Analyze Data Using AI Agents for Research

Use Oracle AI Agents for Research to analyze authorized Oracle Health Real-World Data (RWD) and approved customer-provided data to develop research.

Oracle Life Science AI Data Platform's (AI Data Platform) AI Agents for Research is a capability that users access from an approved AI client (ChatGPT Business or Claude Enterprise) through a model context protocol (MCP) connector. Researchers can submit questions about Oracle Health Real-World Data (RWD) to build cohorts and conduct outcomes analyses. It is designed to analyze your research questions and develop answers in the form of cohort counts, result tables or charts, datasets, and summaries. Depending on the Agent's understanding of your initial question, the Agent may respond with a follow-up question for clarification.

Before AI Agents for Research can be used, an authorized IT administrator must configure identity and access, set up the AI Data Platform MCP connector or plugin in an approved AI client, and make it available to authorized users. See the Oracle Life Sciences AI Data Platform Administrator Guide for more information.

Users can then find the connector or plugin in the AI client and connect to it. Once connected to AI Agents for Research, you can start asking the Agent your research questions. AI Agents for Research develops answers by analyzing data from RWD and Bring Your Own Data (BYOD) datasets, and can generate outputs such as tables, charts, summaries, and derived datasets.

AI Agents for Research has the following capabilities:

  • Building a Cohort
  • Running Outcomes or Comparison Analyses
  • Reasoning over BYOD
  • Saving Derived Datasets
  • Multi-Turn Chat Support
To use AI Agents for Research:
  1. Open an approved AI client, such as ChatGPT Business or Claude Enterprise.
  2. Access the configured AI Data Platform connector or plugin and made available by your IT administrator, and connect to the connector or plugin.
  3. Write your question and select Submit to ask the question to the AI client. The AI client analyzes your question and answers.
  4. Review the AI Agent's answer and ask clarifying questions or present prompts as needed.
  5. Save your interim data by asking the AI Agent to save the data to a specified location. The AI Agent does not save interim work automatically.
When asking a research question, use the following techniques to receive more detailed results:
  • Use natural language to ask questions. The AI Agent may show back-end tool calls while the workflow is running, but you do not need to call each tool directly.
  • Specify a data source for the AI Agent to use or ask the Agent to show the proposed dataset before analyzing. BYOD datasets must not be joined directly to RWD.
  • Follow-up the AI Agent's answer with prompts to inspect the result, refine the workflow, or prepare the output for reuse.

Note:

AI Agents for Research is a learning tool used to help you analyze research and develop a data plan. All AI Agent responses must be checked for accuracy.

See the questions and prompts below for an example of questions and prompts to ask the AI Agent.

How many patients since 2015 have been diagnosed with Chronic Kidney Disease Stage 3-5? give the total, then give the count for each discrete stage (stage 3, stage 4, stage 5) then for the total patient count for stage 3-5 how many have had the following test: EFGR OR Serum Creatine? 
I want to understand the sample size of Interstitial lung disease in total and broken out by drug induced ILD and pneumonititis ILD. From those three cohorts what are the number of patients that have a prescription for Enhertu.
Create a cohort of adults with type 2 diabetes without complications. 
Include patients with at least one diagnosis in the last 24 months. 
Exclude patients with type 1 diabetes.
Compare 12-month persistence for patients initiating Drug A versus Drug B. 
Show the analysis plan before running it.
Save this cohort as a derived dataset.
Using my uploaded claims dataset [dataset_name] identify GLP-1 users who discontinued therapy. Discontinuation is defined as no GLP-1 claim within 90 days after the member’s last GLP-1 claim. Summarize aggregate discontinuation counts by medication.

Parent topic: Analyze Oracle Health Real-World Data