Background/Purpose: New user designs are typically preferred in pharmacoepidemiology to avoid bias. The optimal implementation of a new user design in electronic medical record (EMR) data, where the concept of ‘enrollment’ does not necessarily exist, has not been well studied.

Methods: We used data from PCORnet, a large U.S. comparative effectiveness research network consisting of multiple EMR data sources from both academic and community practices, to compare various new user definitions in 5 PCORnet datamarts, focusing mainly on rheumatoid arthritis (RA) but examining other conditions (e.g. vasculitis) as exemplar use cases. The first prescription appearing in the data for medications of interest (e.g. methotrexate [MTX], biologic, targeted therapy) was identified in the EMR prescribing data. Five new user definitions with varying requirements for the amount and type of prior EMR data needed to construct a ‘baseline’ period were evaluated. New user definitions tested included 0) no requirement for any prior data; 1) >6 months (m) from first medical inpatient or outpatient medical encounter of any type; 2) >6m from first prescription [Rx] for any medication; 3) >6m from first medical encounter for disease indication (e.g. RA, vasculitis); 4) >6 m from first prescription for any disease-specific rheumatologic therapy. The rate of hospitalized infection, where a time-dependent hazard was expected, and herpes zoster, where minimal time-dependent hazard was expected, was examined according to various new user definitions. Infections were ascertained over 6- and 12-month follow-up periods, using a first-exposure carried forward approach.

Results: A total of 6621 RA patients initiated one of 9 unique DMARDS, biologics or targeted medications. Mean (SD) age was 55(16) years, 78% women, 43% MTX, 38% TNFi biologics; 19% non-TNFi therapy, 38% oral glucocorticoids. The proportion of person-time represented in the new user analysis across the five definitions was: 100% (base case, 1st Rx is index date with no prior data required), 83% ( >6m from first visit), 57% ( >6m from first Rx of any type), 71% ( >6m from first RA visit), and 29 % ( >6m from first RA drug), respectively. The rate of hospitalized infection generally was numerically higher, although < 1-2/100py difference, using 6 month vs. 12 month followup. The crude rate of serious infections by exposure was sometimes but not always numerically higher using the more specific new user definitions (3 and 4) but meaningfully reduced exposure time, especially for 1st line therapies (e.g. methotrexate) and in diseases where sample size was small to begin with (e.g. vasculitis). In contrast, rates of herpes zoster varied minimally by new user definition.

Conclusion: A range of tradeoffs exist in how to best apply new user definitions to multi-specialty EMR data that will affect patient selection and outcome ascertainment. More rigorous and specific definitions are likely preferred for most pharmacoepidemiology studies to avoid bias but must be balanced against feasibility and may need to differ by disease and by outcome.

« Back to 2019 ACR/ARP Annual Meeting

ACR Meeting Abstracts - https://acrabstracts.org/abstract/comparison-of-medication-new-user-definitions-in-multi-specialty-emr-data/