A comparison of structured data query methods versus natural language processing to identify metastatic melanoma cases from electronic health records

Abstract

The relative efficacy of natural language processing (NLP) of text reports compared to structured data queries for identifying patients from electronic health records (EHRs) with metastatic cancer remains unclear. Such identification is critical for identifying and recruiting potential study candidates for cancer trials, particularly trials of cancer chemotherapy. For such purposes, we performed a direct comparison between NLP and structured data query methods for identifying patients with metastatic melanoma. Using EHR data from two large institutions, we found that NLP of text reports identified close to three times as many patients with metastatic melanoma compared to a structured data query algorithm (1,727 vs. 607 patients). Using an external tumour registry, we also found NLP had much higher sensitivity than structured query for identifying such patients (67% vs. 35%). Our results emphasise the importance of employing NLP criteria when identifying potential cancer study candidates with metastatic disease.