Organizing the world's skin photos?
An AI-powered dermatology tool is a rare example of a true healthcare D2C play
Last month, at Google I/O 2021, the company pulled back the curtains on its latest initiative within healthcare — an AI-powered dermatology tool for identifying common skin conditions. It’s a web-based application that asks users to take pictures of their skin, nail, and hair issues, and answer a series of questions about their symptoms. By leveraging an AI model that’s been trained on 65,000 images of different conditions, the tool narrows down the possible issues. Importantly, it has been designed to take smartphone photos as input and does not require any specialty imaging.
The data from the two papers Google has published on the application are impressive. One was a validation study covered in Nature Medicine, which found that the AI model was non-inferior to dermatologists and superior to primary care physicians (PCPs) and nurse practitioners (NPs) in distinguishing 26 common skin conditions. Another study published in JAMA Network Open evaluated the product as an assistive tool for primary care, and found that it increased the accuracy of diagnoses from 48 to 58% for PCPs and 46 to 58% for NPs. Google plans to launch a pilot of the product later this year, though it will only be available in the EU as a Class I medical device and not in the US. The product will not be marketed as a diagnostic, presumably to avoid higher regulatory hurdles.
Latest example of Google Health’s focus on AI-driven screening
So first of all, what should we make of this announcement and how does it fit in with Google Health’s other initiatives? On the surface, the dermatology application lines up well with Google Health’s other publicly announced projects, which include screening tools for tuberculosis, breast cancer, lung cancer, age-related macular degeneration, and diabetic retinopathy. Though the type of data input is different across the various indications — from X-rays, to mammograms, to OCT scans, to smartphone photos — the underlying approach is the same: training AI models using large datasets to be able to screen for a given pathology. Google’s foray into dermatology appears to be confirmation of their broader strategy in healthcare.
The decision to focus on AI-driven screening tools is unsurprising, as it fits firmly within Sundar Pichai’s vision of Google becoming an “AI first” company. It’s also sensible because it provides Google with the opportunity to move into the unfamiliar field of healthcare and establish a foothold using its extensive tech expertise. To date, Google has been intentional about only developing products which do not require FDA clearance or approval (unlike some other tech giants like Apple), which has allowed the company to sidestep regulations that would require deeper organizational change and investment.
Since most of Google Health’s initiatives are still in the R&D stage, however, important questions remain around how these products will be used and sold in the real world. Due to information asymmetries and stakeholder inertia, among other reasons, healthcare is far from a perfect market where the best products naturally win out. AI-driven screening tools, despite being grounded in technology, will need to be paired with a commercial strategy that takes into account the unique considerations of the industry and each market. The announcement of this new dermatology tool came with few details about the launch strategy, so we can only speculate about what it will entail.
One approach is marketing to clinicians
The two papers published on the application suggest that Google is considering marketing it to clinicians — the validation paper in Nature Medicine highlights the potential of the AI model for “assist[ing] general practitioners in diagnosing skin conditions”, while the JAMA paper directly evaluates the effectiveness of the application in supporting PCPs and NPs. From an organizational standpoint, this makes sense, as it’s the same approach Google appears to be taking with their tool for diabetic retinopathy. Having a unified commercial strategy across all of Google Health would allow for obvious economies of scale.
I am skeptical, however, of this approach. It’s no secret that pharmaceutical companies allocate boatloads of money to sales and marketing — in fact, big pharma spends more on clinician marketing alone every year than the entire GDP of Iceland. Though drugs aren’t a perfect analogue for Google’s new dermatology tool, the point still stands — taking a B2B angle with this product would require a massive and unprecedented investment in medical sales and marketing for the company.
But it’s not just the dollar figures which should give Google pause. Based on how the product has been studied, Google would be marketing it to primary care physicians, who comprise the largest group of medical professionals in most developed countries. This also makes them one of the most challenging clinician call points. For companies which have a robust pipeline of products for primary care, the additional effort of reaching these physicians may pay off over time. For Google Health, it’s unclear that any of its other projects would benefit from a salesforce in primary care — its ophthalmology programs are targeted at ophthalmologists, while its oncology tools are designed for radiologists.
Qualitative feedback on the product also does not suggest that buy-in from PCPs is a given. Despite the improved accuracy of diagnoses when using the tool, the average PCP only found the product “moderately” useful (feedback from NPs was slightly more positive). One potential reason for the tepid response is that PCPs fear the tool would disrupt primary care workflows which are already under strain. For context, the median duration of a primary care visit is 15.7 minutes, of which only 5 minutes is spent discussing the patient’s chief complaint. The data suggest that their concern may be warranted, as even within the artificial confines of a retrospective study, the AI model increased review time for each case by 6-9%. More research may be needed to assess the tool’s impact on real-world workflows and convince clinicians of its value.
A better approach is going direct-to-consumer
Given these challenges, perhaps Google should consider an alternative strategy to launching this product instead. The biggest differentiator of Google Health’s new dermatology tool — within the field but also within its own portfolio — is that it is trained to work on user-captured photos. This obviates the need for clinician involvement, and opens the door for a direct-to-consumer approach which isn’t possible with Google Health’s other projects.
Excitement about the D2C model in healthcare has waned over the last few years, as new companies have found themselves unable to compete with well-funded incumbents given sky-high customer acquisition costs. But Google isn’t just another anonymous startup. It has one of the most powerful consumer brands for any industry, and also the most frequently visited website in the world. A seamless integration of the dermatology tool with Google Search itself would allow the product to have the biggest impact, and also usher in a new age of personalized health information accessible through the Internet.
According to the recent announcement, Google sees almost 10 billion searches each year on the topic of skin, nail, and hair issues. It’s an indication of the tremendous demand Google already sees from consumers for health-related queries, and incorporating this dermatology product into Search would allow it to tap into billions of potential users on day 1 (ignoring regulatory approvals). But it would also align with the direction that Search is moving, by taking information currently part of third-party sites and pulling it into the Search experience.
More broadly, it would mark an important milestone in the democratization of healthcare for patients. Medical information has been accessible online (e.g., WebMD, MayoClinic) for decades, and in recent years, there’s been a surge in ‘symptom checker’ apps which try to identify issues based on specific input from the patient. But none of them have reached anywhere close to the number of people that Google can, and create a health information platform that unites both personalization and scale.
It is worth noting that none of Google Health’s other projects at the moment lend themselves to a D2C strategy. However, perhaps this product will act as a blueprint for the development of screening models for other conditions which are trained using only user-captured input. GI disorders like irritable bowel syndrome are one potential area, and a group of MIT scientists has already started building a database of fecal photos with this in mind.
My hope, as a consumer, is that Google Health will decide to prioritize D2C initiatives over B2B products across the board moving forward. This is the reason why the announcement of the new dermatology tool is important — because it is the first inkling that an alternative, consumer-centric strategy for Google Health may be a possibility. After all, providing a family of automated screening tools that is universally accessible and personalized for each user is far more aligned with Google’s mission than selling software to clinicians — and far more exciting.
Great writeup! If Google Health were to go D2C, what do you think should be their business model?