An observational study is a research design in which exposures, treatments, and health outcomes are recorded without researchers assigning the therapy through the study protocol. It can investigate care under routine conditions, disease courses, or associations between exposures and outcomes. Because the treatment is not randomized, careful control and transparent description of potential biases are particularly important.
Designs and objectives
Common observational designs include cohort, case-control, and cross-sectional studies. A cohort study follows groups over time and is suitable for describing incidences or time-dependent risks. A case-control study begins with individuals with and without the outcome and compares prior exposures; it can be efficient for rare events. Cross-sectional studies record characteristics at a single point in time and describe prevalences, for example.
Data collection can be prospective or use existing registry, healthcare, or medical record information. Crucial is an adequate specification of the target population, data source, exposure, comparison group, endpoint, and observation period. A study plan should additionally define how data quality, data protection, data lineage, and changes in definitions are controlled. Observational data can complement Real-World Evidence, but do not alone provide an automatic causal statement.
Bias, confounding, and analysis
Unlike in a randomized clinical trial, the treatment decision may be associated with prognostic factors. This confounding by indication arises, for example, when individuals with more severe disease more frequently receive a certain treatment. Additionally, selection into a data source, incomplete follow-up, misclassified exposures, or differently recorded outcomes can distort the results.
Countermeasures begin with a plausible design and clinically justified variables. Methods such as matching, stratification, regression, or propensity score methods can account for measured confounding factors, but they do not eliminate unknown or poorly recorded confounders. Predefined analyses, sensitivity analyses, and a transparent presentation of missing data improve evaluability. The EMA guideline on missing data emphasizes for confirmatory trials that missing data are a potential source of bias; the same principle is also methodologically significant for observational data.
The temporal alignment deserves particular attention. The start of observation, start of exposure, measurement of covariates, and end of follow-up must be defined in such a way that individuals are not assigned to a group based on information that only becomes available later. A traceable analysis plan also describes the treatment of competing events, the rules for censoring, and the limits of causal interpretation.
Differentiation from the non-interventional study (NIS)
Observational study is a methodological umbrella term. It describes that researchers observe rather than assign a treatment, and it encompasses different epidemiological designs and data sources. A non-interventional study, abbreviated NIS, is by contrast a regulatory defined term for medicinal product studies. According to Regulation (EU) No 536/2014, it is not considered a clinical trial if the medicinal products are prescribed in accordance with the marketing authorisation, the assignment is not predefined by a trial protocol, and no additional diagnostic or monitoring procedures are applied.
Thus, a NIS can be an observational study, but the terms are not congruent. Whether a planned data collection fulfills the regulatory requirements must be examined on a case-by-case basis depending on the purpose, treatment decision, additional procedures, and national requirements. GVP Module VIII covers non-interventional Post-Authorisation Safety Studies in particular detail; however, according to this guideline, a PASS can also be interventional.
The methodological designation thus primarily answers the question of how data are generated and analyzed. The regulatory classification additionally decides the applicable legal framework and the necessary procedures. Both perspectives should be documented prior to the start of data collection and re-examined in the event of substantial changes.
Relevance for clinical trials
Observational studies complement clinical trials by being able to capture patients, treatment patterns, and outcomes under routine conditions. For the subsequent use of the results, the protocol, data management, and analysis objective must align. Particular attention should be paid to the suitability of data sources, the completeness of follow-up, defined endpoints, the data protection basis, and the clear separation between observational data collection and interventional trial.
Full-service CROs such as Mediconomics support the study plan and data management concept, the definition of exposures and endpoints, as well as the statistical planning to control measured confounders. Further services include quality controls of data lineage, monitoring or risk-based quality oversight, medical writing, and documentation for pharmacovigilance or regulatory processes.
Frequently Asked Questions (FAQ)
How does an observational study differ from a randomized study?
In an observational study, the treatment is not assigned by the study protocol. As a result, the protective mechanisms against confounding resulting from randomization are missing; the analysis must account for this limitation.
Are propensity scores a substitute for randomization?
No. They can balance observed factors included in the model. Unknown, unmeasured, or erroneously recorded confounders remain a potential cause for bias.
Does every observational study fall under Regulation (EU) No 536/2014?
No. The regulation applies to clinical trials and excludes non-interventional studies. Which other legal and ethical requirements apply depends on the specific design, the data, and the Member State.
Regulatory references
- Regulation (EU) No 536/2014 – differentiates clinical trials from non-interventional studies.
- GVP Module VIII “Post-authorisation safety studies” – contains requirements and recommendations for PASS with a focus on non-interventional PASS.
- EMA/CPMP/EWP/1776/99 Rev. 1 “Guideline on Missing Data in Confirmatory Clinical Trials” – describes missing data as a potential source of bias.
- ICH E9 “Statistical Principles for Clinical Trials” – provides principles for planned statistical analysis.