To test the effectiveness of messages designed to reduce vaccine misperceptions and increase vaccination rates for measles-mumps-rubella (MMR).
A Web-based nationally representative 2-wave survey experiment was conducted with 1759 parents age 18 years and older residing in the United States who have children in their household age 17 years or younger (conducted June–July 2011). Parents were randomly assigned to receive 1 of 4 interventions: (1) information explaining the lack of evidence that MMR causes autism from the Centers for Disease Control and Prevention; (2) textual information about the dangers of the diseases prevented by MMR from the Vaccine Information Statement; (3) images of children who have diseases prevented by the MMR vaccine; (4) a dramatic narrative about an infant who almost died of measles from a Centers for Disease Control and Prevention fact sheet; or to a control group.
None of the interventions increased parental intent to vaccinate a future child. Refuting claims of an MMR/autism link successfully reduced misperceptions that vaccines cause autism but nonetheless decreased intent to vaccinate among parents who had the least favorable vaccine attitudes. In addition, images of sick children increased expressed belief in a vaccine/autism link and a dramatic narrative about an infant in danger increased self-reported belief in serious vaccine side effects.
Current public health communications about vaccines may not be effective. For some parents, they may actually increase misperceptions or reduce vaccination intention. Attempts to increase concerns about communicable diseases or correct false claims about vaccines may be especially likely to be counterproductive. More study of pro-vaccine messaging is needed.
Comments
Assessing Vaccine Messages: Statistics Matters
What outcome was used?
Vaccine favorability was analyzed on a six point scale, but vaccine favorability as a proxy for vaccination is binary; vaccination is given or withheld. Statistically, the distinction matters.
Ordinal logistic regression was used to predict the six-point vaccination favorability outcome. The odds ratio for ordinal logistic represents a one-point difference. In Table 3, the authors report that parents who were initially unfavorable toward vaccination had an odds ratio of 1.20 for disease images vs. control. The intuitive interpretation of this value is that the odds of future vaccine favorability are 1.2 times greater for parents shown such a photo than for control parents after adjusting. However, because the six-point scale was used the odds ratio really means that the odds of vaccine favorability being one point higher on the six point scale are 1.2 times greater for parents shown a photo than for control.
Furthermore, the ordinal odds ratio must be identical for each step of the six-point scale. If the increase were actually among parents who already favored vaccine, the results are meaningless. On the other hand, if the increase were primarily among parents who were initially somewhat suspicious, then the impact of the study may be greater than reported.
What should happen now?
Typically statisticians request that investigators not perform multiple analyses, to reduce the risk of type I error (finding statistical significance when no difference exists in the population). In this case, I propose that the authors perform new analyses. Time and money have been invested in this well-designed study, and we should glean the full potential impact of these important results.
For the "intuitive" odds ratio, the six-point outcome should be categorized into two responses to create a binary outcome, those favorable or unfavorable to vaccination at followup, and analyzed with logistic regression among parents initially unfavorable toward vaccination. As a secondary analysis, it would be useful to limit the results to those parents who were somewhat unfavorable or neutral toward vaccination, excluding parents strongly opposed to vaccination. Two related analyses would also help; effect modification by autism training, and number-needed-to-treat.
Significance
The results of the reanalysis are not trivial. The intervention which looked most favorable was a photo of a sick child. The Center for Disease Control and Prevention publishes Vaccine Information Sheets which are provided with each vaccination in the nation. Adding a photo to each sheet might modify vaccination rates and save lives.
Conflict of Interest:
None declared