Compare methods of identifying individuals with rare neuro-oncologic diagnoses using electronic health records (EHRs).

Maintaining accurate databases facilitates identification of people with rare neurologic conditions such as neurofibromatosis type 1 (NF1), neurofibromatosis type 2 (NF2), and schwannomatosis (SWN) for inclusion in research. International Classification of Disease (ICD) codes are used to identify people with particular conditions in clinical care and research. It is important to understand opportunities and limitations of methods used to collate EHR data for research.

Applied three methods to identify patients with NF1, NF2, or SWN seen at Johns Hopkins University (JHU): (1) ICD-10 code Q85.0% query in Epic Slicer-Dicer; (2) Epic Clarity database search via the Center for Clinical Data Analysis (CCDA) at JHU; and (3) a manually curated database of patients evaluated in clinic. We compared the number of people identified as having NF1, NF2, or SWN through each mechanism, using the manually curated database as the gold standard.

Epic SlicerDicer returned 2,034 patients. The CCDA search yielded 1,010 patients. The manually curated dataset identified 925 patients with confirmed NF1, NF2, or SWN. Validation efforts revealed that 85 patients identified through CCDA did not meet criteria for NF1, NF2, or SWN, yielding a false positive rate of 8.4%. SlicerDicer does not allow comparison of person-level data; however, our assessment suggests at least a 50.3% false positive rate.

Automated EHR searches with ICD-10 codes are efficient, but may be inaccurate. Manually curated databases require intensive time and expertise, and are limited by required manual updates of the data. Efforts to integrate the accuracy of curated databases with the efficiency of automated data updates from the EHR using precision medicine analytics enable optimization of data collection for clinical research. This is particularly important for rare neuro-oncologic conditions with complex diagnostic criteria.