Colonoscopy is the predominant method for screening for colorectal cancer in the US. Yet, its effectiveness in screening is limited by variation in performance. For example, the rate at which physicians detect cancer precursors called adenomas during a colonoscopy has been shown to vary three-fold from one physician to another. A patient whose colonoscopy is performed by a physician with a low adenoma detection rate has a higher risk of subsequent colorectal cancer. Our proposal centers on measuring, understanding, and improving colonoscopy quality. The major innovation of this work is to use natural language processing (NLP) to measure the quality of colonoscopy. NLP is a field of computer science in which a computer is trained to "read" text to identify relevant data We developed and validated the first NLP-based computer software application (C-QUAL) that analyzes colonoscopy and associated pathology reports. Our primary quality measure is adenoma detection rate because it is a common, validated measure that is linked to colorectal cancer incidence. However, we also use a number of secondary quality measures. We applied C-QUAL to almost 25,000 colonoscopy reports in one health system and found large variation in physician's performance on the quality measures. Building on this prior work, our goal is to use C-QUAL to measure colonoscopy quality across a spectrum of US practice environments, to understand what drives variation in colonoscopy quality, and to improve colonoscopy quality.
In Aim 1, we propose to use the C-QUAL tool to measure performance in 4 diverse health care systems. This will be one of the largest assessments of the variation in adenoma detection rates and will span different geographic regions, payment systems, and practice settings.
In Aim 2, we seek to understand why there is variation in quality. We will survey providers at the 4 health care systems about factors that might affect quality. We will link those survey results to the adenoma detection rates assessed in Aim 1 and look for key associations. It is assumed, but not proven, that providing feedback to physicians on colonoscopy quality will improve care.
In Aim 3, we assess whether feedback to physicians does drive quality improvement and, building on Aim 2, explore which types of physicians may respond to feedback. Our proposal is the first to use this innovative method to measure colonoscopy quality and to use the quality scores to decrease the variation in colonoscopy performance. Together the results of the 3 aims are consistent with the NCI's focus on improving the quality of colorectal cancer screening.
Prior research has shown that the quality of colonoscopy varies from physician to physician. It is important to address low quality because patients who see physicians with lower quality colonoscopies have a higher risk of developing colorectal cancer in the future. To help address poor quality of colonoscopy, we propose to use a computer software program to measure colonoscopy quality, survey physicians to understand why there is variation in quality, and provide physician feedback to stimulate quality improvement.