Confounding in Observational Studies Explained

Abstract

Practical and ethical constraints mean that many clinical and/or epidemiological questions cannot be answered through the implementation of a randomized controlled trial. Under these circumstances, observational studies are often required to assess relationships between certain exposures and disease outcomes. Unfortunately, observational studies are notoriously vulnerable to the effect of different types of "confounding," a concept that is often a source of confusion among trainees, clinicians and users of health information. This article discusses the concept of confounding by way of examples and offers a simple guide for assessing the impact of is effects for learners of evidence-based medicine.

Non-steroidal anti-inflammatory drugs and the risk of a second hip fracture: a propensity-score matching study

“…We previously reported that age, female gender, comorbidities, and the prolonged use of analgesics and anti-inflammatory medications are all significant risk factors for a second hip fracture after hip fracture surgery, but that bisphosphonate therapy was protective [ 3 ]. However, in an observational study, there are many confounders that usually distort the relationship between an exposure and an outcome [ 20 – 22 , 25 ]. The PSM method is useful in these circumstances because it provides a neutral weighting formula that yields unbiased estimates of the effects of treatment.…”

“…During the last decade, interest in determining the degree of association and/or causal link between medications and the risk of a fragility hip fracture has grown [ 17 , 19 ]. However, establishing the degree of association and/or causal link is difficult because observational studies are notoriously vulnerable to the effect of different types of confounding [ 20 ]. It is well recognized that the estimate of causality obtained by comparing cases with controls could be prejudiced because of problems such as selection bias or other systematic errors [ 20 – 22 ].…”

“…However, establishing the degree of association and/or causal link is difficult because observational studies are notoriously vulnerable to the effect of different types of confounding [ 20 ]. It is well recognized that the estimate of causality obtained by comparing cases with controls could be prejudiced because of problems such as selection bias or other systematic errors [ 20 – 22 ]. Rosenbaum and Rubin developed and popularized the propensity-score matching (PSM) method for observational (non-experimental) studies, in which only a small subset of controls comparable to cases must be selected [ 21 ]; PSM reduces the selection bias by balancing groups on the probability of being treated based on specific covariates [ 22 ].…”

The association between infantile microcytic anemia and attention deficit hyperactivity disorder, a case-control study

“…The results of the stratification according to SES suggest that SES might also be a partial confounder in the association between any microcytic anemia and ADHD, as the associations weakened in each of the SES strata as compared to the association in the entire sample. A confounder is a factor that is associated with the exposure that independently affects the risk of developing the outcome [ 37 ]. Our data indeed showed associations between the exposure (any microcytic anemia) and both age at data collection and SES, as microcytic anemia prevalences were different among the different strata.…”

Association Between Workarounds and Medication Administration Errors in Bar Code-Assisted Medication Administration: Protocol of a Multicenter Study

“…We will try to reduce confounding by applying multivariate regression analyses (eg, hospital type, type of ward). However, in this type of observational study design, residual confounding may always remain [ 52 ].…”