Key takeaways:
- OpenAI and Color Health, a cancer treatment manager, developed an application that can speed up development of treatment plans for cancer patients.
- The app uses GPT-4o to read through treatment protocols and pair them with patient data to come up with a customized treatment plan that is much faster and more complete than those from human providers.
Cancer care manager Color Health is partnering with OpenAI to speed up the development of cancer treatment plans for patients, where every delay can mean the difference between life or death.
They developed an application that uses OpenAI’s GPT-4o to identify missing diagnostics and generate tailored workup plans, enabling health care providers to make evidence-based decisions on cancer screening and treatment.
To screen, diagnose and treat cancer is “notoriously complex and time-consuming,” according to an OpenAI blog post. It said a treatment delay of just four weeks can raise mortality risk by 6% to 13%.
By using OpenAI’s APIs, Color Health integrates patient medical data with clinical knowledge to create a copilot application that formulates comprehensive treatment plans for health care providers to review and use in patient care.
“Color’s vision is to make cancer expertise accessible at the point and time when it can have the greatest impact on a patient’s health care decisions,” said Othman Laraki, CEO of Color Health.
The copilot application processes patient information, such as family history and risk factors, along with clinical guidelines and data from trusted sources. The technology extracts information buried within inconsistently structured and phrased documents, such as PDFs or clinical notes.
It answers key questions like, “What screenings should the patient be doing?” to identify missing diagnostics and generate a personalized screening plan, including the necessary documentation for diagnostic workups.
A clinician reviews and, if necessary, modifies the application’s output before presenting it to the patient. This clinician-in-the-loop workflow ensures the accuracy and safety of the generated plans, which can then be integrated into the patient’s existing treatment plan.
Initial testing showed that the copilot application was able to find 4x more missing labs, imaging or medical results. It also takes an average of five minutes to analyze patient records and identify gaps. “Without the copilot, data is fragmented and can lead to weeks of delay,” they said.
Next, Color Health is partnering with the University of California, San Francisco’s Helen Diller Family Comprehensive Cancer Center to conduct a retrospective evaluation followed by a targeted rollout, potentially integrating the copilot into clinical workflows for all new cancer cases at UCSF.
Color Health is taking a measured approach to rolling out the copilot, starting with an initial phase-in for its own clinicians and applying the tool to a limited number of cases. The goal is to provide AI-generated personalized care plans, under physician oversight, for over 200,000 patients through the second half of 2024.
Color Health’s collaboration with OpenAI began in 2023, focusing on using AI to enhance cancer patient care. During their initial exploration, Color Health tested the performance of GPT-4 and GPT-4o in extracting information from complex documents like PDFs of clinical guidelines. OpenAI provided guidance on using retrieval-augmented generation (RAG) to improve output quality and streamline clinical documentation processing.
