This study explores whether an LLM, a type of artificial intelligence (AI) that processes and generates text in response to written questions and prompts, can help improve clinician decision-making in Kenya. LLMs have the potential to assist clinicians in making more evidence-based and informed decisions, especially when time and resources are limited. This study will evaluate the effectiveness of an LLM-based clinical decision support tool at Penda Health, a primary health care provider in Kenya.

The study aims to evaluate if using an LLM can help reduce the number of patients who need to return to health care providers for unresolved health problems or need emergency care. We are also evaluating whether this tool can improve care for certain conditions, such as high blood pressure, diabetes, malnutrition in young children, and antibiotic prescribing in infectious diseases. Since these conditions are common but often go untreated or misdiagnosed, providing clear and accurate information to health care workers could make a significant difference.

To test this, patients visiting Penda Health clinics will be assigned to one of two groups. In one group, clinicians will use the LLM to support their decisions and clinicians in the other group will not use the LLM. After their visits, patients will be contacted on days 3 and 14 to check if their symptoms have improved, if they had to seek additional care, or if they had other safety concerns. We will also collect information about how satisfied patients felt with their care.

An independent panel of medical experts will review how well the LLM’s advice matched safe and effective clinical practices. We will determine if using the LLM influences health care workers’ decisions to refer patients for more advanced care. Additionally, we will review changes in the frequency of antibiotics and malaria medication prescriptions. Finally, we will look at how patients feel about the LLM-assisted care compared to regular care, particularly in terms of clarity and thoroughness.

By the end of this study, we hope to understand how well this AI tool works in primary care settings in Kenya and whether it can safely support health care workers and improve patient care.