World Congress at ACG2017

Simultaneous Plenary Session 4C: IBD

80 - Can an Automated Classifier Program Outperform Manual Review of Colorectal Biopsy Reports? A Quality Improvement Initiative

Wednesday, October 18
9:40 AM - 9:50 AM
Location: W414 (Level 4)



Category: Practice Management       

Bijun S. Kannadath, MBBS, Matthew Meriwether, MD, MBA, Andrew Herman, MD, Harsh D. Patel, MD, James Stone, MD, Danielle M. Stone, MD, Nirav C. Thosani, MD, MHA, Sushovan Guha, MD, PhD
University of Texas Health Science Center at Houston, Houston, TX
Introduction: A key quality metric for Gastroenterology practice is the appropriate follow-up colonoscopy rate of high-risk patients. High-risk is defined as patients who had a pathology report of malignancy and/or high-grade dysplasia following biopsy of colorectal polyp(s) and/or mass(es). However, manual review of all polyp pathology reports is subject to human errors. Our project was to determine if an automated lexicon-based classifier can be implemented for the detection and categorization of high-risk patients.

Methods: During the study period investigators manually reviewed a total of 2623 polyp/mass pathology reports and classified them as malignancy or high-grade dysplasia. All records were then deidentified. The automated classifier program (ACP) was written and implemented in Python. From this database, 500 records were randomly selected to train the ACP. The classifier used the presence of clinical pathologic terms indicating the diagnosis for detection and classification of records. The remaining 2123 records were used to test the performance of the classifier compared to manual review (MR) and classification. All instances, in which the assigned classifications of the ACP and MR disagreed, were referred to two senior authors for final review.

Results: Performance of automated classifier program (ACP) was compared against manual review (MR) in a set of 2123 pathology reports. For the detection of high-grade dysplasia, the ACP outperformed MR with an absolute detection rate of 94.7% (54/57) and only two false positives. In comparison, MR had an absolute detection rate of only 71.9% (41/57) with four false positives (Fisher’s Exact Test p-value < 0.002). For malignancy, both the ACP and MR had an equal absolute detection rate of 94.6 % (35/37) with one false positive report each.

Discussion: ACP outperformed MR in detection of high-grade dysplasia and was equivalent to MR for detection of malignancy. Labor intensive tasks like manual review of large numbers of records are fraught with potential for errors due to human fatigue. Automation of such tasks can result in large savings of time and resources. Automation can thus enable divisions to rapidly review large numbers of colorectal biopsy reports with minimal resource and time requirements. Our program is easily portable and quick (runtime < 2 seconds) as well as highly accurate, and can be implemented widely for the routine classification of colorectal pathology reports.

Supported by Industry Grant: No


Table 1. Summary of Results



























































Study Period November 2014 to March 2016
Total Patients 2623 (Training Set: 500 and Testing Set: 2123)
High-grade dysplasia (Total Positives = 57)
  Number Detected Number Missed Detection Rate
Manual Review 41 16 71.9%
Automated Classifier Program 54 3 94.7%
Fisher’s Exact Test p-value < 0.002
 
Malignancy (Total Positives = 37)
  Number Detected Number Missed Detection Rate
Manual Review 35 2 94.6%
Automated Classifier Program




 





35 2 94.6%






Citation: . CAN AN AUTOMATED CLASSIFIER PROGRAM OUTPERFORM MANUAL REVIEW OF COLORECTAL BIOPSY REPORTS? A QUALITY IMPROVEMENT INITIATIVE. Program No. 80. World Congress of Gastroenterology at ACG2017 Meeting Abstracts. Orlando, FL: American College of Gastroenterology.

Bijun S. Kannadath

Graduate Research Assistant
University of Texas Health Science Center at Houston
Houston, Texas

Presentation(s):

Send Email for Bijun Kannadath


Assets

80 - Can an Automated Classifier Program Outperform Manual Review of Colorectal Biopsy Reports? A Quality Improvement Initiative



Attendees who have favorited this

Please enter your access key

The asset you are trying to access is locked. Please enter your access key to unlock.

Send Email for Can an Automated Classifier Program Outperform Manual Review of Colorectal Biopsy Reports? A Quality Improvement Initiative