|Home | About | Journals | Submit | Contact Us | Français|
Byung-Kwang Yoo, M.D., Ph.D., Division of Health Policy and Outcomes Research, Department of Community and Preventive Medicine, University of Rochester, School of Medicine and Dentistry, 601 Elmwood Avenue, Box 644, Rochester, NY 14642, Phone: (585) 275-3276, Fax: (585) 461-4532, ude.retsehcor.cmru@ooY_gnawK-gnuyB
This study measured short-term responsiveness of influenza vaccine demand to ongoing influenza epidemic, analyzing a 5-year period since 2000 during which influenza seasons varied regarding the timing and severity of the epidemics, and vaccine supply. It specifically tested the hypothesis that weekly influenza epidemic change is positively associated with overall annual influenza vaccine receipt as well as daily vaccine receipt.
Cross-sectional survival analyses from the 2000–2001 to 2004–2005 influenza seasons among community-dwelling elderly using the Medicare Current Beneficiary Survey (MCBS) (unweighted/weighted N= 2,280–2,822/7.7–9.7 million per season). The outcome variable was daily vaccine receipt. Covariates included the biweekly changes of epidemic and vaccine supply at nine census-region levels.
In all five seasons, biweekly epidemic change was positively associated with overall annual vaccination (e.g., 2.7% increase in 2003–2004 season) as well as earlier vaccination timing (p<.01). For example, unvaccinated individuals were 5–29% more likely to receive vaccination, after a 100% biweekly epidemic increase.
Accounting for short-term epidemic-responsiveness in predicting demand for influenza vaccination may improve vaccine distribution and the annual vaccination rate, and might assist pandemic preparedness planning.
Seasonal influenza disproportionately affects the elderly. Approximately 90% of the 36,000 influenza-related deaths and 63% of the 226,000 influenza-related hospitalizations per year in the U.S. occur among the elderly.1, 2 Despite the significant disease burden of influenza disease and the benefits of influenza vaccination,3, 4 influenza vaccination coverage levels among the non-institutionalized elderly have fluctuated between 60% and 67% since 1997.5
Influenza vaccine coverage rates tend to be lower when vaccine supply delays and/or shortages occur as observed in several seasons since 2000 (Table 1).6–8 Fortunately, vaccine production capacity has improved, in part because the number of influenza vaccine manufacturers supplying the U.S. market has increased from three in 2004 to five in 2007.9 During this period, the vaccine supply has increased from 57 million to 130 million doses. However, problems in supply remain as evidenced by the suspension of 46 million doses from one manufacturer during the 2004–2005 season,9, 10 and a problem with regulation compliance by manufacturers.11
Despite progress in vaccine production capacity, supply and demand for influenza vaccine remains a “chicken and egg” problem7 in which fluctuating or sporadically low demand for the vaccine leads vaccine manufacturers to reduce their supply or exit the market.7, 9 Several policy options have been proposed to address this problem by motivating vaccine manufacturers, distributors and providers to remain in the system.9 These include extending vaccination efforts into January and beyond, reducing the financial burden for patients and providers,9 and instituting a public “buy back program” for unused doses to reduce financial risk for manufacturers.12
These options could also include improvement in the distribution of influenza vaccine because often a large number of doses remain unused even during seasons with vaccine supply problems.9 It is difficult to predict vaccine demand, particularly late-season demand, for two reasons. First, manufacturers and distributors take vaccine orders as early as January of the prior season because the vaccine production takes eight or nine months.9 Secondly, demand may decrease when influenza epidemic activity is perceived to be mild, while demand may increase when an influenza epidemic is perceived to be severe or occur early.7 Epidemic activity usually affects late-season demand because influenza activity peaks after January in most seasons (84% of the seasons from 1976 to 2006).13
Although the effect of ongoing influenza epidemics on vaccine demand was suggested qualitatively by Layton and associates,7 to the best of our knowledge it has not been measured quantitatively in the literature. Previous studies have noted long-term responsiveness of influenza vaccine demand to epidemic activity with a one-year lag, e.g., past year’s epidemic level.14,15,16
The purpose of our study was to measure short-term responsiveness of influenza vaccine demand to ongoing influenza epidemic levels and timing. This information could improve efficiency in distribution of influenza vaccine, particularly after the onset of an epidemic. Knowledge of this association might also help predict short-term, late-season vaccine demand in different geographic areas, enabling better vaccine distribution and redistribution, thus improving the overall vaccine coverage level. Furthermore, measurement of short-term responsiveness to epidemic activity might be helpful in pandemic planning because of the possibility of insufficient vaccine supply compared to the demand for vaccine,17 and the potential for policies that target younger populations who may experience higher attack rates than in seasonal influenza epidemics.18
Specifically, we tested the hypothesis that weekly influenza epidemic change is positively associated with overall annual influenza vaccine receipt as well as daily vaccine receipt. We analyzed a 5-year period since 2000 during which influenza seasons varied with respect to the timing and severity of the epidemics, and vaccine supply.
We conducted a cross-sectional survival analyses of the Medicare Current Beneficiary Survey (MCBS) and associated claims data to examine the association between influenza epidemic levels and influenza vaccine receipt in subsequent weeks among the Medicare elderly population.6 We focused on five vaccination seasons between September 2000 and May 2005. The 2000–2001 season was chosen as the starting season because it was marked by a severe nationwide vaccine supply delay.10, 19 The 2001–2002 season included a moderate vaccine supply delay, the 2003–2004 season had a moderate relative supply shortage later in the season due to unusually high demand, and the 2004–2005 season was marked by a severe supply delay and shortage.10, 19–21
We defined two study populations: the “entire population” and an “unvaccinated subpopulation prior to the start of the epidemic.” The “entire population” was defined as individuals aged 65 or older and continuously enrolled in Medicare Part B from September 1 to May 20, including those who were alive on September 1 but died between September 1 and May 20, in any given season. Medicare managed care enrollees were excluded because dates of vaccination were not available. We also excluded enrollees who resided in a skilled nursing facility because they presumably have less control over vaccination receipt or timing than community dwelling elderly. The effect on the annual vaccination rate was calculated using the “entire population” as a denominator.
The second study population was created for a survival analysis, by additionally excluding the individuals who were either vaccinated or deceased prior to the start of an influenza epidemic. For individuals who died after the start of the epidemic, we excluded their observations after their date of death, keeping them in the analyses up to their death. The sample size of these “unvaccinated subpopulations” ranged between 2,280 and 2,822 (unweighted), and 7.69 million and 9.74 million (weighted) (rows 1–5, column 2 in Table 2). The epidemic start date, which varied at the nine census region level, was determined as the first date of a week when the percent positive in the regional influenza viral surveillance laboratory data reached 5%.22, 23 These data indicate the percent of specimens testing positive for influenza type A and B, collected by state/county public health laboratories and some large medical centers.22 We used epidemic data at the nine census region level, not those at the state level because state level weekly epidemic data showed larger variance and were less accurate in capturing epidemic activity trends in some states with smaller populations.
Our survival model’s outcome variable was the number of days from an influenza epidemic start date until vaccination in each season. To measure this outcome, we used Medicare physician supplier and outpatient claims up to May 20 when the latest regional epidemic ended.
The key independent variable was the biweekly epidemic activity change, i.e., the change rate in the past two weeks, compared to the prior two weeks. Epidemic activity was measured by weekly viral surveillance data [% positive for influenza] at the nine census region level. We examined the association between vaccination and the epidemic activity change, applying a two-week lag because the waiting time for Medicare elderly to see their primary care provider for a checkup was 12.4 days on average in 2001 and 2003.24
Another weekly varying covariate was the change of vaccine supply, measured by the number of vaccine doses distributed in the previous four weeks at the nine census region level.10 This covariate was created with the assumption of a linear increase over time and equal vaccine distribution proportional to the regional populations of all ages using the original monthly national data.10 Because of the difference in the epidemic start date across nine census regions, estimated effects of vaccine supply in survival models reflected vaccine supply variations across regions as well as across different time periods. In other words, “day 1” in a survival model varied across the nine regions. Other covariates were individual factors suggested by the literature,14–16, 25 listed in Table 2.
A Weibull survival model was used to analyze time to vaccination because it fit with the data better than other parametric survival models. It accommodates decreasing baseline hazards, and thus reduces the potential confounding effect of a seasonal trend of vaccination patterns, e.g., a decreasing trend after the start of an epidemic.
Hypothesized associations between a covariate and daily vaccination, i.e., earlier vaccination timing, were tested with an estimated hazard ratio. The possible positive association between a covariate and daily vaccination does not necessarily address the question of whether the covariate contributes to improvement in the overall annual vaccination rate, throughout an entire epidemic period. This question was addressed by measuring the effect on the annual vaccination rate in two ways: (a) the effect of all observed changes in a covariate, and (b) the effect of one unit of change, i.e., (b-i) a 100% biweekly increase in epidemic and (b-ii) a 1 million dose increase in the regional vaccine supply.
The effect of all observed changes in a covariate was estimated based on the difference in the annual vaccination rate (i.e., the cumulative hazard) between two cases predicted by an estimated Weibull model: a case reflecting all observed changes in the covariate, and a hypothetical case in which this covariate did not change throughout an epidemic period. It should be noted that a hypothesized positive hazard ratio of biweekly epidemic change indicates both positive and negative effects on daily vaccination during one season, i.e., the increasing epidemic changes will increase daily vaccination up to an epidemic peak, but the declining epidemic changes will decrease daily vaccination after an epidemic peak. The cumulative former positive effect is likely to be greater than the cumulative latter negative effect during one season, because the vaccination time-trends generally decrease after the onset of an epidemic. Therefore, accounting for the offsetting negative effects after an epidemic peak, the cumulative effect of epidemic changes on the annual vaccination is likely to be positive, but could be very small in magnitude.
Sensitivity analyses included the various measures of ongoing epidemic activity and definitions of an epidemic period. Epidemic measures included (a) change rates and (b) the absolute levels of the epidemic with 1–4 week(s) lag. The epidemic was measured in two ways: (i) % positive in laboratory data, and (ii) mortality rates due to pneumonia and influenza.22, 23
Table 1 presents influenza vaccination rates prior to and during an influenza epidemic period among the “entire population”. Among the entire community-dwelling elderly, approximately 70 percent of total influenza vaccinations, based on survey responses, were noted on Medicare claims records (columns 6 and 7). The majority of vaccinations occurred prior to the start of an epidemic (columns 4–6).
Among this entire population, the vaccination rate during an epidemic period ranged from 0.54% to 18.3% (column 5). This rate and its proportion among the annual vaccination rate tended to be higher when an epidemic started early and/or a vaccine supply problem occurred such as in the 2000–2001, the 2003–2004, and the 2004–2005 seasons.
Table 3 shows the effects of biweekly influenza epidemic activity change and weekly incremental influenza vaccine supply on daily influenza vaccination during an influenza epidemic period. For instance, in the 2000–2001 season, a hazard ratio (HR) of 1.21 (row 1, column 4) indicates that a Medicare beneficiary was 21% (p<.001) more likely to receive an influenza vaccine during the day, conditional on not being vaccinated up to that day, when influenza activity increased by 100% in the past two weeks, compared to the prior two weeks.
Despite the large variations in the vaccination rates during an epidemic period across five seasons (column 5 in Table 3), the association between an epidemic activity change and daily vaccination was comparable and robust in all seasons in terms of the HR magnitude and its statistical significance level (p<.01).
An addition of one million vaccine doses within four weeks in a beneficiary’s residence region was associated with a 70%–160% (1.70–2.59 in HR) increase in the likelihood of daily vaccination in all four seasons estimated (p<.001) (far right column). Since there was no increase in vaccine supply after an epidemic starting in mid-December in the 2002–2003 flu season, an estimated HR is unavailable for this season.
The magnitude of the HR for vaccine supply effect was larger during the 2000–2001 and 2004–2005 seasons when there was a severe supply delay and shortage, respectively, compared to the other two seasons (p<.05). These results could be interpreted to mean that distributed vaccines were more likely to be used during a season with a severe vaccine supply problem.
Table 4 shows the effects of biweekly influenza epidemic change and weekly incremental influenza vaccine supply on annual influenza vaccination rates. The results of the increased daily vaccination presented in Table 3 suggest that, when the epidemic level increased rapidly and/or additional vaccine supply was available, vaccination occurred earlier during an epidemic period.
The effects on the annual vaccination rate were calculated, using the entire population as a denominator, to be comparable with the annual vaccination rates in columns 1 and 2 in Table 4 (same as columns 6 and 7 in Table 1), while Table 3 presents the effects on the daily vaccine receipt where a denominator is the subpopulation unvaccinated prior to an epidemic start.
Effects of all observed changes were listed in columns 4 and 6 in Table 4. For instance, in the 2000–2001 season, if there was no epidemic increase (i.e., constant epidemic activity level throughout), the annual vaccination rate would have declined by 2.42 percentage points (p<.001; row 1, column 4) which accounted for approximately a quarter of the observed vaccinations administered during the same epidemic period (9.81% in column 3).
Effects of the observed epidemic changes ranged in magnitude from 0.19% in the 2002–2003 season, when an epidemic started late with a low activity level, to 2.72% in the 2003–2004 season when an epidemic started early with a high activity level. The effect of the observed vaccine supply changes was a 7.31% increase in the annual vaccination rate in the 2000–2001 season, a season characterized by a severe supply delay.
The effect of a one unit increase in a covariate on the annual vaccination rate is presented in columns 5 and 7 in Table 4. For example, in the 2000–2001 season, if the epidemic activity kept increasing by 100% every two weeks throughout the epidemic period, the annual vaccination rate would have increased by 1.69% (row 1, column 5). In reality, because the epidemic level increased by more than 100%, the effect of observed epidemic changes (2.42% in row 1, column 4) was greater than the hypothetical case of 1.69%.
The effect of a one unit increase in vaccine supply on the annual vaccination rate was smaller in the 2000–2001 season (a 5.11% increase) characterized with a severe vaccine supply delay compared to the effect in the 2004–2005 season, which was marked by a severe vaccine shortage (a 7.79% increase) (rows 1 and 5, column 7). On the other hand, the effects of all observed vaccine supply changes in the former season were greater (a 7.31% increase) than that in the latter season (a 2.50% increase) because of the relatively larger amount of vaccine available in the former season (rows 1 and 5, column 6).
Sensitivity analyses generally yielded comparable results.
Our results are consistent with prior studies,14–16 Specifically, the results confirmed our hypothesis that short-term influenza epidemic activity change was positively and strongly associated with: 1) overall annual influenza vaccination; 2) earlier vaccination timing, i.e., vaccine receipt within two weeks of increases in the epidemic. These findings were seen across all five seasons which differed considerably in epidemic levels, timing, and vaccine supply. All prior studies, using different measures of severity of influenza epidemics, have reported moderate associations between epidemic levels and vaccination in the subsequent year.
Three additional points should be noted in interpreting these findings. First, our estimates of the effect of epidemics and vaccine supply on vaccine receipt are likely conservative due to our use of claims data which underestimate vaccination rates relative to patient survey data (as shown in columns 1–2 in Table 4). Vaccination rates based on survey data are not generally thought to be very accurate, since they tend to have higher sensitivity but relatively low specificity.26, 27 Nevertheless, assuming that survey data are valid and that vaccination timing was similar between survey data and claims data, our estimated effects would increase by 40%–55%.
Second, although the effect sizes for the pattern of the epidemic are moderate, it is likely that an earlier and more severe epidemic would yield even greater effects. For instance, an ongoing epidemic was associated with as large as a 2.72% increase in the annual vaccination rate during the 2003–2004 season when the epidemic started early and the resulting unusually high demand caused a moderate relative vaccine supply shortage later in the season.10, 19–21 If there had been no incremental vaccine supply during the epidemic period in this season, the annual vaccination rate would have decreased by as much as 3.97% among the entire community-dwelling Medicare elderly (row 4, column 6 in Table 4).
Third, our study estimated the epidemic effects on vaccination only among a sub-population of persons who were unvaccinated prior to an epidemic start, accounting for one-third to half of the entire community-dwelling elderly population. Prior studies have examined the past year’s epidemic effects among the entire population. The sub-population we examined tended to have a lower propensity for influenza vaccination, suggested by their delayed vaccination timing and lower rates of past season’s influenza vaccination receipt, compared to those vaccinated earlier and excluded from our survival analyses. We analyzed this sub-population in order to assess ways to improve the vaccination rate among this at risk (and potentially less motivated) population.
Our study has several limitations. First, we assumed that vaccine supply at regional/local levels was proportional to the regional populations. The estimated effect of vaccine supply might differ if more detailed regional/local supply data were available. However, any random measurement error in vaccine supply would bias the vaccine supply effect towards the null.
Second, we were unable to evaluate other potential contributing factors on vaccination behavior such as vaccination recipient knowledge of either epidemics or vaccine (through mass media, health care providers or social networks), and experience from prior influenza seasons in terms of vaccine shortages/surpluses and disease severity.
Our results have policy relevance for estimating both short-term and long-term vaccine demand for seasonal influenza, and may have implications as well for pandemic influenza preparedness. These findings could improve efficiency in redistributing influenza vaccine doses after the onset of an epidemic through improved estimates of immediate demand in different geographic areas. Consequently this might improve overall annual vaccine coverage levels, reducing influenza disease burden. More efficient redistribution would also decrease vaccine wastage, which would reduce the financial loss for manufacturers or public expenditures for vaccine “buy back programs.”12 Also, because the vaccine supply is fixed in the very short-run, a policy to stimulate short-run demand, e.g., by more aggressive outreach programs, would be useful particularly in a lower epidemic activity region. Multiple studies have demonstrated the benefit of patient reminder/recall or outreach for improving influenza vaccination rates.28, 29
Our models could also improve demand predictions for the subsequent season. If the previous year’s demand could be separated into a “baseline-demand component” and an “unusual-demand component” caused by unusual epidemic severity and/or timing, manufacturers’ predicted production amounts might better meet the subsequent season’s demand. This would minimize financial risks of both manufacturers and public “buy back programs,” due to excessive production, and help ensure the survival of the manufacturers (and vaccine production capacity) in the market over the long-run.
Caution is needed in extrapolating our estimates of demand responsiveness to seasonal influenza to pandemic influenza. Uncertainty in many factors, such as the risk of infection, disease severity, vaccine effectiveness, vaccine availability, and guidelines to prioritize target populations, may influence the demand for vaccination to a great extent.
Influenza vaccination is positively associated with weekly changes in influenza epidemics and vaccine supply. Accounting for short-term demand for vaccination based on these changes in an epidemic might improve the distribution of influenza vaccine, increase annual vaccination rate, help stabilize vaccine supply, and could assist preparedness planning for pandemic influenza.
This study is supported by by National Institute of Health (NIH)/National Institute of Allergy and Infectious Disease (NIAID) (1K25AI073915, Principal Investigator (PI) Dr. Byung-Kwang Yoo)
We thank Jay Bhattacharya, MD, PhD, for his support in conception and interpretation of data.
ContributionB.-K. Yoo conceived and supervised the study, conducted analyses, and led the writing., M. Kasajima assisted with the study and analyses., K. Fiscella assisted with the study and analyses., N. M. Bennett assisted with the study and analyses., C. E. Phelps assisted with the study and analyses., P. G. Szilagyi conceived and supervised the study.
Statement of institutional review board approval
This study was approved by University of Rochester Research Subject Review Board RSRB: (RSRB00015882; Principal Investigator (PI): Byung-Kwang Yoo)