Gopanandan Parthasarathy1, Rocio Lopez2, John McMichael3, Carol A Burke4. 1. Department of Medicine, Cleveland Clinic, Cleveland, Ohio, USA. 2. Center for Populations Health Research, Cleveland Clinic Foundation, Cleveland, Ohio, USA. 3. Department of General Surgery, Cleveland Clinic, Cleveland, Ohio, USA. 4. Department of Gastroenterology and Hepatology, Cleveland Clinic, Cleveland, Ohio, USA.
Abstract
BACKGROUND AND AIMS: Serrated polyposis syndrome (SPS) is common but under-recognized and is associated with an increased risk of colorectal cancer. The diagnosis is based on the World Health Organization (WHO) criteria and is inclusive of the cumulative number of lifetime serrated polyps. We used natural language processing (NLP) to extract colonoscopy and pathology data from the electronic medical record (EMR). The aim of this study was to assess feasibility of using an NLP-based SPS tool to identify patients with SPS. METHODS: NLP was used to extract data from 323,494 colonoscopies performed in 255,074 distinct patients between August 1998 and March 2016 to identify individuals who met SPS criteria. The accuracy of diagnosis of SPS was assessed by manual review of the EMR. RESULTS: Of 255,074 patients, 71 were identified as meeting 1 WHO criteria for SPS. Manual review confirmed the diagnosis of SPS to be accurate in 66 cases (93%). Erroneous diagnosis in the remaining 5 cases occurred because of duplicate polyp data by NLP extraction. Only 25 of 66 patients (38%) were diagnosed with SPS by a clinician in the EMR. Of these, SPS was diagnosed by NLP at least 2 years before the clinician in 5 of 25 patients (20%). CONCLUSIONS: In a large cohort, NLP accurately identified SPS in over 90% of cases, most of which were not previously recognized. NLP can assist in collating colonoscopy and pathology data across multiple procedures in the same patient to make an accurate and earlier diagnosis of SPS.
BACKGROUND AND AIMS: Serrated polyposis syndrome (SPS) is common but under-recognized and is associated with an increased risk of colorectal cancer. The diagnosis is based on the World Health Organization (WHO) criteria and is inclusive of the cumulative number of lifetime serrated polyps. We used natural language processing (NLP) to extract colonoscopy and pathology data from the electronic medical record (EMR). The aim of this study was to assess feasibility of using an NLP-based SPS tool to identify patients with SPS. METHODS: NLP was used to extract data from 323,494 colonoscopies performed in 255,074 distinct patients between August 1998 and March 2016 to identify individuals who met SPS criteria. The accuracy of diagnosis of SPS was assessed by manual review of the EMR. RESULTS: Of 255,074 patients, 71 were identified as meeting 1 WHO criteria for SPS. Manual review confirmed the diagnosis of SPS to be accurate in 66 cases (93%). Erroneous diagnosis in the remaining 5 cases occurred because of duplicate polyp data by NLP extraction. Only 25 of 66 patients (38%) were diagnosed with SPS by a clinician in the EMR. Of these, SPS was diagnosed by NLP at least 2 years before the clinician in 5 of 25 patients (20%). CONCLUSIONS: In a large cohort, NLP accurately identified SPS in over 90% of cases, most of which were not previously recognized. NLP can assist in collating colonoscopy and pathology data across multiple procedures in the same patient to make an accurate and earlier diagnosis of SPS.