Research Protocols

Healthcare GenAI Research Ideas

We have compiled a list of healthcare GenAI research ideas for your consideration. The degree of innovation in these ideas may vary. Some ideas are brand-new, and some already have publications in the literature. Regardless, an applied GenAI research project is about solving a specific problem or answering a specific question. You will need to clearly define the specific research question, design experiments to scientifically measure the effect of GenAI, and generate new evidence for how GenAI can benefit healthcare.

Comparative Effectiveness Research (CER) ideas for evaluating GenAI LLMs in clinical settings:

Choose any task along the healthcare delivery pathway that requires decision-making; use a GenAI model (e.g., ChatGPT) to analyze the case and provide additional information for making decisions; evaluate the accuracy of AI analysis; determine whether the AI analysis is helpful for the task. You may compare different LLMs, commercial or open-source, or both types. The study should include enough cases for statistical analysis of the results. Before carrying out a prospective study, it is recommended to do a retrospective study first with historical data to determine the study's feasibility.
The task may be the diagnosis of one or multiple diseases.
The task may be treatment selection or personalized medication for one or multiple diseases.
The task may be risk prediction for early detection of one or multiple diseases.
The task may be the management of one or multiple diseases.
The task may be patient education and engagement regarding one or multiple diseases.

Ideas for doctors to turn their best practices into GenAI for broader impact:

If you have developed any best practices and want to share them within your department, hospital, or medical community at large, you can fine-tune an open-source LLM with your data and conduct a CER study to prove that the new GenAI capability can help other doctors improve healthcare deliveries. You can compare your fine-tuned LLM to the top commercial LLMs (ChatGPT, Gemini, Ernie). Ideally, your specialized fine-tuned model will perform better than the best general-purpose LLMs for the patient populations you serve.
Your best practices may focus on diagnosis, treatment, disease management, prevention, early detection, etc. Fine-tuning a top open-source LLM like Llama or Gemma is a proven approach to create GenAI abilities from specialized data. Your LLM can be deployed in your environment to optimize or automate your healthcare delivery in compliance with patient data regulations.
After turning your best practices into GenAI in the form of your own fine-tuned LLM, you can study the effectiveness of your GenAI in disseminating the best practices across your clinical research network (CRN).
Furthermore, you can include physicians from community and rural health centers in your CRN, enabling patients in low-resource areas to access a high standard predictive care. Within CRN, you can study whether and how your GenAI can improve the participating doctors’ routine healthcare delivery. The study can measure care access and patient outcomes in different populations, characterizing the progress toward establishing an equitable learning health system (LHS). Your study results may demonstrate for the first time the effectiveness of the GenAI-powered LHS over CRN in your clinical focus.
Once your GenAI-LHS-CRN service model is proven, you may share your innovation with the global medical community, build international collaboration to study whether and how this GenAI-LHS-CRN service model can improve predictive healthcare access and patient outcomes in other parts of the world, and ultimately help reduce global healthcare disparities.

Ideas for doctors to bring GenAI capabilities into practices:

If you want to apply GenAI in your healthcare practices but cannot use the chatbots in the public domain, you can fine-tune an open-source LLM and deploy it in your environment to comply with patient data regulations. You can conduct a CER study to prove your LLM’s GenAI capabilities are effective in helping your practices. You may also compare your LLM to the best performing commercial LLMs like ChatGPT, Gemini, and Ernie.
The GenAI capabilities you can bring into practices include diagnostic prediction, treatment selection prediction, symptom checking prediction, disease management decision prediction, disease risk prediction, etc.

Ideas for medical students doing thesis research:

During clinical training in rotation or residency, any medical student can use one or multiple LLMs to analyze patient cases and study whether GenAI’s information is helpful to decision-making. The student can compare the decisions before and after consulting GenAI analysis. The decision-making may involve diagnosis, treatment selection, risk screening, disease management, etc. This CER study may include one or multiple diseases. The study results will provide real-world evidence for what diseases the latest versions of LLMs may have a beneficial effect on in the healthcare task under study.
GenAI chatbots can provide guideline-like knowledge for students to learn about every aspect of each disease, diagnosis, treatment, prevention, etc. However, GenAI knowledge is expected to contain some random errors. Any student can generate GenAI knowledge from a top LLM and use a second LLM to identify potential errors from the first LLM.
From errors found in GenAI knowledge across a series of diseases, the medical student can further characterize the errors and try to find potential rules for correcting the errors.

Ideas for patient communication tasks:

Applying LLMs to automate responses to patient inquiries.
Applying LLMs to manage referrals: sending referrals, tracking the status of referrals.
Applying LLMs to generate personalized educational materials and instructions for patients regarding their conditions and treatments.
Applying LLMs to automate patient follow-up messages and surveys to assess patient satisfaction and outcomes.

Ideas for clinical documentation tasks:

Apply LLMs to transcribe and summarize physician-patient interactions, automatically generating clinical notes and discharge summaries.
Apply LLMs to assist in ensuring that clinical documentation is complete, accurate, and compliant with coding standards.

Ideas for quality improvement tasks:

Developing GenAI agentic process to monitor patient care, including diagnoses and treatments, alert potential errors, and make correction recommendations.
Developing GenAI agentic process to analyze patient data to predict outcomes, identify at-risk patients, and suggest preventive measures.
Developing GenAI agentic process to automate analysis of patient data to identify trends and provide insights for decision-making.

Ideas for GenAI applications in hospital administrative tasks:

Applying LLMs to automate appointment tasks: scheduling, reminders, cancellations, etc.
Applying LLMs to automate medical coding: translating clinical documentation into standardized codes required for billing and insurance claims.
Applying LLMs to identify Diagnosis-Related Groups (DRGs) automatically from EHR records.
Applying LLMs to review and validate billing claims for errors or inconsistencies.
Applying LLMs to ensure that clinical practices and documentation comply with healthcare regulations and standards.
Applying LLMs to create materials for employee training and education.