gms | German Medical Science

Background: The classifications of epidemiological study designs (e.g. cohort studies) in medical research are usually based on inherent design features (e.g. number of groups, sampling method, measurements before and after intervention). In a broad definition, a “registry is a collection—for one or more purposes—of standardized information about a group of patients who share a condition or experience”. Considering these definition shows that registries have no inherent design features and consequently these cannot be used to classify the study design.

Objective: The objective was to propose a classification for registry-based studies, which is based on features of the statistical analysis and features of the registry.

Methods: We systematically analyzed existing schemes for the classification of study designs. We critical assessed the applicability of usual classification criteria (e.g. temporariness of groups, measurement time points).

All criteria that were not fully applicable because a registry has no inherent design features were adapted. The development of the criteria had two premises. Firstly, the criteria must be completely based on the analysis features (e.g. selection of patients, incorporation of time points in the analysis) and characteristics of the registry (e.g. process of data collection) to ensure universal applicability to all registry data. Secondly, the classification criteria had to result in the common epidemiological study designs.

Results: The following analysis features were developed:

• Concurrency of exposure/intervention and outcome (assessment at same time points)
• Comparative vs noncomparative (same exposure/intervention in all participants)
• Allocation of intervention/exposure
• Exposure-based or outcome based sampling
• Collection of exposure/intervention and outcome data (exposure/intervention data collected prior to outcome data)
• Measurements before and measurements after the intervention incorporated in the analysis
• Number of measurements before and after the intervention incorporated in the analysis
• Comparison in the same participants (repeated measures)

The distinction between retrospective and prospective can be applied to different parts of a study (e.g. data collection, planning of analysis). Some parts might be performed prospectively, and other parts might be performed retrospectively. Registries are prospective regarding the collection of data on exposure/intervention and outcomes. In contrast, details of the analysis of registry data are mostly planned retrospectively (e.g. patients included in the analysis, analysis method). Therefore we did not use retrospective and prospective for classification.

The criteria lead to a classification algorithm that includes the following study designs.

• Cross-sectional study
• (non-concurrent) Cohort study
• Controlled-before-after study
• (nested) Case-control study
• Before after study
• Interrupted-time-series
• Non-randomized trial
• (cluster) Randomized controlled trial

Discussion: We suggest a classification of registry-based studies that results in designs in accordance with the classical epidemiologic study designs. The key element of this is that the study designs are classified based on analysis and registry features.

Practical implications: Our classification can contribute to the harmonization of labeling of registry based studies. Thus, it can avoid misinterpretation of study results and increase the utility and acceptance of registry based studies in evidence based health care. Furthermore, it can support the identification of the best analysis method/study design that can be prepared with the available data in the registry. For example, if it is possible to prepare a controlled before-after study to prove the effect of an intervention, the data should not be analysed as cohort study.