Millions of uninsured, underinsured, and Medicaid-insured Americans receive essential care within safety-net health systems. Despite their important role, these systems face pressures that can undermine high-quality patient care and limit innovation, including challenges adopting digital tools that may help manage costs, address provider burnout, and improve patient outcomes.
Many safety-net health systems face barriers to digital innovation, which contributes to concerns about a growing “digital divide.” But there is early evidence that artificial intelligence (AI)–supported tools may aid in care delivery — and contribute to improved patient outcomes — in safety-net environments. In one study, AI was used to identify patients at a safety-net hospital who were at high risk of readmission after discharge. The hospital’s population health team contacted these patients to provide outpatient management, and as a result, all-cause 30-day readmissions — and readmission disparities — went down, improving patients’ survival, narrowing equity gaps, and saving money.
Evidence on patients’ perspectives of technology like AI — especially regarding how it should be designed and used in care — remains limited. The research that does exist seldom reflects the lived experiences and priorities of patients who are underrepresented, underserved, or otherwise vulnerable. Early research suggests that patients see value in AI tools like digital scribes (i.e., tools that listen to patient–provider conversations and automatically generate medical notes) that allow for better, more engaging care experiences with their health care providers. Alternatively, the safety-net systems themselves rank administrative efficiency as the top benefit, which is indeed an urgent issue given budgetary pressures due to H.R. 1 cuts. To ensure both patients’ and providers’ voices are heard, safety-net systems need to employ inclusive approaches to vetting and adopting technology.
Ensuring Digital Tools Improve Care and Outcomes for Underserved Groups
When designed and implemented well, digital tools can help reduce inequities by improving communication, navigation, and follow-up for patients who face language, literacy, and access barriers. For example, a study using generative AI to rewrite hospital discharge summaries into more patient-friendly language improved readability, an approach that could make postdischarge instructions more usable for patients with lower health literacy.
But without careful design and governance, digital health tools can introduce or worsen inequities. Evidence suggests that underserved people may be especially likely to experience harms or reduced benefits when these tools are poorly designed, evaluated, or implemented. Algorithms trained on data that underrepresent certain communities may systematically underidentify their needs or risks. Researchers have found that algorithms that use health care costs as a proxy for medical need can underestimate illness burden for Black people, which can lead to fewer Black patients flagged for high-risk care management programs, for example. Likewise, tools with complicated informed consent processes or hard-to-access interfaces (e.g., small font sizes or poor sound quality) or tools that assume digital literacy can create additional barriers to receiving needed care.
Because safety-net systems disproportionately serve patients at high risk for the abovementioned harms, the stakes for avoiding AI-related risks are especially high in these settings. Yet safety-net organizations often have fewer resources for testing whether tools work as intended in their settings, conducting ongoing performance monitoring, and rapidly remediating issues. This makes it harder after deployment to detect and respond to bias or privacy issues and other unintended harms.
Safety-net systems need straightforward support in choosing technologies that deliver excellent, equitable care experiences and outcomes but do not add significant financial or operational burden — which is critical as operating and patient care margins narrow. Multistakeholder involvement (which must include patients) can help ensure systems do not prioritize efficiency at the expense of patient experience or equity goals, or adopt tools that perpetuate or worsen existing disparities.
Challenges and Opportunities in Safety-Net Health Systems
Systems with low digital maturity (such as limited health information technology and data infrastructure) and those located in disadvantaged and rural areas face hurdles to adopting digital technologies. Frameworks that guide equitable technology selection and implementation exist, but they can be hard for safety-net and other systems with limited resources to use in practice, requiring technical expertise or supports that can exceed staffing, time, or data capabilities.
Individual systems may lack the capacity to assess whether digital tools work equitably on their own, making shared approaches important. Learning collaboratives already support the sharing of metrics, methods, and data, but have not typically been applied to evaluating digital tools. Leveraging these models could allow for shared learning while reducing the burden to individual systems.
Looking Ahead
Digital tools are rapidly evolving and scaling. As this occurs, we must focus on three things:
- Using shared methods, evidence, and patient input to guard against equity-related harms from digital innovations, including algorithmic bias, accessibility barriers, and environmental impacts that can disproportionately affect underserved groups.
- Leveraging customizable digital tools to specifically target the unique needs and clinical challenges that underserved populations disproportionately face, with the goal of improving experience and outcomes.
- Ensuring that digital innovations are not only available to communities that can afford them, but to everyone. To make that possible, safety-net systems will need shared partnerships and practical implementation supports to adopt and evaluate digital tools for equity, safety, and ability to deliver high-quality care.
Together, these challenges point to the need for practical, equity-centered partnerships and supports that help safety-net health systems adopt, test, and sustain digital tools that truly improve health care and outcomes for the communities they serve.
