The design begins by classifying people according to an exposure recorded in the past, then reconstructs their subsequent follow-up during a defined historical period. This ordering allows researchers to compare later outcomes between exposed and unexposed groups rather than selecting participants based on whether the outcome already occurred. It supports assessment of temporal association.
Confounding and selection bias can make the observed difference between exposed and unexposed groups difficult to interpret as a causal effect. Confounding concerns the influence of other factors on the comparison, whereas selection bias reflects how inclusion or group formation may affect the study population. Recognizing both limitations helps prevent overstatement of medical conclusions.
This design is particularly useful when a prospective study would be impractical or would require many years of follow-up. Existing medical records, registries, and databases allow investigators to examine historical exposure and outcome information without waiting for future events to occur. That efficiency can make treatment, adverse-effect, prognostic, and risk-factor questions more feasible.
Investigators first identify the population and classify individuals by exposure status using existing records. They then specify the historical follow-up period, determine which later health outcomes are recorded, and compare outcome frequency, incidence, or risk between exposure groups. Medical records, registries, or databases provide the underlying information, so the study depends on how completely those sources document events.
In medicine, the approach can examine treatment effectiveness, adverse effects, prognostic factors, and risk factors. Depending on the available data, researchers may compare outcome frequency, incidence, or risk across exposure groups. These results can contribute to clinical decision-making and public health policy, especially when historical data offer evidence that would be difficult to obtain prospectively.
Incomplete records can limit accurate identification of exposures, follow-up information, or later outcomes, weakening the comparison between groups. Confounding and selection bias may also remain and restrict causal interpretation. Findings are therefore most appropriately treated as evidence of an association that can inform decisions, rather than as automatic proof that an exposure caused an outcome.