Live Attenuated or Inactivated Influenza Vaccines and Medical Encounters for Respiratory Illnesses Among US Military Personnel FREE

Military personnel are prone to outbreaks of respiratory illness such as influenza for a variety of reasons, including crowding and stressful conditions.^1- 3 Before the availability of an influenza vaccine, the military population experienced high mortality and morbidity during such outbreaks. Trivalent inactivated vaccine (TIV), administered intramuscularly, was first developed and tested in the military in the 1940s and has been used annually since the 1950s to prevent influenza and its complications.⁴

In 2003, a live attenuated influenza vaccine (LAIV) with the same antigenic characteristics as TIV was formulated for intranasal application and approved for use among healthy adults. Service members were immediately targeted for LAIV use by the US Department of Defense (DOD) because of the ease of vaccine administration and availability early in the season. During the TIV vaccine shortage in 2004, the DOD agreed to preferentially use LAIV to increase the availability of TIV.⁵ Although TIV remained the predominant vaccine until the 2006-2007 season, LAIV has increasingly become the preferred vaccine for service members while TIV is reserved for those with higher risk for respiratory diseases or contraindications to LAIV.^6- 7

Recent clinical trials comparing LAIV with TIV suggest that LAIV has superior efficacy over TIV among young children.^8- 10 However, among healthy adults, similar¹¹ or lower¹² levels of protection were found for LAIV compared with TIV. In the military population, the only studies conducted to determine the comparative effectiveness of these vaccines have been performed among basic trainees.^13- 14 In this young adult population, LAIV was found to be equally effective compared with TIV, although the estimate for LAIV was derived from a single training site. Because basic trainees differ from other service members with respect to their exposure risk and immunization history, the generalizability of these results is limited. To better inform US military influenza vaccination policy, vaccine effectiveness needs to be defined among personnel who compose the majority of the military population, those who are not basic trainees.

To fulfill military health surveillance duties and provide relevant health information about postlicensure effectiveness of LAIV in a young healthy population, we investigated the incidence of health care encounters for pneumonia and influenza illness among active-duty service members eligible for influenza vaccination who were stationed in the United States during the 2004-2005, 2005-2006, or 2006-2007 influenza season.

The Defense Medical Surveillance System (DMSS) is a large relational database containing longitudinal data about demographic characteristics, occupations, immunizations, and medical encounters for US military service members. Data collection begins at the time of entry into service and continues throughout the military career.¹⁵ Race/ethnicity is self-reported and stored in the demographics data set. Active-duty service members have universal health care coverage and must seek care through the military health care system, which allows near complete capture of medical history in the DMSS.

Using the DMSS, we identified yearly cohorts of service members on active duty for at least 1 month during the surveillance period (September 1 to April 30 for each year of interest). Inclusion criteria required being stationed in the United States at the beginning of the influenza season. Because LAIV is indicated for adults as old as 49 years, we included only those adults aged between 17 and 49 years as of September 1 of 2004, 2005, or 2006. Members were excluded if they received more than 1 dose of influenza vaccine in the season of interest, had a medical exemption from influenza vaccination, or were undergoing basic training at the beginning of the risk period (defined below). Recruits were defined with service-specific training duration (Army, 8 weeks; Air Force, 6 weeks; Marine Corps, 12 weeks; Navy, 7 weeks) and age requirements (Army, 16-43 years; Air Force, 16-28 years; Marine Corps, 16-30 years; Navy, 16-36 years). Pregnant women were excluded because they can only receive TIV.

Furthermore, vaccine-naive cohorts were longitudinally constructed to represent populations with no immunization in the prior 1 or 2 seasons. Immunization status of service members was determined by evidence of 1 dose of LAIV or TIV during the season of interest using data routinely maintained in the DMSS. Eligible service members with no documentation of influenza vaccination during the season of interest were considered unimmunized. For the immunized group, the risk period began 15 days after the date of immunization and continued to the end of the surveillance period or a censoring event. For the unimmunized group, person-time at risk began 15 days after the median date of immunizations in the immunized group and continued to the end of the surveillance period or a censoring event.¹⁶ A censoring event occurred when a member began overseas deployment, left active duty, had a medical encounter for the outcome of interest (pneumonia or influenza), or left military service during the risk period.

The Armed Forces Health Surveillance Center has been directed by military authorities to conduct public health surveillance of respiratory infectious diseases and evaluation of related protection measures. According to 45 CFR 46.101/102, this activity does not constitute research, and institutional review board examination is not required. No external funding was used to conduct this investigation, and contents have been cleared for public release by the US Army Center for Health Promotion and Preventive Medicine.

The outcome of interest was defined as a health care encounter during the risk period resulting in a primary diagnosis code consistent with pneumonia or influenza based on codes 480 through 487 from the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM). Only the first event per individual was counted each season, and person-time was censored on the date of diagnosis. Incident hospitalizations were reported as a secondary outcome.

The incidence rate of pneumonia and influenza per 1000 person-years was calculated for the unimmunized, TIV-immunized, and LAIV-immunized groups. Crude incidence rate ratios (IRRs) comparing the TIV-immunized and LAIV-immunized groups with the unimmunized group were calculated. For the adjusted analysis, a multivariate Poisson regression model and a propensity-based matching model were both used.

A multivariate Poisson regression model was used to adjust for known and potential confounders. Variables included in the final model contained age; sex; service branch; history of ever having been hospitalized; and, in the 12 months prior to influenza season, the number of medical encounters, number of influenza-like illness medical encounters, and history of influenza immunization. An influenza-like illness event was defined as an inpatient or outpatient medical encounter with 1 of the ICD-9-CM diagnosis codes that have previously been shown to be associated with culture-confirmed influenza illnesses among the US military population.¹⁷ Adjusted IRRs and 95% confidence intervals (CIs) were determined comparing LAIV- and TIV-immunized groups separately with the unimmunized group. Vaccine effect was calculated as 1 minus the adjusted IRR. A 2-sided significance test at the level of .05 was used with an adjustment for multiple comparisons using the Bonferroni procedure. For trend analysis, the Cochran-Amitage trend test was used.

To better balance the covariates in the immunized and unimmunized groups, a propensity-based matching analysis was also conducted.¹⁸ The propensity score was calculated based on age, sex, service branch, history of ever having been hospitalized, medical encounter history, immunization history, and influenza-like illness encounter history in the previous year. Each immunized service member was matched to the closest unimmunized service member whose propensity score differed by less than 0.10.¹⁹

With an incidence rate in the control group between 10 and 20 incidences per 1000 person-years, the power to detect a 10% to 30% change in this rate among immunized groups is between 93% and 99% (α = .05). All analyses were performed using SAS 9.1.3 (PROC GENMOD and PROC LOGISTIC; SAS Institute, Cary, North Carolina).

The total cohorts were 1 061 728 for 2004-2005, 1 041 264 for 2005-2006, and 1 067 959 for 2006-2007 (Figure 1). Immunization rates ranged from 51.9% in the 2004-2005 to 78.4% in the 2006-2007 influenza season. The proportion of immunized persons receiving LAIV increased from 33.5% in the 2004-2005 influenza season to 47.9% in the 2006-2007 season.

The demographic characteristics of the cohort in the 2006-2007 influenza season are presented in Table 1. Overall, unimmunized service members were more likely to be in the Navy and Marine Corps, whereas the immunized group had a high percentage of Army and Air Force service members. Compared with the immunized group, the unimmunized group had a higher percentage of 40- to 49-year-olds, was less likely to have had medical encounters in the prior year, and was less likely to have been vaccinated in the prior season. With propensity-score matching, the standardized differences in the reconstructed cohorts (Figure 1) decreased to less than 10% (Table 1), albeit with fewer members than were in the total cohort (eg, the 2006-2007 cohort with LAIV matched to TIV had 618 896 members out of a total of 837 230) (Figure 1).

Weekly patterns of incident pneumonia and influenza encounters were in synch with surveillance data from the Centers for Disease Control and Prevention monitoring isolates that were culture-positive for influenza in the general population (Figure 2). Furthermore, the seasonality of morbidity related to pneumonia and influenza in our population was consistent with mortality data from the Centers for Disease Control and Prevention (data not shown).

The incidence rate of health care encounters for pneumonia and influenza was highest in the unimmunized group each season (19.4, 10.9, and 11.7 per 1000 person-years for the 2004-2005, 2005-2006, and 2006-2007 seasons, respectively) (Table 2). The LAIV immunized group had the next highest incidence with rates of 18.3, 10.6, and 11.1 per 1000 person-years for the 2004-2005, 2005-2006, and 2006-2007 influenza seasons, respectively (P < .001). The lowest incidence was found in the TIV immunized group (8.6, 7.8, and 8.0 per 1000 person-years for the 2004-2005, 2005-2006, and 2006-2007 seasons, respectively; P < .001).

The incidence rates of hospitalizations for pneumonia and influenza were highest in the LAIV immunized group for each of the 3 seasons, and the adjusted incidence rate in this group was significantly higher than that in the unimmunized group during the 2004-2005 season (adjusted IRR, 2.14; 95% CI, 1.30 to 3.54; P = .01) but not during 2005-2006 (IRR, 1.18; 95% CI, 0.69 to 2.01; P = .49) or 2006-2007 (IRR, 1.11; 95% CI, 0.71 to 1.74; P = .53).

The effect of vaccination for each individual vaccine group was highest during the 2004-2005 influenza season, followed by the 2005-2006 season, and lowest during the 2006-2007 season. The estimated effect of vaccination with LAIV ranged from 10.7 (95% CI, 2.72 to 18.1; P = .03) in 2006-2007 to 20.8 (95% CI, 12.3 to 28.5) in 2004-2005 using the adjusted Poisson regression model and 5.9 (95% CI, −9.2 to 18.9) in 2005-2006 to 11.8 (95% CI, 0.85 to 21.5; P = .04) in 2006-2007 using propensity-based matching (Table 3). In reference to the unimmunized group, the effect of vaccination with TIV was higher than LAIV, with an estimated effect ranging from 28.4 (95% CI, 21.9 to 34.3) to 54.8 (95% CI, 51.3 to 58.1) for TIV and from 10.7 (95% CI, 2.7 to 18.1) to 20.8 (95% CI, 12.3 to 28.5) for LAIV.

The effect of vaccination was also analyzed in vaccine-naive cohorts. For the 2005-2006 and 2006-2007 seasons, there were 329 471 and 168 663 persons, respectively, in these cohorts. The incidence rates of pneumonia and influenza were similar between the vaccine-naive cohort and the total cohort for the unimmunized group (Table 2). However, incidence rates were lower among LAIV or TIV recipients in the vaccine-naive cohort compared with the total cohort (eg, LAIV during 2005-2006 season, 7.7 per 1000 person-years in vaccine-naive cohort vs 10.6 per 1000 person-years in total cohort). This yielded a higher effect of vaccination for both LAIV and TIV in the vaccine-naive cohort compared with the total cohort, particularly for LAIV. Among the vaccine-naive cohorts, estimates for LAIV and TIV effect were statistically similar in both the 2005-2006 (P = .53) and 2006-2007 (P = .56) seasons.

The effect of vaccination was compared with vaccination history (years of being vaccine-naive, 2 years vs 1 year vs none) for the 2006-2007 season. The estimates were obtained from the total cohort and vaccine-naive cohorts during the 2006-2007 season (Table 3). The correlation between years of being vaccine-naive and the effect of vaccination was statistically significant for LAIV (P = .04) but not for TIV (P = .63). Similar correlations were seen in propensity-matched cohorts (data not shown).

We followed a large healthy population through 3 influenza seasons and assessed health care use, immunization status, demographic information, and person-time at risk. To the best of our knowledge, this is the first systematic assessment and comparison of the effect of influenza vaccination in a population of healthy young adults with high vaccination coverage. Results of this multiseason surveillance report suggest that annual immunization could reduce influenza-related morbidity during seasons with varying levels of influenza activity.

Immunization with TIV was associated with a significantly lower incidence rate of clinical encounters for pneumonia and influenza during the 2004-2005 through 2006-2007 seasons. In an attempt to reduce the selection bias in observational studies, we performed a propensity-matched cohort analysis.²⁰ The consistent estimates for TIV between the total and propensity-matched cohorts argue that TIV administration was not biased among the known covariates included in the model. This was expected because TIV was the predominant vaccine and ubiquitously distributed in most military treatment facilities for use in the event of a local outbreak. On the other hand, LAIV was used only in select services in the 2004-2005 influenza season and did not equal the level of TIV administration until the 2006-2007 season. Thus, we observed a similar reduction in the incidence rate for the LAIV immunized group only during the 2006-2007 season, in either the total or propensity-matched cohort. In addition to limited LAIV administration, other factors might have affected the effectiveness of LAIV, such as the change in formulary from frozen to refrigerated form before the 2006-2007 season. Our results from 2006-2007 were consistent with the findings of the pivotal efficacy trial for LAIV in healthy adults.²¹ Furthermore, our design enabled us to compare TIV and LAIV effect in the same setting.

A recent meta-analysis review of influenza vaccine effectiveness in healthy adults suggested that the intranasal vaccine is less effective than TIV.²² Similarly, Ohmit et al^12,23 recently reported LAIV to be less efficacious in reducing culture-confirmed influenza infection compared with TIV in a randomized control trial conducted during the 2004-2005¹² and 2005-2006²³ influenza seasons. It is worth noting that approximately 50% of the members in the Ohmit et al trial were immunized in the prior season (comparable with our propensity-score matched cohort) (Table 1). These findings are supported by our results. Reports of equal efficacy or effectiveness for LAIV were mostly from infants and young children with limited history of influenza vaccination.^8- 10

Live attenuated influenza vaccine was found to have an effect similar to TIV in the vaccine-naive cohort. This suggests that preexisting vaccine-induced immunity may play a role in determining the effectiveness of LAIV. We think that this modulation may occur because the attenuated virus from LAIV must undergo self-replication in the individual receiving the vaccine in the presence of neutralizing antibodies. Previous studies have reported that children and adults who were seropositive at baseline were less likely to have a serologic response to LAIV compared with those who were seronegative participants.^24- 26

To assess preexisting immunity, we used vaccination history in previous years as a surrogate. The correlation between the effect of LAIV and vaccination history in previous years was in contrast to a lack of correlation between TIV effect and vaccination history. This unique feature of LAIV may not only explain its apparent lack of effect in annually immunized nonrecruits, but also predict its higher relative effect in different populations such as young recruits and children.

Estimates of vaccine effectiveness using observational data depend on many factors, such as virulence of the circulating virus, degree of strain match between the vaccine and the circulating virus, herd immunity,²⁷ specificity of outcome measures,²⁸ completeness and accuracy of vaccination records, and case ascertainment. Among the 3 seasons, the 2004-2005 season was the only season when mortality due to pneumonia and influenza exceeded the epidemic threshold.²⁹ For the military population, the combination of relatively low vaccine coverage and high influenza activity resulted in a relatively high incidence rate of pneumonia and influenza, particularly in the unimmunized group. Consequently, estimated effect of immunization during the 2004-2005 season was higher for TIV compared with the other 2 seasons investigated.

Although our surveillance benefited from a large number of members with detailed demographic and medical information, several factors limit the validity and interpretation of our findings. The first limitation is the confounding issue. Even though we used propensity-based matching to reduce bias associated with the covariate distribution, we did not have information about other important confounders such as smoking status.

The discrepancy in the results between the multivariate regression model and the propensity-based matching analysis suggests that propensity-based matching better adjusted for confounding than the traditional regression method. Propensity-based matching has been shown to be advantageous over multivariate regression models in that it provides the degree of matching between 2 comparison groups in terms of their covariate distributions and avoids extrapolation used in a regression model.³⁰ We used this method to analyze both the total and vaccine-naive cohorts. We included only the variables used in the multivariate regression model to maximize the number of propensity-matched pairs.³¹ Although conditioning on the propensity score can result in biased estimation of vaccine effects, our use of IRR as the measure for count data did not introduce any bias into the estimation of treatment effect.³² Interestingly, estimates with this method were very similar to those from the regression model in vaccine-naive cohorts, in contrast to different estimates in the total cohorts. This may be due to more variability of the preexisting immunity in the total cohort as compared with the vaccine-naive cohort.

The second limitation is the use of ICD-9-CM codes for assessment of clinical outcomes instead of laboratory-confirmed influenza infections or complications. A previous study indicated that specificity was 48% for influenza diagnosis and 12% for pneumonia diagnosis.¹⁷ We included pneumonia as an outcome because it is a major complication associated with influenza. A similar outcome definition was used for a recent study of vaccine effectiveness in a hospital setting among community-dwelling elderly individuals.³³ In addition, we only included health care encounters with a primary diagnosis of pneumonia or influenza, which helped increase the specificity of our outcome. Among the codes we used, approximately 95% of the reported cases had either specific codes for influenza (487.0, 487.1, 487.2) or the code for pneumonia with nonspecific causes (486).

Another limitation is the possibility of misclassification of exposure (vaccination status), although a recent assessment on the quality of vaccination data such as those for anthrax vaccine has been favorable.³⁴ Since immunization status was based on administrative data instead of medical record review, the possibility exists for people in the unimmunized group to have actually been immunized without documentation in our database.

We found that hospitalization rates related to pneumonia and influenza were higher in the LAIV-immunized group compared with the unimmunized or TIV group in the 2004-2005, 2005-2006, and 2006-2007 seasons. This has been reported previously for the 2004-2005 season.^8,35 Our report further revealed that the hospitalization risk decreased to insignificant levels in the later 2 seasons. It is not possible to know if the increased risk in hospitalization seen during the 2004-2005 season was causally linked to LAIV immunization, but it is biologically plausible to have a severe outcome as a result of mild respiratory infection.³⁶ In any case, this elevated risk is countered by the overall low rate of hospitalization in the military health system such that, on average, around 30 excess incident hospitalizations were expected to occur among those who took LAIV compared with TIV.

These results suggest that in a highly immunized adult population, TIV may be more effective than LAIV for the prevention of pneumonia- and influenza-related morbidity. Live attenuated influenza vaccine may be more appropriate for those with no prior immunization, such as military recruits. Continued assessment is needed of TIV and LAIV effectiveness during influenza seasons with a mismatch between circulating virus strains and the vaccine components. Because our population is highly immunized against influenza on an annual basis, results from this report may not be generalizable to the entire US adult population but could be useful for nonmilitary adult populations where vaccination rates are high. Additional efficacy trials in this population or effectiveness studies using laboratory-confirmed influenza infections may be warranted.