The Impact of Influenza Vaccinations on the Adverse Effects and Hospitalization Rate in the Elderly: A National Based Study in an Asian Country

Objectives To examine the risk of adverse effects of special interest in persons vaccinated against seasonal influenza compared with unvaccinated persons aged 65 and above. Methods We retrospectively observed 41,986 vaccinated elderly persons and 50,973 unvaccinated elderly persons in Taiwan from October 1, 2008, through September 30, 2009, using the National Health Insurance database. Neurological and autoimmune disorders and one-year hospitalization rates and in-hospital mortality rates were analyzed according to the vaccination status. Propensity score analysis was used to assess the relationship between adverse outcomes, hospitalization rates, and vaccination status. Results 45% of the elderly received influenza vaccination. Multiple logistic regression showed that the probability of being vaccinated was related to more patients visiting for URI symptoms (odds ratio (OR), 1.03; 95% CI, 1.02–1.03), men (OR, 1.15; 95% CI, 1.12–1.17), increased age (OR, 1.02; 95% CI, 1.02–1.03), and more comorbidities (OR, 1.2; 95% CI, 1.17–1.23). There were no statistical differences in neurological and autoimmune diseases between the vaccinated and unvaccinated individuals using propensity score analysis, but vaccinated persons had a reduced hospitalization rate of 19% (odds ratio [OR], 0.81; 95% CI, 0.77–0.84) for the first six-months and 13% for one-year of follow-up (OR, 0.87; 95% CI, 0.85–0.9). Conclusions Based on data from the one-year follow-ups among 93,049 elderly persons in Taiwan, reassuring results for selected neurological and autoimmune diseases were found among the vaccinated individuals after adjusting other factors. Influenza vaccination decreased the risk for hospitalization. Public health strategies must continue to improve the influenza vaccination rate among the elderly with information based upon tangible evidence.


Introduction
Influenza, which typically occurs during the late fall through early spring in the United States, is associated with high rates of morbidity and mortality, especially for the elderly [1]. Approximately 36,000 deaths from 1990 to 1999 and 226,000 hospitalizations from 1979 to 2001 were associated with influenza epidemics [2,3]. Annual influenza vaccinations have been proven to prevent influenza infection, prevent further complications from influenza, reduce hospitalization rates, and shorten hospital stays among the elderly [4,5].
Although the protective effect of influenza vaccinations have been proven, concern about influenza vaccination safety is often cited by the mass media or persons who decline to undergo vaccination. Several neurological disorders have been reported to be associated with influenza vaccinations. Bell's palsy, Guillain-Barre syndrome (GBS), pericarditis, and nephritic syndrome have been noted in individuals with influenza vaccination in case reports [6,7,8]. This is especially important for GBS, a relatively rare condition that was causally linked to the 1976 swine-influenza A (H1N1) vaccination campaign in the United States [9]. Results from the development of H5N1 vaccines revealed that adjuvants may reduce the amount of antigen needed to create an adequate immunological response which may further provoke autoimmune diseases [10,11,12]. A recent study revealed mild increased risk for Bell's palsy, paraesthesia, and inflammatory bowel disease was noted in individuals with H1N1 influenza vaccinations in Sweden [13]. However, detailed data for autoimmune diseases and neurological disorders in Asian countries remain unknown. In order to achieve higher influenza vaccination coverage for persons aged 65 years or above, adverse events risk evaluation and new strategies may be needed [14].
Given the suggestive but limited data on influenza vaccinations and the risk of autoimmune diseases and neurological disorders, we conducted a retrospective cohort study using the National Health Insurance (NHI) Research Database in Taiwan, a unique nationwide, population-based claims data system with information on the medical history of all citizens.

Ethics statement
This study was initiated after approval by the Institutional Review Board of Buddhist Dalin Tzu Chi General Hospital, Taiwan. Since all identifying personal information was stripped from the secondary files before analysis, the review board waived the requirement for written informed consent from the patients involved.

Database
The National Health Insurance program, which provides compulsory universal health insurance, was implemented in Taiwan in 1995. It enrolls up to 99% of the Taiwanese population and contracts with 97% of all medical providers. The database contains comprehensive information on insured subjects, including dates of clinical visits, diagnostic codes, details of prescriptions, and expenditure amounts. This study used the Longitudinal Health Insurance Dataset for 2008-2009 released by the Taiwan National Health Research Institute. The patients studied did not signifi-

Study Population
All patients with a recorded age of 65 and above in the dataset between October 1, 2008 and Dec 31, 2008 were included. We excluded those with any type of adverse outcomes, such as Bell's palsy, multiple sclerosis, Guillain-Barre syndrome, hypoesthesia, sudden hearing loss, rheumatoid arthritis, and type I diabetes.
A total of 93,049 individuals were included. Each patient was tracked from his or her index ambulatory visit between October 1, 2008 and September 30, 2009 to identify outcomes. These patients were then linked to the administrative data for the 2008 to 2009 period to calculate cumulative risk, with cases censored for patients who drew back guarantees from the National Health Insurance Program or were still healthy without defined events at the end of follow-up.

Definitions of hospitalization
Hospitalization for all causes between the two cohorts during the study period were recorded [4].

Definitions of covariates
Patients were characterized by age, gender, number of outpatient visits for URI during the six months before the vaccination date (for the vaccinated group) or the first day when the free vaccination was offered (for the unvaccinated group), comorbidities, geographic region and urbanization of residence, and individual socioeconomic status (SES). The comorbidities of each patient included existing heart disease, lung disease, diabetes, stroke, renal disease, and catastrophic illnesses (cancer and rare diseases) [18]. Baseline comorbid conditions were defined as having any inpatient or outpatient diagnosis of the above mentioned diseases during the preceding year prior to October 1, 2008. This study used income-related insurance payment amount as a proxy measure of individual SES at the time of diagnosis. The individuals were classified into three groups: (1) low SES: lower than US$571 per month (New Taiwan Dollar (NT$) 20,000); (2) moderate SES: between US$571-1,141 per month (NT$20,000-40,000); and (3) high SES: US$1,142 per month (NT$40,001) or more [19]. The geographic regions and urbanization of residence were classified as previously described [20,21].

Statistical analysis
The SAS statistical package, version 9.2 (SAS Institute, Inc., Cary, NC), and SPSS version 15 (SPSS Inc., Chicago, IL) were used for data analysis. Pearson's chi-square test was used for categorical variables, demographic characteristics (gender, region, and urbanization of residence), comorbidities, and individual socioeconomic status in vaccinated and unvaccinated individuals. Continuous variables, such as age, and URI OPD# were analyzed with two-sample t-tests in this series. Multiple logistic regression was used to estimate the predictors for receiving a flu vaccination based on each individual's observed variables, which included URI OPD# in the prior six months, gender, age, comorbidity,  Table 3. Adverse events and hospitalization rates among the study population for one-year of follow-up*.   geographic region and urbanization of residence, and individual SES. The cumulative risk of neurological outcomes/autoimmune diseases was estimated as a function of time from initial treatment. To adjust for the potential bias caused by self-selection, propensity score was used [22]. To obtain the propensity score in this study, patient characteristics were entered into a logistic regression model to predict selection for vaccination. These characteristics were age, gender, URI OPD# in the prior six months, comorbidity, geographic region and urbanization level of residence, and socioeconomic status. The effect of influenza vaccination on the selected neurological and autoimmune diseases was analyzed with each quintile. The Mantel-Haenszel odds ratio was estimated in addition to performing the Cochran-Mantel-Haenszel x 2 test.

Diseases and vaccination status
In order to clarify the benefit of vaccination, the odds ratio of hospitalization for all cases of vaccinated and unvaccinated individuals was performed using propensity score analysis. A p,0.05 was considered statistically significant in the regression models.

Results
By December 31, 2008, virtually all vaccination activity had been completed, with a cumulative 41,986 people vaccinated (45.1% of the study population). Table 1 shows the vaccine coverage rate and distribution of demographic characteristics, URI OPD#, comorbidities, geographic regions and urbanization of residence, and socioeconomic status for the two cohorts. Individuals aged 71 and above were more often vaccinated than those aged 65 to 70 years. More men than women were vaccinated (46.9% vs. 43.4%). Individuals with more comorbidities were more likely to be vaccinated. Vaccine coverage was greater in those who lived in central and southern Taiwan Table 2 shows the predictors for receiving an influenza vaccine. Visitors of an outpatient clinic for URI before the study period in the prior six months were more likely to receive an influenza vaccine (odds ratio (OR), 1.03; 95% CI, 1.02-1.03). Male gender (OR, 1.15; 95% CI, 1.12-1317) and old age (OR, 1.02; 95% CI, 1.02-1.03) were more likely to be vaccinated. Individuals with more comrobidies were more likely to be vaccinated than those without any comorbidities (OR, 1.2; 95%, 1.17-1.23). Compared with those in northern Taiwan, individuals who lived in central (OR, 1.31; 95% CI, 1.26-1.36) and southern Taiwan (OR, 1.15; 95% CI, 1.11-1.12) were more likely to receive a flu vaccination. Residence in suburban (OR, 09; 95% CI, 0.87-0.93) and urban areas (OR, 0.87; 95% CI, 0.84-0.91) were negative predictors for receiving vaccination. Furthermore, individuals with moderate SES were more likely to be vaccinated (OR, 1.1; 95% CI, 1.07-1.14), compared with those with low SES. Table 3 shows the adverse outcomes between the vaccinated and unvaccinated individuals. Among the neurological and autoimmune disease, seasonal preponderance was found in GBS among the individuals with or without vaccination: 41% of GBS patients in the vaccinated group and 42% of GBS patients in the unvaccinated group developed the disease in spring (March to May).

Univariate analysis
The risks of neurological and autoimmune diseases were further analyzed according the first six-months and one-year of follow-up after the vaccination campaign. During the first six-months of follow up, there was no statistical difference in GBS (p = 0.144), Bell's palsy (p = 0.083), multiple sclerosis (p = 0.218), an/hypoesthesia, or paraesthesia (p = 0.848), sudden hearing loss (p = 0.744), rheumatoid arthritis (p = 0.461), and type I diabetes (p = 0.342) between the two groups. There was also no statistical difference in GBS (p = 0.144), Bell's palsy (p = 0.083), multiple sclerosis (p = 0.218), an/hypoesthesia, or paraesthesia (p = 0.848), sudden hearing loss (p = 0.744), rheumatoid arthritis (p = 0.461), and type I diabetes (p = 0.342) between the two groups after one-year of follow up in univariate analysis. During the first six-months of follow-up, the hospitalization rate was 10.2% for vaccinated individuals and 11.3% for unvaccinated individuals (p,0.001). After one-year of follow-up, the vaccinated group had a lower hospitalization rate (19.8% vs. 20.5%, p,0.001).

Propensity score analysis
Due to the unequal distribution of baseline characteristics between the vaccinated and unvaccinated groups, propensity score analysis was performed. Figure 1 shows that most of the baseline characteristics were well balanced within each propensity quintile. During the first six months of follow-up, there was no significant difference in neurological and autoimmune diseases between the vaccinated and unvaccinated individuals using propensity analysis (Table 4). After one year of follow-up, the vaccinated individuals didn't incur higher risk for neurologic diseases or autoimmune diseases compared with the unvaccinated group after adjusting for other factors (Table 4).

Discussion
Influenza epidemics occur yearly and have a significant impact on the health of the elderly and medical expenditures. Although public health vaccination is freely available in Taiwan, up to 55% of the elderly did not receive an influenza vaccination. Concerns about the adverse effects of influenza vaccination may be one of the barriers. More clinics visiting for URI symptoms, men, increased age, and number of comorbidies were independent predictors for those receiving an influenza vaccination. In this series, the increased risks were not evident for selected neurological and autoimmune diseases for the first-six months and one-year of follow up among the vaccinated individuals. Furthermore, influenza vaccination reduced the rate of all causes for hospitalization.
The strengths of our study are that it is based upon a large population (n = 93,049), the population experienced almost complete follow-up of neurological and autoimmune outcomes (99%) and had routine monitoring of diagnosis accuracy by the National Health Insurance Bureau of Taiwan. We found the elderly in Taiwan to have an influenza vaccination rate of 45%. There were also significant differences patient characteristics between the vaccinated and unvaccinated groups. Multiple logistic regression showed that more visiting clinics for URI symptoms in the prior 6 months, men, increased age, number of comorbidies, and residence in central/southern Taiwan were positive predictors for influenza vaccination. However, residence in urban/suburban areas and high SES were negative predictors for vaccination. These predictors, which are in agreement with previous series in Taiwan, could be used to find the target population in order to increase the future influenza vaccination coverage rate by the National Health Research Institution [18,23]. The quality of the risk-adjustment technique when analyzing administrative infor-mation is an important issue. There is a significant difference in the baseline characteristics between the vaccinated and unvaccinated groups. Propensity scores were used to stratify the patients into five groups with similar scores in order to reduce the effects of selection bias between the vaccinated and unvaccinated groups [22,24,25]. Our results suggest that the influenza vaccine did not incur higher risk for neurological and autoimmune diseases. Influenza vaccination reduced the risk of hospitalization by 13% in elderly persons using propensity score analysis for one-year of follow-up. The effect of the vaccine on the hospitalization rate was more evident (19%) during the first six months than in the rest of the study period.
Several studies reported adverse events following influenza vaccination. After vaccination with the 1976 H1N1 influenza vaccine, 8.8 excess GBS cases per million vaccine doses were noted [9]. Substantial elevation of FBS risk within 6 weeks after vaccination and a low percentage of reporting other potential causes among the GBS cases supported the causal relationship between the influenza vaccination and GBS. A previous study reported that the H1N1 influenza vaccination incurred higher risk for Bell's palsy, which may be contributed to a phenomenon of Guillain-Barre syndrome [13]. However, other large-scale studies have questioned the association of Guillain-Barre syndrome and influenza vaccination [26,27,28]. Under the Emerging Infections Program (EIP) surveillance system for near-real time GBS surveillance throughout the 2009 to 2010 pH1N1 vaccination program, recent reports suggested that H1N1 influenza vaccine from 2009 to 2010 was associated with 0.8 excess GBS cases per million pH1N1 vaccines doses [29,30]. The EIP system was featured by active case-finding and classification based on clinical criteria rather than administrative data codes alone, patient interviews for assessment of antecedent sources [31]. However, the lack of temporal clustering evidence and the possible exposure to other factors in over half of the GBS reports precluded the conclusion of the causal relationship between the H1N1 vaccine and GBS. In our series, there is a seasonal preponderance for GBS in both vaccinated and unvaccinated groups. 41% to 42% of GBS instances developed in the spring, which is in agreement with the report by Lyu et al [32]. After the six-months and one year of follow-up, our data did not reveal a statistically increased risk for Guillain-Barre syndrome in elderly individuals with an influenza vaccination.
Previous literature questioned the association of the influenza vaccination and autoimmune diseases [33,34,35]. Autoimmune diseases, such as reactive arthritis and polyarteritis nodosa, were reported after influenza vaccination. One possible mechanism is molecular mimicry when a structure similarity exists between viral antigens or other components of the vaccine. This phenomenon may trigger autoimmune responses [34]. Adjuvants were proven to reduce the amount of antigen, improve the magnitude, and prolong the duration of the immune response in H5N1 vaccines [10,36]. However, lupus autoantibodies could be induced by adjuvants, such as the incomplete Freund's adjuvant, pristine, or squalene in vaccines [11,12]. In our series, Fluvirin, Vaxigrip and Fluarix were all non-adjuvanted vaccines and there was no significant difference in autoimmune disorders between the vaccinated and unvaccinated individuals. Up to 88% of elderly persons have one or more chronic conditions [37]. Chronic diseases or inflammation may be a risk factor for cardiovascular events, especially for elderly persons [38]. Systemic respiratory tract infections may aggravate this impending obstructive vessel and lead to acute myocardial infarction and stroke [39]. Poor physical function and impairment of judgment after influenza infection may result in falling or traffic accidents [40]. Poor compliance in the control of preexisting chronic diseases might exacerbate a disadvantaged condition, such as hypertension and diabetes. All of the above mentioned adverse effects could be erased by influenza vaccination. For the elderly, influenza vaccination could reduce hospitalization rates and shorten hospital stays [4,5].
Barriers and predictors for influenza vaccination have been explored. Among the elderly, individuals with threat-responsiveness or perceived risk (prior influenza vaccination, or prior outpatient visits for flu-like respiratory conditions), men, persons aged 70 to 74 years, and those with chronic disease were more likely to have an influenza vaccination in Taiwan [18,23]. Among the elderly, influenza vaccination coverage levels were about 65.6% to 66.3% between 2006 and 2009 in the United States, and this is still far away from the goal of Healthy People 2010 [14]. In Taiwan, the influenza vaccination coverage levels ranged from 38 to 50% [23,41]. There is ample room for further vaccination rate increases among older individuals. Our series validated the protective effect of influenza vaccine, and the risks of adverse effects from the vaccine were not significantly elevated. Some strategies to support hospitalized patients, such as personalized postcards or phone calls, are effective. Home visits and facilitators may be effective in increasing influenza vaccination rates for the elderly in the community. Influenza vaccination programs for healthcare workers in hospitals or clinics is also worthy of a trail [42,43].
Consideration must also be given to the potential limitations of the present study that may have influenced the observed conclusions. First, the diagnoses of neurological and autoimmune disease, and any other comorbid conditions, are completely dependent on ICD codes among the administrative data. However, the National Health Insurance Bureau of Taiwan randomly reviewed the charts and interviewed patients in order to verify the diagnosis. Hospitals with outlier chargers or practice could undergo an audit, with subsequent heavy penalties for malpractice or discrepancies. The diagnosis of GBS were further verified by the neurology specialists and the validity of coding for GBS in administrative data is confirmed [44]. The diagnosis of multiple sclerosis, rheumatoid arthritis and type I diabetes were confirmed by catastrophic disease dataset released from the NHIRD. Second, the database does not contain information on tobacco use, dietary habits, and body mass index, which may also be risk factors for neurological events and may contribute to hospitalization for cardiovascular diseases. We hope to further link the NHIRD with primary survey data in the future. Third, we used in-hospital mortality as a proxy of estimation of mortality rate. There may have been a few individuals with severe disease who chose not to seek medical help. However, the barriers to medical access is almost negligible because the National Health Insurance system in Taiwan allows patients to visit any clinic or hospital freely without referral by a general practitioner and cover up to 90% of the hospitalization costs.
In conclusion, 45% of the elderly aged 65 years or older received influenza vaccinations in 2008. Based on data from one year of follow-up, we did not find statistically increased risks for neurological and autoimmune diseases among the vaccinated individuals after adjusting for other factors. Influenza vaccinations decreased the risk for hospitalization by 13% in elderly patients during the one-year of follow-up. Based on these reports, we could try to provide more information based upon tangible evidence on influenza vaccination, particularly its effect on preventing epidemic outbreaks and the protective effects of hospitalization to those who do not think the vaccine is necessary and for those who are concerned about the adverse effects.