Smartphone apps for detecting skin cancer are not very accurate, but over-the-counter accessibility may be the future of many medical tests.
There are now several smartphone applications (apps) that make claims regarding skin cancer detection. Each app evaluates photographs of skin lesions and provides information as to the likelihood of malignancy. These apps are relatively inexpensive when compared to a doctor’s visit and are not subject to validation or government oversight. Although the limitations of such apps are disclosed, some consumers may be misled by the conclusions of these untested algorithms and may be harmed in the process. A team at the University of Pittsburgh set out to test the effectiveness of four such apps that were publicly available.
The researchers submitted digital clinical images of 60 melanoma lesions (44 invasive; 16 in situ) and 128 benign lesions (lentigo, nevi, seborrheic keratoses, hemangioma, dermatofibroma) that had been obtained before biopsy and were histologically confirmed by a dermatopathologist. Three of the apps used algorithms that assessed the risk of the lesion. The lesions were assessed as follows:
- App 1: “problematic” versus “okay”
- App 2: “melanoma” versus “looks good”
- App 3: “high risk,” “medium risk,” or “low risk”
The fourth app claims to submit the image to a board-certified dermatologist (identity unknown) for review.
Sensitivity ranged from 6.8% to 98.1%. App 3 had the lowest sensitivity, even when “medium” risk was considered positive (54.2%; 95% confidence interval, 40.8%–67.1%). App 4 had the highest sensitivity (98.1%; 95% CI, 88.8–99.9) compared with the others. Specificity ranged from 30.4% to 93.7%; app 3 had the highest specificity. The positive and negative predictive values ranged from 33.3% to 42.1% and from 65.4% to 97.0%, respectively.
Comment: In this study, apps 1, 2, and 3 are considered diagnostic tools, and app 4 is essentially a store-and-forward dermatology tool. App 4 was the most expensive and had the highest sensitivity and the lowest specificity. Although one may be drawn to compare the performance of these tools with that of a professional device like Melafind (sensitivity, 98.3%; specificity, 9.9%), usage is very different. These smartphone apps are aimed at the consumer audience, whereas Melafind is used for lesions prescreened by dermatologists. Overall, we need to understand the goal of these devices. High sensitivity, which is probably desirable in a consumer-based technology, will undoubtedly drive up demand for office visits in this already access-limited field. One can envision a secondary agenda, wherein the smartphone automatically locates available dermatologists based on the GPS position of the phone, perhaps becoming a marketing tool for physicians. Nevertheless, medicine is moving toward over-the-counter accessibility, as even HIV tests, which used to be considered incredibly sacrosanct, can now be found in local drug stores.
—by Hensin Tsao, MD, PhD in JournalWATCH