(I) New Tools for Pharmacoepidemiological Studies: Developing a Lightweight Modular Analytics Framework Based on a Nordic Common Data Model (NCDM) and a General Model of Person-Time-Event Data

Saturday, August 27, 2022

8:00 AM – 6:00 PM CEST

Location: Hall E South

Publication Number: 659

Presenting Author(s)

Morten Andersen, MD, PhD
– Professor, University of Copenhagen, University of Copenhagen, Copenhagen, Hovedstaden, Denmark

Co-Author(s)

Reeha Sharon, PhD
– Copenhagen Centre for Regulatory Science, University of Copenhagen, Copenhagen, Denmark
Sarah Kristiansen
– Postdoctoral researcher, data manager, Pharmacovigilance Research Center, Department of Drug Design and Pharmacology, University of Copenhagen, Copenhagen, Denmark, Vanløse, Hovedstaden, Denmark
Mia Aakjær
– University of Copenhagen, K�benhavn �, Hovedstaden, Denmark

Background: The complexity of real-world administrative and healthcare data makes pharmacoepidemiological studies challenging. In multi-country studies, harmonisation of data structure and content using a common data model is part of the solution. Subsequent data processing, creating the analysis datasets, performing the statistical analyses and producing results, may also benefit from harmonisation and streamlining.

Objectives: To develop and implement a lean and flexible data model and a lightweight framework of modular analytics programs that can be used for pharmacoepidemiological studies.

Methods: The framework is currently implemented in SAS and based on a Nordic Common Data Model (NCDM) with added analytic metadata containing 1) a specification of study parameters, 2) a concept dictionary defining the mapping of codes (for diagnoses, procedures, etc.) to concepts and 3) a variable dictionary defining variables in the analytic dataset based on concepts and time windows.
A prerequisite for using the framework is an initial extract-transform-load step where data are structured in person-time-event datasets corresponding to residence episodes, clinical events (healthcare encounters, cancer and death records with diagnosis and procedure codes), drug events (pharmacy dispensings), and lab events (laboratory measurements). The data model can be extended with ad hoc person-time-event datasets (e.g., vaccinations).
Modular programs are designed to run in a logical sequence. First, codes from different systems are translated into broader concepts. Second, the study population is selected, identifying exposed persons and comparators, applying inclusion and exclusion criteria and sampling controls if relevant. Finally, the analysis dataset is created, applying metadata definitions of confounder variables, criteria for follow-up, outcome events and censoring.
In the data analysis step, automated programs output descriptive frequency tables, regression tables and figures in pre-specified layouts. Each program has a built-in system, which saves logs and creates quality check output for easy inspection.

Results: The NCDM analytics framework has been developed over the past 5 years and deployed in several recent studies. The overall programming and data processing time in studies is considerably reduced. Standard operating procedures and built-in checks ensure quality compliance throughout data management and analyses.

Conclusions: Using a modular analytics framework provides an efficient way of performing pharmacoepidemiological studies using administrative and healthcare data. The simplicity of the framework enables solving some of the complexities of real-world data.