An observational study is also called an epidemiological study,^[1] in which researchers observe and document exposures and outcomes without carrying out any changes or interventions.^[2] The researchers then try to determine if and how the observed exposures and outcomes are related. When one increases, does the other decrease? If variables move together, they’re said to be “correlated”. This allows a researcher to predict the behavior of one variable from others. For example, HbA1c is strongly related to diabetic complications.^[3] If someone’s HbA1c is high, it’s reasonable to predict that they’re at higher risk for problems from diabetes. But this doesn’t mean glycated hemoglobin is directly causing diabetic complications. It’s a byproduct of the high blood sugar and other metabolic changes that happen when someone has diabetes, and diabetes is the downstream cause. The chemical changes in hemoglobin have very little to do with the actual problems that diabetes causes because it’s just a marker.

This highlights a very important point. Observational studies usually can’t establish causality. If the researchers could wave a magic wand and de-glycate all the hemoglobin in a person’s body, they still wouldn’t solve the underlying problem: the person would still have diabetes. Correlation is not necessarily causation, and observational studies usually can’t prove causality. While there are some ways to strengthen causal inference from observational studies at the study design or analysis level,^[4][5][6] the average observational study does not allow researchers to infer causality and only provides predictive power.

There are several types of observational studies, including cross sectional, case-control, case-crossover, retrospective and prospective cohort studies.