(137) Study Scripts Supporting Multiple Common Data Models

Sunday, August 28, 2022

8:00 AM – 1:45 PM CEST

Location: Hall C

Publication Number: 1182

Presenting Author(s)

Rosa Gini, PhD
– Agenzia Reg Della Sanit, Firenze, Firenze, Italy

Co-Author(s)

Olga Paoletti
– ARS Toscana, Florence, Italy, Italy
Romin Pajouheshnia, PhD
– Assistant Professor (formerly), University of Utrecht, The Hague, Zuid-Holland, Netherlands
Patrick Souverein
– Utrecht University, Netherlands
Nicolas H. Thurin, PharmD, PhD
– Scientific Officer, Bordeaux PharmacoEpi, Université de bordeaux, Bordeaux, France
Vera Ehrenstein, MPH, DSc
– Aarhus University, Aarhus, Denmark
Manuel Barreiro-De Acosta
– Spanish Working Group on Crohn’s Disease and Ulcerative Colitis- GETECCU, Spain
Miriam C. Sturkenboom, PharmD, PhD, FISPE
– prof. dr., University Medical Center Utrecht, Utrecht, Utrecht, Netherlands
Lia Gutierrez, MPH, BScN
– Senior Epidemiology Director , RTI Health Solutions, University of Texas School of Public Health, Houston, BARCELONA, Barcelona, Spain
Susana Perez-Gutthann, MD, PhD, MPH, FISPE, FRCP
– VP Epidemiology, RTI Health Solutions, BARCELONA, Barcelona, Spain

Background: Multinational studies can be performed across a distributed network of data sources by converting the original data into a common data model (CDM), and then running a common analysis script at each site. Multiple CDMs have been successfully deployed, but analysis scripts designed to run against one CDM cannot directly work for another CDM. It is therefore useful to develop scripts that can support multiple CDMs. In a past study, the analytic pipelines of multiple CDMs, including OMOP and Sentinel, were conceptually mapped to a sequence of transformation steps: (T1) conversion to the CDM, resulting in a CDM instance; (T2) study variable creation, resulting in datasets of observations on the study population: this is the first step of a study script; (T3) application of study design, resulting in analytic datasets; and (T4) statistical analysis, resulting in datasets of study results. In practice, scripts can be structured according to this sequence of steps. In such scripts, step T2 alone needs adaptation to the different CDMs

Objectives: To evaluate the feasibility of retrieving quantitative metadata (age and sex distribution) from data sources mapped to different CDMs

Methods: In the European project MINERVA, we simulated a data source with sex and dates of birth and data source entry and exit. We converted the data source to 4 commonly used CDMs: OMOP, ConcePTION, Nordic and TheShinISS and developed an R analysis script to calculate annual sex and age distributions of the population. Four versions of step T2 were programmed, one per CDM, to generate the same output. The next steps (T3 and T4) were designed to run on the output of T2, and were programmed just once. The script was run against the 4 conversions of the simulated data source, and the resulting 4 outputs were merged, to test whether they were the same. Finally, the script was run against two real instances of CPRD and ARS, converted, respectively, to the ConcePTION and TheShinISS CDM.

Results: The script took a few hours to develop and is loaded in a GitHub repository ( https://github.com/ARS-toscana/MINERVA_samplescript). The 4 runs in the simulated data source resulted in generation of the same output dataset (distributions). The script ran successfully against the two data sources and correctly calculated their annual age and sex distribution.

Conclusions: It is possible to structure study scripts in a common sequence of steps. This minimizes the effort to adapt them to multiple CDMs, because one step only (T2) requires adaptation. Structuring scripts this way has the potential to support collaboration in studies, by enabling the use of multiple CDMs.