Safety of the Trivalent Inactivated Influenza Vaccine Among Children: A Population-Based Study

Background  To our knowledge, there are no published population-based studies on the safety of the inactivated trivalent influenza vaccine among children.

Objective  To screen a large population of children for evidence of increased medical visits in the 2 weeks after influenza vaccination compared with 2 control periods. Secondary analyses included shorter risk periods and restricted age categories.

Design  Self-control screening analysis. Children vaccinated from January 1, 1993, through December 31, 1999, were randomly divided into 2 equal groups. In group 1, risks of outpatient, emergency department, and inpatient visits during the 14 days after vaccination were compared with the risks of visits in 2 control periods. Significant plausible medically attended events identified in group 1 were then analyzed in group 2, using the same 2 control periods. Medically attended events significant in both groups were considered potentially associated with vaccination and were assessed by medical record review.

Setting  Five managed care organizations in the United States.

Participants  Children younger than 18 years who received an influenza vaccination in one of the managed care settings (N = 251 600).

Main Outcome Measure  Among vaccinated children seen for a medically attended event, the odds of the visit occurring in the 2 weeks after vaccination vs during 1 of the 2 control periods.

Results  Study participants incurred 1165, 230, and 489 different diagnoses during the 14 days after vaccination according to the outpatient, emergency department, and inpatient data, respectively. Four diagnoses were positively associated with the vaccine in both groups 1 and 2: impetigo, dermatitis, uncomplicated diabetes mellitus, and ureteral disorder not otherwise specified. After medical record review, impetigo (9 cases) in children 6 to 23 months old remained significantly associated with vaccination.

Conclusion  This large screening safety study did not reveal any evidence of important medically attended events associated with pediatric influenza vaccination.

Over the last few years there has been a growing interest in expanding the recommendations for influenza vaccination of children. In the past, annual vaccination with the trivalent inactivated influenza vaccine (TIV) was recommended for children aged 6 months and older with increased risk of complications due to influenza.¹ Because of an increased morbidity from influenza in infants and young children,^2,3 in 2003 these recommendations were changed by the Advisory Committee on Immunization Practices to recommend influenza vaccination for all children 6 to 23 months old. Few studies exist, however, that assess the safety of TIV among children. In the last 15 years, the only controlled studies with TIV have been designed either to assess outcomes, such as the prevention of otitis media,^4-6 or to serve as the control arm for studies of the live attenuated intranasal influenza vaccine.⁷ To our knowledge, no large postmarketing studies have been conducted looking for rare adverse events associated with the TIV used in children.⁸ Such studies have been important in demonstrating the safety of varicella vaccination⁹ and the occurrence of intussusception in the week after rotavirus vaccine (Rotashield) administration.¹⁰ The purpose of this study was to screen a large cohort of children who had received the TIV for evidence of increased risk of medically attended events (MAEs) during the 2 weeks after vaccination.

A self-control screening analysis, based on the case-crossover method,¹¹ was conducted to examine the risk of MAEs after influenza vaccination. All MAEs that occurred at least once in a 14-day window after influenza vaccination were explored. The Vaccine Safety Datalink provided data for this analysis, a Project established in 1991 by the Centers for Disease Control and Prevention, Atlanta, Ga, to evaluate the safety of vaccines.¹² Specifically, this project involves linking large, computerized databases containing information on vaccination history, medical encounters (inpatient and outpatient records), and demographics from 8 managed care organizations located across the United States. For this investigation, data from 5 of the sites were analyzed: Kaiser Permanente of Colorado, Denver; Northwest Kaiser Permanente, Portland, Ore; Group Health Cooperative, Seattle, Wash; Northern California Kaiser Permanente, Oakland; and Southern California Kaiser Permanente, Panorama City. The institutional review boards at each of the managed care organizations approved this study.

Influenza season was defined as the period from October 1 through April 30, and 3 separate sets of analyses were conducted with outpatient, inpatient, and emergency department (ED) data. Vaccines administered October through April from January 1, 1995, through December 31, 1999, were included in the outpatient and ED analyses; vaccinations from January 1, 1993, through December 31, 1999, were used for the inpatient analyses. The study was restricted to individuals younger than 18 years who received an influenza vaccination and were continuously enrolled in the managed care organization 28 days before and 28 days after receiving the vaccination. For children who received 2 doses of influenza vaccine during a season, only the first vaccine was included in the primary analysis; second seasonal influenza vaccinations were analyzed separately in a subanalysis. Individuals could have appeared in the analysis more than once if they were vaccinated in more than 1 season. Multiple vaccinations of an individual are treated independently in the analysis.

Medically attended events were defined by International Classification of Diseases, Ninth Revision, Clinical Modification(ICD-9-CM) codes and aggregate codes. The aggregate codes are based on groupings that were developed by Finkelstein et al¹³ of individual ICD-9-CM codes related to specific symptoms. Twenty-six aggregate codes were used in this study, including pharyngitis, upper respiratory tract infection/common cold, sinusitis, bronchitis, pneumonia, cellulitis/skin infection, asthma, limb soreness, rash, headache, epilepsy, urticaria, and allergic reaction (the complete list is available on request from the authors). The ICD-9-CM codes for fever and malaise, and the ICD-9-CM codes comprising the aggregate codes for cellulitis/skin infection, limb soreness, headache, urticaria, and allergic reaction were also analyzed separately.

Outpatient and ED analyses assessed vaccinations over 5 years (1995-1999), representing 4 complete influenza seasons (October 1, 1995, through April 30, 1999) and 2 half influenza seasons (January-April 1995; October-December 1999). Inpatient analyses spanned 7 years (1993-1999), representing 6 complete influenza seasons and 2 half influenza seasons. Outpatient visits included any walk-in and/or ambulatory care visits to general practitioners or specialists within the managed care organization. For the primary analysis, the risk of an MAE was explored in a 14-day risk period after vaccination in the outpatient, inpatient, and ED data sets. Children with MAEs that occurred between days 1 and 14 after vaccination were considered exposed cases. Medically attended events that occurred on the vaccination date (day 0), however, were excluded from the analysis. On day 0, it is hypothesized that individuals are more likely to be diagnosed as having an illness for which they would not normally seek medical attention. For example, a child may visit the clinic to receive his or her influenza vaccination, and during this visit, the physician may identify an illness (eg, rash) or may record a chronic condition (eg, asthma). Conversely, an individual may visit the clinic for an illness, and out of convenience, the physician may administer the vaccination. Either scenario could create a bias that leads to overinflated risk estimates. Because excluding day 0 from this analysis could miss important acute serious reactions, we looked for specific diagnoses during the 0- through 2-day window in our secondary analyses (allergic reaction, urticaria, cellulitis, limb soreness, malaise, and fever).

Risk estimates for MAEs that occurred in the 14-day period after vaccination were generated with 2 separate control (unexposed) periods: days 15 through 28 before the vaccination date (control period 1), and days 15 through 28 after the date of vaccination (control period 2). Each distinct MAE was analyzed separately, and an individual was included in the analysis if he or she experienced the event in either the 14-day risk period or in 1 of the 2 control periods. Within these individuals, intervals were compared, allowing each case to serve as its own control. This provided a means to control for unmeasured confounders, such as existing health disorders (eg, asthma and diabetes mellitus), race/ethnicity, educational level, birth order, and environmental tobacco smoke exposure. Conditional logistic regression was used to generate the risk estimates (odds ratios [ORs]), treating risk period, and control period of each case as a matched pair (or stratum). The resulting matched ORs (mORs) can be interpreted with the following statement: given that an individual experienced an MAE, what are the odds that it occurred in a postvaccination risk window rather than in a control period? With this method, only discordant pairs were analyzed; individuals who experienced an MAE in both the postvaccination period and control period (defined as concordant pairs) were dropped from the analysis.¹⁴ Medically attended events from models that would not converge owing to insufficient case numbers were also dropped from the analysis.

Before the logistic models were implemented, each data set was randomly divided into 2 groups. In group 1, we used the logistic models to determine the risk of an MAE during the risk window by comparing it with the risk of the MAE in the 2 control windows. Statistically significant (P≤.05), medically plausible MAEs identified in group 1—with either control group—were then analyzed in group 2 using the same 2 control periods. Hence, up to 4 mORs were calculated for each MAE. Requiring that any significant findings from the first sample be confirmed in the second sample served to limit the number of associations found to be significant by chance alone (type I error). Medically attended events with at least 1 significant mOR in both groups 1 and 2 were considered potential vaccine-associated adverse events.

Despite our large sample size, important rare outcomes could be missed owing to power issues. To screen for such outcomes, positive associations with both liberalized P values (.05<P≤.20) and mORs greater than or equal to 2.5 in group 1 were also examined in group 2 for all 3 medical settings. Other secondary analyses performed include exploring the risk of MAEs in 0- through 2- and 1- through 3-day postvaccination windows (with similar length control periods) and after the second seasonal influenza vaccination. Because of national considerations for expanding universal influenza vaccination to include 6- to 23-month-olds when this study was designed, the risks among children in this age group within each of the exposure windows (1-14, 0-2, and 1-3 days) were also explored. Since the 6- to 23-month-olds were a small population (n = 8476 vaccinations) dispersed over a large number of distinct outpatient events (n = 285) during the 14-day risk window, the cohort was not divided into the 2 samples for these analyses because of concerns for statistical power.

At the conclusion of the analyses, medically plausible positive associations were examined further with medical record reviews. The Vaccine Safety Datalink Project staff at each site followed a standard protocol to verify influenza vaccination status, visit diagnosis, and date of the medical visit and to determine if the significant MAE was the reason for the visit or a secondary finding (eg, stable asthma coded during a well-child care visit). The reason for the visit was determined by review of the caregivers’ narrative (eg, chief complaint of well-child care visit or assessment of upper respiratory tract infection), and incorrect visit and vaccination dates from the electronic data were corrected. Visits occurring in both the exposed periods and the control periods were audited. Copies of the visit notes and the completed medical record audit tool were sent to the principal investigator (E.K.F.); all medical records were reviewed by the principal investigator to verify the accuracy of the medical record abstraction.

The outpatient and ED analyses included data from 6 influenza seasons across 4 managed care organizations, constituting a cohort of 128 679 children who had received a total of 221 484 distinct influenza vaccinations. For the inpatient analyses, a fifth managed care organization (Southern California Kaiser Permanente) and 2 influenza seasons were added, resulting in a cohort of 251 600 children who had received 438 167 distinct vaccinations. The sex distribution of the total cohort was approximately equal (47% female and 53% male), and the mean (SD) age was 10 (4.6) years. Approximately 29% (n = 72 445) of the children were identified as having a high risk for influenza morbidity by ICD-9-CM codes.

In total, there were 41 383 clinic visits, 1621 ED visits, and 2214 hospitalizations within the 14-day risk period after vaccination. These medical encounters represented 1165, 230, and 489 distinct MAEs in the outpatient, ED, and inpatient data, respectively. Each of these outcomes was evaluated with 2 separate control groups, resulting in 2330, 460, and 978 different conditional logistic models for each of the 3 respective medical settings. The estimates generated from the outpatient models revealed 560 MAEs with an mOR greater than 1; 660 MAEs with an mOR less than 1; and 271 MAEs with an mOR equal to 1.

In the outpatient setting, 44 MAEs demonstrated significant associations against 1 or 2 of the control periods in sample 1. Of these, 10 were deemed medically implausible, including obesity, tuberculin test reaction, infestation, insect bite, dislocated knee, fracture of carpal bone, sprain of the neck, sprain of the elbow, sprain of the lumbar region, and sprain of the shoulder. Of the 34 medically plausible associations, 11 demonstrated a significant association (1 positive, 10 negative associations) against either of the control periods in sample 2. These 11 MAEs are listed in Table 1.

Only one medically plausible MAE—uncomplicated diabetes mellitus—demonstrated significant positive associations against either control period in sample groups 1 and 2. In the outpatient setting against control period 1, the mORs for uncomplicated diabetes mellitus (ICD-9-CM: 250.0x) were 1.25 (95% confidence interval [CI], 1.00-1.55) and 1.41 (95% CI, 1.12-1.78) in the 2 respective sample groups. Owing to the large number of visits for uncomplicated diabetes mellitus (n = 623), we limited our medical record audit to all visits occurring in sample 2 (n = 295; 173 exposed and 122 controls). This process revealed that 85% of these clinic visits were for routine well-child diabetes mellitus care or for reasons unrelated to diabetes mellitus (eg, upper respiratory tract infection, pharyngitis, well-child care). When these were removed from the data set, the mOR dropped from the original value of 1.41 to a negative effect of 0.76 (95% CI, 0.42-1.38).

The remaining 10 significant mORs in the outpatient setting were negative. Therefore, they were less likely to occur in the risk period after vaccination rather than in the control period. Sinusitis, upper respiratory tract infection/common cold, asthma, chronic rhinitis, and otitis media demonstrated 4 significant negative associations; the estimates for bronchitis, pneumonia, allergic rhinitis, and dyspnea/respiratory abnormalities were significant in 3 of the 4 comparison groups; dermatitis demonstrated 1 significant mOR in each sample group (Table 1).

Twenty-five MAEs with mORs of 2.5 or more in the outpatient setting were identified using the liberal significance level (.05<P≤.20). Of these potentially large associations, only renal and ureteral disorder not otherwise specified (NOS) (ICD-9-CM: 593.9) demonstrated a positive association against 1 of the control groups in samples 1 and 2. The mORs for renal and ureteral disorder NOS were 3.50 (95% CI, 0.73-16.85) and 5.00 (95% CI, 0.58-42.79) for control group 2 in samples 1 and 2, respectively. Renal and ureteral disorder NOS, however, was a broad diagnosis code used for disparate clinical conditions. The medical record reviews (n = 26) demonstrated that there was a wide variety of unrelated diagnoses in this group, including children without renal disorders (n = 2). As with the uncomplicated diabetes mellitus code, many (27%) of these visits were for well-child care or unrelated minor acute illnesses. When the analysis was restricted to visits for a renal or ureteral disorder NOS, the OR was no longer significant (mOR = 2.00; 95% CI, 0.60-6.64).

In our secondary analysis of the 0- through 2-day period, many roll-up codes had positive mORs: limb soreness (mOR range, 6.6-7.5), malaise (mOR range, 4.8-50.0), and allergic reaction (mOR range, 2.6-4.6). None of these were significant when the analysis was repeated for 1 through 3 days (suggesting that events on day 0 are responsible for these positive mORs) nor were any significant for ED or inpatient visits. Medical record reviews at Kaiser Permanente Colorado, found that these codes were often used for conditions unrelated to vaccination. Patients treated at the clinic for acute limb pain (eg, patellar tendon contusion, knee pain, foot strain) were given an influenza injection during the visit and coded as having limb soreness, while patients with known peanut or other food allergy had these diagnoses coded at the visit during which the influenza injection was administered. The associated risk estimates for the inpatient and ED settings and second seasonal vaccinations followed a similar pattern to the outpatient, 14-day risk-period analysis—there were no significant positive associations and many mORs were below 1.

Because the sample was not split for the analysis of the 6- to 23-month age group, positive associations in comparisons against either control period were evaluated with medical record review. While the 6- to 23-month age group also demonstrated several significant negative associations, 2 risk estimates were significantly above 1. The OR for other atopic dermatitis (ICD-9-CM: 691.8) in the outpatient setting against control period 1 was 1.94 (95% CI, 1.08-3.48). The risk estimate for impetigo (ICD-9-CM: 684) in the outpatient setting against control period 2 with the 3-day risk window was 8.0 (95% CI, 1.00-62.5) (Table 2). The medical record review of the possible atopic dermatitis cases (n = 50) demonstrated that 62% of the visits were attributed to follow-up visits among children with ongoing atopic dermatitis, children without verified dates, or visits unrelated to atopic dermatitis. It was difficult to determine if the remaining visits represented either a new onset or an exacerbation of the condition. When these visits were excluded, the mOR was no longer significant (mOR = 2.17; 95% CI, 0.82-5.70). In contrast, the medical record reviews of the possible impetigo cases revealed that all of the MAEs (n = 9) were confirmed incident cases. Among these cases, however, the site of the impetigo varied (the finger, lip, ear, scrotum, arms, leg, and buttocks); none occurred at the site of injection.

The purpose of our study was to conduct a screening analysis of children in the Vaccine Safety Datalink Project data set who had received a TIV to determine if vaccination is associated with any unsuspected adverse MAEs during the 14 days after vaccination. In our primary analysis, only 1 MAE—uncomplicated diabetes mellitus—was positively associated with influenza vaccination. Based on medical record review, this association represented a tendency to be seen for routine well-child diabetes mellitus care in the weeks after influenza vaccination rather than a worsening of diabetes mellitus (eg, ketoacidosis, poor glycemic control). Similarly, only 3 positive associations surfaced in our various subanalyses that excluded day 0 (impetigo, atopic dermatitis, and renal and ureteral disorder NOS), 2 of which were not considered true vaccine-associated MAEs after medical record review (atopic dermatitis and renal and ureteral disorder NOS). It is difficult to draw conclusions about the association of impetigo in the 3 days after influenza vaccination in children younger than 2 years. This result is based on a few cases (8 in the exposed period and 1 in the control period) and could not be confirmed in a second sample because of the few children 23 months old or younger in our cohort. We are planning to investigate this possible association further in a larger study of TIV safety of 6- to 23-month-old children that is under way. Overall, our investigation—to our knowledge, the largest TIV safety study performed to date—found no worrisome serious adverse MAEs during the 14 days after vaccination with TIV.

Studies in which vaccinated children are compared with unvaccinated children can be biased by “confounding by indication,”^15,16 in which a child is vaccinated because of an increased risk of an adverse outcome. Among asthmatic patients, children with severe asthma are more likely to be vaccinated against influenza than are children with mild asthma.¹⁷ Vaccinated asthmatic patients are also 1.5 to 4 times more likely to be seen for an asthma exacerbation^18,19 or acute respiratory tract infection²⁰ after influenza vaccination than are unvaccinated asthmatic patients when severity of illness is not controlled for in the analysis. By using self-control methods such as conditional Poisson or logistic regression, this confounding is reduced.^18,19 The conditional logistic method used in our study, in which different periods are compared among the same vaccinated subjects, adjusts for both known confounders such as age, sex, and season and unknown confounders such as severity of illness, race/ethnicity, and birth order.^21,22

Our study found that children vaccinated against influenza were significantly less likely to be seen for many illnesses during the postvaccination period than during the control period. Similar negative mORs have been seen in other vaccine safety studies using self-control methods.^18,19 We feel there are at least 4 possible explanations for these findings. First, children are likely to be vaccinated only when considered healthy by parents and physicians. The probability of becoming ill in the weeks after vaccination is, therefore, likely to be less than their average risk over a given period (the “healthy vaccinee effect”). Second, persons undergoing vaccination are informed that there are certain adverse effects, such as low-grade fever, limb soreness, and irritability that may follow vaccination. It may be that families are less likely to contact the health care system for these expected symptoms during the period after vaccination compared with the control periods. This “expected symptoms effect” could reduce the odds of MAEs after vaccination. Third, children may come to the clinic for an acute illness, such as an upper respiratory tract infection or mild wheezing, may receive an influenza injection during the visit, and are then less likely to return in the subsequent weeks. This “recent visit effect” may reduce the number of clinic visits for upper respiratory tract–related illnesses in the days after vaccination. Fourth, vaccination with TIV may induce a nonspecific immunologic response, such as interferon stimulation, that would transiently protect against other minor acute viral infections.

While these 4 explanations may explain the negative mORs seen for minor acute illnesses, it is unlikely that they would mask the presence of serious adverse MAEs (eg, seizures, encephalopathy, intussusception) after vaccination. Our investigation found no worrisome serious adverse MAEs after vaccination with TIV. Using conditional logistic regression, 68 discordant pairs are needed to detect an mOR of 2.0 (α = .05; 1−β = .80).²³ For fever, asthma, convulsions, and dyspnea (symptoms that have been commonly reported to the Vaccine Adverse Event Reporting System after influenza vaccination), we had 1094, 498, 81, and 510 cases in our study, respectively. While we had enough statistical power to detect meaningful associations with these outcomes, we conducted additional analyses with a liberalized significance criterion (.05<P≤.20) to identify potentially important large associations (mOR ≥2.5) that may have been overlooked with our primary criteria owing to lack of statistical power. In doing so, renal and ureteral disorder NOS surfaced as a potential vaccine-related event but was deemed unrelated to influenza vaccination after medical record review.

From our secondary analysis of events occurring on days 0 through 2, it is clear that our method cannot be used to screen for outpatient events occurring on day 0. Limb soreness, malaise, and allergic reaction had strongly positive mORs in comparisons made for days 0 to 2 in the outpatient setting. Similarly, the aggregate codes that had protective mORs in the 14 days post vaccination also had positive mORs on day 0 (eg, upper respiratory tract infection mOR range, 4.9-5.7; pneumonia mOR range, 1.9-2.5; data not shown). Physicians will often administer an influenza injection during a clinic visit for a minor acute illness such as an upper respiratory tract infection, or may code a chronic condition such as a food allergy during a visit specifically for vaccination. For this reason, screening studies that do not test a specific hypothesis should exclude day 0 in their analyses.

Our study has several limitations. The investigation is based on automated data used by managed care organizations for administrative purposes rather than for research. However, based on the 380 medical record audits completed for this study, little misclassification occurred. Influenza vaccination was confirmed in 91% of audited medical records, and visit diagnoses were correct in 93% of outpatient visits. Our study assessed MAEs and, therefore, would miss minor adverse events managed at home. Finally, since our screening methods involved doing many comparisons, it is not surprising that medically implausible diagnoses (eg, insect bite or sprain of neck) were associated with influenza vaccination owing to type I errors. Though we split our cohort in 2 and required confirmation of possible associations in the second sample, it is possible that the positive associations found were due to chance alone.

Despite these limitations, to our knowledge, our investigation represents the largest population-based study designed to evaluate the safety of the pediatric TIV. In deciding whether to expand current recommendations on influenza vaccination to include all children, national policy groups will consider 4 main issues: vaccine efficacy, safety, cost-effectiveness, and the feasibility of immunizing all children each year. Our study provides important information on the safety of this vaccine for children. We found no evidence to suggest an increased risk of MAEs during the 2 weeks after pediatric influenza vaccination. Should national policy recommendations expand to include all children, ongoing studies should be undertaken to further evaluate this vaccine.

The TIV is thought to be safe for children: the most common events reported to the Vaccine Adverse Event Reporting System include fever, asthma, convulsions, and dyspnea. Despite the wide use of this vaccine, to our knowledge, there are actually no published studies evaluating its safety in a large population of children. This study found no worrisome signals of serious adverse MAEs in the 2 weeks after influenza vaccination. Our findings support the conclusion that the use of TIV in children is safe.

Correspondence: Eric K. France, MD, MSPH, Kaiser Permanente Colorado, PO Box 378066, Denver, CO 80237-8066 (eric.k.france@kp.org).

Previous Presentation: Preliminary findings of this study were presented to the Advisory Committee on Immunization Practices; October 17, 2002; Atlanta, Ga; and to the Institute of Medicine Panel on Vaccine Safety; March 13, 2003; Washington, DC.

Accepted for Publication: May 7, 2004.

Financial Interest: Dr France has a grant from Wyeth Lederle Vaccines. Dr Jackson has a support grant from Aventis.

Funding/Support: This study was supported by the Centers for Disease Control and Prevention through an agreement with America’s Health Insurance Plans.

Acknowledgment: We acknowledge the work of the data management staff at the participating managed care organizations for their work in putting together the Vaccine Safety Datalink files, the assistance of Simon Hambidge, MD, in reviewing medical records and the manuscript, and the support from staff at the Centers for Disease Control and Prevention and the America’s Health Insurance Plans.