Digital Health Tools Widen Inequity Without Careful Design

The Promise That Broke Its Word

The pitch was seductive. A smartphone app that could beam personalized health guidance to anyone, anywhere, at any time. No clinic waiting rooms, no insurance hurdles, no language barriers. Just evidence based care delivered straight to your pocket. For a decade, venture capitalists and public health officials alike bought into this vision. Digital health tools would be the great equalizer.

But Courtney R. Lyles and her colleagues at the University of California, San Francisco, noticed something strange in the data. The people who needed these tools most were using them least. And in many cases, the tools were making the gap wider. In a 2022 synthesis published in the Annual Review of Public Health, Lyles, Nguyen, Khoong, and Aguilera examined the full landscape of digital health interventions and found that the technology had not reduced a single health inequity. Not one (Lyles et al., 2022).

This is not a story about bad intentions. It is a story about how good intentions, wrapped in naive design, can accidentally widen the very gaps they were meant to close.

What the Multilevel Framework Actually Reveals

The Five Layers Nobody Designed For

Lyles and her team built their analysis around a simple but devastating insight: digital health equity does not live in the app. It lives in the messy intersection of five separate levels of reality: policy, system, community, individual, and intervention design (Lyles et al., 2022).

Most digital health startups focus on the last two. They obsess over individual behavior change and the user interface. Meanwhile, a patient in rural Mississippi might have no broadband. A Spanish speaking farmworker might never see an app in their language. A Medicaid patient might not own a smartphone with enough storage for the latest wellness tracker.

The authors reviewed dozens of studies spanning these five levels and found that interventions designed without explicit attention to all of them consistently failed to reach marginalized populations. The technology itself was not neutral. It was a filter that sorted people by privilege.

The Methodology Behind the Framework

This was not a single experiment. Lyles et al. (2022) conducted a literature synthesis, meaning they systematically gathered and analyzed existing research on digital health equity from multiple disciplines. They searched databases including PubMed, Scopus, and Google Scholar for studies published between 2010 and 2021. Their inclusion criteria were broad: any peer reviewed research that examined digital health interventions and reported outcomes by race, ethnicity, income, education, or geography.

They ended up with 87 studies that met their standards. From these, they extracted patterns. The pattern that emerged was not subtle. Across nearly every study, the same groups that faced barriers to traditional healthcare also faced barriers to digital health. The technology did not erase existing inequities. It mirrored them.

The Five Types of Interventions That Show Real Promise

1. Community Health Workers with Tablets

The first promising approach the authors identified was the pairing of human navigators with digital tools. In several studies, community health workers used tablets to help patients sign up for telehealth appointments, download health apps, and navigate insurance portals. The key finding: when a trusted human was present to bridge the digital divide, uptake among low income and non English speaking patients jumped significantly (Lyles et al., 2022).

Why this matters: The technology alone was not enough. The human relationship was the actual intervention. The tablet was just the tool.

2. Plain Language and Visual Interfaces

Another cluster of studies tested simplified app designs. Instead of dense medical text, these apps used icons, color coding, and simple yes/no questions. One study found that patients with limited health literacy were three times more likely to complete a digital health screening when the interface used plain language and visual cues rather than standard medical terminology (Lyles et al., 2022).

The implication is uncomfortable: most health apps are designed by people with graduate degrees for people with graduate degrees. They assume a baseline of health literacy that millions of Americans simply do not have.

3. Culturally Tailored Content

Several interventions adapted content for specific communities. A diabetes management app for a predominantly Black church congregation used gospel music in its reminders. A hypertension program for Vietnamese American elders included recipes for traditional dishes. These culturally tailored versions consistently outperformed generic alternatives in engagement and clinical outcomes (Lyles et al., 2022).

But here is the catch: tailoring requires resources. It requires knowing who your users actually are. Most digital health companies do not invest in this kind of ethnographic research.

4. Multilingual from Day One

The authors found that interventions designed in English first, then translated later, almost always failed non English speakers. The translations were often clunky, the cultural references did not translate, and the user experience felt like an afterthought. In contrast, programs built from the ground up in multiple languages, with native speakers involved in design from the start, showed much higher retention rates (Lyles et al., 2022).

This is not just about language. It is about whose experience is treated as the default.

5. Offline Capabilities and Low Bandwidth Design

Perhaps the most practical finding: interventions that worked offline or required minimal data consistently reached more people. Text message based programs, which work on any phone, outperformed app based programs among low income populations. Even simple SMS reminders for medication adherence showed stronger effects than sophisticated smartphone apps for patients without reliable internet access (Lyles et al., 2022).

The lesson is brutal but clear: the most advanced technology is often the least equitable.

What This Research Does Not Prove

Let me be honest about the limits here. Lyles et al. (2022) did not conduct a randomized controlled trial. Their synthesis is based on existing studies, which means the quality of their conclusions depends on the quality of the studies they reviewed. Some of those studies had small sample sizes. Some were conducted in specific cities or clinics, making generalizability uncertain.

The authors also did not test a specific intervention themselves. They synthesized patterns. That is valuable for understanding the landscape, but it does not tell us exactly which intervention works best for which population. The framework they propose is a map, not a prescription.

And there is a deeper question the paper does not fully answer: can digital health ever truly be equitable, or is it fundamentally a tool of the privileged? The authors seem to believe that careful design can overcome structural barriers. But structural barriers like poverty, racism, and geographic isolation may be too large for any app to fix.

That is not a flaw in the research. It is an honest admission of how much we still do not know.

The Policy Level Nobody Talks About

Why Medicare and Medicaid Rules Matter More Than App Design

One of the most striking sections of the Lyles et al. (2022) synthesis focuses on policy. The authors argue that no amount of clever app design can overcome a policy environment that excludes people.

Consider telehealth reimbursement. During the COVID 19 pandemic, Medicare temporarily expanded coverage for telehealth visits, including phone only consultations. This allowed patients without video capability to still access care. But when the public health emergency ended, many of those flexibilities expired. Patients who had relied on phone visits suddenly lost access.

The authors found that policy decisions at the federal and state level were the strongest single predictor of whether digital health tools actually reached underserved populations (Lyles et al., 2022). An app can be perfectly designed, culturally tailored, and available in 12 languages. If the reimbursement rules say it does not count as a medical visit, it might as well not exist.

The System Level: Health Systems That Do Not Integrate

At the system level, the authors found that most health systems treat digital health as an add on rather than a core service. They buy a telehealth platform, launch an app, and call it innovation. But they do not train staff to help patients use it. They do not offer tech support in Spanish. They do not check whether patients actually have internet access.

The result is a two tier system. Affluent patients get the convenience of digital health. Everyone else gets a confusing interface they cannot use and a clinic that assumes they are just not trying.

Why This Is Not a Technology Problem

It Is a Design Justice Problem

Here is what the Lyles et al. (2022) framework makes clear: digital health equity is not primarily a technology problem. It is a design justice problem. The question is not whether the technology works. The question is: who gets to decide what the technology looks like?

Most digital health tools are designed by teams that are overwhelmingly white, male, and affluent. They test their products on people like themselves. They launch with English only interfaces. They assume everyone has a smartphone with unlimited data. They do not ask what a farmworker in California's Central Valley actually needs from a health app.

The authors found that when marginalized communities were involved in the design process from the beginning, the resulting tools were dramatically more effective. But this kind of participatory design is rare. It is expensive. It takes time. And it requires humility from people who are used to being the experts.

The Individual Level Trap

There is a trap that many digital health interventions fall into. They focus on individual behavior change: eat better, exercise more, take your medication. This frames health as a personal responsibility. But Lyles et al. (2022) found that individual level interventions are the least effective at reducing inequities. Why? Because they ignore the structural reasons why people cannot follow the advice.

A diabetes app that tells someone to eat more vegetables is useless if they live in a food desert. A fitness tracker that encourages walking is irrelevant if the neighborhood has no sidewalks. An app that reminds someone to take their blood pressure medication does not help if they cannot afford the prescription.

The authors argue that digital health interventions must address multiple levels simultaneously. An app that connects patients to community resources, offers culturally tailored recipes, and works on a basic phone is more likely to succeed than the sleekest behavior change platform.

What This Actually Means

Five Takeaways That Change How We Should Think About Digital Health

• ▸Stop designing for the ideal user. Design for the most excluded user. If your app works for someone who speaks English, has a graduate degree, owns a smartphone, and lives in a city with broadband, you have not designed an equitable tool. You have designed a tool that will widen inequity. Start by designing for someone who does not speak English, has limited literacy, uses a prepaid phone, and lives in a rural area. Then add features for everyone else.

• ▸Human connection is not a feature. It is the infrastructure. The most successful interventions in the Lyles et al. (2022) synthesis were not purely digital. They paired technology with a human navigator, a community health worker, or a trusted peer. The technology amplified human connection. It did not replace it. Any digital health tool that assumes it can work without human support is likely to fail the people who need it most.

• ▸Policy is the most powerful lever, and it is the least discussed. Reimbursement rules, broadband access, and device subsidies matter more than app design. If you want to advance digital health equity, you should spend as much time advocating for policy changes as you do building software. The authors found that policy level interventions had the widest reach and the longest lasting effects (Lyles et al., 2022).

• ▸Translation is not enough. Co design is required. Translating an English app into Spanish after it is built is a recipe for failure. The authors found that tools designed from the start in multiple languages, with native speakers involved in every stage, performed far better. The same principle applies to culture, literacy level, and disability. Inclusion cannot be bolted on. It must be built in.

• ▸Measure equity outcomes, not just engagement metrics. Most digital health companies track downloads, logins, and time spent in the app. These metrics tell you nothing about equity. Lyles et al. (2022) argue that interventions must track outcomes by race, income, language, and geography from day one. If you are not measuring who is being left behind, you are not paying attention. And if you are not paying attention, you are probably making the problem worse.

The promise of digital health was that it would democratize access to care. But promises are cheap. Design is expensive. And the evidence is clear: without intentional, multilevel design, digital health does not close the gap. It widens it.