(202) Systematic Definition Development of Variables of Interest for Studies Using Administrative Databases

Background: Validating clinical codes and algorithms for identifying variables of interest in administrative health data is indispensable for real-world evidence generation. US Food and Drug Administration guidance suggests clearly defining both a conceptual definition, based on clinical concepts, and an operational definition, describing how to identify it in a database, and additionally describing the rationale and potential impact on outcome misclassification.

Objectives: This research project aims to describe the methodology of creating a conceptual definition based on clinical guidelines and a standardized, validated, human- and machine-implementable operational definition based on validation studies for variables of interest. We targeted the International Classification of Diseases, Ninth and Tenth Revision, Clinical Modification (ICD-9-CM and ICD-10-CM, respectively) and the associated procedure coding systems.

Methods: To develop a conceptual definition, a clinical expert reviewed variable of interest definitions from clinical guidelines and clinical trials, and settled discrepancies between sources. Subsequently, a targeted literature review was conducted to search for validated operational definitions. If a validation study that reflected the conceptual definition was found, the operational definitions, code lists and positive predictive value, sensitivity and specificity were extracted and evaluated. In absence of a validation study, the operational definition and code lists were developed based on the conceptual definition and subject matter expertise. A clinical coder and expert reviewed all code lists to ensure codes reflected coding and clinical practices.

Results: To date, this approach has been used to develop over sixty variables of interest for thrombosis research. Key lessons in conceptual definition development included resolving differences between international clinical guidelines and various clinical trials. Key lessons in operational definition development included acknowledging the necessity of predefining the use of the variable of interest as an outcome or as a covariate, selecting a validation study that matches the conceptual definition, and emphasizing clinical experts and coder reviews. Challenges included discrepancies between validated operational definitions from the literature and with the conceptual definitions, absence of a validation study, and coding systems changes over time (from ICD-9-CM to ICD-10-CM, addition of new codes).

Conclusions: The systematic approach used to develop conceptual and operational definitions enhances their reproducibility, replicability and acceptance of real-world evidence using administrative databases.