Use of an Electronic Medical Record (EMR) Database to Identify a Real-World Cohort of US Patients (Pts) with Myelodysplastic Syndromes (MDS)

Introduction: Clinical trial data for MDS pts may not be representative of the general MDS pt population because some pts may be ineligible for clinical trials or may decline to participate. The GE Centricity™ EMR database (GE Healthcare IT, Princeton, NJ, USA), which is anonymized and compliant with the Health Insurance Portability and Accountability Act of 1996, contains clinical practice data from > 38 million US pts from 1994 onward (Asche CV, et al. ISRN Cardiol. 2011;2011:924343). Pt diagnoses in the database can be determined using International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes; however, cohorts selected this way may include pts without MDS due to coding errors. We accessed the GE EMR database to identify a large, real-world cohort of MDS pts and compared them with MDS pts reported to the well-established, population-based Surveillance, Epidemiology, and End Results (SEER) Program.

Methods: The GE EMR database is generally representative of the US population and contains data from single-physician offices to large practices and networks, covering 49 US states. Of > 30,000 participating clinicians, two-thirds are primary care physicians; the remainder are specialists. Using this database, we assembled a retrospective cohort of MDS pts with data entered during January 2006–February 2014. MDS pts were initially identified by the presence of ≥ 1 MDS-specific ICD-9-CM diagnosis code (codes: 238.72–238.75). Several different approaches were developed, evaluated, and applied to refine this cohort by excluding pts with a lower likelihood of MDS (Figure).

Results: The Initial Cohort comprised 9,645 pts with ≥ 1 MDS-specific ICD-9-CM code; 31% of pts received supportive treatment (hematopoietic growth factors, iron-chelating therapy, or transfusions) and 8% received active treatment (azacitidine, decitabine, or lenalidomide). However, only 9% of pts in this cohort received transfusions; fewer than expected. This may be due to unrecorded transfusions occurring outside the EMR system, a limitation that may apply to other EMR databases. The Initial Cohort included 563 pts with ≥ 2 MDS-specific ICD-9-CM codes (Figure; 1st Modified Cohort); 15% of these pts had received transfusions; considerably higher than in the Initial Cohort. To exclude pts unlikely to have MDS in the Initial Cohort, the inclusion criteria were modified to include only pts with ≥ 1 MDS entry plus an entry for MDS-specific active treatment (N = 1,208) (Figure; 2nd Modified Cohort); supportive treatment was insufficient for inclusion as it is not MDS-specific. As diagnosis of MDS requires hemoglobin (Hb) measurement and bone marrow (BM) aspirate or biopsy prior to diagnosis, inclusion criteria were refined to include only those pts with either ≥ 2 MDS entries or ≥ 1 entry for MDS plus ≥ 1 of the following: prior MDS-specific active treatment; ≥ 2 Hb tests ≤ 1 year prior to code entry; or ≥ 1 BM procedure ≤ 1 year prior to code entry (N = 5,623) (Figure; 3rd Modified Cohort). This cohort was then further refined by excluding pts whose EMR included terms indicative of a non-diagnosis of MDS, such as “rule out,” resulting in a Final Cohort of 5,162 pts most likely to have MDS (Figure; Final Cohort) whose baseline characteristics were generally comparable to MDS pts reported to the SEER Program during 2001-2011 (Table). In this Final Cohort, 35% received supportive care alone while 13% received active treatment; unrecorded transfusions may have contributed to underreporting of supportive care. Treatment patterns and the impact of variables such as age, insurance type, geographical location, and comorbidities will be presented.

Conclusions: By leveraging the GE Centricity EMR database and evaluating different approaches for the ascertainment of MDS pts, we identified a large, real-world cohort of MDS pts whose baseline characteristics are comparable to MDS pts from the SEER Program. Although characteristics between the modified cohorts were similar, the Final Cohort was selected based on clinical and diagnostic features associated with the diagnosis of MDS in real-world clinical practice. Our methodology can inform other investigators interested in utilizing EMR databases for cancer outcomes research, and the cohort we identified will be useful in the further characterization of treatment patterns and outcomes of MDS pts who are more representative than participants of clinical trials.