Health Apps Can Embed Bias Before Coding Begins
Digital health tools such as period trackers and fitness apps are often built on datasets that underrepresent certain groups, leading to biased algorithms and potential harm. The authors argue that these biases can create self-reinforcing loops that worsen health inequities. They call for inclusive conversations and design decisions before any code is written.
The article notes that digital health spans period trackers, fitness tools, mental-health quizzes, wearables and diagnostics, frequently powered by AI. Their training data and creators' assumptions shape them. Skin-diagnosis systems, for instance, have long relied heavily on lighter skin images, a known problem for nearly ten years that can lead to errors for darker-skinned patients.
A U.S. hospital algorithm illustrates the stakes: it referred extra care using spending-based risk scores. Fewer than 18% of recipients were Black, though over 46% would have been expected. Because Black patients often spend less due to socioeconomic factors and health-system mistrust tied to systemic racism, they needed greater illness to be flagged.
If these concerns are addressed, patients from underrepresented groups could benefit from tools that better reflect their bodies, histories and needs. Developers, funders and health systems may face pressure to involve marginalized communities earlier and examine hidden assumptions. Without such changes, biased algorithms could reinforce existing inequities, potentially affecting diagnosis, referrals and trust in care. The impact may be greatest for people already underserved, while inclusive design could improve safety and usefulness for broader populations.