While I agree with Dr. Wolfe's overall assessment that MelaFind may be a tool to support sound clinical decision making, I would like to clarify and expound upon several points raised in his recent article, "Inside the Data for Multi-Spectral Imaging for Melanoma Detection."1 I critically reviewed and presented the results of the MelaFind pivotal trial at the FDA Panel meeting held last year. The MelaFind pivotal trial included over 1,300 patients, 1,800 lesions and 23 investigators from seven clinical sites in the US.2 All melanomas in the study were difficult to identify clinically and early (45 percent were in situ, and the median Breslow thickness of the invasive melanomas was just 0.365mm) and over 30 percent were considered to be unlikely melanoma (one to 33 percent probability).
The first point that must be addressed is dermatologist sensitivity. Many studies have documented the sensitivity of dermatologists in detecting melanoma to be in the 70 percent range. In 1992, we published a paper in which we reviewed 1,784 melanomas in which the clinical diagnosis did not mention “melanoma” in 33 percent of cases, and there are a number of other papers that show the sensitivity of clinical diagnosis of melanoma to be between 66 and 70 percent, indicating that clinicians and dermatologists fail to diagnose melanoma clinically in a significant number of cases.3 Currently, histopathology is the only gold standard for measuring diagnostic performance.
The goal of the MelaFind pivotal trial was to demonstrate unprecedentedly high sensitivity at detecting early melanoma while at the same time showing superior specificity to clinicians. Sensitivity of investigators could not be measured in the pivotal trial because all lesions required biopsy and histopathology evaluation to determine sensitivity and specificity of MelaFind. Therefore, a target threshold of at least 95 percent sensitivity was selected before the start of the study as the hurdle for clinically meaningful success of MelaFind. The sensitivity of MelaFind was over 98 percent (detecting 112 of 114 melanomas with a 95.1 percent lower confidence bound), and its specificity was 9.5 percent versus 3.7 percent for dermatologists (p < 0.02) with respect to the primary aim population.
The biopsy ratio of MelaFind was 12:1 and the biopsy ratio of dermatologists was 12.5:1. This represents a significant difference in that the specificity and biopsy ratios of the biopsying physicians occur in the context of approximately 70 to 80 percent sensitivity, whereas the specificity and biopsy ratios of MelaFind occur in the context of over 98 percent sensitivity. In a corollary study performed by Dr. Suephy Chen of Emory University of 110 dermatologists who reviewed three high resolution images (close-up, dermoscopic, and distant) and extensive clinical histories of 65 melanomas and 65 non-melanomas taken from the MelaFind pivotal trial, the average sensitivity of dermatologists was 72 percent. On average, 18 of 65 melanomas were not diagnosed.
This study also proved that different dermatologists biopsy different atypical pigmented lesions, therefore, the pivotal trial was not biased, as Dr. Wolfe suggested. In fact, there were many lesions (over 100) that were biopsied with the clinical diagnosis “not melanoma,” therefore, these were not suspicious lesions. The findings, then, are indeed able to be generalized to the types of clinically atypical pigmented lesions that we see in practice.
MelaFind provides an objective analysis (“positive” or “negative”) based on the degree of three dimensional morphological disorder. By incorporating the MelaFind result with the clinical ABCD and other features of pigmented lesions, the level of diagnostic performance of the study dermatologists could be increased significantly, perhaps to a level of 95, percent which was achieved by four of the 110 dermatologists with a specificity equivalent to that of MelaFind.
In melanoma, errors of omission, not of commission, are the most grave. The data that I reviewed and presented at the MelaFind FDA Panel meeting suggest that MelaFind can help clinicians detect more melanomas at the earliest most curable stages by performing biopsies of the most appropriate atypical pigmented skin lesions. Taken together, and in the context of the challenges we face every day with respect to early melanoma detection, the data strongly suggest that MelaFind can provide significant benefit to our patients.
— Clay J. Cockerell, MD, FAAD
Dr. Wolfe responds:
Dr. Cockerell's detailed analysis of the statistical data is truly appreciated. I agree that MelaFind may be a tool to support sound clinical decision making, though I add and emphasize when it is used properly and by appropriately trained dermatologists. My goal in reviewing the data on MelaFind and the FDA panel's analysis was to help dermatologists understand the strengths and potential limitations of the device. The notion of potential bias was raised by the FDA. This does not suggest that the trial results are unreliable or flawed overall; however, those assessing the data must understand them in context of a trial attempting to utilize a diagnostic tool for cancer.
A lingering concern for the MelaFind device, if cleared for marketing, is how it may be used by non-dermatologists (whose sensitivity in the visual diagnosis of melanoma is lower relative to dermatologists) and whether certain clinicians may rely too heavily on the device. I am sure all agree that any diagnostic tool leading to improved diagnosis of melanoma is one that will stand all skin care providers in good stead.
Reader feedback is welcome. Comments, suggestions, corrections, and letters to the editor can be sent to Paul Winnington via e-mail: firstname.lastname@example.org.