Before and After Study

A before and after study, also known as a pre post study, is a research design that measures outcomes in the same group of participants or settings before and after an intervention is implemented, without a separate contemporaneous control group. The change in outcomes from the pre period to the post period is taken as the apparent effect of the intervention, although other changes over the same period cannot be separated from it.

Before and after studies are widely used designs in health services research, quality improvement, public health, and education because they are practical, inexpensive, and easy to implement. They are often the only feasible design when randomization is not possible, when the intervention is applied to an entire population or system simultaneously, or when the research aims to evaluate a policy or program that has already been implemented. However, their results must be interpreted cautiously because the absence of a control group makes it impossible to distinguish the intervention's effect from other changes occurring simultaneously.

The design involves three phases. In the pre intervention phase, the researcher collects baseline data on the outcome of interest over a defined period. In the intervention phase, the new treatment, policy, program, or practice change is implemented. In the post intervention phase, the same outcome is measured again using the same methods and instruments.

The primary analysis compares the pre and post measurements. When the same participants are measured twice, paired tests are used, such as the paired t test (for continuous outcomes) or McNemar's test (for binary outcomes); when different people are sampled in each period, such as consecutive admissions before and after a change, the two periods are compared as independent groups. The difference between the pre and post means (or proportions) is the observed change, which is often reported as the intervention's effect even though it also contains the other changes described below.

The fundamental limitation of this design is the absence of a control group to account for factors other than the intervention that might have caused the observed change. These threats to internal validity include temporal trends (the outcome was already improving or worsening before the intervention), regression to the mean (extreme baseline values naturally move closer to the average over time, regardless of intervention), maturation (participants change over time due to natural development or aging), contemporaneous events (other changes in the environment coincide with the intervention and affect outcomes), and testing effects (the act of measuring participants at baseline influences their subsequent behavior or scores).

Several design modifications can strengthen before and after studies. Interrupted time series analysis uses multiple data points before and after the intervention to model the underlying trend and detect changes in level or slope attributable to the intervention. Adding a non equivalent control group (one that does not receive the intervention but is measured at the same time points) creates a controlled before and after study, which provides stronger evidence than an uncontrolled design.

Before and after studies are appropriate for evaluating system level interventions that cannot be randomized, for preliminary assessments of promising interventions before investing in a randomized trial, for quality improvement projects where the primary goal is practical improvement rather than causal inference, and for policy evaluations where the policy has already been implemented universally. They are not appropriate when strong causal evidence is required, when randomization is feasible, or when the outcome is known to be subject to substantial temporal trends or regression to the mean.

A common mistake is interpreting the pre post difference as proof that the intervention caused the change, without acknowledging the threats to internal validity. Any observed improvement may be partially or entirely due to regression to the mean, temporal trends, or concurrent changes unrelated to the intervention. Researchers should explicitly discuss these threats and, where possible, present additional evidence (such as time series data or comparisons with untreated settings) to support a causal interpretation.

Another error is failing to use the same measurement methods at both time points, which introduces measurement bias. If the outcome assessment changes between the pre and post periods (for example, due to a change in diagnostic criteria or recording practices), the apparent intervention effect may be an artifact of measurement differences. Finally, when the same participants are measured at both time points, researchers sometimes calculate sample size as if the pre and post measurements were independent, when in fact they are paired, leading to incorrect power calculations.

A hospital implements a new electronic medication reconciliation system and measures medication discrepancy rates in 500 consecutive admissions during the three months before implementation and 500 consecutive admissions during the three months after. The discrepancy rate drops from 34% to 21% (absolute reduction: 13 percentage points, p < 0.001). However, the researchers note that a concurrent hospital wide patient safety campaign was launched during the same period, making it impossible to attribute the entire reduction to the reconciliation system alone. They recommend a controlled study comparing the system's performance with that of similar hospitals that did not implement it.