DR. FEIGAL: Thanks very much. Actually, in the spirit of disclosure, I should mention that my first contact with the FDA was when I was asked to come and make a presentation to an advisory panel. Later, I served on a panel and one thing led to another. So I think this is just as a fair warning to those of you on the committee that there sometimes are adverse--I don't know of they are adverse, but there are unpredictable career effects of getting involved with the FDA.
Let me begin with a task which is both pleasant but one which also reflects our deep appreciation of the service of two outgoing members of the committee. What I would like to present this morning is a letter and a plaque recognizing the contributions of Dr. Smathers and Dr. Destouet.
Let me just read the letter signed by Dr. Henney. "I would like to express my deepest appreciation for your efforts and guidance during your term as a member of the Radiologic Devices Panel of the Medical Devices Advisory Committee. The success of this committee's work reinforces our conviction that responsible regulation of consumer products depends greatly on the participation and advice of the nongovernmental health community.
"In recognition of your distinguished service to the Food and Drug Administration, I am pleased to present you with the enclosed certificate."
Let me present these to you right now.
DR. FEIGAL: Let me just make a couple of introductory comments to say how much I appreciate the committee's grappling with this issue with us. Today's focus will really be on one particular application and whether or not that application meets the standards which are required and whether your recommendation on whether the approach suggesting in this application should be successful.
As you know, it has not been entirely straightforward to identify a regulatory path for approval for this technology. There are relatively few screening technologies used in healthy people that have paid their dues and have shown that there is actual clinical benefit. Pap smear is an example of that.
Recently, there was an approval of an automated Pap-smear reader and the same types of issues arose which is how much could we rely on small datasets and detailed technical analyses of the performance of such equipment and at what level did we need assurance that it would produce the same kinds of sensitivity and specificity and predictive accuracy that the Pap smear read by the human reader, all the same issues of inter-reader variability.
That was a technology that was approved with a dataset of about 33,000 slides. It was one where we were able to establish the relative sensitivity and specificity. The points that are made about changes in this technology being evolutionary are quite correct. That is one of the great difficulties in deciding when you ask for more data than some of the physics data or small datasets.
Different approaches have been suggested for this technology. As you are well aware, an initial approach was suggested that these technologies might be similar enough that it would be straightforward to demonstrate that there was agreement between two technologies in study designs that wouldn't even tell you why there was disagreement and, in fact, if there was high enough agreement, if the two results, if the two side-by-side technologies always led to the same result, you really couldn't argue about substantial equivalence.
But if one of the technologies was superior, all you would know was that there was a discrepancy between the two. You would not have any information unless you modified the study in some way as to which technology even had better sensitivity, and then the factors that lead to the variability which have already been alluded to, would be factors that would make it difficult to even know how much agreement there was because of the technology and where the disagreement comes from.
There are other approaches. There are approaches which do not answer the question of whether or not this technology will identify cancers that screen film misses. Those types of technologies are where you identify patients who have already been referred because of a suspicious screen film and are now being evaluated in a more diagnostic setting.
Mammography devices that are on the market now are not separately labeled for diagnosis and screening but, clearly, that is a population. That type of study design, while it is a very good source for abnormal exams, it does not provide any insight into what is seen as one of the great promises of techniques with greater resolution which is the ability to improve the detection rate. It is a real question about when it is that we should forego this kind of information.
The final approach, and one that is, obviously, the most challenging and difficult, is to evaluate a new technology in a screening setting where you have the ability to assess the technology where it will be used.
In our regulatory letters, the initial approach of the agency to suggest agreement studies was realized to be too narrow an approach. There still remain manufacturers who are interested in pursuing agreement studies, in pursuing the 510(k) route of approval. And there are other ways to get a 510(k) approval than an agreement study. You also could do that with a ground-truth study.
The reason that we proposed the PMA as an alternative to the 510(k) was because we felt that the substantial experience that many of the manufacturers already had could be put together in a PMA application which does not require that the application be complete because there is flexibility to extend some of the study requirements, some of the things you would like to know about the technology, into the postmarketing period with postmarketing commitments.
We realize that each company had a slightly different approach to the way they were collecting data and studying their equipment. What we were attempting to communicate, and I don't think we entirely succeeded, was that, in fact, the 510(k) mechanism is still open if someone wishes to complete enough of their studies to demonstrate it substantially equivalent, or they can use a PMA route if they wish to come in with data with a postmarketing commitment and, if that bring the technology to the market more quickly, then they have to weigh the business decision about the relative long-term and business effects of being in the 510(k) or the PMA stream.
As you may be aware, even those types of decisions are not forever because there are times when PMAs are downclassified and technologies are used in different ways.
We essentially attempted to make this an option for the companies, to look at the information that they had and to choose the regulatory pathway that they wished to come forward with. As you look at this single application today, and as the company and as the FDA scientists present their perspectives on this, pay relatively less attention to whether this is setting a paradigm for how all companies should proceed because I can guarantee you we will be back with different types of information trying to accomplish the same end with other applications.
Our goal is to get these products into the marketplace as quickly as possible and to allow regulatory flexibility to allow companies to choose the pathway that they wish to choose.
Without any further comments, I think we should begin with the morning. Our first speaker this morning on introductory matters is the capable Dan Schultz.
DR. SCHULTZ: Welcome members of the panel and members of the audience.
Would like to take this opportunity, once again, to welcome you here and, again, thank you for helping us through what has obviously been a somewhat difficult and convoluted problem.
As you can see from the title of my introductory slide, what I intend to talk about very briefly is where we are today and how we got there. The fact that the fonts are different is not necessarily an accident. I think the important thing, really, is for us to move forward. I think that is important both to the agency and to the women of America.
We can spend some time going over how we got there, but I think that will be the brief part of this presentation.
I would like to say that, normally, as Division Director, I don't get to make these kinds of sort of detailed remarks. It is normally that I get to get up and give a couple of sort of perfunctory introductory remarks, but when we were deciding on who was going to give this talk today, for some strange reason, there were not a lot of volunteers so here I am.
Briefly, and I think everybody in this room knows this probably as well as I do, and Dr. Feigal just went over some of it, the history of this product dates back to the early '90s. In 1995, we had our first panel meeting regarding digital mammography. The panel recommended agreement as an alternative to large screening trials for the very reasons that have been talked about previously, that those trials are time consuming. They are costly.
There was a feeling at that time that, since this was, in fact, mammography, that that agreement paradigm would have a chance to show enough information to be able to determine that the two technologies were, in fact, substantially equivalent.
In 1996, the agency incorporated that paradigm into its guidance document. Between 1996 and 1998, we actually had the opportunity to look at data, to talk to companies, to look at different protocol ideas and, basically, came to the conclusion that, whether we liked it or not and, in fact, this is not something that we looked at with great glee because, in fact, the agreement paradigm would have been the simplest albeit not providing as much information as could be obtained in other ways, as Dr. Feigal mentioned.
But, the bottom line was that the agreement paradigm, at least as we proposed it in the guidance document, doesn't work.
In 1998, we asked you back here to discuss, once again, whether there were alternative clinical-study options. Again, as has been previously discussed, there were a number of opinions that were provided. There was a lot of good information and, in fact, one thing I would like to correct is that the agency did not ignore those ideas and those recommendations.
We looked at all of them extremely carefully and, in fact, I think if you look closely at what we are suggesting today, there are elements of all of those opinions.
September, 1999, FDA issued a letter to sponsors, again, as Dr. Feigal mentioned, trying to be specific to and requesting that each sponsor come in to discuss their individual applications given the fact that sponsors were and are in different points on the developmental curve.
So the letter was, in fact, directed at the sponsors that we had had discussions with regarding digital mammography and the letter suggested that there might be an alternative pathway to the market through the PMA process.
December 16, 1999 is where we are today and we are having the third digital panel. But this time, I think, there is a big difference. We, today, are actually going to be looking at an individual marketing application. For the first time, the panel will be asked not to talk about theories, not to talk about regulatory paradigms, not to talk about a variety of different approaches, but actually to look at data. We believe that that is a significant step forward.
Where do we go from here? It has been mentioned that the revised guidance has not yet been issued. All I can tell you, at this point, is that we are working on it. It will come out in the next millennium, hopefully early in the next millennium but it will, I guarantee--and you can quote me on this--I guarantee that there will be a revised guidance in the next millennium. Thank you.
A number of issues, again, without belaboring the point, one of the questions that has been raised is why is mammography different. It is, in fact, the only imaging technology currently indicated for both diagnosis and screening. While not a significant risk in the traditional sense of high-risk devices, I think the risk of this device is based upon the fact that it is, in fact, relied upon by millions in the United States for the early detection of breast cancer.
As Dr. Feigal mentioned, we know that this technology saves lives and we know that it leads to increased breast conservation, both of which we consider to be extremely important issues for American women.
Other issues that have already been touched upon, why ground truth versus agreement. Again, our idea, somewhat, I agree, simplistically, a few years ago, was that if you got perfect agreement you would, in essence, be mimicking ground truth. Unfortunately, what we have discovered since then is that anything less than perfect agreement does raise questions and, in fact, the poorer the agreement, and we all know the reasons why that agreement has not been as good as what we would have like to have seen, the poorer the agreement, the more questions are raised.
Another issue that has been brought up on several occasions is the issue of enriched trials versus screening trials. What we think is that enriched trials do provide adequate information, at least for the diagnostic component of mammography and do, in fact, provide some important information on screening. However, we still believe that the screening trial is, in fact, a more sensitive measure of the ability to detect the earliest lesions and that that is, in fact, probably the most important aspect of mammography and, therefore, one that needs to be looked at in some form at some point in the developmental process.
Finally, last but not least, PMA versus 510(k). Very simply, they answer different questions. They ask different questions and they are meant to answer different questions. The PMA process looks at each individual device and the determination is made as to whether that device is, in fact, based on its own merits, safe and effective whereas the 510(k) process essentially lumps the technology together and looks at whether or not products are substantially equivalent.
How substantial that equivalence needs to be is basically dependent on the device itself and how critical those differences between devices really are. We feel that the PMA process provides us with some increased flexibility. We think that the labeling for and individual PMA can be tailored to reflect the data for that individual device. We think that the PMA lends itself to a regulatory paradigm which includes both a premarket component with gives us enough reassurance to put this device on the market as well as a postmarket component which answers some of the more difficult, harder-to-answer questions over a longer period of time once the device has actually been put on the market.
We also think that, as counterintiutive as it might seem, in fact, for this and for some other technologies, that the PMA process may, in fact, provide a faster route to market while still maintaining the control and the data requirements that are necessary to assure the American public that these devices will do what they say they will.
As Dr. Feigal has mentioned previously, again, and I just reiterate this one more time, we are not completely closing the door to 510(k). We think it is going to be difficult. We don't want that to be swept under the rug. Based on our experience over the last few years showing that these two products could be equivalent is not going to be an easy task, but the 510(k) process does remain open and does remain an option for those companies that wish to pursue it and we would be more than happy to discuss with any company, the ones that we have talked to so far, the ones that we haven't talked to, what those options might be and listen to their ideas on how to get their product to market.
Finally, while all this may be very interesting and we could have long discussions and long debates on whether or not some of the ideas that have been presented today are right or wrong, in essence, the discussion and the comments that I have made so far are somewhat irrelevant for today's purpose.
Today, we are here to discuss an individual PMA and the questions that we are going to be asking you are not looking at whether the FDA is right or wrong. Today's questions are, in fact, and these will be read to you in a slightly different form later on, but, basically, I tried to summarize them: does the data for this PMS provide reasonable assurance of the safety and effectiveness of this device; does the labeling for this PMA clearly and accurately reflect what is known and unknown about this device; and does the developmental plan, in its totality for this PMA, provide women and caregivers the data necessary to make informed decisions.
We look forward to your deliberations. We look forward to your recommendations. And, once again, we thank you for being here and helping us with this very difficult problem.
DR. GARRA: Thank you, Dr. Schultz.
We are now going to proceed with the sponsor's presentation of the PMA, itself. The first speaker will be Scott Donnelly, General Electric's Vice President for Global Technology Operations. He will be followed by Dr. Edward Hendrick, the principle investigator from Northwestern University.
G.E. MEDICAL SYSTEMS PRESENTATION OF P990066
Introduction, Device Description, Non-Clinical Studies
MR. DONNELLY: Good morning. I would like to thank the panel for their time.
This morning, we are going to present the Senographe 2000D which is G.E.'s digital-mammography
My name is Scott Donnelly. I am the Vice President of G.E.'s Medical Systems Global Technology Operation. As such, I am an employee and shareholder in the General Electric Company.
What I will be presenting this morning is an overview of the device and the technology as we have implemented it in our full-field digital-mammography machine at which point I will turn over the presentation to Dr. Ed Hendrick who will present the results from our clinical trials and studies.
This is a very brief overview. G.E. Medical Systems in addition to being a developer of mammography, both in conventional as well as, now, digital systems also is in the business of the design and development of other X-ray equipment, both fluoroscopic and radiographic including other digital X-ray technology.
We also are a major developer, manufacturer, of CT, mR, ultrasound, PET and nuclear-medicine machines. Additionally, we provide solutions for managing that diagnostic information in terms of picture archival systems and radiological information systems.
I will do an introduction and overview of the product and a device description as implemented in the technology that we have selected and a brief overview of the performance of the product in terms of its physics and engineering, and then Dr. Hendrick will cover the clinical data and also the postapproval study which we are proposing.
I think it is important when you look at our full-field digital-mammography product that it is really based on our current platform for analogues, film-screen mammography. The gantry, patient position and acquisition system is actually quite similar to what we do today in analogue film-screen mammography. Once you go into the digital world, we are actually levering quite heavily a product we call our G.E. Advantage Windows Platform which currently is used for out CT and mR product as well as in our other digital X-ray products.
It is important to note, as indicated in the PMA, that the indications-for-use statement is for both diagnostic and for screening applications, so we are seeking approval for both diagnostic and screening use of this machine.
The device description, on the left-hand side of the chart, shows the acquisition platform. If you look at the gantry, it is actually very, very similar to our current film-screen mammography product. The gantry, and, therefore, the patient positioning and the way the technician would use the equipment is the same, the only difference, really, being that, in place of a film-screen buckey, you now have a digital detector that is used in place of the film. I will go into some details later that explain how we have implemented the digital detector.
New to the system is the acquisition work station. The acquisition work station is used to collect all the electronic data that is generated by the detector to do the image manipulation and image processing to generate the image. It is also used as a link to the rest of the information system.
Additionally, what you have, and one of the advantages of the technology, is the ability to do an immediate review of the exam to do basic quality-assurance checking to make sure the positioning was done properly and that the parameters were such that you received a good-quality film.
After that is done, you use the acquisition work station to then send the data over to a laser camera to generate, for purposes of our clinical trials, hard copy review which is then reviewed on a conventional viewbox as you would with film-screen mammography.
The acquisition process is actually very similar to film screen. The patient positioning and exam setting, demographic data, is entered in a very similar fashion. Where the difference is is because the acquisition process is very fast. In a very short time, you can acquire a series of acquisitions with the patient and then go over and review each of those acquisitions to insure quality is there.
In a post process, you will be able to send that data to a laser camera to generate the hard-copy review in the same form as a film screen. So one of the advantages of the technology is that it does dramatically increase the speed at which you can do the acquisition and it also provides you immediate feedback in terms of quality assurance, hopefully reducing the number of retakes based on later film processing which would show a gross positioning error or that something went wrong during the acquisition that resulted in a poor-quality image.
Those retakes can be taken immediately upon review of the QA.
The detector is the new technology involved in this digital mammography. Effectively, you have photons coming into the detector the same as you would in a film-screen system. But, instead of a film screen, you have a solid-state digital detector that has a cesium-iodide layer across the top which brings the photons in and, through a crystalline structure, converts those into light.
That light then comes out and is placed directly onto an amorphous silicon panel which I will describe in the next chart in more detail which basically takes and converts the light to an electronic charge. And then you have read-out electronics which take the charge and scan across the panel and, therefore, take all the digital charge out of the panel, convert that to digital data and send that to the analysis work station where it is processed and the image is generated from that data.
The detector is manufactured with semi-conductor technology. We start with a basic glass substrate. It is important that there are a number of digital technologies. This is unique to the G.E. detector.
After you take the glass substrate, you use conventional semi-conductor manufacturing processes to lay down an amorphous silicon array with 100-micron pixel sizes across the entire field of the array. You then have electronics which scan from the individual rays. The charge would be stored here after it is converted from photons to light.
The signals are extracted on three panels. They are not extracted on the fourth panel in order to minimize the distance and how close you can get to the chest wall so there are no electronics or connector across the front allowing perfect access in against the chest wall.
On top of that is deposited the cesium-iodide scintillator which, again, is a crystalline structure that converts the X-ray photons into light. And then, as you scan out, there is actually an electronics assembly that mounts to the back side of the glass substrate so all the electrons are swept out and converted to digital data to be transmitted to the acquisition work station.
If you look at the nonclinical data on the system, our intent was to take what was available today in screen-film mammography and improve the most important characteristics that are necessary to get good image quality when doing a mammography screening or diagnostic exam. And so the critical functions of dynamic range, modulation transfer function or the spatial resolution of the image, the contrast and the signal-to-noise ratio I will address in this presentation.
In the end, the detective quantum efficiency which is really the overall measure of how effectively you convert from X-ray to an image is discussed in some detail.
This chart shows the comparison between how a digital detector responds versus what you see in a conventional film screen. On the right-hand side is the sensormetric response that you would see in a typical film-screen mammography system. There is actually a considerable differentiation with very small changes in dose that is here in this region which is where normal tissue would be. A film screen performs quite well in this region.
Where you don't see as great a differentiation between the amount of exposure and, therefore, the sensitivity to different absorption of X-ray turning into a very significant change in optical density with a relatively flat line is in this area which would be a very dense area, either against the chest wall or glandular region of the breast or, at the other end of the spectrum, at the very high end, which might be typical of a very low contrast area, let's say, against the skin line.
One of the advantages of the digital detector response is if you look at the digital detector response through that same dose region, it is very linear from very low dose areas that would be typical of a high-density breast or chest wall all the way up to very low density on the skin line.
Of course, in terms of optical density, the equivalent is the number of electrons that are converted as a function of that dose. So what you see is a very, very linear response across the whole range which gives you some superior physics to what you see in a film-screen system.
In the end, the most important thing for us, of course, is detectability. Detectability is a function of both the spatial resolution--so this is a very high spatial resolution moving to the lower spatial resolution and we start to see some blur but, also, and very importantly, the amount of noise that is in with the image.
So, if you see a very, very high-noise environment, even though you may have very high spatial resolution, it is very difficult to extract the signal you are interested in from the noise environment on the display. As you move to a lower noise environment, sometimes even a lower spatial resolutions may be more detectable in terms of the radiographer's ability to extract a signal from the image.
The measurements which we used, which quantifies the overall performance in terms of detectability is what we call the detective quantum efficiency. This takes into consideration both the balance between spatial resolution, or MTF, or the amount of noise that is in the image.
So the most important thing, in terms of maintaining and achieving a very high DQE is that, regardless of whether you have a digital detector or screen-film system, the amount of signal to noise that impinges upon that detector is the same. This is the system now X-rayed that is propagated from the tube through the patient and is received at the detector.
So the really important thing to optimize the image quality and/or patient dose is a function of how efficiently you convert the signal and to do that in such a fashion that you do not induce noise into the image. So it is a very good measure of both signal and noise.
One of the things that having high DQE gives you is also the tradeoff now to decide do you want to have the same image quality with a lower dose or do you want to have improved image quality by the same dose because they are relative, since really what you are talking about is a ratio of the signal to the noise in the system.
This chart takes measurements which we have made on the Senographe 2000D digital system versus published data on an existing and typical film screening. So what you see on the left-hand side is the percentage of detectable quantum efficiency or how efficient is the detector at converting signal into either electrons or, in the case of digital, sensormetric response.
So you see that, in a digital system, you have a much higher DQE across the entire range of line pairs in terms of spatial frequency. This is the region of interest in terms of clinical benefit for mammography. So you can see, across that whole region, you have substantially improved DQE as compared to a typical film screen.
This chart has similar information except instead of selecting two of the same dose, you look across the entire range of dose from very, very low dose to high dose, you see the response of the digital detector is actually very linear and quite flat across the whole range until you get to extremely low dose down in this end.
In terms of noise conversion, what you want to have is to not contribute additional noise. There is some noise called quantum noise which is inherent in the X-ray generation. What you don't want to do is contribute any more noise to that image through the conversion process.
If you look at--even in extremely low doses, you have the noise of the X-ray and you don't have any contribution of additional noise in a digital detector until you are down, approximately--the quantum and the detector-contributed noise become equivalent down at about a dosage of 0.8 mR which is almost an order of magnitude below the region of clinical interest because you are not going to see dosages down to 0.8 to be clinically important in a mammography machine.
So that is kind of a summary of the data that we have for you in terms of the physics and the engineering that we have incorporated in the full-field digital-mammography machine. I think it leverages quite well our long history in mammography taking advantage of a system that is already out in clinical use on a widespread basis.
We have invested and generated a lot of time and tried to come up with a digital detector that has superior physics and to make sure that we leverage our signal-processing expertise both in other digital radiography as well as CT and mR to try to optimize and take advantage of that digital-conversion technology.
The end results, of course, are to be proven in the clinical studies. I will introduce Dr. Hendrick who will take you through the results of the clinical trials using the G.E. F50M machine.