Evaluation of Adjunctive Ultrasonography for Breast Cancer Detection Among Women Aged 40-49 Years With Varying Breast Density Undergoing Screening Mammography

Key Points Question Does the performance of adjunctive ultrasonography for breast cancer detection among women undergoing screening mammography change according to breast tissue density? Findings In this secondary analysis of a randomized clinical trial, screening mammography alone demonstrated low sensitivity, whereas adjunctive ultrasonography improved sensitivity both in dense and nondense breasts. Meaning These findings suggest that adjunctive ultrasonography has the potential to improve the detection of early-stage and invasive cancers across both dense and nondense breasts and to mitigate the flaws of mammographic screening.


Overview 74
Background and outline 75 Breast cancer is the most frequent type of cancer for females. The death rate from 76 breast cancer is higher than that from any other cancer for the age group 30-69. Among 77 others, the age group 40-49 has recently been showing a marked increase in the 78 incidence of breast cancer. Therefore, there is an urgent need to countermeasures for 79 reducing the death rate from breast cancer among women in this age group. 80 Mammography, which is a global standard of breast cancer screening, is low in 81 accuracy when used for breasts with high mammary gland density. High density breasts 82 are often seen at ages less than 50 and more frequent among Japanese women than 83 among Western women. Ultrasonography has high accuracy when used for checking 84 breasts with high mammary gland density. Attempts to use ultrasonography as a means 85 of breast cancer screening have already begun to be made. However, neither the 86 specifications of devices for breast ultrasonography nor its procedure and image reading 87 techniques have yet been standardized. Furthermore, the accuracy and usefulness of 88 ultrasonography as a means of breast cancer screening have not been endorsed. 89 This study is designed to compare the effect in reducing the death rate from breast 90 cancer between two groups of women, i.e., between women screened with standardized 91 ultrasonography + mammography and women screened with mammography without 92 ultrasonography.  Since 1994, the incidence of breast cancer has been higher than that of any other 135 female cancer. Among other, the incidence of this cancer in females aged between 40 136 and 49 is expected to further rise from now on following changes in the lifestyle of 137 Japanese people [1,2,3]. Premenopausal breast cancer, developing at age 40- 49, has 138 been sharply increasing in recent years, and this cancer is related to estrogen, one of the 139 female hormones. It is said that women with lower age of menarche, no history of 140 delivery and delay in menopause are more likely to develop breast cancer because of 141 prolonged exposure to estrogen [3]. However, primary prophylaxis, designed to remove 142 these factors, is not realistic. Under the current trend for less child delivery by 143 individual women, a valid means of preventing death from breast cancer is secondary 144 prophylaxis designed to facilitate early detection of breast cancer, i.e., development, 145 spread, utilization and improvement of a valid method of breast cancer screening. 146 Breast cancer screening with mammography has been shown in multiple 147 randomized comparative studies in Western countries to be useful in reducing the death 148 rate from breast cancer among women aged over 50, and has been adopted as a global 149 standard. However, the usefulness of this way of screening in reducing the death from 150 breast cancer is lower for women aged 40-49 than for women aged over 50 [4,5,6]. 151 Furthermore, mammography is less accurate for screening of breasts with high 152 mammary gland density [1]. High dense breasts are more often seen in women younger 153 than 50 and are more frequent in Japanese women than in Western women [8]. 154 Furthermore, since the peak age for onset of breast cancer is 40-49 in Japanese women 155 in contrast to the peak age (70-79) in Western women, screening of Japanese women 156 using mammography alone is not expected to be satisfactorily useful [1,2]. 157 Ultrasonography has high accuracy when used for checking breasts with high 158 mammary gland density. Attempts to use ultrasonography as a means of breast cancer 159 screening have already begun to be made. It has been reported that screening with a 160 combination of mammography and ultrasonography has elevated the breast cancer 161 detection rate [9,10,11]. However, neither the specifications of devices for breast 162 ultrasonography nor its procedure and image reading techniques have yet been 163 standardized. Furthermore, the usefulness of ultrasonography as a means of breast 164 cancer screening has not been endorsed. 165 Nowadays, in both Japan and foreign countries, a prerequisite for performing 166 "cancer screening as a means of countermeasure" under public initiatives is availability   The primary endpoint of this study is inter-group comparison of the sensitivity, 217 specificity and detection rate. The secondary endpoint is inter-group comparison of the 218 cumulative incidence of advanced* breast cancer during the follow-up period. The most important indicator of the usefulness of cancer screening is the death 224 rate from a given type of cancer in the population studied. However, in view of the 225 natural history of breast cancer, the four-year period of strategic research is too short to 226 explore a significant inter-group difference in this indicator. Therefore, a system will be 227 established to allow long-term follow-up of the survival/death of the subjects of this 228 study in both groups after completion of this strategic research.

Study design 231
This study is designed to compare the accuracy and usefulness of breast cancer 232 screening of women aged 40-49 between two groups, i.e., the intervention group 233 screened with a combination of mammography and ultrasonography and the 234 non-intervention group screened with mammography. To this end, the below shown two 235 groups are incorporated into this study, with individuals or clusters serving as units.

236
(1) Ultrasonography combined group (mammography + ultrasonography or 237 mammography + clinical breast examination + ultrasonography) 238 (2) Ultrasonography uncombined group (mammography alone, or mammography 239 + clinical breast examination) 240 The targeted number of subjects is 60,000 for each group (120,000 subjects in 241 total). Of these subjects, 50,000 subjects from each group will receive two sessions of 242 breast cancer screening at an interval of 2 years.

243
The targeted number of subjects was set on the basis of the forecast as to the 244 capability of elevating the sensitivity (the primary endpoint) by the use of 245 ultrasonography in combination with conventional mammography.

246
In analysis of the intermediate period cancer (cancer detected before the next 247 session of screening after the latest session of screening yielding a judgment of 248 "abnormality-free"), using the data from the Miyagi Prefecture Regional Cancer that screening is continued for 10 years at intervals of 2 years, the cumulative death rate 295 from breast cancer will be 170/100,000 population and 110/100,000 population for the 296 two groups, respectively. To detect this difference as a statistically significant difference 297 at a significance level of 5% (both-sided) and with a detective power of 80%, 55,000 298 subjects are at least needed for each group. Therefore, if screening can be continued for 299 10 years, a statistically valid analysis of a decrease in death rate from breast cancer 300 following the addition of ultrasonography will be possible by enrollment of at least 301 60,000 subjects to each group.

302
Study designs available for obtaining scientific evidence include individual 303 randomized study, cluster randomized study, cluster non-randomized comparison, etc.

304
From the viewpoint of conducing a high-quality study, validity seems to be highest for 305 the individual randomized study design by which individual subjects, having given 306 informed consent, are assigned at random to one of the ultrasonography-combined 307 group and the ultrasonography-uncombined group. A study design with the second 308 highest validity will be cluster randomized study design by which assignment to the 309 ultrasonography-combined group or the ultrasonography-uncombined group is made at 310 random in units of local community (town or village) or workshop (place of work). If 311 randomized assignment is difficult, cluster non-randomized comparison (a study design 312 by which comparison with the non-intervention group is made, although assignment is 313 not made) has to be selected.

314
Another possible study design will be an independent study in which only one 315 type of screening is made (either ultrasonography-combined screening or 316 ultrasonography-uncombined screening), but this study cannot be viewed as a 317 comparative study. For the present study, it is desirable to adopt a study design capable 318 of providing scientific evidence as far as possible. For this reason, participation in this 319 study will be accepted with emphasis laid on the organizations or facilities which intend 320 to perform the study by means of individual randomized comparison, cluster 321 randomized comparison or non-randomized comparison (greatest emphasis laid on the 322 first study design and progressively less emphasis on the next two designs).      Each participating organization/facility must satisfy the following requirements:

366
(7) Preferably being linked to a highly precise regional cancer registration program.

408
(3) Judgment is made by a physician experienced with breast ultrasonography.

409
(4) Categorization for ultrasonography is used when a judgment is made.

555
(2) Survey of inhabitant register: A survey is conducted as to dislocation (from a given 556 local community), death and date of death of individual subjects.

557
(3) Use of population statistic data for purposes not initially planned: As to the subjects 558 confirmed to be dead by a survey of inhabitant register, an application is filed with the 559 Ministry of Health, Labour and Welfare (MHLW) and the Ministry of Internal Affairs 560 and Communications to obtain approval of the use of the population statistic data for 561 purposes other than the initially planned purposes. If approved, the death slip or tape is 562 accessed to identify the cause of death.

563
(4) Comparison with the regional cancer register data: An application is filed with the 564 regional cancer registry organization located in the same district as the screening facility  Follow-up is made using two methods, i.e., direct contact (interview and mail) 570 with the subject (or her relative) and comparison with external data.

571
When communicating directly with subjects (or persons involved), at the point at which 572 20 months have passed since they underwent the first breast cancer screening, a sealed letter 573 shall be sent to the subjects to encourage them to undergo a second breast cancer screening. 574 During the second screening, the presence/absence of breast cancer diagnosed after the 575 first screening is checked. For subjects who do not receive the second screening, the 576 presence/absence of breast cancer diagnosed after the first screening is investigated over 577 telephone.

578
Comparison with external data means checking for move-outs or death of subjects 579 by means of access to the inhabitant register, access to the death slip to confirm death 580 (and its cause) and access to regional cancer register to check for beast cancer among 581 the subjects. The step to obtain approval of each comparison to external data is taken by 582 the Central Data Center, and comparison is practiced after approval is obtained (refer to 583 "8.5. Follow-up" for details).   issuing record deletion certificates, use of Vital Statistics, and cross-checking with local 646 cancer registry data, view these data, and carry out cross-checking.

647
In view of the natural history of breast cancer, the four-year period of strategic 648 research is too short to explore a significant inter-group difference in this indicator. 649 Therefore, a system will be established to allow long-term follow-up of the   Explanation about the study to obtain informed consent is given to each subject 671 orally and in writing, as a rule. Informed consent in writing from the subject is 672 indispensable.

673
The following are major pieces of information to be supplied to the subject (refer 674 to Appendix 1, 2, 3, 4 and 5).

675
(1) Objectives, significance and outline of the study;

676
(2) Item and method of investigation: Screening results and detailed tests results  If the subject has sustained health hazards directly associated with this study, no 707 special compensation will be made. The person in charge will deal with such hazards 708 within the framework of healthcare under health insurance.    data for purposes other than initially planned purposes and comparison to regional 803 cancer registry data (submission of application for approval, access to data, and 804 comparison of these data with the data from screening subjects); 805 (13) Conducts statistical analysis of data from this study using the methods set forth in 806 "11. Statistical analyses" and reports the results to the Research Leader. The Research Leader and the co-researchers may plan additional compound 824 studies if deemed necessary. After securing coherence with the present study, a plan 825 for each additional study is prepared.

826
The primary analysis in the original report revealed that the sensitivity was 827 significantly higher in the intervention group than in the control group. However, results 828 according to breast density were not described in detail. Study addressing issues related 829 to performance of each modality according to differences in breast density was   subjects from both groups as to survival/death after the end of the study will be 843 established, and the data collected from such follow-up will be analyzed and reported.

844
Although the results about standardization of ultrasonic breast cancer screening 845 procedure are not major outcomes from this study, they will be useful in improving the 846 accuracy of breast cancer screening in Japan. For this reason, co-researchers will be able 847 to publish these secondary outcomes and the processes of such an aspect of this study in  The authorship of each academic paper related to this study will be decided     (2) In principle, mediolateral oblique imaging and craniocaudal imaging shall be performed for both breasts.
[The minor changes in the Japanese, but After securing coherence with the present study, a plan for each additional study is prepared.

71
This is a comparative study to evaluate the diagnostic accuracy and efficacy of a breast cancer screening method 72 in women aged 40 to 49 years between a group using mammography in combination with ultrasonography 73 (intervention group) and group using mammography alone (non-intervention group). Hence, the following two 74 groups will be established individually or using a cluster as a unit: 75 (1) Group using ultrasonography in combination: (mammography + ultrasonography, or mammography + 76 clinical breast examination (CBE) + ultrasonography) 77 (2) Group without ultrasonography: (mammography alone, or mammography + CBE) 78

Objectives 79
To evaluate the diagnostic accuracy and efficacy of breast cancer screening using ultrasonography. 80

Study design 81
The following two groups will be set to evaluate the efficacy of breast cancer screening using ultrasonography in 82 women aged 40 to 49 years: 83 1) Group using ultrasonography in combination (intervention group) 84 (mammography + ultrasonography, or mammography + CBE + ultrasonography) 85 2) Group without ultrasonography (non-intervention group) 86 (mammography alone, or mammography + CBE). 87 For scientific evidence, designs such as an individual randomized control study, cluster randomized control study, 88 and non-randomized control study can be assumed. From the viewpoint of performing a high-quality study, an 89 individual randomized control study is the most appropriate followed by a cluster randomized control study. If 90 random allocation cannot be done, a non-randomized control study has to be chosen, and a method to compare with 91 the non-intervention group without randomization has to be considered. A total of 42 organizations in 23 92 prefectures in Japan will participate in the study. 93  If study expenses can be obtained, the study period may be extended based on the plan. 170

171
In this study, with the use of the central data monitoring system, it will be verified whether the study is conducted 172 safely and in accordance with the protocol, as well as data are correctly collected. After the start of study 173 enrollment, the central data center will, in each month, tabulate the input status of data sent from the Electronic 174 Data Capturing (EDC) shuttle at each site for the quality control (QC) of the data. The central data center will 175 perform periodic logical checks and request the study sites to report cross-checking with source documents with 176 respect to error data. Entry and transcription mistakes identified through cross-checking with source documents will be corrected by the persons in charge at the study sites. In addition, "data monitoring reports" based on the data 178 collection/ accumulation status and the results of data monitoring will be prepared twice a year and notified to the 179 data monitoring committee. The quality assurance (QA) of the data will be implemented by on-site audits. 180

Definitions and Data Handling Rules for Analyses
Instruction at the first screening * Those in the green frame will be analyzed for the first screening. At the time of analysis, sensitivity and specificity cut off at this line will also be examined.
(  Subjects who do not fall into either of the above Items 1) Had Breast Cancer or 2) Did not have Breast Cancer 230 will be defined as Had unknown status. Subjects whose medical examination data in a report of breast cancer are 231 not grasped will be classified into unknown. 232

Data review in a blinded manner (Case ascertainment and validation) 233
Subjects with a diagnosis of breast cancer in this study will be defined as those with breast cancer recorded in the 234 column for "Diagnosis" in the report of detailed examination results at the examined medical institution and images 235 and pathological data supporting the diagnosis of breast cancer can be understood. After the input function of the 236 EDC installed at each participating site is locked, the central data center should request each study site to submit 237 copies of source documents on subjects with a diagnosis of breast cancer. In addition, the EDC screen capture of 238 the database and documents, in which the randomized groups are blinded, for review should be also sent to the 239 It is a given fact that the classifications ( (1) Table 1. Blinded review and data check in subjects with a diagnosis of breast cancer for classifying them into detected and interval cancers (source documents and EDC data) Step 1 Use SAS to perform a logical check of essential parameters in EDC data.
Step 2 Blind the randomized groups recorded in case result reports. Review the case by checking screening, detailed examination and the chronology of medical examination/ logics of the course. For 1) subjects in whom breast cancer determined once is denied based on the results of a medical examination at a cancer base hospital, and 2) subjects who are found to be not having breast cancer because of mistakes with transcription in the CRF and input in the column for diagnosis, clean the data based on the facts, and specify the absence of breast cancer. Retain not only the SAS program but also correspondence tables and source documents. * With regard to missing data on TNM Classification and histological classification, make inquiries again to the participating site and medical institution.
Step 3 Cross-check the results of a review assessment and SAS program calculations.
Step 4 If the results in Step 3 are inconsistent, search for the reasons, and handle the matter through processes from an examination, discussion to agreement.
Step 5 Determine the classification to be used for the analysis through the abovementioned examination processes.
Comments For classification between screen-detected cancer and interval cancer, both screening data at the study site and data on a history of screening recorded in the questionnaire from by the study participants will be used so that the screening and questionnaire data have equal values.
However, for some subjects with a complicated data structure, for example, 1) when there are both biennial screening specified in the study and annual screening, 2) the results of several follow-ups are entered as detailed examination data, or 3) subjects who report that they have undergone several screening sessions and/or medical examinations not specified in the study in the questionnaire form, accuracy cannot be sufficiently secured only by classification based on data logical checks using analysis software, SAS. Hence, overall verification activities such as cross-checking between source documents and inputted EDC data will be essential so that the source documents of subjects with a diagnosis of breast cancer should be reviewed.

Pooling data for analyses 281
First, the appropriateness of a pooled analysis in subjects who are randomized individually or by cluster will be 282 examined. 283 In this analysis, study participants assigned by individual randomization and cluster randomization will be 284 analyzed. Cluster randomization in the study will be as follows: Communities or healthcare providers are a 285 randomization unit; explanation and consent are given individually; interventional effect is on an individual basis; 286 and subjects with a diagnosis of breast cancer are those with objective data from a medical examination. 287 For the estimation of the primary endpoints, generalized estimating equation (GEE) with a compound symmetry 288 structure as a working correlation matrix will be used. Analyses including only subjects who are randomized 289 individually or by cluster will be performed, and it will be reviewed that the point estimates of the primary 290 endpoints are similar. Furthermore, a model including the randomized groups as well as an interaction between the 291 randomization methods (either individual or cluster) and randomized groups as an explanatory variable will be 292 applied to subjects pooled from the two analyses, and it will be evaluated that the interaction effect is not 293 statistically significant. If there is no interaction, a pooled analysis including subjects who are randomized 294 individually and by cluster will be carried out. 295

Analysis set 296
Efficacy analyses will be performed in the intention-to-treat (ITT) set. The ITT set will be a set of participants 297 enrolled in the study excluding (1) non-randomized enrolled subjects, (2) ineligible subjects, (3) subjects withdrawn 298 consent, (4) double-entry subjects, and (5) subjects without data that should be analyzed for demonstrating efficacy 299 (subjects without first screening data). 300 After all input, correction and sending functions of EDC for study data at the study site are locked, logical check 301 and inquiries to applicable study sites will be made based on data received by the EDC system at the central data 302 center. The final review of the concerned subjects will be carried out based on confirmation, responses and reports 303 from the study sites and source documents. 304

Preparation of SAS data set for analysis (Data Source) 305
When preparing an analysis data set, the below-mentioned verification should be conducted. 306

Verification of compliance with randomization 307
After the EDC input function at the site is locked, the key will be opened. System numbers issued by the central 308 randomization and the randomized groups, and system numbers inputted in the EDC at the site and the randomized 309 groups will be cross-checked using data received by the EDC system at the central data center, and inquiries on the 310 following items should be made to each study site: 311 (1) Verification of subjects incompliant with randomization who underwent screening different from that 312 centrally randomized. 313 (2) Ve rification of a system number that a number issued by central randomization is not entered in the EDC. 314 (3) Verification of subjects for whom numbers different from those assigned by the central data center are 315 inputted in the EDC. 316 Age at the time of enrollment and time (month) from the previous screening. History of screening, menstruation 351 status, feeding history, family history, history of breast surgery, and history of mammary gland disease. Time 352 (month) from the previous screening, which is a continuous variable, will be categorized. The age of first 353 menstruation, number of pregnancies, frequency of childbearing, age at the first delivery, and number of 354 first-degree relatives with breast cancer will be categorized in accordance with the Gail model for performing an 355 international comparison. 356 For characteristics of baseline demographic variables, descriptive statistics will be calculated to describe the data. 357 Continuous variables will be summarized with a number, mean, standard deviation, range and mode. Categorical 358 data will be presented as percentages. For methods of comparisons of the distributions of baseline factors, 359 differences in categorical data will be assessed by Fisher's exact test and the Chi-square test, and differences in 360 continuous data by the Mann-Whitney's U-test. All tests will be two-sided. The 95% CIs will be calculated using the binomial distribution method. 365 At the statistical analysis/ data monitoring committee meeting held on October 28, 2013, consensus was reached 366 that subjects having an unknown status would be excluded from the analyses of sensitivity and specificity. 367 7.2.2 Method of the primary analysis taking account of correlations within the cluster 368 All "screen-detected cancer," "interval cancer" and "subjects without cancer" at the first screening to be used for 369 calculating sensitivity/ specificity, the primary endpoints, are binary outcomes. As an analysis method for 370 estimating the primary endpoints in the two groups, GEE taking account of correlations within the cluster will be 371 used. For estimation of the parameters, robust variance estimator with a compound symmetry structure as a 372 working correlation matrix will be employed. 373 Proportion (percentage) To be rounded to the nearest tenth and displayed up to one digit. 95% CI for the rate (percentage)

Detection
To be displayed as ‰

Descriptive statistics
No. of subjects To be displayed as integers.
Mean, standard deviation, median Significant digits will be rounded to the nearest tenth and displayed up to one digit.

Minimum
The same number of figures as significant digits will be displayed.

Maximum
The same number of figures as significant digits will be displayed.

Interquartile range
The same number of figures as significant digits will be displayed.

Test/ estimation P value
Rules specified by a journal, to which results will be submitted, will be followed. The value will be rounded off to the nearest thousandth and displayed up to three decimal places. However, when it is less than 0.001, it will be displayed as < 0.001.

Significance level
A two-sided test will be performed with a significance level of 5%.
If other values are used for significance levels, they will be mentioned at each time. The confidence coefficient of interval estimation will be two-side 95%.
CI Significant digits will be rounded to the nearest tenth and displayed up to one digit.

378
For the statistical analyses in this study, under the supervision of both statistical analysis/ data monitoring 379 committees, data sets and SAS programs to be applied to the statistical analyses will be prepared, and analysis 380 operations will be carried out. If data handling is not specified in the analysis plan or other relevant documents, a 381 request for decision and consultation should be, as necessary, made to the statistical analysis/ data monitoring 382 committees, and data should be processed upon consensus. 383 The validity of the analysis SAS program should be ensured by review, preparation, execution, validation and 384 verification of an SAS program for each analysis using double programming. Practical analysis operations will be 385 performed by two staffs. One has experience with SAS for 14 years (Tohoku University, Ying-Fang Zheng M.D., 386 Ph.D., CCRP ® (Certification of Clinical Research Professionals by SoCRA), and the other has experience with SAS 387 for 21 years (J-CRSU, Akihide Inoue, SAS Certified with SAS Base Programmer for SAS ® 9).

Section
Original Text Revised Text Reason for change To make text clearer. To make text clearer.