Performance analysis of a machine learning flagging system used to identify a group of individuals at a high risk for colorectal cancer.

Yaron Kinar,Steven A Narod,Bernard Levin,Varda Shalev,Revital Kariv,Pinchas Akiva,Ran Goshen,Eran Choman,Frank T Kolligs

doi:10.1371/journal.pone.0171759

Abstract

Individuals with colorectal cancer (CRC) have a tendency to intestinal bleeding which may result in mild to severe iron deficiency anemia, but for many colon cancer patients hematological abnormalities are subtle. The fecal occult blood test (FOBT) is used as a pre-screening test whereby those with a positive FOBT are referred to colonscopy. We sought to determine if information contained in the complete blood count (CBC) report coud be processed automatically and used to predict the presence of occult colorectal cancer (CRC) in the setting of a large health services plan. Using the health records of the Maccabi Health Services (MHS) we reviewed CBC reports for 112,584 study subjects of whom 133 were diagnosed with CRC in 2008 and analysed these with the MeScore tool. The odds ratio for being diagnosed with CRC in 2008 was calculated with regards to the MeScore, using cutoff levels of 97% and 99% percentiles. For individuals in the highest one percentile, the odds ratio for CRC was 21.8 (95% CI 13.8 to 34.2). For the majority of the individuals with cancer, CRC was not suspected at the time of the blood draw. Frequent use of anticoagulants, the presence of other gastrointestinal pathologies and non-GI malignancies were assocaitged with false positive MeScores. The MeScore can help identify individuals in the population who would benefit most from CRC screening, including those with no clinical signs or symptoms of CRC.

Highlights

Detection of colorectal cancer through active screening has been shown in several studies to reduce mortality [1,2]
Classical screening methods include direct visualization and sampling through colonoscopy or sigmoidoscopy or indirectly through the detection of occult blood in the feces. The latter screening method is based on the tendency for many colorectal cancer (CRC) to bleed slowly and persistently and this may lead to a state of iron deficiency anemia
It is of interest to determine whether or not complete blood count (CBC) reports, used singly or in series, can be scanned automatically and interpreted using a machine learning based algorithm in order to identify a subgroup of individuals who have a higher than average probability of harboring an occult CRC

Summary

Introduction

Detection of colorectal cancer through active screening has been shown in several studies to reduce mortality [1,2]. Classical screening methods include direct visualization and sampling through colonoscopy or sigmoidoscopy or indirectly through the detection of occult blood in the feces. The latter screening method is based on the tendency for many CRCs to bleed slowly and persistently and this may lead to a state of iron deficiency anemia. It is of interest to determine whether or not CBC reports, used singly or in series, can be scanned automatically and interpreted using a machine learning based algorithm in order to identify a subgroup of individuals who have a higher than average probability of harboring an occult CRC. We generated a risk score for each member of a cohort of individuals at risk for colon cancer, evaluated whether or not they developed colon cancer by linkage to the Israel Cancer Registry and we reviewed the charts of all patients diagnosed with CRC and a sample of false positives

Methods

Results

Discussion