Using natural language processing to improve efficiency of manual chart abstraction in research: the case of breast cancer recurrence

Abstract

The increasing availability of electronic health records (EHRs) creates opportunities for automated extraction of information from clinical text. We hypothesized that natural language processing (NLP) could substantially reduce the burden of manual abstraction in studies examining outcomes, like cancer recurrence, that are documented in unstructured clinical text, such as progress notes, radiology reports, and pathology reports. We developed an NLP-based system using open-source software to process electronic clinical notes from 1995 to 2012 for women with early-stage incident breast cancers to identify whether and when recurrences were diagnosed. We developed and evaluated the system using clinical notes from 1,472 patients receiving EHR-documented care in an integrated health care system in the Pacific Northwest. A separate study provided the patient-level reference standard for recurrence status and date. The NLP-based system correctly identified 92% of recurrences and estimated diagnosis dates within 30 days for 88% of these. Specificity was 96%. The NLP-based system overlooked 5 of 65 recurrences, 4 because electronic documents were unavailable. The NLP-based system identified 5 other recurrences incorrectly classified as nonrecurrent in the reference standard. If used in similar cohorts, NLP could reduce by 90% the number of EHR charts abstracted to identify confirmed breast cancer recurrence cases at a rate comparable to traditional abstraction.

Keywords: breast cancer recurrence; chart abstraction; natural language processing.

Figure 1.

Architecture of the natural language processing (NLP) system used to identify recurrent breast cancer diagnoses among Group Health patients, Pacific Northwest, 1995–2012. cTAKES, Clinical Text Analysis and Knowledge Extraction System.

Figure 2.

Breast cancer recurrence hits per calendar day for Group Health patients, Pacific Northwest, 1995–2012, found by the natural language processing system's clinical module in electronic charts for random samples of A) 16 breast cancer patients without recurrence, and B) 16 breast cancer patients with clinically confirmed recurrence.