(AL) Use of Provider Notes in Novel Identification of Cancer at The Gastroesophageal Junction

Publication Number: 1036

Background: Amid rising incidence of gastroesophageal junction (GEJ) cancer, identification of cases in real world datasets has proved difficult. While the GEJ has differing pathological features from the esophagus and stomach, there are no International Classification of Diseases (ICD) codes for cancer at the GEJ.

Objectives: To develop a method for use of electronic health record (EHR) provider notes to identify patients with, and diagnostic patterns in, GEJ cancer.

Methods: This study used Optum Pan-Oncology ® EHR data (United States). The study cohort included adult patients with an ICD code for cancer of the stomach or esophagus between March 31 2015-2020. GEJ cases were identified using two definitions in Natural Language Processed (NLP) provider notes: (1) Primary GEJ Cohort: presence of a “Term” related to cancer and “Location” related to the GEJ, and (2) Subgroup: patients in the primary GEJ cohort with additional presence of a confirmatory note called a “Sentiment.” Those identified were then assessed for subsequent ICD and provider note diagnoses in their patient timelines.

Results: Out of 41,370 patients with an ICD code for esophageal or gastric cancer, 567 (1.4%) had a provider note for GEJ cancer in the 30 days prior to the ICD diagnosis. 2,526 (6.1%) had a provider note for GEJ cancer on or in the 30 days following the ICD diagnosis.

After applying exclusion criteria, 1,869 of these 2,526 patients met the primary GEJ cohort definition. 76.4% and 56.7% of these patients had a subsequent diagnosis for esophageal or gastric cancer respectively. 69.6% and 37.5% of these patients had subsequent provider notes that additionally met the “Term/Location” and “Term/Location/Sentiment” GEJ cancer definitions, respectively.

72% of provider note events in the dataset were not associated with a “Sentiment.” Of patients in the primary cohort, 649 (34.7%) met the subgroup definition. 80.7% and 62.4% had a subsequent diagnosis for esophageal or gastric cancer respectively. 72.1% and 55.8% of this subgroup had subsequent provider notes that additionally met the “Term/Location” and “Term/Location/Sentiment” definitions, respectively.

Conclusions: Compared to the broader “Term/Location” definition, the more restricted “Term/Location/Sentiment” definition captured a higher proportion of patients who had additional GEJ provider notes following their first, but identified fewer patients overall. In the absence of a gold standard, these explorations and sample size considerations were important in understanding real world clinical patterns and creating a meaningful and novel algorithm for identifying GEJ cancer in EHR records.