During a recent COVID-19 Evidence Accelerator (EvAcc) Therapeutics Parallel Analysis meeting¹ we saw the synergy and creativity that arises when multiple partners work on the same data challenges. As all of us listened (and have been listening) to data partners express frustration over wrangling medication use codes, an idea emerged: “Why don’t we all put our heads together on this issue and solve it together?” Such collaboration is a goal of the Evidence Accelerator, and a key data challenge in real-world studies is the inclusion of medication use data (from prescribing to dispensing).

Medication lists enter the EvAcc analyses of real-world data (RWD) when, for example, we consider treatments a patient may have received in the progression of the disease or when we describe the course of a specific treatment and collect information on other medical conditions and medications that may be relevant to our specific study question. (These are often covariates or potential confounders in propensity score modeling.) The task is daunting because there are many medications, formulations, coding systems, and opportunities for misspellings, non-standard abbreviations, truncations of names, and other issues. In short, data partners have been putting hours of time into getting the medication use data in shape.

The data challenge is complex. One of the Accelerators (Keith Marsolo, Duke University) explained, “Within the [Electronic Health Record] EHR, information on medication exposure can be found in four broad domains: prescribing/orders, procedures, administrations, and dispensing. (Self-reported medications and those delivered in home health visits tend to be slotted into one of those areas.)  Depending on the study question, it may be necessary to look in any/all of those domains to gain a complete understanding of exposure. (A caveat: The domains may be incomplete. For example, a health system may only have dispensing information for surgery patients that covers the 90 days prior to surgery.)  Data within these domains will be coded in a variety of formats: CPT/HCPCS for procedures, NDC for dispensing, NDC or RxNorm for administrations (depending on site) and RxNorm for prescribing/orders.”

There are, of course, additional complexities related to the coding standard that are enough to make your head spin, underscoring that the EvAcc community needs to work together to standardize approaches to medication use information and agree on practices. Fortunately, the EvAcc community has a wealth of expertise from which to draw!

Some points to consider:

Medication use data includes prescribing/orders, procedures, administrations, and dispensing, as well as self-reported use and other events. These considerations can, and should, be incorporated in our conceptual and operational definitions and, as Keith so rightly points out, will vary with the research context. For example, when we are describing practice, we may rely on the prescribing orders; when we are using medications to indicate co-morbidities, we may combine prescribing and physician orders with self-report, administrations, dispensing, etc.; and when we are seeking to describe the result of medication exposure, we may prefer administration and dispensing data. These will vary with the specific question and the state of the field. Regarding Keith’s point about the multiple coding systems, it will be useful to explore how others have developed code to use the medication data in different types of analyses. And then, perhaps, we could tackle additional nuances and complexities.

Looking forward to seeing the result of this collaboration.

The ad hoc Medication Use Definitions Working Group kicked off on February 1. The limited duration workgroup will develop resources to guide use of medication data in the EvAcc, and to share those resources with the broader research community. This effort is intended to save time and energy, allowing us to focus on critical scientific questions. Contact EvidenceAccelerator@reaganudall.org for more information.

^{1 The COVID-19 Evidence Accelerator includes two primary objectives: 1) convening the COVID-19 real-world data community to share lessons learned and keep pace with new developments and 2) pursuing discrete, prioritized research questions in our Parallel Analysis format. In the Parallel Analysis approach, data partners agree on a common protocol and query their data sources to generate information, which is then presented side-by-side, or in parallel}.