsy during planned surgery. Of the 33 available pairs, 32 had adequate tissue for exploratory studies of change in breast tissue collagen fibers using SHG microscopy. With SHG microscopy, breast tissue collagen fibers appear wavy or straight with thick bundles and thin strands (Fig. 2). Descriptive statistics for three collagen features (length, width, and straightness) and within-subject variance at the baseline measure are shown in Supplementary Tables S2 and S3. Unadjusted, mean straightness declined at 6 months (P = 0.032), but attenuated slightly after adjusting for baseline BD, months on AI, BMI, and study site (P = 0.053; Fig. 2). No significant changes in width or length were detected from baseline to 6 months (all P > 0.2).

All 50 subjects who started sulindac were included in the toxicity analysis. The most common toxicities were attributed to AI therapy and included musculoskeletal symptoms in 11 patients (22%), insomnia in 2 patients, and increased depression in 1 patient (Supplementary Table S4). Grade 2 or higher AEs possibly, probably, or definitely attributed to sulindac included 7 (14%) patients with gastrointestinal side effects: abdominal pain (3), heartburn/indigestion (2), diarrhea (1), and nausea (1). An additional 3 patients complained of grade 2 rash or pruritis of unclear etiology (6). Seven patients had grade 2 or 3 hypertension (14%; 95% CI, 6.6–26.7). One of the 2 patients with grade 3 hypertension had the event during an intracranial hemorrhage serious AE. Two patients initiated new antihypertensive medications on study, whereas the remainder identified during a clinic visit resolved following home blood pressure monitoring. Because earlier studies suggested that sulindac was more renal sparing than other NSAIDs (35), clinical blood pressure was examined at three time points during the study. No change was observed in either arm for mean arterial pressure, diastolic, or systolic blood pressure over 12 months (Supplementary Fig. S3).

Of the 4 patients who experienced a serious AE, 2 were possibly related to sulindac, including a patient with an underlying cerebral amyloid angiopathy who developed an intraparenchymal hemorrhage, and a patient who developed acute pancreatitis requiring hospitalization. Both, as well as 2 patients with gastrointestinal (GI) pain, discontinued sulindac therapy.

In this open-label study of 150 mg sulindac twice daily for 12 months, we found that postmenopausal women with a history of hormone receptor–positive breast cancer on AI therapy experienced a significant decrease at 12 months in relative and absolute MRPD. In a nonrandomized but similar population of women on AI therapy only, no significant change in MRPD was observed after 12 months. This is the first study to evaluate the effect of 12 months' use of an NSAID on BD and the first to support a possible effect of sulindac to decrease BD in postmenopausal women on AI therapy.

Only one randomized placebo-controlled trial has examined the effects of an NSAID on BD. In a 6-month study of 325 mg/day of aspirin or placebo, McTiernan and colleagues (25) found no effect of aspirin on absolute change in percent mammographic BD (PMD) at 6 months in a sample size nearly twice the size of the current study. Like the aspirin study, we observed no effect of sulindac on absolute change in MRPD at 6 months, with evidence for a small change in relative MRPD (4.2%). Both studies enrolled postmenopausal women with similar mean baseline BD (i.e., 18.3% aspirin study, 21.8% this study). In contrast, the current study enrolled only women on AI therapy, a group excluded from the aspirin study and the treatment duration was longer at 12 months. The current study also used a noncontrast MRI BD measure (MRPD) shown to demonstrate high reproducibility and low variability between measures (27).

Unlike the aspirin study, a limitation of the current study is that it was not randomized, nor placebo controlled. And while a “placebo effect” is unlikely with a quantitative imaging endpoint, the significant decrease with sulindac could be due to chance alone. Further, while we routinely monitored AI use in both arms and considered time on AI with baseline BD and any change in BMI as potential confounders, we did not conduct AI pill counts for dose adherence. As such, we cannot rule out the possibility that differences between the two groups may be due to imbalances in adherence to AI dose.

Alternatively, longer exposure to sulindac (12 months) and the high dose may explain differences between this and the aspirin study. Also dissimilar to the aspirin study, which reported no effect on circulating estrogen levels (26), sulindac was combined in this study with an AI. Whether sulindac acts independent of AI to reduce BD or by enhancing suppression of PGE₂-mediated aromatase gene expression in the adipose of this predominantly overweight population is unknown. Greater activity of NSAIDs for breast cancers occurring in overweight/obese women is supported by observational findings, though whether the benefit is greater for hormone-dependent cancers remains unclear (36). The potential for an effect of NSAIDs on BD in overweight women is further supported by findings that aromatase is upregulated in inflammatory foci in tissues of overweight women and that stromal adipocytes from overweight/obese women promote focal desmoplastic-like changes to the ECM (16). Whether or not these changes link BD to increased breast cancer risk is, however, unknown. In addition to effects on PGE₂, sulindac is pharmacologically distinct from aspirin (37, 38), including COX (PGE₂)-independent activity (37, 38) that may account for differences in the two studies.

It is appreciated that the ECM is functionally important in breast tumorigenesis, including evidence that more aligned and stiffer collagen fibers correlate with increased COX-2 expression, promote tumor growth and invasion, and predict worse patient outcomes (18, 39, 40). Poorly understood, however, are the biochemical and physical ECM properties of “at-risk” breast tissue and how inflammation influences the ECM, relates to BD, and BD-related cancer risk. Our finding that collagen fiber straightness measured by SHG decreased after 6 months with sulindac is promising for the potential ECM-modulating effects of NSAIDs. However, the lack of similar paired biopsy specimens from the observation group and small sample size to relate changes to BD are clear limitations. Recognized insensitivity of CT-FIRE to differentiate wavy from straight collagen fibers presents measurement constraints. For example, while we observed a statistically significant decrease in straightness after 6 months, the absolute change is quite small. This is despite what appears visually as large differences (Fig. 2). Anecdotally, on unblinding, the appearance of the stroma by H&E light microscopy and by SHG was noticeably “more disorganized” in the post sulindac samples. It is worth commenting that the magnitude of the differences in collagen fiber straightness observed here, although very small, is consistent with other studies that have shown for example that while mean collagen straightness is significantly higher in ductal carcinoma in situ (DCIS) with inflammation or with central necrosis compared with DCIS without inflammation or central necrosis, the absolute differences are very small and of the same magnitude to what we observe for pre and post sulindac (41).

Unlike endocrine therapies, NSAIDS are widely used medications in adults and are generally well tolerated. Recognizing common GI intolerance and rare, but serious adverse events with chronic use, the need to understand the relative benefit of NSAIDs for breast cancer prevention remains. Currently, we are unable to address if a small decrease in BD with sulindac after 12 months would relate to a reduction in breast cancer events or breast cancer mortality. There is limited evidence on which to draw a relationship between the magnitude of a reduction in BD by any intervention and breast cancer risk reduction. The only randomized controlled trial evidence indicating that a reduction in BD correlates with a reduction in breast cancer incidence are from secondary analyses of the International Breast Cancer Intervention Study (IBIS-I) that compared tamoxifen to placebo in high-risk women for breast cancer chemoprevention (42). Findings from IBIS-I were the first to show tamoxifen use reduces BD (43). In follow-up analyses, Cuzick and colleagues explored change in BD using change in PMD intervals of 5% in participants with >10% baseline BD who received tamoxifen or placebo (44). By 18 months, PMD decreased with tamoxifen by 7.9% (95% CI, 6.9–8.9) compared with a 3.5% reduction (95% CI, 2.7–4.3) with placebo revealing a significant difference between the two groups (P < 0.001). Subsequently, the group found that women who received tamoxifen and who had a 10% or greater absolute reduction in BD by 18 months experienced significantly lower odds of developing breast cancer (OR = 0.32; 95% CI, 0.14–0.72) when compared with women who received tamoxifen but had no change in BD. For women with “5%” reduction, a nonsignificant reduced odds ratio of 0.90 and a wide confidence interval (0.40–2.04) was observed compared with women with no PMD change. These results support that PMD can serve as a predictor of response to tamoxifen (biomarker) in the prevention setting. However, in an accompanying editorial (45), Dr. Normal Boyd raised important points about interpreting these findings as well as for relating a reduction in PMD to a cancer prevention benefit. These are important points to reconsider here.

First, although the findings from IBIS-I support an effect of tamoxifen to reduce PMD, the qualitative PMD measurement used in the study was recognized as highly subjective and prone to error. As noted by Boyd, measurement error combined with a small sample size could result in an underestimation of the effect of tamoxifen on PMD and also in evaluating relationships between the reduction in PMD with tamoxifen and a decrease in breast cancer incidence. In this context, it is worth noting the magnitude of BD change with tamoxifen is strongly dependent on two highly correlated factors: menopausal status and baseline BD. To illustrate this point, participants in IBIS-I (43) who were 55 years or older (postmenopausal) had an average baseline PMD of 28.53% and experienced an absolute net reduction in BD after 54 months of tamoxifen of 1.1%. Despite an equivalent sample size, women who were ≤45 years had an average baseline PMD of 48.81% and experienced a 13.4% absolute net reduction in BD. The null to small change with tamoxifen for effect on PMD in the postmenopausal setting is similar to what is reported for AIs used exclusively in postmenopausal women. Although it is certainly plausible that the underlying factors that influence BD in pre- and postmenopausal women differ, we suspect that a more likely explanation for failure to relate endocrine therapy effects on BD in the postmenopausal setting is the low sensitivity of mammography to small changes in women with lower baseline BD. This motivated our efforts to develop and validate a quantitative MRI measure of BD (27, 32). Our results here support noncontrast breast MRI as a sensitive measure to detect small changes in BD.

Secondly, although we are encouraged by these results including the larger declines in MRPD in women with heterogeneous or extremely dense breast (high-risk group), as discussed by Boyd, we cannot conclude that simply reducing BD will reduce the risk of breast cancer absent understanding of how sulindac or other NSAIDs affect the biology that links BD to breast cancer risk. The significance of our results depends on replicating findings that sulindac reduces BD and on relating reductions in BD with tissue changes in the factors that link BD to cancer risk. From our perspective, this includes gaining better understanding on hypothesized effects of PGE₂ on breast tissue including “proestrogenic” activity via aromatase gene upregulation in adipose and pro-desmoplasia effects on the ECM and their contribution to BD and to cancer risk.

In summary, this study provides the first evidence for an effect of the NSAID sulindac to reduce percent BD after 12 months of use. Improvements in AI-associated stiffness with sulindac (to be reported elsewhere) and findings of a decrease in breast tissue collagen straightness, a suspect tissue risk biomarker, add evidence for the potential benefits of sulindac for breast chemoprevention. However, although we are encouraged by the results and high adherence to sulindac in this breast cancer patient population (i.e., a high-risk and highly motivated patient group), NSAID-related toxicities, including the more common GI side effects observed in this study, remain a challenge. These results support continued efforts to investigate NSAIDs for their effects on BD and to understanding how BD and modulation of BD relates to cancer risk.

C. Huang reports grants from NIH during the conduct of the study. D. Roe reports grants from NCI during the conduct of the study. P. Chalasani reports grants from Pfizer. P. Chalasani also reports other support from Oncosec, Asthenex, Zeno Pharmaceuticals, Eli Lilly, and Novartis, as well as personal fees from HOPA, Horizon CME, and Health Advances outside the submitted work. A.T. Stopeck reports grants from NCI during the conduct of the study, as well as grants and personal fees from Amgen and personal fees from AstraZeneca outside the submitted work. No disclosures were reported by the other authors.

P.A. Thompson: Conceptualization, resources, formal analysis, supervision, funding acquisition, methodology, writing–original draft, project administration, writing–review and editing. C. Huang: Conceptualization, resources, formal analysis, supervision, investigation, methodology, writing–original draft, project administration, writing–review and editing. J. Yang: Formal analysis, supervision, writing–original draft, writing–review and editing. B.C. Wertheim: Formal analysis, writing–review and editing. D. Roe: Formal analysis, supervision, writing–review and editing. X. Zhang: Formal analysis, writing–review and editing. J. Ding: Formal analysis, methodology, writing–review and editing. P. Chalasani: Supervision, project administration, writing–review and editing. C. Preece: Data curation, investigation, methodology, writing–review and editing. J. Martinez: Formal analysis, writing–review and editing. H.-H.S. Chow: Formal analysis, supervision, writing–review and editing. A.T. Stopeck: Conceptualization, formal analysis, funding acquisition, investigation, methodology, project administration, writing–review and editing.

The study was supported by funding from the NCI CA1615301. We acknowledge the biostatistical consultation and support provided by the Biostatistics and Bioinformatics Shared Resource at the Stony Brook Cancer Center. We also thank Dr. Matthew Conklin at University of Wisconsin at Madison for his assistance in establishing the Second Harmonics Generation Microscopy and implementation of CT-FIRE. The work was supported by funding from the NCI (grant number R01CA161534). Trial Registration: ClinicalTrials.gov Identifier: NCT00245024.

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.