Improving Care Access and Navigation Among Underserved Patients

Oren K. Fix, MD, from the University of North Carolina at Chapel Hill School of Medicine, described the digital gap in patient care, created by a lack of access, affordability, digital literacy, health literacy, which has further exacerbated disparities in health outcomes. While the COVID-19 pandemic led to increased use of both telehealth and patient portals that improved doctor-patient communication, self-management, disease awareness, medication adherence, and even measurable improvements in disease biomarkers, these tools are not available to everyone, and in some patient populations, the digital shift worsened disparities. One in four Americans do not have the digital literacy skills or access to technology needed to engage in telehealth or patient portals. Digital equity is necessary for health equity, and internet access should be included as a social determinant of health that is screened regularly. Digital navigation could be built into care delivery, while also preserving non-digital access pathways and providing interpreter services and screen readers for websites. Clinics can engage community partners such as libraries, community colleges, and senior centers to increase digital literacy at the population level. At the policy level, affordable connectivity and technical support is needed. As a first step, Epic Cosmos, which collects deidentified data on digital access by 300 million patients, is being analyzed by various patient-level factors (e.g., race/ethnicity, age, rural, insurance, area deprivation index, social vulnerability index) to identify barriers and implement solutions that will start to close the digital gap.

Ashley Spann, from Vanderbilt University Medical Center, discussed opportunities to use AI to address health disparities. It is crucial that users first understand the risk of bias in AI model development, with the need to select input data that will lead to accurate and actionable predictions without introducing new ethical issues. For disease detection, AI can be used to scan EHR data for word frequency in a single patient chart or across charts. This approach can be used to identify individuals who might have undiagnosed disease. AI can also be used to improve diversity in clinical trial recruitment and to expand and accelerate referral to clinical trial coordinators. The RecruitEHR dashboard at Vanderbilt pre-screens patients based on demographics, taking 15 days from identification to randomization compared to 50 days for manual screening, and with higher accuracy (less frequent selection of ineligible candidates). In another example, AI is being used to turn the EHR into the trial coordinator, which handles candidate identification, screening, risk stratification and randomization, and produces decision aids for patient assessment. With this approach, target enrollment can be scaled, up to 7200 participants in this case. Finally, AI can be used to improve healthcare access in the setting of limited resources, through chatbots that prompt patients to improve adherence or provide support. Studies have shown that these tools can provide patients with misinformation and guardrails are needed to ensure AI tools do not exacerbate disparities or cause patient harm.

Prof Seth Gross, Sr, from NYU Langone Health, shared some cautionary tales and limitations of using AI in patient care, advising clinicians to be mindful of becoming over-reliant on AI tools and risking the loss of practical skills, critical thinking, clinical creativity, and ethical decision-making. AI is programmed to satisfy the user, in some cases by providing a wrong answer (hallucination). AI also tends to overgeneralize when used at scale. Human oversight is needed in all cases. In patient care, additional concerns revolved around data privacy and security and implementation of tools with false information. Healthcare providers need better training, starting in medical school, to use AI safely and reduce the risk of medical errors. Health systems must ensure that AI tools are safe, secure, and private, that the data inputs and algorithms have low bias and high transparency, and that tools are consistently monitored to ensure accuracy. Health systems, healthcare providers, and tool developers can be at risk of “AI liability” for medical malpractice, ordinary negligence, and product liability.

Even before overcoming the digital gap, healthcare providers must overcome distrust among vulnerable patient communities and work to build relationships that will improve access to GI and hepatology care. Joan Culpepper-Morgan, MD, of NYC Health + Hospitals/Harlem, discussed how doctors today are more often questioned about their competence and ethics, and healthcare systems are met with skepticism about corporate and government influence. The insurance industry and the view of healthcare as a consumer product have further eroded trust, creating a more adversarial relationship between doctors and patients. Black, Hispanic, rural, and female patients report feeling judged, dismissed, and labeled non-adherent, when in reality they face significant barriers to care and need both a partner and a plan. In this setting, active listening leads to shared decision-making, which leads to trust. Use of patient navigators, plain language and graphic instructions, and community outreach can improve the care experience and health outcomes. It is important that providers acknowledge generational and personal trauma and cultural beliefs about healthcare, and understand the source of patient fears about pain and the loss of control. Concordant and culturally competent care provided by a diverse workforce has been shown to improve patient trust, adherence to care, and healthcare utilization.