Advanced Methods Webinar - Measurement in Administrative Health Data: Case Definitions, Algorithms, and Validation Studies
Despite not being collected for research purposes, administrative health data (e.g., outpatient physician billings, hospitalizations, and prescription drug data) are increasingly being used in epidemiological analyses. Unlike a setting where the researcher can directly measure whether a study participant is living with a given condition (primary data collection), one may need to rely on what is available in the administrative data to make such assessments. For example, the occurrence of a health or clinical outcome may be captured through a combination of diagnostic codes (e.g., ICD-9/ICD-10-CA), procedure codes (e.g., CCP/CCI), and/or drug identification numbers (e.g., DINs).
In practice, a researcher searches for the aforementioned codes within a person’s administrative data. These queries require that the codes appear at a certain frequency, within a given time-window, in specific datasets. In conjunction with the codes, these additional criteria comprise what is often referred to as a ‘case definition’ or a ‘case-finding algorithm’; the terms ‘variable definition’ or ‘algorithm’ are used when speaking more generally about measurements derived from administrative data. Given the expanding use of these population-based data sources for research, it is important to equip researchers with the knowledge to use such measures in a rigorous and consistent manner.
This webinar will focus on measurement in epidemiological studies that use linked administrative health data. Specifically, this 1.5-hours session will:
- Unpack the various elements of case definitions/algorithms in population health research; an emphasis will be placed on the selection and interpretation of definitions/algorithms that are tailored to a given jurisdiction.
- Introduce the conceptual framework of a validation study (i.e., ‘gold’ standard measures and validation sub-samples).
- Provide a theoretical overview of metrics (e.g., sensitivity, specificity, positive and negative predictive values) which quantify the amount of bias (misclassification) that may be introduced when using a given definition/algorithm.
View recorded presentation below.
Taylor McLinden, PhD, is the Scientific and Quality Assurance Officer (Epidemiology and Population Health Program) at the BC Centre for Excellence in HIV/AIDS (BC-CfE) in Vancouver. He completed his PhD (Epidemiology) in the Department of Epidemiology, Biostatistics and Occupational Health at McGill University. During his doctoral training in Montreal, Taylor developed methodological expertise in longitudinal data analysis, causal inference, and missing data. At the BC-CfE, he currently leads initiatives focused on education and capacity building in the area of epidemiological methods. Presently, his efforts relate to facilitating the use of linked administrative health data for HIV research.