Before a vaccine is approved for general use, its protective efficacy must be demonstrated, usually in a double-blind, randomized clinical trial, the gold standard for scientific validity [1]. Randomization assures lack of bias in allocation of the exposure (vaccine), whereas blinding assures lack of bias in ascertainment of the outcome (infection). Nevertheless, there are a number of disadvantages, both practical and scientific, to randomized clinical trials to assess the efficacy of vaccines [2]. Because large samples and relatively prolonged follow-up may be necessary for adequate statistical power, these studies are extremely expensive. To limit costs, they often are conducted in select populations with an unusually high incidence of the infection of interest. These and other factors, such as the carefully controlled conditions of an experimental study, may lead to questions about the generalizability of the results of such trials to target populations that differ from that in which the trial was conducted. In addition, because clinical trials of experimental vaccines usually are conducted for only a relatively short period, the efficacy of the vaccine over time rarely is assessed. Case-control studies, a type of nonexperimental study that may be subject to a number of potential biases, nonetheless may be very useful to answer questions about the efficacy of a vaccine in actual practice (also termed “effectiveness”) once it has been approved, including the vaccine’s effectiveness in subgroups of patients and its effectiveness over time [2, 3]. I will demonstrate that it is possible to incorporate strategies to minimize bias and to assess the potential effects of bias to help to assure the validityof case-control studies ofthe effectiveness of vaccines. Methods to minimize bias in case-control studies include prospective identification of a consecutive series of potential case subjects, random selection of controls from a list of potential controls, matching of controls to cases on potential confounders such as age and socioeconomic status, and use of statistical techniques, such as stratification or logistic regression, to adjust for differences in potential confounders between cases and controls. However, even with the use of such techniques, one cannot be sure that bias has not affected the results. Although it is not possible to allocate subjects randomly in observational studies (leading to uncertainty about the effects of unrecognized confounders), use of “sham” outcomes (which should not be affected by the exposure of interest) and of “sham” exposures (which should not affect the outcome of interest) provide an opportunity to assess whether bias might have affected the results of the study [6]. In a case-control study of the effectiveness of pneumococcal polysaccharide vaccine (PPV23), we showed that the vaccine was effective against infections caused by serotypes included in the vaccine; however, using identical methods to identify cases with invasive pneumococcal infections and matched controls without pneumococcal infections, the vaccine was not effective against infections caused by serotypes not in the vaccine (a sham outcome) [7]. If the difference between cases with infections due to serotypes in the vaccine and their matched controls in the odds of having received PPV23 were due to bias, one might expect to see a similar (erroneous) difference between cases infected by serotypes not in the vaccine and their matched controls. We also assessed the odds of having received influenza vaccine (a sham exposure), a vaccine that was also indicated for all subjects. Unlike for PPV23, there was no statistically significant difference in the odds of having received influenza vaccine between cases and matched controls, additional evidence that the apparent effectiveness of PPV23 was not a result of bias [7]. W e also conducted a case-control study to assess the effectiveness of varicella vaccine over time [8,9]. The efficacy
Read full abstract
Jul 2, 2023
A M Kostinov + 2
Open Access
