(172 days)
MammaPrint is a qualitative in vitro diagnostic test service, performed in a single laboratory, using the gene expression profile of fresh breast cancer tissue samples to assess a patient's risk for distant metastasis.
The test is performed for breast cancer patients who are less than 61 years old, with Stage I or Stage II disease, with tumor size
The MammaPrint service is a microarray based gene expression analysis of a tumor. The analysis is based on several processes: isolation of RNA from frozen turnor tissue sections, DNA'se treatment of isolated RNA, linear amplification and labeling of DNA'se treated RNA, cRNA purification, hybridization of the cRNA to the MammaPrint microarray, scanning the MammaPrint microarray and data acquisition (feature extraction), calculation and determination of the risk of recurrence in breast cancer patients.
The MammaPrint analysis is designed to determine the gene activity of specific genes in a tissue sample compared to a reference standard. The result is an expression profile, or fingerprint, of the sample.
The correlation of the sample expression profile to a template (the mean expression profile of 44 tumors with a known good clinical outcome) is calculated and the molecular profile of the sample is determined (Low Risk, High Risk).
Here's a breakdown of the acceptance criteria and the study information for the MammaPrint® device, based on the provided text:
MammaPrint® Device Performance Study Analysis
1. Acceptance Criteria and Reported Device Performance
The provided document primarily focuses on demonstrating substantial equivalence to a predicate device (Agendia BV's MammaPrint (K070675)) rather than setting explicit, quantifiable clinical acceptance criteria for de novo approval. However, it does present analytical performance targets relative to the predicate and clinical outcomes from supporting studies.
Here's a table summarizing the implicit acceptance criteria (based on predicate performance) and the reported device performance from the analytical studies:
| Acceptance Criteria / Performance Metric | Target (derived from predicate or equivalency) | Reported Device Performance (New MammaPrint HD) |
|---|---|---|
| Analytical Performance | ||
| Repeatability and Precision (MammaPrint index Standard Deviation) | At least as good as predicate (≤ 0.030) | At least as good as predicate (0.030) |
| Repeatability and Precision (MammaPrint index Variance) | At least as good as predicate (≤ 0.001) | At least as good as predicate (0.001) |
| Concordance in MammaPrint outcome (HD vs. LD microarray) | Within 97.7% technical accuracy of predicate | 98.9% |
| MammaPrint Index difference between scanners (mean, median, standard deviation) | Within predicate's accepted variance (1.96 * 0.030) | Falls within accepted variance |
| Accuracy of classifying as High Risk or Low Risk | Better than predicate (97.7% accuracy, 1.1% false negative) | 98.9% (0.5% false negative classification) |
| Percentage of "Borderline Samples" | Not explicitly stated as a target, but reported for context | Less than 5% |
| "Borderline Samples" Classification Accuracy | Not explicitly stated as a target, but reported for context | Approximately 90% (10% chance of false classification) |
2. Sample Size and Data Provenance for Test Set (Analytical Performance)
- Precision Evaluation (Method Validation):
- Sample Size: 3 samples with different outcome levels (high, low, borderline) were run repeatedly over 20 days. Two replicates per sample level per day. Also included High Risk Control (HRC) and Low Risk Control (LRC). Total 120 (3 samples x 2 replicates x 20 days), plus controls.
- Data Provenance: Not explicitly stated, but likely internal laboratory data from Agendia BV. Retrospective, as these were established samples.
- LD and HD Microarray Comparison:
- Sample Size: 98 historical MammaPrint service samples.
- Data Provenance: Samples from the period 2004 through 2007. Origin country is not explicitly stated in this section, but the company is based in the Netherlands. This is retrospective data.
- Scanner Performance Comparison:
- Sample Size: 26 newly hybridized slides (representing 104 samples). This included three samples with varying risk profiles, repeated, and LRC/HRC controls.
- Data Provenance: Likely internal laboratory data from Agendia BV. Retrospective.
For Clinical Performance, the document refers to four published studies. These studies serve as the clinical evidence for the device's utility, rather than a single "test set" in the context of analytical validation. The details provided are for each study:
- Nature Paper (1): 78 patients
- NEJM Paper (2): 151 patients
- MammaPrint Paper (3): Not a clinical study for patient outcomes, focuses on reproducibility of (1) and (2).
- Transbig Paper (4): 302 patients
The provenance of these clinical studies would generally be international (e.g., European for Transbig). All are retrospective analyses of patient cohorts.
3. Number of Experts and Qualifications for Ground Truth (Test Set)
For the analytical performance studies, "ground truth" refers to the known characteristics of the samples (e.g., whether they were designed to be high, low, or borderline risk, or previously classified by the predicate device). No human experts are used to establish ground truth in these analytical comparisons; instead, it relies on the pre-determined characteristics of the control samples or the results from the predicate device.
For the extensive clinical studies referenced (Nature, NEJM, Transbig), the ground truth for patient outcomes (e.g., metastasis-free survival) would have been established through long-term clinical follow-up by medical professionals, including oncologists and pathologists. The number and specific qualifications of these experts are not detailed in this 510(k) summary, as it refers to published peer-reviewed literature. Typically, these studies involve panels of clinicians (e.g., pathologists, oncologists) determining initial diagnosis and patient outcomes.
4. Adjudication Method for the Test Set
- Analytical Performance: No adjudication method as human interpretation is not involved in comparing array results or scanner outputs. The comparisons are quantitative.
- Clinical Performance (Referenced Studies): The 510(k) summary does not detail the specific adjudication methods used in the referenced clinical studies (e.g., for outcome endpoints like metastasis). Such details would typically be found within the full publications.
5. Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study
No MRMC study is mentioned or implied. MammaPrint is a gene expression profiling test, which is an automated, quantitative assay. Human readers do not directly interpret the "cases" (gene expression profiles) in the same way they would an image. The device provides a quantitative risk score (Low Risk, High Risk) that physicians use to make decisions, but the device itself does not involve human interpretation for its output. Therefore, a study demonstrating how human readers improve with AI vs. without AI assistance is not applicable here.
6. Standalone Performance Study
Yes, the entire submission describes the standalone performance of the MammaPrint device (the algorithm/system itself) in classifying tissue samples as High Risk or Low Risk based on gene expression profiles. The analytical performance metrics (precision, reproducibility, classification accuracy) are all measures of the algorithm's performance in isolation from human interpretation.
7. Type of Ground Truth Used
- Analytical Performance:
- Known outcome levels for control samples (high, low, borderline).
- Results from the predicate device (FDA cleared Low Density microarray).
- Clinical Performance (Referenced Studies):
- Long-term clinical follow-up data: Metastasis-free survival.
- Clinicopathological factors (e.g., tumor size, lymph node status, age).
8. Sample Size for the Training Set
The document mentions: "The correlation of the sample expression profile to a template (the mean expression profile of 44 tumors with a known good clinical outcome) is calculated and the molecular profile of the sample is determined (Low Risk, High Risk)." This "template" of 44 tumors appears to be a core component of the algorithm's development.
The broader clinical studies referenced (e.g., Nature Paper (1) with 78 patients) were likely instrumental in the initial development and refinement (training) of the 70-gene profile itself. The MammaPrint Paper (3) describes the "Development of MammaPrint" and "Converting a breast cancer microarray signature into a high-throughput diagnostic test," implying further training and optimization steps using data from earlier studies.
Therefore, the initial "training set" for defining the gene profile involved patient cohorts (e.g., the 78 patients in the Nature paper, or the 44 specific "good clinical outcome" tumors).
9. How the Ground Truth for the Training Set Was Established
For the "44 tumors with a known good clinical outcome" that form the template, the "known good clinical outcome" would have been established through:
- Long-term clinical follow-up: Patients whose tumors formed this template would have had a prolonged period without recurrence or distant metastasis.
- Clinicopathological data: Standard clinical assessments (e.g., pathology reports, surgical findings) would have confirmed the characteristics of these tumors and the patients' disease stage.
For the broader training of the 70-gene profile, as implied by the referenced papers, the ground truth was established by:
- Clinical outcome data: Patients were followed for several years (e.g., 5-10 years) to determine their metastasis-free survival or overall survival.
- Pathology: Original biopsy/surgical pathology diagnoses, including tumor type, grade, and lymph node status, were used.
- Expert Consensus: While not explicitly stated for individual training cases in this summary, the clinical outcomes and pathological assessments that defined "good clinical outcome" and "poor clinical outcome" would have been made by qualified medical professionals (pathologists, oncologists).
§ 866.6040 Gene expression profiling test system for breast cancer prognosis.
(a)
Identification. A gene expression profiling test system for breast cancer prognosis is a device that measures the ribonucleic acid (RNA) expression level of multiple genes and combines this information to yield a signature (pattern or classifier or index) to aid in prognosis of previously diagnosed breast cancer.(b)
Classification. Class II (special controls). The special control is FDA's guidance document entitled “Class II Special Controls Guidance Document: Gene Expression Profiling Test System for Breast Cancer Prognosis.” See § 866.1(e) for the availability of this guidance document.