Article
Merging differently operationalized predictors of outcome when combining cohorts from separate sources
Search Medline for
Authors
-
Dominikus Stelzer
- Institut für Medizinische Biometrie und Statistik, Universitätsklinikum Freiburg, Medizinische Fakultät, Albert-Ludwigs-Universität Freiburg, Freiburg, Deutschland
-
Susanne Weber
- Institut für Medizinische Biometrie und Statistik, Universitätsklinikum Freiburg, Medizinische Fakultät, Albert-Ludwigs-Universität Freiburg, Freiburg, Deutschland
-
Julia Ortner
- Lehrstuhl für Controlling, Johannes Gutenberg-Universität Mainz, Mainz, Deutschland
-
Peter R. Galle
- I. Medizinische Klinik und Poliklinik, Universitätsmedizin der Johannes Gutenberg-Universität Mainz, Mainz, Deutschland; Cirrhose Centrum Mainz, Universitätsmedizin Mainz, Mainz, Deutschland
-
Frank Lammert
- Klinik für Innere Medizin II, Universitätsklinikum des Saarlandes, Homburg, Deutschland
-
Marc Nguyen-Tat
- I. Medizinische Klinik und Poliklinik, Universitätsmedizin der Johannes Gutenberg-Universität Mainz, Mainz, Deutschland; Cirrhose Centrum Mainz, Universitätsmedizin Mainz, Mainz, Deutschland
-
Andreas Schwarting
- I. Medizinische Klinik und Poliklinik, Universitätsmedizin der Johannes Gutenberg-Universität Mainz, Mainz, Deutschland; ACURA Rheumazentrum Rheinland-Pfalz, Bad Kreuznach, Deutschland
-
Erik Farin-Glattacker
- Institut für Medizinische Biometrie und Statistik, Universitätsklinikum Freiburg, Medizinische Fakultät, Albert-Ludwigs-Universität Freiburg, Freiburg, Deutschland
-
Harald Binder
- Institut für Medizinische Biometrie und Statistik, Universitätsklinikum Freiburg, Medizinische Fakultät, Albert-Ludwigs-Universität Freiburg, Freiburg, Deutschland
-
Erika Graf
- Institut für Medizinische Biometrie und Statistik, Universitätsklinikum Freiburg, Medizinische Fakultät, Albert-Ludwigs-Universität Freiburg, Freiburg, Deutschland
| Published: | August 27, 2018 |
|---|
Outline
Text
Many projects financed by the Innovation Fund of the Federal Joint Committee (G-BA) have in common that, while prospective data acquisition is performed for the intervention cohort, information regarding the control cohort must be obtained from secondary data, such as healthcare data. However, combining cohorts from different sources requires merging variables, both outcome and predictors of outcome, which may be operationalized rather differently. This can result in bias, if not taken into account. Here we focus on the issue of differently operationalized predictors, which could, e.g., be addressed by employing error model approaches.
We present these models and discuss their usefulness in two of our latest applications, the studies SEAL and Rheuma-VOR. In SEAL (“Structured Early Assessment of Asymptomatic Liver Fibrosis and Cirrhosis”, grant no. 01NVF16026) we investigate, to which extent the early diagnosis of chronic liver diseases is improved upon by introducing an early detection program as part of the regular Check-up 35. Data for the program participants, which constitute the intervention cohort, is collected by the physician using electronic case report forms (eCRF). This is compared to a control cohort, for which only generic healthcare data is available. Similarly, the intervention in the Rheuma-VOR study aims to promote early detection and treatment of chronic inflammable rheumatic diseases. Here, evidence is gained by contrasting prospectively acquired data with data obtained from the National Database of the German Collaborative Arthritis Centres (“Kerndokumentation”), which is maintained by the German Rheumatism Research Centre Berlin (DRFZ).
To investigate the effect of differently operationalized outcome predictors, we conduct a simulation study using various settings. The results are summarized and discussed with regard to their relevance for the SEAL and Rheuma-VOR projects.
The authors declare that they have no competing interests.
The authors declare that an ethics committee vote is not required.
