Color Health, a company that’s spent a decade trying to untangle access to healthcare, is now wielding OpenAI’s GPT-4o to do something genuinely difficult: build personalized cancer screening and treatment plans. The new copilot application, detailed today, is already being used to ingest messy, real-world patient data — family histories buried in PDFs, risk factors scattered across clinical notes — and identify gaps in diagnostics. It then generates a tailored workup plan for a clinician to review. Color serves over 7 million patients and partnered with the American Cancer Society in 2023 to tackle cancer, which remains the second leading cause of death in the U.S. and the top driver of healthcare costs.
The technical challenge here isn’t trivial. The team was particularly impressed with GPT-4o’s ability to extract and normalize information from inconsistently structured documents, including hundreds of pages of clinical guideline diagrams. They developed a method using GPT-4 Vision to interpret screenshots of these complex care pathways, a task that would otherwise require painstaking manual review. The system also generates the administrative scaffolding for care, like insurance pre-authorizations and medical necessity documents. Color’s CEO, Othman Laraki, frames this as making expertise accessible at the critical moment of a patient’s decision-making, while emphasizing that OpenAI’s HIPAA-compliant standards are non-negotiable for safety and privacy.
A doctor is still very much in the loop here. Every output is evaluated and can be edited by a clinician before it reaches a patient, with those edits feeding back to refine the model. This isn’t a chatbot replacing a physician; it’s software taking a first pass at a grindingly slow administrative process. Dr. Keegan Duchicela, a Color primary care physician, pointed to the constant evolution of guidelines and unclear individual risk factors as a core problem this addresses. More than a third of Color’s patients require earlier or different screening than standard guidelines suggest.
The stakes are measured in weeks of lost time. Research indicates that a four-week delay in cancer treatment correlates with a 6–13% higher mortality risk. Dr. Allison Kurian, a Stanford oncologist, underscored the real-world consequence: many patients arrive at their first oncology appointment without a complete diagnostic workup, losing precious time while clinicians drown in administrative burden. Color’s bet is that by shrinking the weeks-long process of documenting a workup, they can change the starting line for treatment before a patient ever sees an oncologist.
