Access to dermatologists is often limited, leading to around 60% of skin, hair, and nail
issues being treated by non-specialists. This study will evaluate the effectiveness of an
AI dermatology decision support tool in assisting primary care providers (PCPs) with the
diagnosis of skin conditions. AI-based image analysis has been shown to enhance
diagnostic accuracy, particularly for non-dermatologists. Previous studies have primarily
focused on dermatologists, but AI could be more beneficial for PCPs, as it has been shown
to improve their diagnostic accuracy and agreement with dermatologists.
Globally, about 1.9 billion people suffer from skin diseases annually, with 1 in 3
Americans seeking dermatological care from non-specialists. Skin-related issues make up a
significant proportion of visits to general practitioners and emergency departments. AI
has proven effective in diagnosing skin conditions such as melanoma and other
inflammatory diseases, and studies indicate that AI tools can enhance diagnostic
accuracy, particularly for non-dermatologists.
The Belle AI tool, which will be used in this study, employs a convolutional neural
network trained on over 500,000 images to identify over 2,000 skin conditions. It
provides image match scores to help physicians identify conditions and offers a protocol
for second opinions from board-certified dermatologists. The study aims to assess the
tool's utility in real-time clinical settings, with potential improvements in triage
accuracy, referral quality, and cost savings.
This study is supported by the Advanced Research Projects Agency for Health (ARPA-H) and
will be one of the first to prospectively examine AI's impact on dermatology decision
support in primary care.
The study aims to evaluate the accuracy and utility of the Belle AI dermatological
reference system in a real-world clinical environment, in partnership with Urban Health
Plan (UHP). Key endpoints include assessing diagnostic utility and accuracy compared to a
final diagnosis from a dermatological review committee, as well as gathering feedback
from primary care providers and physician extenders on their experiences with the AI. The
sponsor will also analyze the cost implications of the system's use to demonstrate its
value in frontline medicine.
Participants will use a smartphone app to capture images of their skin conditions, which
will then be analyzed by the AI. Participants will receive financial incentives for
submitting images after their initial visit. A follow-up appointment will be scheduled
two weeks after the initial consultation, though some visits may be canceled based on the
AI analysis.
Participants will be included if the participants present with a primary dermatological
complaint and can provide informed consent. Exclusions apply to those unable to comply
with procedures or pediatric participants with genital conditions for privacy reasons.
Upon entering UHP, participants register and are triaged. Those with qualifying
dermatological conditions will be approached for recruitment by a study coordinator, who
will explain the study and obtain consent. Participants will download the Belle Image
Capture App to their smartphones, where a Study ID code linked to their EMR will be
created, ensuring privacy.
During the initial appointment, providers will examine the patient, document their
history, and diagnose the condition. The BellePro Physician App will be used to capture
images and generate a differential diagnosis, which the provider will review before
making a final diagnosis. Participants will be scheduled for a follow-up visit, and the
study coordinator will notify providers of any received images captured through the app.
Beginning seven days after enrollment, push notifications will prompt participants to
submit images using the app. The coordinator will follow up with participants who do not
respond, aiming for a clear image within a specified timeframe. Upon receipt, the
provider will reassess the diagnosis using updated AI analysis. Decisions regarding
follow-up appointments will be communicated by the coordinator.
If a follow-up appointment is deemed unnecessary, participants will still be asked to
submit images on Day 14. The coordinator will follow up similarly to ensure compliance.
The study spans from Day 1 (initial clinic visit) to Day 14-18, when final images or
follow-up appointments will occur. Case notes will be updated continuously in
eClinicalWorks, determining whether cases are resolved or require ongoing care.
Primary care providers at UHP will undergo onboarding, including an electronic intake
survey and training on the BellePro Physician App via group video chat. Providers will be
trained on the app's use, and their feedback will be collected in an exit survey to
evaluate their experiences and willingness to continue using the AI system.
Given the complexity of dermatological diagnoses, a review committee of senior
board-certified dermatologists will confirm diagnoses from the study. The review process
involves three phases: initial image assessment, review of redacted medical records, and
consideration of AI analysis results. The committee's consensus will determine the final
diagnosis, which will be documented for analysis. Cases lacking a unanimous decision will
be excluded from the study's final evaluation. Reviews will occur once enrollment is
complete.
