2
OPPORTUNITIES AND EARLY EVIDENCE FOR GENERATIVE ARTIFICIAL INTELLIGENCE IN HEALTH AND MEDICINE

Early experience with generative AI (GenAI) has shown promise with potentially reducing clinician burden and delays in care (Albrecht et al., 2024; Tierney et al., 2024). GenAI’s potential also extends beyond daily patient care to support biomedical discovery in areas like drug development, diagnostics, and clinical trial management. It also shows promise in areas like health equity and patient support network engagement. We discuss these specific applications below.

CHART SUMMARIZATION

Medical chart abstraction and summarization is an essential and often time-consuming part of clinical practice, and it is an area where GenAI has shown early promise for reducing burden. Recent research supports that, when clinicians are asked to summarize a chart, it takes about 7 minutes, whereas a GenAI large language model (LLM) can perform a similar task in less time. Furthermore, when comparing clinician-generated summaries to GenAI summaries, clinicians preferred the GenAI summaries more than half the time (Van Veen et al., 2024). GenAI-created summaries have also been deemed comparable to physician-created summaries in terms of completeness, correctness, and trustworthiness (Schoonbeek et al., 2024).

NOTE DOCUMENTATION

GenAI has shown potential in assisting clinicians with ambulatory clinic note generation. Using an ambient recording of the clinic visit, GenAI LLMs can process the recording and generate an encounter note that follows a designated

template format. In one evaluation of the approach, clinicians at the Permanente Medical Group and their patients reported favorable experiences with GenAI-supported note generation. They also reported that the approach generated high-quality note drafts for editing and was linked to reduced documentation time (Tierney et al., 2024). In contrast, a study from Atrium Health found no overall differences in electronic health record (EHR) or financial metrics with the use of the GenAI-supported note generation approach, although there were some small signals of benefit among low-volume clinicians, high utilizers of the GenAI-supported note generation approach, and family medicine clinicians (Liu, Hetherington, Dharod, et al., 2024). Despite these mixed results, GenAI-generated notes are being rapidly adopted (Epic, 2024; Tierney et al., 2024), so it will be important to continually assess the overall value of this approach as it becomes more widespread.

DRAFT PATIENT MESSAGING

GenAI can also assist with routine patient communications. A study performed at the University of California, San Diego (Tai-Seale et al., 2024), explored the benefits GenAI might pose for administrative clinical workflow. Physicians that participated noted GenAI-assisted patient message responses relieved clinician cognitive burden by providing an empathetic draft. However, the study also found that using GenAI was associated with an increase in read time and longer replies. Another study from the Mayo Clinic showed that nurses saved approximately 30 seconds per message when using GenAI-drafted responses, a small but valuable time savings over the course of a shift (Cacciaglia, 2024). These studies suggest that human input is still necessary as there are large improvements that need to be met with GenAI tools.

POPULATION HEALTH AND RESEARCH

GenAI is showing early potential to assist with population health analytics and research. Data queries can be conducted using conversational language as prompts, which the GenAI tool translates into data queries (Epic, 2024). GenAI can also help manage the large and ever-growing corpus of medical literature and research-derived evidence. GenAI can help clinicians stay abreast of the latest evidence-based practices and clinical guidelines by synthesizing and summarizing current research findings accurately, ensuring that patient care decisions are

informed by the most current scientific knowledge. However, this use of GenAI is also subject to hallucinations, so development and evaluation of this capability will need to mitigate that risk.

PRIOR AUTHORIZATIONS AND INSURANCE COMMUNICATIONS

Another source of clinical administrative burden relates to the common work of requesting prior authorization for a clinical test or trying to contact insurance companies on behalf of a patient who had a test denied. GenAI has the potential to reduce the burden of these tasks by helping draft request letters and denial appeals. Rather than foraging for the specific data that support the denial appeal, GenAI tools can abstract them from a patient’s chart and compile it into an appeal letter (Kirby, 2024).

CLINICAL DECISION SUPPORT

While GenAI’s current role in enabling clinical decision support (CDS) is nascent, in part due to concerns related to errors and bias (Ratwani et al., 2024), there are emerging roles for GenAI in the creation of explainable CDS (Liu, McCoy, and Peterson, et al., 2024) and in assessing and optimizing CDS to make it more useful and less burdensome (Liu, McCoy, and Wright, et al., 2024). Over time, we anticipate that the use of GenAI in enabling clinical decision making via supportive approaches will continue to evolve.

CLINICAL TRIALS

GenAI has significant potential in enhancing clinical trial processes by automating tasks like trial design, evidence summarization, and regulatory compliance. It has the potential to streamline participant recruitment, improving enrollment speed and diversity (Carroll and Anderson, 2024; Gangwal et al., 2024; Merk et al., 2018; Ng, 2024). For example, its capabilities to standardize and analyze unstructured text from clinical notes or reports hold potential for increasing the efficiency and reliability of research data (Hutson, 2024). GenAI also supports patient education by presenting enrollment information in multiple languages and comprehension levels.

DRUG DISCOVERY AND REPOSITIONING

In drug development, GenAI can facilitate the identification of novel molecular structures, potentially enabling faster development of new therapeutics, such as targeted antibody therapies (Marinov et al., 2024). Early evidence also indicates the capacity to identify not only new molecules that might be created, but new applications for existing medicines (Yan et al., 2024).

DIAGNOSTICS AND DISEASE MONITORING

GenAI also has the potential to accelerate and improve diagnostic efficiency and accuracy. Early evidence shows promise in improving diagnostic accuracy, such as differentiating forms of dementia (Xue et al., 2024), and to assist clinicians in developing differential diagnoses in cases that require complex diagnostic reasoning (Kanjee et al., 2023). GenAI applications may also have potential for enhancing monitoring of conditions like hypertension through enabling improved patient engagement (Andreadis et al., 2024).

HEALTH EQUITY

GenAI has the potential to significantly impact health equity by addressing various disparities and barriers to health care access. First, given current access limitations for many people and populations, equity can be positively affected by GenAI solutions that can either directly enable care interactions or indirectly increase the availability of clinicians to see more patients due to decreased burden.

Second, GenAI’s capabilities can also bridge communication gaps between health care providers and patients. Such tools can translate educational materials and health information into multiple languages, enabling non–English speaking patients to receive information in their preferred language. This has the potential to improve understanding and even build trust between a patient and a provider, which can lead to higher treatment adherence levels. GenAI can also tailor health care information and interventions to the specific cultural contexts and social determinants of health affecting individual patients or communities. This personalized approach acknowledges and addresses the broader factors influencing health, from socioeconomic status to environmental conditions, thereby promoting more equitable health care delivery.

Third, GenAI—as previously discussed—can improve reimbursement for health care services through prior authorization and claims adjudication generation. In addition, GenAI tools such as patient-facing chatbots can improve appointment

scheduling processes, allowing more patients to find health care encounters that match their schedules and circumstances (Clark and Bailey, 2024). Both of these applications may make health care services more accessible and reduce administrative burdens that disproportionately affect lower-income patients.

Finally, GenAI trained on large datasets can also incorporate information gleaned from training data and help identify and illuminate health disparities that exist across populations. By analyzing and summarizing data, GenAI can uncover patterns of inequity in disease prevalence, treatment outcomes, and access to care. This insight can guide health care providers and policy makers in targeting interventions and resources to underserved communities, thereby addressing gaps in care. That said, GenAI could worsen equity as well, given the risks of using non-representative or biased data in generating its output or applying its output in inequitable ways. As such, careful evaluation and monitoring of GenAI models and their training sets is needed to mitigate this potential risk.

Although these applications of GenAI hold promise in improving health equity, empiric evidence of this impact is currently absent and validation of these approaches is needed. Accordingly, it will be important to continually monitor the use of GenAI in the context of health equity and address any inequities to which it may be contributing.

PATIENT AND SUPPORT NETWORK ENGAGEMENT

GenAI can potentially increase patients’ and their support networks’ engagement in health care by providing personalized, accessible, and interactive tools for health management and education. By leveraging its capabilities in natural language processing and data analysis, GenAI can transform complex medical information into understandable, patient-friendly formats. This demystification of medical jargon plays a crucial role in helping patients, their families, and their broader support network comprehend diagnoses, treatment plans, and health guidelines, leading to better informed and more engaged patients.

Specifically, GenAI can assist through personalized patient education. GenAI systems can tailor educational content to individual health conditions and learning preferences, making information more relevant and engaging. For example, GenAI can generate patient information for a personalized diet plan, exercise suggestions, and information on managing blood sugar levels, all in easy-to-understand language. Additionally, GenAI-powered chatbots and virtual health assistants can provide 24/7 support to patients and their support network by answering queries, providing health tips, and reminding patients about medication and appointments. This constant availability is especially beneficial

for chronic disease management, where ongoing monitoring and adherence to treatment are crucial. Importantly, these chatbots and health assistants will require ongoing clinical oversight and monitoring to ensure adequate performance. For family members and other members of a patient’s support network that serve as a caregiver, GenAI can provide educational resources and support tools, helping them understand and manage their loved one’s health conditions effectively. Finally, GenAI-enabled platforms can connect patients and networks with additional support networks and communities, providing emotional support and shared experiences with others in similar situations.