|Home | About | Journals | Submit | Contact Us | Français|
Vaccination is critical to controlling disease transmission during a pandemic, yet little is known about how the public’s risk perception and willingness to be immunized evolve as pandemics progress. We sought to evaluate longitudinal trends in risk perceptions and vaccination intentions during the 2009 H1N1 pandemic.
We performed a longitudinal analysis of risk perception and vaccination intention from 10 waves of a survey (May 2009 to January 2010) from a national sample of U.S. adults.
Self-reported perceived risk of becoming infected with H1N1 paralleled H1N1 activity throughout the pandemic’s first year. However, intention to be vaccinated declined from 50% (May 2009) to 16% (January 2010) among those still unvaccinated (27% were vaccinated by January). Respondents that reported previous vaccination against seasonal influenza reported significantly higher H1N1 vaccination intentions (67% versus 26%, p<0.001).
Reported intention to be vaccinated declined well before vaccine became available and decreased throughout the pandemic year. To the extent that prior vaccination for seasonal influenza vaccination is a strong correlate of H1N1 risk perceptions, encouraging seasonal influenza vaccination may benefit pandemic preparedness efforts.
Vaccination is one of the most effective means of controlling illness due to influenza. At no time is control more critical than when a new strain of influenza emerges, causing a pandemic. Willingness to be immunized against a novel strain of influenza appears to change over time,1 yet little is known about how willingness changes with perceived risk as a pandemic evolves from an early threat of unknown severity to a more mature threat with known parameters. To be effective in limiting spread of the disease, strategies to optimize vaccination rates during a pandemic must take into account the public’s changing perception of risk to target those who are hesitant to be vaccinated. Opportunities to gain an in-depth understanding of vaccination behaviors are hampered by the infrequency of pandemics and the lack of longitudinal data on risk perceptions.
Last year’s H1N1 pandemic offered an opportunity to study evolving vaccination behaviors. On March 28, 2009 a 9 year-old California girl became the first confirmed case of H1N1 influenza in the United States.2 Within one month, the U.S. declared a public health emergency;3 within two months the World Health Organization declared a level 6 pandemic (the highest possible).4 Fortunately, by August 2009 it became clear that the death rate from H1N1 was about 0.1–0.3%, comparable to that expected from seasonal influenza, according to a report by the President’s Council of Advisors on Science and Technology.5 However, there was ongoing concern that up to half of the population could become infected, with deaths expected to range from 30,000–90,000. By September 2009, a vaccine for the novel H1N1 strain had been approved by the Food and Drug Administration.6 In anticipation of this development, the American Committee for Immunization Practices (ACIP) released criteria for vaccine prioritization in late July 2009.7 Vaccination began in October 2009. Although production delays impeded initial distribution, by December most states had adequate supplies of the 2009 H1N1 vaccine.8
Despite the availability of a safe and effective H1N1 vaccine6 and an unprecedented public health campaign to promote its use, uptake by December 2009 was disappointingly low. Of the entire U.S. population, only 24% were vaccinated; the number was only somewhat higher (33%) for priority vaccination groups.9
The failure to achieve widespread vaccination uptake has many possible causes. We speculate that some of the failure reflects the public’s evolving perceptions of H1N1 risk. Previous research has highlighted subjective risk as a key predictor of vaccine uptake.10 Yet, almost nothing is known about how the public’s perception of novel risks such as H1N1 track the actual evolution of a pandemic, or whether variations in risk perceptions explain subsequent patterns in vaccination intentions and behaviors. Previous studies have examined the relationship between risk perceptions and intentions at a single point in time,11, 12 but a longitudinal perspective is more policy-relevant, since risk perception evolves over time.
In this paper, we chronicle the U.S. public’s evolving risk perceptions from the 2009 H1N1 influenza and the relationship of these perceptions to vaccination intentions. We drew on the capacity of RAND’s American Life Panel (ALP) to field frequent interviews to a common sample in order to generate a detailed, longitudinal perspective of risk perceptions and vaccination intentions throughout the 2009 H1N1 pandemic. Our research questions were as follows: (1) How did perceived risk of H1N1 infection and of death given H1N1 infection relate to measured disease activity in the U.S. population?; (2) Did perceived risk of H1N1 correlate with reported intention to be vaccinated against H1N1?; and, (3) What other policy-relevant factors were associated with H1N1 vaccination intentions over time?
We analyzed data from online surveys of U.S. adults aged 18 to 91 years participating in the ALP. ALP panelists are recruited from respondents to the University of Michigan's longstanding Survey of Consumer Attitudes. Panelists agree to respond to surveys regularly in exchange for financial compensation. The panel includes both online and offline populations by supplying WebTV to participants who lack Internet access. A detailed description of ALP is available at http://www.rand.org/labor/roybalfd/american_life.html and a comparison to the Current Population Survey is available at https://mmicdata.rand.org/alp/index.php/Main_Page.13 The survey was fielded in 10 waves beginning on the following dates (number of completed responses follows in parentheses): (1) May 26, 2009 (N=2070), (2) June 8, 2009 (N=1979), (3) June 22, 2009 (N=1987), (4) July 6, 2009 (N=1932), (5) July 20, 2009 (N=1874), (6) August 3, 2009 (N=2081), (7) August 17, 2009 (N=1942), (8) September 22, 2009 (N=2090), (9) November 19, 2009 (N=2368), and (10) January 22, 2010 (N=2504). We included in our study only panel members who were eligible to participate in all 10 waves. The study protocol was approved by the institutional review board of the RAND Corporation, Santa Monica, California.
Every survey wave included three core questions. Intention to be vaccinated against H1N1 was measured by asking respondents to estimate the “chances” that they would get an H1N1 vaccine, indicating their estimate on a clickable, visual probability scale from 0 to 100% (see figure 1 from the supplemental materials). This response mode has been used successfully in past studies of intention14, 15 and other expectations,16–18 and has been shown to improve aggregate-level predictions of future actions relative to binary “yes–no” measures.14, 15 For the November 2009 and January 2010 waves, respondents who had already been vaccinated against H1N1 were not asked about intention to be vaccinated, but their intention was coded as 100% once they reported vaccination. All respondents were asked to estimate the risk of getting H1N1 influenza over the next month, as well as the risk of death conditional on getting H1N1 influenza, using the same visual scale from 0 to 100%.
Other key questions asked about past receipt of influenza vaccine and risk factors that would place the respondent in a priority group for receipt of H1N1 vaccine.7 Demographic variables included age, gender, race, income, education, and state of residence, which was recoded as region of residence, corresponding to the ten Health and Human Services Regions.19 Respondents self-classified their race and ethnicity using categories defined by the American Life Panel based on the U.S. Census classifications.20 Race and ethnicity data are collected when panelists enter the American Life Panel and quarterly thereafter, for the purposes of describing panelist demographics.
To construct an objective, longitudinal record of influenza activity against which to compare risk perceptions and intention to be vaccinated, we used publicly available data from the CDC.21 Each week, the CDC reports information collected through the US Outpatient Influenza-like Illness Surveillance Network on percent of outpatient visits to health care providers for influenza-like illness (ILI). This network comprises 3,000 healthcare providers in all 50 states, the District of Columbia and the U.S. Virgin Islands, reporting over 25 million patient visits each year.22 We also used data from the CDC on laboratory-confirmed influenza hospitalizations starting in October 2009. Prior to October 2009, this information was either only reported by stratified age groups and/or was reported in combination with syndromic surveillance-based hospitalizations.
We used chi-square tests and Fisher’s t-tests to evaluate differences between responders and non-responders. We constructed descriptive analyses and three separate linear regression models with the following outcomes (1) perceived risk of getting H1N1 influenza over the next month, (2) perceived risk of death from H1N1 if infected, and (3) intention to be vaccinated against H1N1 influenza. Covariates included demographic characteristics, membership in a priority group for H1N1 vaccine, receipt of seasonal influenza vaccine in the previous season, region of residence and survey wave. Because of repeated measurements, we used robust standard errors to account for intra-subject clustering. Adjusted means for each covariate were generated from the multivariate models.
Our data for perceived risk of getting H1N1 and dying from H1N1 and vaccination intentions were bounded (between 0 and 100), yet linear regressions assume unbounded dependent variables. As a robustness check, we ran ordered logistic regression models, which account for the bounded nature of the data but ignore the interval nature of participants’ responses. The results from these models were similar to the linear regression models. For parsimony, we only report here the linear regression models that take into account the full cardinal properties of the data.
The response rates for the waves, as defined by the number of complete surveys divided by the size of the selected sample for that wave, varied from 64% (wave 5) to 73% (wave 10). Over the ten waves, there were 19,341 observations from 2530 unique respondents. There were no significant differences between respondents and non-respondents in terms of age, income, or gender. However, respondents were significantly more likely to be white (89% versus 83%, p<0.001) and have a bachelor’s degree or higher (44% versus 37%, p<0.001).
The mean age of respondents was approximately 49 years (± 15 years), and 42% were male (see table 1 in the supplemental materials). About 44% had education with a bachelor’s degree or higher, and 89% were white. Overall, 9% were vaccinated for H1N1 by the end of November 2009, and 27% by the end of January 2010. Twenty-three percent were in an H1N1 high priority group per ACIP guidelines. Of the respondents in high-priority groups, 19% were vaccinated for H1N1 by the end of November and 38% by the end of January. Almost half (44%) of all respondents in wave 1 reported receiving a seasonal vaccine in the previous year.
The mean perceived risk of getting H1N1 influenza in the next month reported by respondents was about 9% during the first wave, slowly increasing over the summer (figure 1, upper left). Perceived risk spiked to 18% from August to September, then declined over the next few months to 11% by the end of January. This curve largely parallels objective influenza activity (percent of outpatient visits made for ILI and hospitalizations).
Perceived risk of death if infected with H1N1 influenza was initially high at 14%, then fell, with a small spike to about 13% in July. It declined from the summer through January 2010 to about 10% (figure 1, upper right).
Intention to be vaccinated against H1N1 (figure 1, lower left) was highest at the beginning of the pandemic at 50%, and steadily decreased over time. The lowest point was during the last wave of the survey, at 16% for respondents who remained unvaccinated. However, because some respondents were vaccinated by January, the denominator for this curve decreases for the final two waves. The upper branch in figure 1c includes the 27% of the sample vaccinated by January (by imputing their intention to be 100% once vaccinated), shifting the overall mean vaccination intention to 37%, which is still substantially lower than mean stated intentions at Wave 1. To provide an alternate characterization of constant cohorts, figure 1d (lower right) stratifies respondents’ responses in earlier waves according to whether they received the H1N1 vaccine by the end of January (figure 1, lower right). Mean intention to be vaccinated was uniformly higher for those who were ultimately vaccinated compared to those who remained unvaccinated. However, the slope of the decline was similar for both groups, as evidenced by the largely parallel curves.
Increased perceived risk of H1N1 infection within the next month (table 1) was associated with female gender, lower income, nonwhite race, lower education, and receipt of seasonal influenza vaccine in the past season. In general, responses during later waves (from July 2009 onwards) were associated with increased perceived risk, with the exception of the January 2010 wave, which was associated with a lower coefficient compared to 5 previous survey waves.
Increased perceived risk of dying if infected with H1N1 (table 2) was significantly associated with lower income, nonwhite race, lower education, and receipt of seasonal influenza vaccine in the previous year. Compared to May 2009, mean perceived risk of dying in each survey wave decreased slightly through January 2010, but remained far higher than the actual risk of death (0.1–0.3%).
Intention to be vaccinated against H1N1 (table 3) was associated with higher age, higher income, higher education, membership in a high priority group for H1N1, and receipt of seasonal influenza vaccine the previous season. Intention to be vaccinated increased with perceived risk of getting H1N1 and perceived risk of dying if infected. However, intention to be vaccinated decreased across progressive survey waves, with the largest decreases starting in September 2009 (when both perceived risk and observed disease activity were substantially increasing).
We took advantage of a unique opportunity to repeatedly survey a national sample of the U.S. population at ten points during the first year of the H1N1 influenza pandemic. Our study is the first to conduct multiple surveys of this type over the course of the pandemic, drawing from a national panel of respondents.
Overall, the public’s perceived risk of getting H1N1 in the month following each survey wave tracked objective markers of H1N1 influenza activity, both outpatient visits and hospitalizations. However, respondents grossly overestimated risk of death from H1N1, with estimates ranging from about 10% to 14%, depending on the time point. Perceived risk of death declined somewhat during the year, but estimates remained 100-fold higher than the 0.1–0.3% estimates cited in the PCAST report released in August 2009.5 These data are in keeping with other studies of perceived disease risk, in which people tend to overestimate risk from rare events that are newsworthy and unusual.23, 24
Despite the exaggerated perception of risk of death, which did not decrease substantially across waves, intention to be vaccinated peaked at our first measurement in May 2009 and declined steadily, even in the face of growing numbers of H1N1 infections during the fall months. In fact, by the time that H1N1 vaccination became available in October–November 2009, many previously motivated members of the public were no longer interested in receiving it, a finding that helps to explain the suboptimal immunization rates achieved by the end of January 2010. The fact that significant declines occurred even as the primary wave of the pandemic was building (though the timing of the pandemic varied by region) implies that people’s vaccination decisions were insensitive to the immediacy of the disease threat. Instead, our data are consistent with the idea that vaccination intentions may have been more directly related to the novelty of the risk25 and declined as public health officials gained a more comprehensive understanding of the magnitude of the threat posed by the disease. Furthermore, even those who early on sought access to the vaccine were often unable to obtain it,1 further exacerbating a disconnect between vaccine supply and demand that likely negatively impacted vaccine uptake.26 These results further underscore the need for improved and timely vaccine supply, because delays both allow disease to spread and appear to decrease people’s interest in taking steps to prevent further spread.
Higher perceived risk of getting infected and dying from H1N1 are significantly associated with higher intentions to receive H1N1 vaccinations. In fact, a 1% increase in perceived risk of getting H1N1 corresponded to a 0.57% increase in intention to be vaccinated. Thus, those with a 10% higher perceived risk of getting H1N1 were 5.7% more likely to intend to be vaccinated against H1N1. A particularly intriguing finding was that lower income and education were significantly related to lower intention to get the H1N1 vaccine, but were simultaneously related to higher risk perception. This may suggest that these groups have particularly high distrust of novel vaccines. Alternately these groups may simply be particularly poorly calibrated in their risk perceptions, perhaps because of lower health literacy and numeracy skills. Either way, it seems clear that addressing the needs of the less educated and less affluent is essential in order to optimize vaccine coverage.
Another strong predictor of H1N1 vaccination intentions was a history of being vaccinated against seasonal influenza in the prior season, even after controlling for age and membership in a high priority group. This finding, which has been observed in previous studies of the H1N1 pandemic,11, 12, 27–31 highlights the need to vaccinate as many people as possible for seasonal influenza each year. Furthermore, almost half of those patients discussing H1N1 vaccination with a provider failed to get a recommendation for getting the vaccine,1 a result mirrored in low numbers of healthcare workers getting vaccinated themselves. Therefore, efforts to inform both patient and provider decision making are likely to yield considerable public health benefits beyond protection against seasonal influenza and are critical to preparedness for pandemic influenza.
However, the logistics of linking seasonal influenza vaccination to pandemic influenza vaccination are challenging. For example in this pandemic, the priority groups for H1N1 vaccine were quite different from those for seasonal influenza vaccine in past years. In the future, this may represent less of a problem because the entire U.S. population older than 6 months of age is now recommended to receive seasonal influenza vaccine annually.32 Developing a targeted and effective public health communication strategy will depend on better understanding why those who have received seasonal influenza vaccine were more likely to intend to be vaccinated against H1N1. This subpopulation may be inherently more compliant with government vaccine recommendations, or may have selectively benefited from more effective communication about influenza vaccine (whether seasonal or H1N1). They may have more accurate perceptions of the safety and effectiveness of influenza vaccine, or a higher perceived need for influenza vaccine. In addition, provider recommendations are well known to be a driving force of patient vaccination behavior,33, 34 and receipt of both seasonal and pandemic influenza vaccination is likely linked to frequency of patient-provider interactions. As a result, better education and increased acceptance of vaccinations by healthcare workers and providers could have positive secondary effects by increasing the likelihood of vaccination recommendations.
The observational nature of our study limits our ability to comment definitively on causal relationships. However, the policy implications of our observations (e.g., need for more timely supply of vaccine; need for better education) do not depend on inferring causality. The study provides frequent assessments (10 surveys in 9 months) via a small set of questions, and we endeavored to use all available data. However, missing data compounded across waves (a cost of administering multiple waves), which prevented repeated-measures analyses of a core group of respondents in a regression analysis. Instead, we examined policy-relevant patterns of risk perceptions (e.g., their correspondence to changes in population infection rates) by focusing on repeated cross-sectional analyses, clustering by respondent for regression models when needed. We relied on respondent self-report of being vaccinated. Other studies have found self-report of vaccination to be reasonably reliable.35, 36 Obtaining objective verification of vaccination would be challenging on such a large scale.
No publicly available measure exists for actual risk of getting H1N1 influenza at any given time. As a proxy for an objective measure of H1N1 activity, we used percentage of outpatient visits for influenza-like illness and hospitalizations, as reported by the CDC. Though these measures should accurately track actual H1N1 activity, the absolute numbers would underestimate the number of H1N1 infections as not all infected people would seek medical care or require hospitalization. Thus, we were only able to compare the patterns of perceived risk of getting H1N1 with patterns of influenza activity, and could not comment on whether our respondents over- or underestimated their individual risk. As well, one of our objective markers could be influenced by perceived risk; thus it might not be surprising that perceived risk of getting H1N1 maps to proportion of visits for influenza-like illness. However, hospitalizations are unlikely to be influenced by risk perception.
The representativeness of our sample is underscored by the very similar H1N1 vaccination rates observed in our sample compared to CDC’s national estimates (27% versus 24%). Using a consistent set of three core questions over time allowed us to examine longitudinal patterns and predictors of perceived risk of becoming infected with H1N1 influenza, perceived risk of death if infected, and intention to be vaccinated.
The public’s perception of risk of getting H1N1 influenza tracked the actual evolution of the pandemic, but the risk of death from H1N1 influenza was overestimated even in light of objective data showing a relatively low risk of death. Despite such overestimates, intentions to receive H1N1 vaccination declined as the pandemic progressed, with decreases observed even as perceived risk (and disease activity) was increasing in September 2009. Those who perceived higher risk of getting H1N1 and dying from H1N1 were more likely to intend to be vaccinated, but by far the strongest predictor for H1N1 vaccination intention was a history of getting seasonal influenza vaccine in the previous year. Because prior seasonal influenza vaccination predicts future vaccination for H1N1, encouraging regular seasonal influenza vaccination (both directly with patients and through support for provider recommendations) appears to be a valuable component of pandemic preparedness strategy.
This study was funded by grant 5R01AG020717-07 from the National Institute on Aging. Dr. Zikmund-Fisher is supported by a career development award from the American Cancer Society (MRSG-06-130-01-CPPB). The funding agency had no role in the design and conduct of the study; in the collection, analysis, and interpretation of data; or in the preparation, review, or approval of the manuscript. We would like to thank Tania Gutsche, BA, and Arie Kapteyn, PhD, for their support of this work. In addition we would like to thank Arthur Kellermann, MD MPH, and Eric Schneider, MD MS, for their review of this manuscript, as well as Amy Maletic, BSc, and Mary Vaiana, PhD, for their assistance in preparing this manuscript (all of the RAND Corporation).
The authors have no conflicts of interest to report.