Randomized Trial of Surveillance with Abbreviated MRI in Breast Cancer Survivors – Does it impact patient anxiety and cancer detection rate?

Purpose: Abbreviated breast MRI substantially reduces the image acquisition and reading times and has been reported to have similar diagnostic accuracy as a full diagnostic protocol but has not been evaluated prospectively with respect to impact on patient anxiety in breast cancer survivors and cancer outcomes. Methods: This prospective controlled trial of parallel design was performed at an academic center on women with a personal history of breast cancer who were randomized into two groups: surveillance with MG or MG plus A-MRI. Primary outcome was anxiety compared between the two and measured by four validated questionnaires at three different time-points during the study. Other parameters including the CDR, abnormal interpretation rate (AIR), and positive predictive value for biopsy (PPV3) were compared between modalities of MG and A-MRI. Tissue diagnoses or 1 year of follow-up were used to establish the reference standard. Linear mixed models were used to analyze anxiety and Fisher’s exact test to compare imaging outcomes. Results: 198 patients were allocated to either MG alone (94) or A-MRI plus MG (104). Anxiety scores in all questionnaires were similarly elevated in both groups (50.99+/-4.6 with MG vs 51.73+/-2.56 with MRI,p>0.05) and did not change over time. MRI detected 5 invasive cancers and 1 DCIS, and MG detected 1 DCIS. MRI had higher incremental CDR(48/1000(5/104) vs MG 5/1000(1/198,p=0.01)) and higher AIR 25%(26/104) vs MG 4.5%(9/198,p<0.00001), with no difference in PPV3:MRI 28.6%(6/21)vs MG 16.7%(1/6,p=0.557). Conclusion: Compared to mammography alone, A-MRI had signicantly higher incremental cancer detection in breast cancer survivors. Despite a higher rate of recalls and biopsies, A-MRI had no adverse impact on anxiety. on intent-to-treat (ITT) analyzed linear mixed models, with Intervention (Mammogram versus Mammogram plus fast MRI), time of assessment (T1, T2, T3), Intervention by Time interaction as xed factors. Models using Maximum (REML) with an unstructured covariance structure to account for correlations among repeated measures over time. A signicant Time by Intervention interaction would suggest that changes in measures over time were different between the interventions; signicant interactions were further analyzed with pairwise least square mean comparisons. Data from missing questionnaires were not imputed because our analytical REML reliable of assumption of missing at random (MAR)[22]. Descriptive statistics were calculated using a spreadsheet software program (Excel, Version 2013, Microsoft). Screening outcomes were compared between groups using Fisher’s exact test. Sample size calculation was based on primary outcome the STAI. There is no generally accepted minimal clinically important difference for the STAI, and a 4-point difference was selected to be a minimal clinically important difference, as it would correspond to a complete difference in one of the 40 items. In order to have 80% power to detect a 4-point difference between the groups, we planned 134 patients per group. Recruitment stopped early due to differences in cancer detection rates. Results considered signicant if p 0.05. Results The following positive predictive value for biopsy recommendations (PPV2), positive predictive for performed (PPV3), and specicity.


Introduction
Women with a prior history of breast cancer (PHBC) often have a high level of anxiety related to breast cancer surveillance [1]. Their actual recurrence rates are estimated in the order of 1-5% per year [2,3], and can be as high as 11% at 5 years and 20% at 10 years after completion of adjuvant chemotherapy [4,5]. Early detection decreases mortality for women with breast cancer [6][7][8]. In women with PHBC, the survival bene t is improved if new or recurrent breast cancer is found on surveillance mammography instead of physical examination [9]. However, mammography has been shown to be less sensitive in women with PHBC, with sensitivity of 65.4% compared with 76.5% in women with no PHBC [10]. Breast MRI is the most sensitive test for detecting breast cancer [11]. Breast MRI is currently recommended for women with personal history of breast cancer and dense tissue or those diagnosed by age 50, as per American College of Radiology (ACR) guidelines [12]. High MRI costs associated with a lack of sites that offer high-level breast MRI limit clinical access to screening MRIs [4]. Compliance with MRI screening has been shown to be low, in the order of 25% [13]. Abbreviated breast MRI (A-MRI), which substantially reduces the image acquisition and reading time, has been reported to have similar diagnostic accuracy as a full diagnostic protocol [4,[14][15][16][17][18]. Currently, A-MRI has not been adopted as the standard for screening for breast cancer and more studies are required to evaluate outcomes.
Prior studies demonstrated that supplementary MRI surveillance in women at high risk of breast cancer does not impact anxiety, cancerspeci c distress or health-related quality of life [1,19]. This is the rst study to our knowledge to evaluate the psychological effect of adding abbreviated MRI to mammography surveillance in women with PHBC.
The primary purpose of the study was to determine if A-MRI in addition to mammography impacted patient anxiety and, secondarily, if it improved cancer detection rate in breast cancer survivors.

Study subjects
This prospective randomized controlled trial of parallel design was performed at a large tertiary care academic medical center and was approved by the hospital's institutional review board (https://clinicaltrials.gov/ct2/show/NCT02244593). The patients' oncologists or surgeons obtained written informed consent. Recruitment began 2/1/2015 and was completed on 4/30/2019. Patients were followed for a minimum of 12 months.
The eligibility criteria included: (a) female patients 18 years or older; (b) personal history of breast carcinoma (including DCIS and invasive ductal or lobular carcinoma); (c) prior unilateral mastectomy or breast conservation surgery; (d) treatment for breast cancer completed; and (e) no symptoms of breast cancer. Patients were excluded if they were considered high-risk (lifetime risk ≥ 25%) [20], were unable to undergo an MRI due to either physical or mental issues (i.e.: severe claustrophobia, allergy to gadolinium, severe renal failure), had bilateral mastectomies, were pregnant or breastfeeding, or had undergone a breast MRI within the last 6 months. Regular surveillance imaging consisted of annual surveillance mammography, irrespective of breast tissue density. All patients had undergone prior mammographic imaging, and some (< 50%) had undergone prior breast MRI imaging.
Eligible patients were randomized in a 1:1 allocation ratio to one of the two arms of the study: 1) surveillance with MG or 2) MG plus A-MRI, with use of permuted blocks of variable length (2, 4, and 6) to ensure that recruiting physicians remained unaware of the randomization. Researchers or study participants were not blinded to their allocation.

Imaging Technique and Interpretation
All mammographic examinations were performed using a full-eld digital technique (Hologic, Bedford, MA, USA) in accordance with national guidelines. Standard two-dimensional craniocaudal (CC) and mediolateral oblique (MLO) views were obtained.
Surveillance MG and A-MRI were reviewed by one of two breast radiologists independently (8 and 20 years of experience) using ACR Breast Imaging-Reporting Data System (BI-RADS) lexicon [21]. For patients in the A-MRI group, mammography and A-MRI studies were usually performed on the same day. Radiologists were not blinded but reported each modality separately according to the imaging modality ndings, with the mammograms interpreted rst. Based on the imaging ndings, additional mammographic images, including diagnostic tomosynthesis, or targeted ultrasound were requested at the discretion of the interpreting radiologist. Findings and management were communicated to the patient by telephone by the reporting radiologist. Subsequent imaging was performed on separate visits, within 3 weeks of the MG or MR. Histologic samples for pathologic diagnosis were obtained under ultrasound (14G, 5-6 cores), stereotactic (10G, 6-12 cores) or MRI (10G, 6-12 cores) guidance.

Anxiety Measures
Patients in both groups were asked to ll out four validated self-report questionnaires that measure anxiety level and overall health (22)(23)(24). Penn State Worry Questionnaire (PSWQ)(22) is a 16-item self-report questionnaire which measures frequency and intensity of worry symptoms. Items are rated on a 5-point scale, with total scores ranging from 16-80. A score between 16-39 indicates low worry, 40-59 moderate worry and 60-80 high worry. The Breast Cancer Worry Scale (BCWS)(23), is a 3-item scale which measures frequency of breast cancer worry and the impact of worrying on mood and ability to perform daily activities. Higher scores indicate greater cancer worry. The State-Trait Anxiety Inventory (STAI) (24), is a 40-item scale that includes 20 items that assess state anxiety (S-Anxiety) (i.e., how the person feels at this moment) and 20-items that assess trait anxiety (T-Anxiety) (i.e., how the personal generally feels). Items are rated on a 1 to 4 scale. The range of scores for each subscale is 20-80, with cut-off scores of ≥ 32.2 and ≥ 31.8 indicating elevated levels of state and trait anxiety, respectively. Both STAI subscales have solid psychometric properties and are sensitive to assessment of longitudinal change. The Health Status Questionnaire 12 (HSQ-12) assesses the impact of health on social, emotional and physical functioning over the past four weeks. Depending on the item, questions are rated of a 3-point, 5-point and 6-point scale, with higher scores indicating poorer health status. The questionnaires were completed upon enrolment during consultation (T0), upon receipt of the MG and/or MRI results (T1), and then 6 months later (T3). T3 questionnaires were mailed to patients and returned to the study coordination center.

Data Collection and Statistical Analysis
Medical records were reviewed to determine patient age, family history of breast and/or ovarian cancer in a rst-degree relative, surgery modality, initial breast tumor stage (TNM), histology, hormone receptor status, months since diagnosis of breast cancer and breast density. Results were compared between the two groups. For malignant or atypical/high-risk lesions, surgical pathologic results were reviewed when available. Imaging follow-up for all patients with benign imaging or pathology was documented with the date of the most recent negative mammogram.
The anxiety measures were analyzed using SPSS Statistics version 25. Analysis was based on intent-to-treat (ITT) principles. Data were analyzed using linear mixed models, with Intervention (Mammogram only versus Mammogram plus fast MRI), time of assessment (T1, T2, T3), and Intervention by Time interaction as xed factors. Models were estimated using Restricted Maximum Likelihood (REML) with an unstructured covariance structure to account for correlations among repeated measures over time. A signi cant Time by Intervention interaction would suggest that changes in measures over time were different between the interventions; signi cant interactions were further analyzed with pairwise least square mean comparisons. Data from missing questionnaires were not imputed because our analytical strategy using REML allowed the estimation of reliable parameters without the need for imputation of the data under an assumption of missing at random (MAR) [22]. Descriptive statistics were calculated using a spreadsheet software program (Excel, Version 2013, Microsoft). Screening outcomes were compared between groups using Fisher's exact test. Sample size calculation was based on primary outcome the STAI. There is no generally accepted minimal clinically important difference for the STAI, and a 4-point difference was selected to be a minimal clinically important difference, as it would correspond to a complete difference in one of the 40 items. In order to have 80% power to detect a 4-point difference between the groups, we planned 134 patients per group. Recruitment stopped early due to differences in cancer detection rates. Results were considered signi cant if p < 0.05.
Imaging modalities (MG, A-MRI), and BI-RADS nal assessment categories for each modality were noted. Imaging ndings and outcomes were documented for all BI-RADS 3, 4 and 5 lesions, including suspicious extra-mammary ndings. Results were compared between MG and A-MRI. A screening examination was considered as positive when additional diagnostic imaging was recommended prior to the next routine screening examination and included BI-RADS 0, 3, 4 and 5. True positive ndings were de ned as a cancer diagnosis within 12 months of a positive screening examination. Imaging studies were considered false negatives if there was a tissue diagnosis of cancer within 12 months of a negative study. The following performance metrics were calculated for each modality: CDR, AIR, biopsy rate, positive predictive value for biopsy recommendations (PPV2), positive predictive value for biopsies performed (PPV3), sensitivity and speci city.

Results
A total of 202 of 1000 patients ful lled the eligibility criteria ( Fig. 1). At enrollment, 94 were randomized to group 1 and 108 to group 2. Of these, four patients from group 2 were excluded because they withdrew from the study before undergoing MRI for different reasons: two patients developed breast cancer metastases, one patient developed sepsis and her doctor decided to postpone contrast injection and one patient opted to withdraw from the study. Accordingly, the study population consisted of 198 patients: 47.5% (94/198) randomized to regular surveillance with MG (group 1) and 52.5% (104/198) to surveillance with MG and A-MRI (group 2). Among the 104 patients from group 2, 82.7% (86/104) patients had both imaging exams the same day and 17.3% (18/104) patients on different days (average 33.2 days (range: 1-147)).

Patients' demographics
Patients' demographics are presented in Table 1. Mean age was 59 years (range 35-80) for group 1 and 58.2 years (range 38-83) for group 2. No clinically important differences in age, family history of breast and/or ovarian cancer, surgery modality, months since diagnosis, breast density, initial tumor histology, stage, or hormone receptor status were noted between the two groups, although a non signi cant higher number of patient with triple negative cancers was observed in the MRI group.

Results regarding anxiety
Mean scores and standard deviations for self-report measures are displayed in Table 2. The intervention groups did not differ signi cantly on any of the baseline self-report measures. Linear mixed models revealed no signi cant Time main effects or Time x Intervention interactions for the worry measures PSWQ (p = .14 and p = .57, respectively) and BCWQ (p = .73 and p = .48, respectively). Analysis of the STAI revealed that the Time main effect and Time x Intervention interaction were not signi cant for trait anxiety (p = .51 and p = .20, respectively). In contrast, there was a signi cant time main effect (p < .001) and Time x Intervention interaction (p = .022) for state anxiety.
Post hoc tests revealed that for both groups, state anxiety scores decreased from Time 1 to Time 2 (ps < .001) but increased from Time 2 to Time 3 (ps < .001). Between group differences were found at T2, with participants in the fast MRI condition reporting lower levels of state anxiety than those in the mammogram condition (estimated mean difference = 2.6 [95% CI, . 13-4.19], p = .037), but was less than a 4 point difference. There was a signi cant Time main effect for self-report health status (p = .008), but the Time x Intervention interaction was not statistically signi cant (p = .10).

Findings according to imaging modality
Outcomes according to imaging modality are presented in Table 3. Among the 302 imaging examinations performed (198 mammograms and 104 A-MRI), 9 mammograms and 29 A-MRI were interpreted as abnormal (17%) (Fig. 2). Four patients had abnormal mammography and A-MRI ndings, three for the same abnormality.

Mammography
There were 198 mammographic examinations performed: 94 for group 1 and 104 for group 2; 95.5% (189/198) were negative or benign (BI-RADS 1 and 2), 4.5% (9/198) were recalled (BI-RADS 0) and 3.0% (6/198) presented ndings suspicious for malignancy (BI- RADS 4) and underwent biopsy. One cancer was detected (Table 4) and no high-risk lesions were identi ed.  (Table 4), all only detected with A-MRI, 2 of which were in patients with original triple negative breast cancer (Fig. 2). Three patients had suspicious extra mammary ndings: one lung mass seen in the right middle lobe on the A-MRI, and two bone lesions seen in the sternum and manubrium, respectively. The lung mass was con rmed to be a metastatic carcinoma from breast primary on CT guided transthoracic lung biopsy, the manubrial lesion was con rmed to be a hemangioma on bone scan and the sternal lesion was con rmed to be a hibernoma on CT guided biopsy. Of the MRI detected breast cancers, none was identi ed on mammography, even in retrospect. No high-risk lesions were detected.

Necessity for full diagnostic or repeat MRI
Three patients required further investigation requiring diagnostic full MRI based on the radiologist's uncertainty of the ndings seen on the A-MRI: 1 had a mass which was benign on assessment (fat necrosis) determined by the full protocol, one was a BI-RADS 3 lesion which showed stability on 12 month follow-up and one had BI-RADS 4B lesion that led to a benign MRI-guided biopsy of Pseudoangiomatous hyperplasia. A fourth patient had motion artifact and required a repeat abbreviated MRI that was normal and of high technical quality.

Follow-up
All 191 patients with benign imaging or pathology results underwent clinical and imaging follow-up at the same center, for an average 24 months (10-56 months). There were no cancers found retrospectively as false negatives on follow-up. 1.57% (3/191) had breast cancer on follow-up, all from group 1. Of the 3 cancers diagnosed on subsequent surveillance imaging, two were diagnosed at 26 & 27 months with MG (DCIS and invasive ductal carcinoma (IDC), T1N0M0) and one was diagnosed at 50 months on MRI IDC, T2N0M0). Two were new cancers in the contralateral breast and one DCIS was in the ipsilateral breast; no cancer was seen retrospectively on initial MG and/or MRI. Of the remaining 188 patients with no cancer diagnosed on follow-up, 98.9 % (186/188) had follow up of 12 months or longer and 1.06 % (2/188) had follow-up of shorter than 12 months. There were no patients lost to follow-up.

Discussion
This prospective randomized controlled trial showed that abbreviated breast MRI was superior to mammography in the detection of cancer in 198 breast cancer survivors; almost 10 times more cancers were detected with breast MRI than MG (48 (5/104) vs 5/1000 (1/198), p = 0.01). Despite higher rates of biopsies and abnormal interpretations, breast MRI was not associated with an increase in anxiety, with average anxiety scores of STAI of 50.99+/-4.6 with MG vs 51.73+/-2.56 with MRI, p > 0.05, 6 months after the study. Anxiety was moderately high in all patients and did not change, whether patients underwent surveillance with MG alone or MG plus A-MRI.
Although a reassuring effect from undergoing A-MRI was not observed, there was no detrimental effect.
Our results support other studies on the impact of breast MRI on anxiety. The Dutch MRI screening (MRISC) study of patients at high risk for breast cancer found that the addition of breast MRI did not affect quality of life or anxiety [19]. In a more recent prospective nonrandomized multicentre study, 1561 women at intermediate and high breast cancer risk were noted to have similar moderate distress levels, and there were no more harmful psychological effects observed between standard MG plus ultrasound as compared with the addition of MRI to standard imaging [23].
A signi cantly higher cancer detection rate was noted in the patients who underwent A-MRI as compared with mammography, despite similar demographics. Our study demonstrated A-MRI had a sensitivity of 100% and CDR 48/1000 as compared to mammography's sensitivity of 14.2% and CDR 5/1000. The abnormal interpretation and biopsy rates were signi cantly higher for A-MRI than mammography, 25% and 18.3% for A-MRI and 4.5% and 3% for MG, respectively. PPV3 was higher with A-MRI than MG, 26.3% vs 16.7%, although this difference did not reach statistical signi cance. When extra-mammary ndings were included, A-MRI offered the bene t of detecting an incidental lung metastasis.
We recognize some limitations of our study. Patients were recruited by their oncologists or treating surgeons, which could have introduced a bias in patient selection. This may have partly explained the high cancer detection rate in the MRI group. Nonetheless, the fact that randomization was blinded mitigated any potential bias of intervention arm selection and there were no clinical differences between the MRI and mammography groups. Because of the high cancer detection rates and minimal effect on anxiety in the MRI group we stopped the clinical trial early. Additionally, some patients could have developed breast cancer after the follow up period, which might have been missed with mammography. Given that the majority patients were followed for over 24 months, this is less likely. Another limitation is that the radiologists were not blinded to the allocation arm, which could have in uenced their reporting of the mammogram, if they knew that an MRI would be done. However, given similar recall rates for mammography within both groups, this is unlikely to have been present. The high biopsy rate in the MRI group may be perceived as a limitation, but this was related to the high CDR with an acceptable PPV3.
However, more research is required to nd ways to further reduce the rate of false positives. There is likely a learning curve with A-MRI and the addition of T2 sequences may help to improve PPV3 without signi cant time cost [15]. We have subsequently adapted an abbreviated protocol to include T2 sequence and two more post contrast sequences to improve the speci city of MRI [28]. Another limitation is that assessment of anxiety was based on self-report questionnaires and limited by the time points in which it is measured. A more objective measure would be to evaluate adherence to follow-up rounds of screening, which may address poor compliance with MRI screening [13]. This is recommended for future study. Also, our study lacked the sample size and enough long-term follow-up to be able to say whether the earlier detection in the A-MRI group led to any difference in survival.

Conclusions
The addition of abbreviated breast MRI to surveillance mammography did not impact patient anxiety in breast cancer survivors, regardless of the signi cantly higher recall and biopsy rates. MRI showed signi cantly higher cancer detection rate compared to mammography alone, which is consistent with recent recommendations. Although further study with larger cohorts is warranted, an abbreviated protocol may be considered for surveillance in this population.       within the centre of the enhancing mass (circle), diagnosing pleomorphic lobular carcinoma in situ, subsequently con rmed at surgical excision to be invasive lobular carcinoma.