Color Health, a genetic testing company, is using OpenAI's modern, affordable, large-scale language models to speed up pre-approval applications for cancer screening diagnoses and provide doctors with pre-treatment workup expertise to get patients started on treatment sooner.
The company is also partnering with the University of California, San Francisco to study how its Cancer Copilot tool might work to surface early warning signs, seemingly contradictory red flags and other relevant details that may be deeply scattered throughout electronic medical records and other patient information.
Why is this important?
The company says that while factors vary depending on the type of cancer, trials of the technology have enabled healthcare providers to analyze patient records in five minutes.
“Primary care physicians tend not to have the time or even the expertise to risk-adjust screening guidelines for people,” said Osman Laraki, co-founder and CEO of Color Health. The Wall Street Journal I will report on Monday.
The UCSF Helen Diller Family Comprehensive Cancer Center is testing Color's Copilot for pre-treatment diagnostic testing of cancer patients by comparing it with a retrospective analysis of their medical records.
The research is still in its early stages, but a Color spokesperson said that if AI can connect the dots and ultimately reduce wait times for cancer treatment, it would be a win for patient care.
In Color's announcement Monday, Laraki said the company designed the tool to address a supply gap in oncology expertise to determine pre-treatment testing for patients with confirmed malignancies.
The goal, he explained, is to provide primary care physicians and other clinicians with an AI service that can determine what tests are needed to inform a patient's cancer treatment, without waiting for the patient to see an oncologist before a pre-treatment diagnostic is ordered and the prior authorization process begins.
“That way, by the time patients see their oncologist for the first time, they're much more likely to be ready to start treatment, hopefully saving them weeks of valuable time.”
Laraki also emphasized the role of clinicians in decision-making when using the tool.
“One of the most important design decisions behind our work is that the tools were built from the ground up around a human-involved model,” he said.
The company said it will first share test results from its initial use case, which focuses on automating the analysis of an individual's background risk factors and applying guidelines to adjust screening plans, with individuals participating in its cancer program, who will then have the opportunity to review the information with their primary care physician.
Color estimates that by the end of the year, doctors using Cancer Care Copilot will be helping more than 200,000 patients with personalized, AI-driven care plans.
Larger trends
Before focusing on tools to help doctors improve outcomes for cancer patients, Color launched its patient-led, proactive testing model in 2015. The tests focused on genes known to increase an individual's risk of cancer, such as BRCA1 and BRCA2 for breast, ovarian and pancreatic cancer.
Within a few years, the unicorn, along with 23andMe and others, had broken down previously unavailable patient barriers to cancer testing by offering low-cost, commercially available, at-home tests that could reveal key genetic risk factors.
Using AI for new decision support services to help PCPs get cancer patients started on treatment faster is a budding field in healthcare AI, with automating physician note-taking and reducing the burden of clinical management making up a large part of mainstream LLM use cases.
However, applying machine learning to health data offers a huge opportunity to improve health outcomes for individuals and populations.
Xin Wang, an assistant professor in the University at Albany's department of epidemiology and biostatistics, said AI could be useful in disease management.
“By analysing patient data over time, AI algorithms can predict risks for individual patients, suggest personalised treatment plans and even alert healthcare professionals to early signs of complications,” he said. Healthcare IT News In January.
“This proactive approach may lead to earlier intervention, better disease management and ultimately improved health outcomes.”
Be on record
“We think the AI technology, the language models, are the best,” said Brad Lightcap, OpenAI's chief operating officer. The Wall Street Journal Story: “We can provide clinicians with the tools to understand medical records, data, test results and diagnoses.”
Andrea Fox is a senior editor at Healthcare IT News.
Email: afox@himss.org
Healthcare IT News is a publication from HIMSS Media.
The HIMSS AI in Healthcare Forum is scheduled to take place in Boston on September 5-6. More information and registration available here.
