P1329 - Initial Outcomes of the Implementation of a Validated Machine Learning Algorithm to Identify Unscreened Individuals at High Risk for Advanced Adenomas or Colorectal Cancer in an Academic Healthcare System

Monday, October 25, 2021

10:30 AM – 4:15 PM PT

Location: Shoreline Exhibit Hall

Has Audio

Presenting Author(s)

Michael Mullarkey, MD

University of Alabama at Birmingham
Birmingham, AL, United States

Michael Mullarkey, MD, William Curry, MD, Douglas Morgan, MD, MPH, Adam Edwards, MD, MS
University of Alabama at Birmingham, Birmingham, AL

Introduction: Despite being one of the most preventable cancers with screening, colorectal cancer (CRC) remains the second leading cause of cancer-related deaths in the United States, and over 30% of US adults are not current with CRC screening. We used a validated machine learning algorithm to flag unscreened patients at high risk for advanced adenoma (AA) and CRC, referred them for expedited colonoscopy, and compared detection rates with those from a cohort of similar patients referred for colonoscopy based on current CRC screening guidelines.

Methods: We conducted a prospective, observational study of patients from our academic ambulatory clinic referred for colonoscopy after being flagged by the machine learning algorithm (algorithm group) and a cohort of age-, race/ethnicity-, and sex-matched patients from the same population referred for screening colonoscopy during the same time interval based on current CRC screening guidelines (standard screening group). Demographics and pathology data were obtained via chart review. Detection rates were compared between groups with the Mann-Whitney U test and Pearson Chi-square test as appropriate.

Results: Over a six-month period, 22 patients identified by the algorithm underwent colonoscopy. A 5:1 sample (n = 109) was obtained of screening colonoscopies from the same time period matched on age, race/ethnicity, and sex. There was no statistically significant difference between the algorithm group and the standard screening group in the rates of detection of AAs (14% vs. 10%, P = 0.624) or CRC (4.5% vs. 1.0%, P = 0.206). Of the adenomas that were detected, there was a trend for them to be larger (10 mm vs. 6 mm, P = 0.110) and more prevalent in the proximal colon (40% vs. 12%, P = 0.095) in the algorithm group.

Discussion: There was a trend toward more AAs and CRC in the cohort identified by the machine learning algorithm to be at high risk, though this did not reach statistical significance. When adenomas are detected, they may be larger and more often found in the proximal colon in the algorithm group. The most significant challenge in our study was getting patients flagged by the algorithm from our open academic healthcare system to undergo colonoscopy, which occurred in only 4% (Table 1). Our data indicates that such an algorithm may be a valuable tool in the outpatient setting, particularly in an integrated healthcare system, though more research is needed to determine its efficacy for CRC screening.

Primary Challenges To Utilization of a Machine Learning Algorithm for Advanced Adenoma and Colorectal Cancer (CRC) Detection in an Open Academic Healthcare System
1. Support required for full-time equivalents for staff to review charts for exclusion criteria and make repeated attempts to contact patients
2. Patients flagged by algorithm who are up-to-date with CRC screening performed outside of the academic healthcare system
3. Significant proportion of patients flagged who decline follow-up colonoscopy

Table: Table 1: A summary of the challenges encountered when implementing use of a machine learning algorithm for detection of patients at high-risk of harboring advanced adenomas or colorectal cancer in an open academic healthcare system

Disclosures:

Michael Mullarkey indicated no relevant financial relationships.

William Curry indicated no relevant financial relationships.

Douglas Morgan indicated no relevant financial relationships.

Adam Edwards indicated no relevant financial relationships.

Michael Mullarkey, MD, William Curry, MD, Douglas Morgan, MD, MPH, Adam Edwards, MD, MS. P1329 - Initial Outcomes of the Implementation of a Validated Machine Learning Algorithm to Identify Unscreened Individuals at High Risk for Advanced Adenomas or Colorectal Cancer in an Academic Healthcare System, ACG 2021 Annual Scientific Meeting Abstracts. Las Vegas, Nevada: American College of Gastroenterology.