Diagnostics & Imaging Week Washington Editor
GAITHERSBURG, Maryland – Makers of full-field digital mammography systems (FFDMs) probably did not expect a great deal of difficulty in their efforts to get their machines reclassified from Class III to Class II devices given the results of the recently completed DMIST (Digital Mammography Imaging Screening Trial) trial. However, manufacturers must have taken heart that the Radiological Devices panel did not recommend stringent post-market study conditions to FDA as a condition of the unanimous recommendation to reclassify.
Elizabeth Krupinski, PhD, the temporary chair of the panel, opened the meeting by reminding participants and observers that computer-aided diagnostic systems (CAD) and tomographic systems were not part of the class of devices proposed for reclassification. Krupinski is a research professor of radiology and psychology at the University of Arizona (Tucson).
The first speaker for FDA was Thomas Gross, MD, the director of the Division of Postmarket Surveillance at CDRH, who reminded the panel that "we do have the legal authority to require manufacturers to conduct post-marketing studies," and presented a briefing on study oversight.
Gross said that a review of 127 PMAs issued between "the beginning of 1998 and the end of 2000" showed 45 that came through with conditions of approval, but according to Gross, CDRH had "limited ability" to track those studies. Part of the difficulty, he said, was a "tremendous turnover" in reviewers, a situation that forced a recognition that "there was a need for change." One of the conclusions drawn by surveillance officials at CDRH was that it had to get those studies in place "by the time the product hit the market."
Another of the fixes was the development of a tracking system for studies that were undertaken by companies as a condition of approval (CoA) for their PMAs, and a third was the addition of epidemiologists to PMA review teams. The "epis" as they are often called, take the lead in developing CoA study design and in reviewing the results.
The 2005 final report on CoA studies noted that the agency could locate the final results of only six of the 45 studies referred to by Gross and that it had received no file on 26 of the studies. The report also noted that 40% of the lead reviewers on these CoA studies had either left the department they were in when the studies commenced or had left the agency altogether.
Gross added that FDA was in position to penalize firms guilty of "extreme failure" to conduct agreed-upon postmarketing studies. The draft guidance in question, dated September 15, 2005, includes a provision for "civil money penalties." Gross told Diagnostics & Imaging Week that "extreme failure" is a state of affairs in which "agreed upon studies . . . are clearly not being conducted with no reasonable rationale."
Bob Phillips, PhD, chief of the radiological devices branch, gave a short review of the history of film/screen mammography systems. He pointed out that such systems had to go through the PMA system only after May 28, 1976, unless the manufacturer could establish equivalence. He noted that "we've been aware of full-field digital mammography systems since the 1980s."
Phillips informed the panel that a "rather large intra- and inter-reader variability" was a problem for some FFDM systems' premarket notification applications, or 510(k)s, and that only four PMA applications have made it through the FDA to date. They are the GE Medical Systems (Waukesha, Wisconsin) Senographe 2000, the Fischer Imaging (Denver) SenoScan, the Hologic (Bedford, Massachusetts) Lorad, and the Siemens Medical Solutions (Malvern, Pennsylvania) Mammomat Novation.
According to Phillips' presentation, manufacturers have implemented five recalls, but he did not say whether those were for entire lines of machines, individual machines or otherwise.
As for the question of why this class of devices should be reclassified into Class II, Phillips said that "our understanding of FFDM technology has improved to the point where we can develop appropriate special controls" that will aid in determinations of substantial equivalence without incurring undue risk.
Sophie Pacquerault, PhD, at the office of science and engineering labs stated in her presentation that the DMIST trial formed much of the basis of the reclassification. This study, which was funded largely by the National Cancer Institute and the American College of Radiology (Reston, Virginia), involved comparisons scans generated by digital and film/screen (analog) systems for more than 49,000 asymptomatic women at 33 sites, 335 of whom were diagnosed with breast cancer as a result.
Pacquerault said that the study, reported in September 2005 in the New England Journal of Medicine, demonstrated "no significant difference in diagnostic accuracy" in the overall population, although the numbers suggested an improvement in detection for women younger than 50 years of age for digital systems. FFDMs also exhibited slightly superior performance in perimenopausal and premenopausal women as well as in scans of breasts with dense tissues, but the only group in which the difference was of statistical significance was the population under the age of 50.
Pacquerault stated that "the callback rate of 8.4%" for both systems in the DMIST study "is similar to or lower than" reported in other settings for mammography in the U.S. "DMIST supports reclassification," she concluded.
Robert Jennings, PhD, of CDRH's lab for medical imaging systems assessment reviewed the risks to health for FFDM systems, but briefly noted the subject of relevant guidances. He stated that the "guidance for software contained in medical devices is already available," but that FDA is still working on guidances specific to 510(k) submissions for FFDMs as well as those for "accessories," such as reviewer workstations.
Jennings also said that in terms of safety, the two types of systems are "essentially the same," but that the average dose for the 337 FFDM units the agency certified between 2000 and 2003 "was about 15% lower than the average dose" for film/screen systems.
In the discussion that followed the presentations, Andrew Zhou, PhD, director of biostatistics at the Seattle Veterans Affairs Medical Center asked if the observed overall equivalence between film and digital was observed at each of the centers as well as for the study population as a whole. The FDA staff said that they were not in possession of such data, but the author of DMIST, Etta Pisano, PhD, the chief of breast imaging at the University of North Carolina (Chapel Hill), stated that she could not say whether the data would show any such variances and that it had not yet been analyzed by center.
Krupinski asked if DMIST involved the use of hard copy or soft copy (electronic) for each comparison, to which Jennings replied that "perhaps 95%" of the FFDM systems involved produced only soft copy images, due in some cases to an inability to print the images.
During the following discussion, Zhou reiterated his desire to see "evidence that diagnostic accuracy" is maintained from center to center for the DMIST results.
Krupinski replied that a study conducted "some years ago" addressing reader variability examined center-based concerns. "I doubt it would be more variable" for digital systems than for film systems, she remarked.
Pisano took to the microphone to discuss several aspects of the DMIST study and argued that "we should probably change tomosynthesis" into a Class II device as well, provided that these machines can be configured so as to provide a two-view mammogram in addition to the traditional cross-section series of scans.
Margarita Zuley, MD, of the Elizabeth Wende Breast Clinic (Rochester, New York), praised the advent of digital technology for breast screening, but commented that even with a given machine, different algorithms can produce strikingly different images of the same breast, which "makes my job harder." She added that "the radiologist is at the mercy of the vendor" in such situations, recommending that FDA consider creating two 510(k)s for digital systems, one for the scanner and one for the algorithm that handles the raw data file.
Pisano said she supported the idea of lending consistency to the algorithm used, arguing that it is "not an overly burdensome thing" for manufacturers to control because the concomitant study work would involve fewer than 100 subjects for each algorithm comparison.
While the panel members voted unanimously to reclassify FFDMs, they did so with the proviso that the guidance suggest ways that industry might deal with the difference in images produced by the different manufacturer's algorithms.