Identifying Late-Stage Cancer and Chronic Kidney Disease Patients for Palliative Care Research and Practice: Computable Phenotypes and Natural Language Processing (S824)

1.Describe the utility of computable phenotypes and natural language processing in a palliative care population.2.Describe the research and practice implications of systematic identification of palliative care patients. Systematic identification of seriously-ill patients allows palliative care researchers and clinicians to test new models of care delivery. Algorithms based on clinical indicators—including natural language processing—can aid in such identification. To develop electronic health record (EHR) phenotypes to identify patients with Stage 4 solid-tumor cancer (CA) or Stages 4-5 chronic kidney disease (CKD). We developed two computable EHR phenotypes to retrospectively identify decedents who had been admitted to an academic medical center in the last six months of life with CA or CKD, respectively. Each search included International Classification of Diseases (ICD) 9 and 10 codes and a date of death 11/07/17-12/31/17 (CA) or 11/26/17-12/31/17 (CKD). Additionally, the cancer search included natural language processing (NLP) searches of notes for indicators of stage 4 CA (e.g., “stage IV,” “metastatic”); the CKD search included glomerular filtration rate (GFR) <30. For each EHR phenotype, we calculated the sensitivity, positive predictive value (PPV), and false discovery rate (FDR). Results of the phenotypes were compared to manual chart review for indicators of late-stage disease among patients admitted to the Oncology and Nephrology inpatient services, respectively. The EHR phenotype identified 116 CA patients, of whom 84 had Stage 4 CA, and 65 CKD patients, of whom 23 had Stage 4-5 CKD. The EHR phenotype for Stage 4 cancer had a sensitivity of 98.8%, PPV of 79.2%, and a FDR of 20.8% when compared to the assessment of the primary oncology services. The EHR phenotype for Stage 4-5 CKD had a sensitivity of 100%, PPV of 47.9%, and a FDR of 52.1% when compared to the assessment of the primary nephrology service. EHR phenotypes can efficiently identify patients with late-stage disease for palliative care.