How AI Can Close the Gap in Evidence-Based Care

Despite America being known as a “melting pot” for its diverse population, healthcare spotlights a more significant issue in diversity. Healthcare is a highly evidence-based practice that uses data in nearly every aspect, from diagnosis to treatment to pharmaceutical and medical device testing. Yet, profound racial disparities have become the “norm” due to a lack of evidence. 

Only 14 percent of daily medical decisions are made based on high-quality evidence. That evidence is derived from 30 percent of the U.S. population, as 70 percent of the U.S. population is excluded from clinical trials. 

While women make up half the population, women’s health has historically lacked funding and research. This could be in part due to the FDA’s policy in 1977, which recommended excluding women of childbearing potential from early phases of drug trials. This ultimately led to a shortage of data on how drugs can affect women until a law came into effect in 1994 that required female participation by the National Institutes of Health. 

Minority patients, including those who are Black, Brown, and Asian, are also rarely included in clinical trials, creating significant evidence gaps that result in less than favorable outcomes or stereotypes and biases reinforced based on outdated, problematic algorithms that lead to misdiagnosis and inappropriate treatments.

Black and American Indians and Alaska Natives (AIAN) have a shorter life expectancy, in addition to the highest rates of pregnancy-related mortality. Native Hawaiian and Pacific Islanders, on the other hand, is a population for which healthcare has still not been able to accurately analyze disparities because there is such a significant data gap. 

Native Hawaiians and Pacific Islanders make up less than 0.2 percent of Massachusetts’ population, for example. What would happen if someone in this demographic was admitted to the emergency room and was resistant to the typical treatment plan for the local demographic? Unfortunately, this scenario happens more often than you’d think. 

Situations like this leave patients susceptible to worsening conditions unless their care team can access larger pools of diverse data. In this example, a Boston ER could ideally pull data from a Hawaiian health system, ensuring the patient receives the most appropriate and personalized care since demographics respond to treatment differently. However, this practice can require hours of clinicians’ time to sift through research to determine which data is best for each patient.

Minorities aren’t alone, either. Rural communities are also misrepresented in evidence collection. Due to a lack of access and awareness, only one-third of clinical trial participants are from rural communities.

Rural health systems are notoriously understaffed and lack resources, which is why providing clinicians with data from metro health systems across the U.S. could not only generate more positive outcomes but, with AI, expedite care decisions and give clinicians valuable time back they would typically spend combing through research. 

Evidence is the key to filling these gaps, and AI is needed to translate evidence into real-world insights. 

Evidence is the transactional unit of health care. It’s how we decide what treatments to give, measure the benefit of those treatments, and ensure we’re providing the proper care to the right patient. Given these disparities in data access, care for minority populations lacks the evidence it needs to inform these decisions. 

Health systems and life sciences companies must reevaluate their approach to data and evidence, filling the data gaps with quality evidence that can better inform clinicians and result in more positive patient outcomes regardless of race, gender, or location. 

Fortunately, recent advancements in anonymous real-world evidence generation and innovations in AI enable companies to reevaluate existing data sets and bridge gaps with additional outsourced data. By doing so, we can increase the evidence available, making the dream of personalized medicine a reality – even for underrepresented patients. 

AI can produce evidence at scale, not just fast. These tools can run hundreds of thousands of studies simultaneously to generate extensive evidence for women, children, and other demographics, such as those with comorbidities and disease-based groups, that are notoriously underrepresented in clinical trials and other research. 

One of the biggest obstacles to evidence generation is the tedious analysis of medical records and the lengthy de-identification process. Generative AI exists to automate impending tasks, one of which is producing real-world evidence. AI can expedite this time-consuming process and provide researchers and clinicians with deidentified data that fills gaps in representation. These data sets can then be compiled and used to fuel large-language models (LLMs) with more accurate, research-grade evidence or supplement missing data for clinical decision-making, ensuring more research is available for minority populations. 

Companies investing in LLMs want to ensure that their model’s data is relevant and evidence-based. Proper evidence generation from published literature or real-world data can solve this problem. Despite some doctors using ChatGPT for clinical decision-making, general-purpose LLMs like ChatGPT are unreliable in healthcare because they aren’t fueled by real-world evidence. The data it’s sourcing from is not based on real-world evidence, thus generating inaccurate outputs. 

AI tools have also been designed to evaluate and enhance data quality, enabling health systems and life sciences companies to identify gaps and take actionable steps to bridge them, ensuring that future healthcare decisions are made with sufficient evidence. For clinicians using LLMs to inform decisions, data evaluation tools can rank the quality of the evidence generated based on how well it matches a patient’s background. It can also generate future care suggestions and include updated data for subsequent use. Data evaluation tools can even inform physicians about how well a provided dataset fits their question, revealing the trustworthiness of suggestions and unpacking any inconsistencies in responses. 

AI will enable us to offer more personalization in healthcare. With evidence generated in minutes, the days of broad-based guidelines will be gone, as researchers and clinicians will have personalized evidence at their fingertips that can transform value-based care. 

As we approach 2025, AI tools must focus on producing quality evidence. Those that do so through transparency, high-quality methodology, and the ability to get a trust-based response from clinicians, will be the ones that succeed long-term and transform evidence-based care.

Photo: Natali_Mis, Getty Images

This post appears through the MedCity Influencers program. Anyone can publish their perspective on business and innovation in healthcare on MedCity News through MedCity Influencers. Click here to find out how.