UC Davis Health is using artificial intelligence (AI) to improve the quality of colonoscopy procedures, better track how doctors are performing and reduce patients’ risk of developing colorectal cancer.
Colonoscopies are a key tool in preventing colon cancer because they allow physicians to detect and remove precancerous polyps before they become cancerous. While increasing access to screening is important, the quality of each procedure is just as critical.
“Not all colonoscopies are equal,” said Juan Carlos Garcia, medical director of gastroenterology clinical services at UC Davis Health. “The goal isn’t just to perform a colonoscopy, but to perform a high-quality colonoscopy that detects precancerous lesions early.”
One of the most important measures of colonoscopy quality is the adenoma detection rate (ADR). ADR represents the percentage of procedures in which a physician finds adenomas, a common type of precancerous polyp.
Research shows that for every 1% increase in physician’s adenoma detection rate, their patient’s risk of developing colorectal cancer after a colonoscopy drops by about 3% and dying from colorectal cancer by approximately 5%.
“The data clearly shows that early detection leads to better outcomes,” Garcia said. “Our goal is to identify areas for improvement and make sure patients receive the highest-quality care possible.”
Closing gaps in data with AI
Until recently, accurately tracking ADR required manual data entry. After each colonoscopy, pathologists had to manually flag findings such as tubular adenomas in the electronic medical record. This extra step sometimes led to missing or incomplete data, making consistent performance measurement difficult.
To address this challenge, gastroenterologists at UC Davis Health implemented an AI-supported tool within Epic, the health system’s electronic medical record. The tool automatically analyzes pathology reports and identifies the number of detected adenomatous polyps. By removing the need for manual input, the system produces more accurate, complete and reliable data.