Radiology Advisory Panel Meeting Hologic Selenia Dimensions 3D System with C-View Software Module FDA Review Robert Ochs, PhD Branch Chief Mammography, Ultrasound, and Imaging Software Branch Division of Radiological Health Office of In Vitro Diagnostics and Radiological Health Center for Devices and Radiological Health Food and Drug Administration
Outline of FDA Presentations Introduction and Background Clinical Study Design: Weijie Chen, PhD Statistical Results: Jingjing Ye, PhD Clinical Significance: Gary Levine, MD, MSE Panel Discussion 2
FDA Review Team Robert Ochs, PhD Gary Levine, MD, MSE Jingjing Ye, PhD Weijie Chen, PhD Frank Samuelson, PhD Helen Barr, MD Lead Reviewer Division of Radiological Health Medical Officer Division of Radiological Health Statistician Division of Biostatistics Scientist Division of Imaging and Applied Mathematics Scientist Division of Imaging and Applied Mathematics Director Division of Mammography Quality Standards 3
Selenia Dimensions 3D System The device is currently approved for breast cancer screening and diagnosis (PMA-number: P080003) The screening exam can consist of full field digital mammography (FFDM) alone or the combination of FFDM with digital breast tomosynthesis (DBT) 4
Premarket Approval (PMA) Supplement Selenia Dimensions 3D System with C-View Software Module Ability to create synthesized 2D views from the 3D data No other significant hardware or software changes Expanded Indications for Use 3D plus synthesized 2D images for screening 3D S = 3D plus synthesized 2D 5
Proposed Indications for Use The Hologic Selenia Dimensions system generates digital mammographic images that can be used for screening and diagnosis of breast cancer. The Selenia Dimensions (2D or 3D) system is intended for use in the same clinical applications as a 2D mammography system for screening mammograms. Specifically, the Selenia Dimensions system can be used to acquire 2D digital mammograms and 3D mammograms. Each view in a screening examination will consist of: a 2D image set, or a 2D and 3D image set, or a 3D image set in combination with a synthesized 2D image set The Selenia Dimensions system may also be used for additional diagnostic workup of the breast. 6
C-View Software Module The C-View software generates synthesized 2D images from the 3D tomosynthesis data The synthesized 2D views are similar to a maximum intensity projection of the 3D data Images will include a C-View mark to alert users that they are not FFDM images (not shown in the example) RMLO 2D FFDM Synthesized 2D RMLO 7
Configuration The default contrast of the synthesized 2D images can be set to low, medium, or high (window/level settings allow further adjustment by user) Medium contrast was used for the reader study Low Medium High 8
Configuration The appearance of the skin line on the synthesized 2D images can be set as less prominent or more prominent The more prominent skin line was used for the reader study Less Prominent More Prominent 9
FDA Review of PMA Supplement Hologic s PMA supplement was reviewed in the context of the proposed indications for use, and the potential impact on the clinical practice of mammography 10
Panel Discussion The panel will be asked to consider: Potential impact on performing and interpreting mammography when using 3D S and not acquiring FFDM Study bias Study exclusions Safety Effectiveness Benefit and risk 11
Next Speaker The next speaker is Dr. Weijie Chen 12
Radiology Advisory Panel Meeting Hologic Selenia Dimensions 3D System with C-View Software Module Clinical Study Design Weijie Chen, PhD Division of Imaging and Applied Mathematics Office of Science and Engineering Laboratories Center for Devices and Radiological Health Food and Drug Administration
Outline General considerations Controlled multi-reader multi-case (MRMC) study ROC methodology Pivotal study Objective to Panel Case selection Readers Image interpretation Mainly related Questions 2 & 3 14
Radiologist Performance in Breast Cancer Screening Is Selenia Dimensions 3D S (3D digital breast tomosynthesis (DBT) in combination with synthetic 2D images) non-inferior to the standard of care 2D FFDM for breast cancer screening? Controlled multi-reader multi-case (MRMC) study ROC methodology 15
Prospective vs. Controlled Retrospective Clinical Studies Prospective Clinical Study Reader decisions affect patient management Multiple readers usually do not read the same case No Enrichment Reading conditions may be difficult to control Real world, multi-institutional Can lead to absolute scale performance estimates e.g., cancers per 1000 women Retrospective Clinical Study Reader decisions do not affect patient management Multiple readers read the same case Enrichment Esp. when prevalence is low Well-controlled reading conditions Some components in clinical reading may be absent Compare two modalities e.g., 3D S vs. FFDM 16
Controlled Retrospective Studies Do not provide absolute measures of performance Possible differences in case mix, reader training and experience, availability of prior images, patient history, reader mindset Provide meaningful and least-burdensome comparison of imaging modalities Have been used to bring to market several imaging devices Full-field digital mammography (FFDM) Compared to screen-film mammography Digital breast tomosynthesis (DBT) Combined with FFDM compared to FFDM alone Computer-aided detection (CAD) Compared to reader alone 17
Receiver Operating Characteristics Clinical decisions Often binary: e.g. call the patient back vs. do not call the patient back ROC methodology Multiple levels of confidence in disease presence Independent of decision threshold, which varies with user experience and training ROC analysis and area under the ROC curve (AUC) are consistent indicators of diagnostic information given to a physician Can be especially useful when one modality has higher sensitivity but lower specificity than the other 18
AUC vs (Sensitivity, Specificity) Example (presented in the panel meeting for the original Hologic PMA P080003, 09/24/2010): FFDM vs FFDM+DBT (original Hologic PMA) Table of changes in diagnostic measures Study ΔSens ΔSpec ΔAUC 1 st Study -0.08 (-0.15-0.00) +0.38 (0.30-0.46) +0.07 (4-11%) 2 nd Study +0.00 +0.19 +0.07 (4-10%) 3 rd Study +0.05 +0.13 (0.00-0.26) +0.07 19
Study Objective: Rationale ACR phantom dose per view Mode Hologic FFDM Hologic 3D S (DBT+Synthetic 2D) Hologic FFDM+DBT Average dose of FFDM* Average dose of S/F* Average dose of CR* MQSA Limit Dose for Single View 1.2 mgy 1.45 mgy 2.65 mgy 1.43 mgy 1.77 mgy 1.90 mgy 3.0 mgy * Collected from medical physicists dose measurements available at the time of MQSA inspections Jan Sept 2012 20
Study Objective: Primary Compare radiologist performance in detecting breast cancer in terms of the area under the ROC Curve (AUC) 2D FFDM (CC & MLO views) -- FFDM vs. 3D DBT (CC & MLO views) plus synthetic 2D views (CC & MLO views) -- 3D S AUC of 3D S is non-inferior to that of FFDM with a non-inferiority margin of 0.05 21
Study Objectives: Secondary AUC for subjects with dense breasts using 3Ds is non-inferior to that of FFDM (noninferiority margin: 0.05) The non-cancer recall rate for 3D S is noninferior to that of FFDM (non-inferiority margin: 5%) 22
Case Selection: Inclusion/Exclusion Criteria Inclusion Criteria Female No contraindication for routine bilateral mammography Exclusion Criteria Subjects who presented with any contraindications to mammographic screening Previous surgical (excisional) biopsy Previous breast cancer Placement of an internal breast marker Breast implants or breast reduction surgery Subjects who were unable to understand or execute written informed consent Breasts too large to be imaged on a 24 cm by 29 cm detector with a single exposure 23
Case Selection: Image Acquisition 3521 subjects from 22 prequalified clinical centers 2985 of the subjects eligible 19.8% (590/2985) of eligible cases were excluded for image Q/C 300 FFDM 171 DBT 137 positioning (both FFDM and DBT) The 302 cases used for the reader study were randomly selected using stratified sampling 3521 subjects imaged 3285 Subjects 2985 eligible 2299 eligible 302 selected 200 at site 590 Q/C 236 Training 100 incorrect tech tables 96 protocol/pilot 1997 not in study Red box represents cases excluded 24
Case Selection: Selection Categories and Truthing Case selection categories Negative screening cases negative by FFDM reader and FFDM+DBT reader at acquisition site Recalled screening cases recalled by FFDM reader or FFDM+DBT reader at acquisition site, but not biopsied Benign biopsy cases - pathology proven benign cases Cancer cases pathology proven malignant cases Truthing: Pathology and one-year follow up 9/302 lost to follow up 1/76 benign lesions found to be cancer by follow-up 25
Case Selection: Case Mix and Enrichment Case Type Current study Original PMA Cancer 77 (25%) 51 (16%) Benign 75 (25%) 47 (15%) Recall 24 (8%) 138 (45%) Negatives 126 (42%) 74 (24%) Total 302 310 The proportion of recall cases was chosen to reduce potential bias in favor of 3D S on noncancer recall rate. 26
Case Selection: Case Mix and Enrichment Case Type Current study Original PMA Cancer 77 (25%) 51 (16%) Benign 75 (25%) 47 (15%) Recall 24 (8%) 138 (45%) Negatives 126 (42%) 74 (24%) Total 302 310 The proportion of cancer cases was increased to gain statistical power. 27
Case Selection: Case Mix and Enrichment Case Type Current study Original PMA Cancer 77 (25%) 51 (16%) Benign 75 (25%) 47 (15%) Recall 24 (8%) 138 (45%) Negatives 126 (42%) 74 (24%) Total 302 310 In the non-cancer categories, the proportion of benign+recall cases decreased (from 60% to 33%) while the proportion of negative cases increased (from 24% to 42%). Differentiating cancer from non-cancers may be easier for both FFDM and 3D S. May bias in favor of showing non-inferiority between FFDM and 3D S. 28
Case Selection: Distribution by Breast Density BI-RADS Breast Density Category Cases (n=302) 1 37 (12.3%) 2 117 (38.7%) 3 118 (39.1%) 4 30 (9.9%) Fatty: 51% Dense: 49% 29
Case Selection: Calcifications/non-calcs Case Type Calcification Non-Calcs Total Cancer 24 53 77 Recall 8 16 24 Benign 24 51 75 Negative 0 126 126 Total 53 249 302 Cases presenting with only calcifications were considered the calcification group. Cases with mass only or both mass and calcifications were considered with the non-calcification group. 30
Radiologists/Readers Fifteen radiologists representing a range of clinical experience participated in the reader study Board certified, MQSA qualified, trained in and familiar with the interpretation of digital mammography, and users of the Hologic SecurView workstation. Reader training Two full days of training on the reading of 3D tomosynthesis with synthesized 2D images prior to the start of the reader study. FDA requested that the statistical analysis be performed with all 15 readers 31
Image Interpretation: Study Design Fully paired acquisition Cross-over reading to mitigate readingorder effect Session 1 (One Month) Session 2 FFDM 3D S 3D S FFDM First half of the cases Second half First half of the cases Second half 32
Image Interpretation: Scoring The following information was collected for each marked lesion: Lesion location Probability of Malignancy (POM) score of 0 to 100 Forced BI-RADS score of 1, 2, 3, 4a, 4b, 4c, or 5 The following information was collected for cases with no marked lesions: POM score of 0 to 100 BI-RADS score of 1 or 2 33
Summary Controlled retrospective Multi-Reader Multi-Case (MRMC) study on an enriched patient dataset 15 Board certified and MQSA qualified radiologists 302 cases (77 cancer, 225 non-cancer) Case dataset is enriched with different proportions of cancers/non-cancers than the general screening population. Two-session cross-over reading separated by a one-month memory wash-out period Primary Objective Show AUC of 3D S is non-inferior to that of FFDM with a non-inferiority margin of 0.05 34
Summary Lost patients could impact performance About 20% of eligible cases were excluded due to image quality control issues. The protocol did not allow for repeat imaging on these cases. Nine of the 302 study cases were lost to follow-up. The higher relative proportions of negative screening cases could favor the non-inferiority hypothesis The study excluded: Previous surgical (excisional) biopsy Placement of an internal breast marker Breast implants or breast reduction surgery Breasts too large to be imaged with a single exposure 35
Next Speaker The next speaker is Dr. Jingjing Ye 36
Radiology Advisory Panel Meeting Hologic Selenia Dimensions 3D System with C-View Software Module FDA Statistical Review Jingjing Ye, PhD Mathematical Statistician Division of Biostatistics Center for Devices and Radiological Health
Overview Reader Study Results Primary endpoint (AUC) Secondary endpoints (Dense breast AUC, noncancer recall rate) Additional endpoints (sensitivity/specificity, subset AUC, cancer recall rate) Additional Robustness Analyses Discussion on Case Distribution Summary 38
Study Samples Completer In Study Not In Study Total Negative 118 126 1213 1339 Recall 24 24 137 161 Benign 74 75 436 511 Cancer 77 77 211 288 Total 293 302 1997 2299 Present the completer analyses results for 293 patients 9 lost-to-follow-up subjects will be addressed in Additional Robustness Analyses 39
Study Endpoints Primary endpoint AUC: 3D S is non-inferior to 2D (Non-inferiority margin 0.05) Secondary endpoints: Dense breast AUC: 3D S is non-inferior to 2D (noninferiority margin 0.05) Non-cancer recall rate: 3D S is non-inferior to 2D (noninferiority margin 5%) Additional endpoints: Calcification and non-calcification AUC Sensitivity and specificity Fatty breast AUC Cancer recall rate Descriptive Statistics 40
Primary Endpoint-AUC Difference (3D S -2D FFDM) 41 41
Primary Analysis-AUC (Reader Averaged), 77 Cancers and 216 Non-cancers Method 2D FFDM 3D S Difference (95% CI lower bound) P-value+ (Non- Inferiority) Nonparametric 0.859 0.895 0.036 (0.011) <0.0001 Parametric 0.866 0.907 0.041 (0.015) <0.0001 DBM* Method used for primary analysis (FDA calculation) Non-inferiority margin is met +One-sided p-value *Dorfman, Berbaum, Metz, 1992 Invest Radiol 27(9):723-31 42
Secondary Endpoint-Dense Breast (AUC difference: 3D S -2D) 43 43
Dense Breast AUC (Reader Averaged), Method 37 Cancers and 106 Non-cancers (Secondary Endpoint) 2D FFDM 3D S Difference (95% CI lower bound) & P-value+ (Non- Inferiority) Nonparametric 0.831 0.881 0.050 (0.005) 0.0001 Parametric 0.848 0.893 0.045 (0.007) <0.0001 DBM* Method used in secondary analysis (FDA calculation) Non-inferiority margin is met & No multiplicity adjustment + One-sided p-value *Dorfman, Berbaum, Metz, 1992 Invest Radiol 27(9):723-31 44
Non-cancer Recall Rate (Reader Averaged) (Secondary Endpoint) Method N 2D FFDM 3Ds Difference (3Ds-2D) 95% CI upper limit+ Non-cancer recall rate 293 46.3% 32.3% -14% -11.3% Random reader and random case bootstrap used (FDA calculation) Non-inferiority margin met +No multiplicity adjustment 45
Sensitivity, Specificity (Reader Averaged) (Additional Endpoint) Metric N 2D FFDM 3D S Difference Sensitivity 77 86.2% 86.8% 0.6% Specificity 216 64.8% 74.8% 10% BIRADS cut-off=4 BIRADS 1, 2, 3 test negative BIRADS 4, 5 test positive FDA calculation 46
Change in Sensitivity, Specificity by Reader 47
Calcification and Non-Calcification Calcification Non-Calcification Total Cancer 24 53 77 Benign 24 50 74 Recall 8 16 24 Negative 0 118 118 Total 56 237 293 48
Calcification and Non-Calcification (Reader Averaged) (Additional Endpoint) Calcification+ 2D FFDM 3D S Difference AUC 0.780 0.816 0.036 Non-Calcification# 2D FFDM 3D S Difference AUC 0.867 0.907 0.040 DBM Method used (Nonparametric AUC, FDA calculation) +Calcification: 24 cancers, 32 Non-cancers #Non-calcification: 53 cancers, 184 non-cancers 49
Fatty Breast (Reader Averaged) (Additional Endpoint) Fatty 2D FFDM 3Ds Difference AUC 0.889 0.912 0.024 DBM Method used (Nonparametric AUC, FDA calculation) 40 cancers and 114 non-cancers 50
Cancer Recall Rate (Reader Averaged) (Additional Endpoint) Method 2D FFDM 3Ds Difference Recall rate # 88.7% 89.4% 0.7% Bootstrap for random reader and random cases (FDA calculation) # Not lesion-specific 51
Positive and Negative Likelihood Ratio (PLR and NLR) (Reader Averaged) Metric 2D FFDM 3D S Difference (95%CI)* PLR 2.75 3.97 1.22 (0.72, 2.03) NLR 0.21 0.17-0.04 (-0.084, 0.016) PLR : greater PPV with 3Ds (statistically significant) NLR : greater NPV with 3Ds (not statistically significant) Results do not depend on prevalence *Bootstrap CI, Unadjusted confidence interval *Biggerstaff, Stat medicine, 2000, 19, 649:663 52
Robustness Analysis 9 subjects lost-to-follow-up (unconfirmed noncancers) 8 in negative 1 in Benign Analyses scenarios Scenario 1: impute all as non-cancers Scenario 2: impute all as cancers Worst-case: impute k of 9 cases with highest POM in 3Ds as non-cancers, and remaining 9-k as cancers, k=1,,8 53
Robustness Analysis Results Data AUC Difference (95% lower bound) Dense Breast AUC Difference (95% lower bound) DBM Method used (Nonparametric AUC) Results robust to unconfirmed non-cancers * 9 lost-to-follow-up subjects as non-cancers + 9 lost-to-follow-up subjects as cancers Non-cancer Recall Rate Difference (95% upper bound) Completer 0.036 (0.011) 0.050 (0.005) -14% (-11.3%) Scenario 1* 0.035 (0.01) 0.049 (0.004) -13.9% (-11.4%) Scenario 2+ 0.034(0.01) 0.049 (0.007) -14% (-11.2%) Worst-case (k=3) 0.031(0.007) 0.043 (-0.0001) -13.8% (-11.2%) 54
Missing Data - Exclusion Q/C Problem 21%(524/2536) non-cancers were excluded for Q/C 19% (66/353) cancers were excluded for Q/C Robust analyses done to examine the impact of Q/C exclusion Imputation Analyses Missing At Random Non-ignorable Missing Results of Primary Endpoint Met Mixed 55
Discussion on Case Distribution The current enrichment of samples may favor non-inferiority design Reweight case distributions according to previous PMA 25% cancers -> 16% Cancers 25% Benign -> 15% Benign 8% Recall -> 45% Recall 42% Negative -> 24% Negative 56
Primary Endpoint AUC Data 2D FFDM 3D S AUC Difference (95% CI lower bound) Scenario 1* 0.861 (0.841, 0.878) 0.896 (0.880, 0.911) 0.035 (0.021) Reweighting+ 0.839 (0.817, 0.858) 0.878 (0.860, 0.894) 0.039 (0.024) Bootstrap for random reader and random cases (FDA calculation) * 9 lost-to-follow-up subjects as non-cancers (25% Cancers, 25% Benign, 8% Recall and 42% Negative) + Reweighting according to P080003 proportions (16% Cancers, 15% Benign, 45% Recall and 24% Negative) 57
Summary of Statistical Analysis Results For completers (excluding 9 lost to follow-up), the prespecified non-inferiority margins were met for the primary analysis of AUC difference (margin 0.05), the secondary analysis of dense breast AUC difference (margin 0.05), and the secondary analysis of non-cancer recall rate (margin 5%) Classifications favored device use: Non-cancer recall rate decreased by 14% (Secondary endpoint) Cancer recall rate increased by 0.7% (Additional endpoint) Likelihood ratios indicate increases in PPV, NPV 58
Summary of Statistical Analysis Results Study conclusions on AUC non-inferiority appeared to be Robust to 9 unconfirmed non-cancers Robust to Q/C exclusions if MAR, but not if non-ignorable missing analysis Robust to choice of case distribution (cancer, benign, recall, normal) 59
Next Speaker The next speaker is Dr. Gary Levine 60
Radiology Advisory Panel Meeting Hologic Selenia Dimensions 3D System with C-View Software Module FDA Clinical Perspective Gary Levine, MD, MSE Mammography, Ultrasound, and Imaging Software Branch Division of Radiological Health Office of In vitro Diagnostics, and Radiological Health Center for Devices and Radiological Health Food and Drug Administration
Overview Clinical Significance Study Design Study Results Benefit vs. Risk Labeling and Reader Training MQSA Issues Summary 62
Background Tomosynthesis is the generation of slice images by digitally processing X-ray images obtained at multiple angles The synthesized 2D images are a mathematical reconstruction from the 3D image set similar to a maximum intensity projection (MIP) 63
Background Advantages of 3D over traditional 2D images Reduces visual clutter by removing overlying anatomic structures Allows for depth localization Original PMA approval studies support the benefits of 2D plus 3D for breast cancer screening and diagnosis 64
Clinical Significance 3D S is a new exam option with a radiation dose that is comparable to current FFDM Using the 3D S option could reduce the radiation exposure and compression time by eliminating the FFDM acquisition required with current 2D plus 3D exams Exam 2D FFDM Hologic 2D FFDM across manufacturers** 3D S (3D DBT plus synthesized 2D) 2D FFDM plus 3D DBT Radiation Dose* 1.2 mgy (single 2D acquisition) 1.43 mgy (single 2D acquisition) 1.45 mgy (single 3D acquisition) 2.65 mgy (2D + 3D acquisition) * Measured using a standard phantom ** Collected from medical physicists dose measurements available at the time of MQSA inspections Jan Sept 2012 65
Study Design The study design is similar to the MRMC studies used to support the initial DBT approval Stratified sampling to obtain representative enriched dataset based on lesion type and breast density 15 Board certified and MQSA qualified radiologists Readers and cases were not used for any prior studies Non-inferiority study since the 3D S exam dose is comparable to existing average 2D FFDM dose 66
Study Design The study population was a mix of routine screening and patients scheduled for biopsy Cancer truth is based on biopsy results Non cancer truth required a one year follow-up mammogram Follow up on 9 of 302 cases is missing Case distribution: 77 Cancers 24 Recalled Screening Cases 75 Benign Biopsy 126 Screening Negatives 67
Study Results 3D S was non-inferior in ROC AUC compared with FFDM Secondary analysis indicate a reduction in the non-cancer recall rate Secondary analysis also compared 3D S to 2D FFDM across: Fatty and dense breasts Non-calcifications and calcifications Sensitivity and specificity Cancer recall rate 68
Benefit vs. Risk Compared with FFDM In clinical practice, 3D S may reduce the number of non-cancer recalls, thereby reducing patient anxiety and radiation dose from additional imaging However, the 3D S exam will not include the standard of care FFDM images, which will impact how mammography is performed and interpreted 69
Benefit vs. Risk Compared with the approved 2D FFDM plus 3D DBT option With 3D S, lower radiation dose compared to the combined 2D FFDM + 3D DBT Shorter breast compression with 3D S 70
Labeling and Reader Training The labeling states that the synthesized 2D are only to be used as an aid to the 3D images This reflects the intended use and study design 71
Labeling and Reader Training Users need to understand that the synthesized 2D is similar to, but not identical to standard of care 2D FFDM Note: The images will also contain a C-View mark The software used to create the synthesized images depends on assumptions of the breast geometry The training should include how synthesized 2D images might look different and how to account for those differences when comparing to prior FFDM exams 72
Labeling and Reader Training The labeling and training should inform radiologists/technologists of situations when to acquire 2D FFDM instead of using 3D S The labeling states the following situations that could result in atypical synthesized 2D images: Large artificial objects (implants, pacemakers, and port-a-caths) Diagnostic paddles Intramammary metallic markers Patient motion Mastectomy view Surgical specimen 73
Labeling and Reader Training The study excluded subjects with an internal breast marker, breast implants, and breasts too large to be imaged with a single exposure These situations may not match the assumptions used by the software and could result in atypical synthesized 2D images FDA expects the labeling to address that the performance of 3D S in these situations was not established 74
Labeling and Reader Training The study excluded subjects with previous surgical (excisional) biopsy This exclusion was by design to avoid confounding the results This exclusion is not addressed in the labeling Consideration for Panel Discussion Question #3 75
MQSA Requirements Tomosynthesis is being treated as a new mammographic modality requiring that: All personnel obtain 8 hours of training in that mammographic modality prior to use on patients Hologic must provide each facility with a QC manual clearly defining initial, periodic, and yearly testing requirements Facilities follow the manufacturer s quality control (QC) manual 76
MQSA Requirements The synthesized 2D view will not be treated as a new modality The training for the synthesized 2D views can be included as part of the 8 hour training for new users of tomosynthesis MQSA will continue using the certificate extension program for the immediate future 77
Summary The study design is consistent with other MRMC mammography studies Difficult to estimate the clinical benefit to risk ratio from an enriched stratified sampled study The results are non-inferior to FFDM, but indicate the likely benefit of using 3D S to reduce the recall rate The dose is comparable with current FFDM, so there is minimal additional risk 78
Summary Hologic needs to provide training to help new users utilize the new technology effectively FDA expects the labeling to address situations where the currently approved FFDM and/or FFDM plus DBT screening options may be more appropriate instead of 3D S 79
Radiology Advisory Panel Meeting Hologic Selenia Dimensions 3D System with C-View Software Module Panel Discussion Robert Ochs, PhD Branch Chief Mammography, Ultrasound, and Imaging Software Branch Division of Radiological Health Office of In Vitro Diagnostics and Radiological Health Center for Devices and Radiological Health Food and Drug Administration
Discussion Topic #1: Potential Impact on Mammography Standard-of-care FFDM will not be acquired with the proposed 3D S (3D plus synthesized 2D) exam option FDA would like the panel to discuss aspects of performing and interpreting mammography that should be considered in the training and/or labeling For example, the panel could discuss the use of 3D S for comparing current and prior exams 81
Discussion Topic #2: Study Bias The study design is similar to the other MRMC studies Potential biases with the current study include: The distribution of cases (cancer, benign, recall, negative) for the non-inferiority hypothesis Cases being excluded from random selection because the image set did not pass QC criteria Lack of confirmation of 8 negatives and 1 benign as non-cancer because they were lost to follow-up FDA would like the panel to discuss any concerns with the potential sources of bias and if the presented analysis on the robustness of the results is sufficient 82
Discussion Topic #3: Study Exclusions The labeling addresses some study exclusions, such as breast implants that could result in atypical images However, there is no information to indicate that subjects with previous surgical (excisional) biopsy will cause atypical images In the original Hologic DBT PMA, subjects with surgical (excisional) biopsy were excluded by design to avoid confounding the results as prior patient studies and reports were not available to the reader The results were considered generalizable for the original DBT PMA since there were no technical concerns with imaging these subjects and the FFDM images would still be available FDA would like the panel to discuss the generalizability of 3D S to subjects with previous surgical (excisional) biopsy 83
Discussion Topic #4 The panel is also asked to discuss any other concerns with respect to the safety, effectiveness, or benefit/risk associated with the proposed expansion to the indications for use 84
Radiology Advisory Panel Meeting Hologic Selenia Dimensions 3D System with C-View Software Module Panel Discussion Questions
Question #1 Hologic is proposing a new screening option that does not include the acquisition of standard of care FFDM images. Please discuss aspects of performing and interpreting mammography that should be considered in the training and/or labeling when FFDM images are not acquired during the screening exam. 86
Question #2 Does the panel have concerns with specific sources of potential bias, such as the case distribution, missing follow-up cases, and exclusions? Does the panel believe the presented analysis of these biases is sufficient? 87
Question #3 The study protocol excluded subjects with previous surgical (excisional) biopsy. Does the panel have concerns with the generalizability of 3D S to subjects with previous surgical (excisional) biopsy? 88
Question #4 Please discuss any other concerns with respect to the safety, effectiveness, or benefit/risk associated with the proposed expansion to the indications for use. 89