It is widely believed that elderly persons have markedly diminished abilities to produce antibody in sufficient quantity after the administration of bacterial and viral vaccines [8, 9]. Although the efficacy of influenza vaccines in elderly persons varies considerably among studies, several have shown a reduced incidence of pneumonia and mortality with these vaccines, particularly in high-risk groups. To better understand influenza vaccine efficacy in elderly persons, we reviewed the English-language literature and did a meta-analysis of the published studies ([10-29]; referred to in text, tables, and figures as studies 1-20 [see Table 1 for a cross-listing of study and reference numbers]). Our meta-analysis confirms the findings of several recent studies on the effectiveness of influenza vaccination in elderly persons. The accumulation of favorable scientific evidence presented here should heighten physician awareness and encourage the universal annual use of influenza vaccine in elderly persons.

Methods

Top Methods Results Discussion Author & Article Info References

Search Criteria

All studies on the protective efficacy of influenza vaccine in elderly persons were identified through a MEDLINE search. The search terms were influenza vaccine, vaccine efficacy, elderly, mortality, hospitalized, and pneumonia. Only cohort observational studies that assessed mortality were included in the meta-analysis. Artificial challenge studies in which live attenuated influenza vaccine was administered intranasally to test vaccine efficacy were excluded from the meta-analysis because they did not duplicate the natural conditions of an influenza epidemic. Comparison of immune responses to vaccine was not one of our goals. We retrieved all of the articles found on efficacy in elderly persons and examined them for references to articles not found in our search. We excluded studies that described only upper respiratory illness scores. If an epidemic was developing while vaccination was in progress and no attempt was made to account for it, we excluded the study from analysis. Studies were also excluded if they contained no descriptions of epidemic or vaccine strains.

Large-scale case–control studies, cost-effectiveness studies, and the single randomized, placebo-controlled trial were considered separately whether or not they included mortality data. They were not included in the meta-analysis.

Comparative Factors

We compared the vaccine and control groups for age; sex; severity of illness; and whether patients were ambulatory or bedridden, nursing home residents or ambulatory patients in the community. We also compared epidemic and vaccine strains to determine whether there was a shift, drift, or match during the study period. We recorded whether studies were randomized, prospective, or uncontrolled. Finally, for outcome measures, we recorded the incidence of upper respiratory illness, hospitalization, pneumonia, and mortality.

Meta-Analysis

For the meta-analysis, we combined results using the method of DerSimonian and Laird [30], which calculates a pooled rate difference, that is, an absolute percentage reduction between the control and treatment groups.

Results were confirmed using the Yusuf-Peto modification of the Mantel-Haenszel technique [31, 32], which yields a combined estimate of the odds ratio under the assumption of homogeneity of the odds ratios across studies and provides a test of significance of the combined odds ratio. ² is the between-study variance [30].

Results

Top Methods Results Discussion Author & Article Info References

Literature Review

We found 20 observational studies on the efficacy of influenza vaccine in elderly persons [10-29]. All studies used a cohort design; none was randomized or placebo controlled. In most instances, the controls were patients who refused vaccination. Nine of the studies were prospective and 11 were retrospective. Table 1 shows the authors, epidemic years, and epidemic and similar vaccine strains for the observational studies; it also shows whether the vaccine and epidemic strains matched closely or differed because of a drift or shift. Not all vaccine strains for each year are shown; only the vaccine strains that were predicted to match the expected epidemic strains are listed.

The size and the sex and age distributions of the vaccine and control groups are shown in Table 2. Group sizes ranged from 17 to greater than 1000; the group size of most studies was greater than 100. Only six studies described sex distribution; in those studies, most of the participants were women. In most studies, the mean or median age of participants was 80 years or more.

Definitions of respiratory illness and criteria for diagnosing pneumonia varied among the studies. The reasons for hospitalization and death during the winter season (for example, whether a hospitalization was for pneumonia or whether death was due to pneumonia) were not always stated.

The type of facility in which the elderly persons resided and the underlying chronic health status of patients were described quantitatively in studies 6 and 11 and qualitatively in studies 1, 3 to 5, and 7. When described, characteristics in vaccinated persons were similar to those in controls. All studies consisted of institutionalized elderly patients, except for study 1, in which metropolitan elderly patients were vaccinated.

Evaluation for illness severity varied in the published studies. For example, in study 1, patients were classified as high-risk or non-high-risk. High-risk patients were those who had received medical care for any chronic disease (such as cardiovascular, pulmonary, renal, meta-bolic, neurologic, or neoplastic disease) during the preceding year. In study 3, patients were classified as either physically disabled or as having chronic symptoms, but specific diseases were not mentioned. Studies 4 and 11 commented on the need to include cardiovascular disease as a risk factor for increasing influenza morbidity and mortality. Studies 5 and 7 involved patients who required skilled nursing care and intermediate-level care.

Study 6 gave each patient a severity score (Department of Medical Sciences score) that was a measure of that patient's physical activity functional status. The higher the score, the more physically and mentally handicapped the patient. Although the mean severity score was slightly higher in the control group, the range was wide and the difference was not meaningful.

Chronic illnesses were specifically described in two studies. In study 4, a random sample showed that the percentage of chronic illnesses in vaccinated patients was similar to that in controls. In study 6, a systematic sample of both groups showed a similar distribution for chronic illnesses.

Long-term use of medications, such as digitalis, oral hypoglycemic drugs, and antiarrhythmic drugs, was mentioned in studies 4 and 6 and was similar in vaccinated patients and in controls in these studies.

Documentation of influenza infection in the patient populations of the studies varied. In studies 1, 3 to 8, and 11 to 20, the investigators documented the presence of influenza virus in their local communities by using viral isolation.

In the articles reviewed, we examined influenza vaccine efficacy by comparing the incidences of respiratory (influenza-like) illness, pneumonia, hospitalization, and mortality in vaccinated patients with those in controls (Table 3).

Strassburg and colleagues [21] reviewed 12 studies and found that vaccine reduced clinical illness by 33% (95% CI, 5% to 48%) and mortality by 74% (CI, 61% to 83%) in institutionalized elderly persons and that it reduced clinical illness by 5% (CI, –6% to 14%) and mortality by 47% (CI, 5% to 70%) in noninstitutionalized elderly persons. These authors used the standardized morbidity ratio, estimated as 1 – SMR, for calculating clinical illness and the Mantel-Haenszel risk ratio for mortality. The information in Table 1,Table 2, and Table 3 reflect only Strassburg and colleagues' [21] nursing home patients and not the patients in the other 11 studies reviewed by these authors.

Meta-Analysis

We did meta-analyses on the 20 observational studies with cohort designs. If the necessary data were available, meta-analyses were completed for respiratory illness Figure 1, pneumonia Figure 2, hospitalization Figure 3, and mortality (Figure 4). Combining the results of the 20 observational studies showed that, in vaccinated patients, the risk for respiratory illness was 44% that of controls (CI, 32% to 61%; P < 0.00001 [data from 9043 patients]); risk for pneumonia was 47% that of controls (CI, 34% to 65%; P < 0.00001 [data from 24 774 patients]); risk for hospitalization was 50% that of controls (CI, 35% to 72%; P = 0.00023 [data from 24 224 patients]); and mortality was 32% (CI, 24% to 44%; P < 0.00001 [data from 29 928 patients]). In other words, the pooled estimate of vaccine efficacy (1 – odds ratio) was 56% (CI, 39% to 68%) for preventing respiratory illness, 53% (CI, 35% to 66%) for preventing pneumonia, 48% (CI, 28% to 65%) for preventing hospitalization, and 68% (CI, 56% to 76%) for preventing death. The studies with extreme outliers in the 95% CIs can be accounted for by small sample sizes or numerators of 3 or less.

View larger version (35K): [in this window] [in a new window]   Figure 1. Meta-analysis of odds ratios by random-effects model (DerSimonian and Laird [30] ) of observational cohort studies for the effect of influenza virus immunization on respiratory illness. The odds ratio pooled estimate for all studies is shown at the bottom of the graph. See Table 1 for reference numbers of studies.

View larger version (31K): [in this window] [in a new window]   Figure 2. Meta-analysis of odds ratios by random-effects model (DerSimonian and Laird [30] ) of observational cohort studies for the effect of influenza virus immunization on pneumonia. The odds ratio pooled estimate for all studies is shown at the bottom of the graph. See Table 1 for reference numbers of studies.

View larger version (24K): [in this window] [in a new window]   Figure 3. Meta-analysis of odds ratios by random-effects model (DerSimonian and Laird [30] ) of observational cohort studies for the effect of influenza virus immunization on hospitalization. The odds ratio pooled estimate for all studies is shown at the bottom of the graph. See Table 1 for reference numbers of studies.

View larger version (46K): [in this window] [in a new window]   Figure 4. Meta-analysis of odds ratios by random-effects model (DerSimonian and Laird [30] ) of observational cohort studies for the effect of influenza virus immunization on mortality. The odds ratio pooled estimate for all studies is shown at the bottom of the graph. See Table 1 for reference numbers of studies.

Study 1a represents a major shift and is included in the meta-analysis for comparative purposes (Figure 2, Figure 3 and Figure 4).

The random-effects model of DerSimonian and Laird [30] was used for the primary analysis because this estimate adjusts for the amount of heterogeneity present. Considerable heterogeneity was encountered in the analysis of respiratory illness (² equals 0.4039; Figure 1). Less heterogeneity was found for pneumonia (² equals 0.0770; Figure 2). No heterogeneity was found for hospitalization (² equals 0.0000; Figure 3) or mortality (² equals 0.0000; Figure 4). The Yusuf-Peto fixed-effect model gave similar estimates with slightly narrower CIs (not shown).

The studies and study subgroups were also divided into those in which the vaccine and epidemic influenza virus strains were defined as identical by the Centers for Disease Control and Prevention (14 studies) or in which the epidemic strain was defined as a drift variant of the vaccine strain (12 studies). A meta-analysis of these two groups showed that vaccine efficacy was considerable for all four outcomes measured (Table 4). The favorable effect was substantial and was statistically significant except for hospitalization with the drift variant (P = 0.064). We excluded studies or study subgroups shown in Table 4 when a major shift was encountered or when no epidemic occurred.

Only two studies were done in noninstitutionalized elderly persons. Consequently, we did not do a meta-analysis comparing institutionalized with noninstitutionalized elderly persons.

Case-Control Studies

Large case–control studies have only recently been done to study the efficacy of influenza vaccine. Foster and colleagues [33] did a large-scale case–control study in which the case-patients were patients hospitalized for pneumonia and influenza. Other case–control studies were done by Barker and coworkers [34] and Strikas and colleagues [35]. Foster and colleagues [33] studied noninstitutionalized patients 65 or more years of age who were seen in 20 acute-care hospitals in southern Michigan. The case-patients were matched by age, sex, race, and zip code to randomly sampled controls in the general population. Questionnaires were completed by 1907 persons (449 case-patients and 1458 controls). Approximately 52% were women, and the mean age—estimated from the authors' data—was about 78 years. During the 3-month peak influenza period, vaccine efficacy was 45% (CI, 14% to 64%; P = 0.009) in reducing the likelihood of hospitalization for pneumonia and influenza. During 3 months before and after the epidemic, when virus activity was low or nonexistent, vaccine efficacy was 21% (P = 0.36); this was not statistically different from zero.

Foster and colleagues also pointed out that comorbid conditions were present in 71.3% of case-patients and only 47.5% of controls. During the epidemic peak period, heart disease increased the risk for hospitalization for pneumonia and influenza by a factor of 2.04. Heart disease was not a risk factor during the nonepidemic period. Lung disease increased the risk for hospitalization by sevenfold to eightfold during both periods. Influenza immunization reduced the risk for hospitalization by a factor of 55% during the epidemic but was not an important factor during the nonepidemic period. Mortality was not assessed.

Fedson and coworkers [36] recently published their findings, which were based on the Manitoba population registry for the 1982-1983 and 1985-1986 winter out-breaks. They did a matched-set analysis in which they matched each case-patient with influenza-related illness with three community controls for age, sex, and residence. They showed that influenza vaccine was effective in preventing 32% to 39% of hospitalizations for pneumonia and influenza, 35% to 65% of hospital deaths from pneumonia and influenza, and 43% to 50% of hospital deaths from all respiratory conditions. Unexpectedly, the vaccine was also shown to be effective in preventing 27% (CI, 7% to 42%) to 30% (CI, 12% to 43%) of deaths from all causes.

The Medicare Influenza Vaccine Demonstration Project was conducted by the Health Care Financing Administration between 1988 and 1992 [37]. Three case–control studies were done [33-35]; one was the study by Foster and colleagues [33], described above. The vaccine was 31% to 45% effective in preventing hospitalization for any type of pneumonia during the 1989-1990, 1990-1991, and 1991-1992 winter seasons. Mortality was not assessed.

Cost-effectiveness Studies

Two recent cost-effectiveness studies have been reported. Nichol and coworkers [38] did a serial cohort study of elderly persons during three winter seasons: 1990-1991, 1991-1992, and 1992-1993. For each season, influenza vaccine was associated with a 48% to 57% reduction in hospitalization for pneumonia and influenza and a 27% to 39% reduction in hospitalization for all acute and chronic respiratory conditions. The cost of hospitalization was reduced for an average direct savings per year of $117 per person. Direct cost savings per year were calculated by subtracting the cost of hospitalization for vaccinated persons plus the cost of the influenza vaccination program from the cost of hospitalization for unvaccinated persons. Mortality from all causes was decreased by 39% to 54% in vaccinated persons.

In a case–control study, Mullooly and colleagues [39] reviewed influenza vaccine effectiveness in a health maintenance organization over nine winter seasons. The vaccine did not reduce in-hospital mortality but was effective in preventing influenza-related morbidity. Mortality after hospitalization was not reported. The vaccine reduced influenza hospitalizations by 30% for high-risk elderly persons and by 40% for non-high-risk elderly persons. A net per person savings of $6.11 accrued to the health maintenance organization for high-risk elderly persons. The overall net per person savings was $1.10 because there was a net cost of $4.82 per vaccination for non-high-risk elderly persons. The net cost savings considered the costs of preventing hospitalizations and outpatient contacts, of vaccine, of promotion, of delivery and wastage, and of adverse reactions. For all elderly persons, earlier cost–benefit and cost-effectiveness studies have been reviewed elsewhere [40].

Randomized, Double-Blind, Placebo-Controlled Study

Only one randomized, double-blind, placebo-controlled study in elderly persons has been reported [41]. Vaccination policies for elderly persons differed in the United States and Holland at the time of the study [42]. Govaert and associates [41] found a 50% reduction in serologically documented influenza and a 53% reduction in clinical influenza in vaccinated persons. The vaccine and epidemic strains were the same. Follow-up for mortality was not a focus of this study.

Discussion

Top Methods Results Discussion Author & Article Info References

Studies of influenza vaccine efficacy differ in their matching of virus strains and their definitions of clinical illness, and they are confounded by the presence of other respiratory microorganisms that cause illness during an influenza epidemic. Studies on vaccine efficacy, therefore, need to be evaluated with caution. The naturally occurring influenza virus strain often varies from year to year, and although the vaccine strain frequently matches the epidemic strain, occasionally it does not. As a result, vaccine efficacy is usually expected to vary according to the match between the epidemic and vaccine strains. However, with stratified analysis, we found efficacy in studies in which the epidemic strain was a drift variant from the vaccine strain to be similar to efficacy in studies in which the epidemic and vaccine strains were the same. When the epidemic strain showed antigenic shift, vaccine efficacy was nil, as expected.

Standard definitions for influenza-like illness, influenza-related pneumonia, hospitalization, and death do not exist, and this confounds comparisons among studies. Determining whether pneumonia is present is one of the most difficult diagnoses in clinical medicine today [43]; pneumonia is frequently confused with congestive heart failure, the acute respiratory distress syndrome, and other disease entities, and this confusion confounds analysis. Reasons for death are also difficult to determine, which further compounds the difficulty of attributing death to influenza.

Appropriate laboratory studies should be done to fully characterize the causative infectious microorganisms responsible for an epidemic of respiratory illness. In a single winter season, several respiratory agents are usually present and causing illness. Microorganisms such as Mycoplasma pneumoniae, respiratory syncytial viruses, para-influenza viruses, and adenoviruses have been shown to circulate simultaneously during influenza virus epidemics [44, 45]. To define the scope of an epidemic, laboratory studies should include influenza virus isolation, isolation for other possible respiratory pathogens, and serologic titers for the suspected pathogens before and after the epidemic. Without these determinations, attribution of morbidity and mortality may be imprecise.

Estimates of vaccine effectiveness are conservative in most studies. During periods of peak influenza activity, other respiratory agents cause acute respiratory disease and occasionally result in hospitalization and death [4, 5, 24, 25]. Vaccine efficacy, therefore, is always underestimated, and case-mortality rates in vaccinated persons are overestimated [21].

Determining the ability of influenza vaccine to reduce upper respiratory illness is also difficult because so many respiratory agents are usually present. Without carefully monitoring for these agents or even for influenza virus alone, it is difficult to accurately determine rates of infection in vaccinated patients and controls.

This question was approached appropriately in the study done by Dowdle and colleagues 20 years ago [6]. They showed that influenza vaccine can prevent clinical influenza-like illnesses during an epidemic. Using groups of 68 to 78 patients, they showed that a vaccine containing influenza A/Aichi/68 (H3N2) reduced influenza-like illness by 40%, fever by 47%, confinement to bed by 68%, positive serodiagnosis of influenza by 78%, and illness plus positive serodiagnosis by 88% compared with groups receiving vaccines that did not contain the epidemic A/Aichi strain. The controls were patients who received vaccines with strains unrelated to the epidemic strain.

The effect of an influenza epidemic is usually greatest after the epidemic has been present for at least several weeks. Mortality, in particular, should be assessed 2 months after the epidemic to see its full effect [4, 5, 13]. This may explain why hospitalization and mortality rates were greater in the nursing home last monitored by Cartter and coworkers [17]. Had these investigators followed the effect of the epidemic at all three nursing homes for the full period of the epidemic, larger differences between the vaccinated and unvaccinated groups might have been seen. Consequently, the follow-up period needs to be long enough to account for most of the cases.

Vaccine efficacy has been predicted to be better in healthy than in infirm elderly persons because the immune response to vaccine is better in the former than in the latter [46, 47]. We could not adequately confirm this because the nursing home studies we reviewed did not provide enough detail on who was infirm and who was not.

The issue of repeated influenza immunization also deserves comment. Hoskins and colleagues [48], in a series of studies in adolescent boys, raised the point that repeated influenza vaccinations might be detrimental for all age groups when a drift variant is encountered during an epidemic. In most of the studies we reviewed, elderly persons had been repeatedly vaccinated, and the vaccine had remained efficacious. These findings suggest that the point raised by Hoskins and colleagues [48] does not apply to elderly persons.

Several limitations were encountered in these studies. Not all elderly persons are adequately represented in nursing home populations because only a small percent-age of elderly persons are in nursing homes [21]. Most of the studies we reviewed were conducted in very old persons and in persons with underlying diseases. Selection bias may occur in the choice of persons vaccinated; Severely ill patients are more likely to be vaccinated because of the perception of need for vaccination by the health care provider. Unequal exposure to the epidemic strain may occur when dissimilar patients are geographically separated in the nursing home; this was noted in several of the studies reviewed [12, 15, 21]. On occasion, documentation of whether a patient was actually vaccinated is poor. When memories rather than written records are used, faulty denominator data may result.

For ethical reasons, randomized, double-blind, placebo-controlled trials are difficult to do in high-risk persons in the United States. Thus, we combined the 20 observational cohort studies and did a meta-analysis to more clearly define influenza vaccine efficacy in elderly persons.

The validity of our meta-analysis may be limited by the issues noted. However, if the vaccine is effective in the predominantly institutionalized population studied, it is even more likely to be effective in the healthier elderly persons residing in the community, who are presumably less likely to have immune system defects.

It is worth emphasizing that the data for mortality, the "hardest" end point, are based on observations in almost 30 000 patients, and they are not statistically heterogeneous. Concern is often raised that meta-analysis may be affected by the tendency to publish the most positive studies. We used the formula proposed by Rosenthal [49] to estimate the number of unpublished "null" studies (studies that found no difference) that would have to exist to contradict the pooled results. There would have to be 952 unpublished null studies of respiratory illness, 153 unpublished null studies of pneumonia, 26 unpublished null studies of hospitalization, and 466 unpublished null studies of mortality to refute the results of this meta-analysis. Therefore, we believe that our findings are unlikely to be the result of publication bias. There is an emerging consensus that, especially for health policy decisions, meta-analysis should not be limited only to randomized controlled trials [50].

Our meta-analysis may also have been affected by the focus on outbreaks to study vaccine efficacy. An outbreak is a potentially negative event for a vaccinated population. In nursing homes in which the vaccination rate is high and no outbreaks develop during a nationwide influenza epidemic, the positive effect of vaccine would go unreported. Underreporting of this positive effect of vaccine would underestimate vaccine efficacy. In fact, outbreaks are the exception rather than the rule in nursing homes because of the high vaccination rate in these institutions (Patriarca P. Personal communication).

The case–control studies reviewed also indicate that influenza vaccination reduces the incidence of hospitalization for pneumonia from all causes. Importantly and unexpectedly, they showed that vaccination reduced the incidences of hospital deaths from all respiratory conditions and from all causes.

In addition to being effective, influenza vaccine is safe. In a controlled study in male veterans that used a cross-over design, Margolis and associates [51] showed that the incidence of local and systemic side effects in the first 48 hours after vaccination was less than 5%.

Despite repeated documentation of the efficacy of influenza vaccine in elderly persons, universal immunization has not been achieved in this group. And yet the acceptance of influenza vaccine has improved substantially. In the early 1980s, 20% to 23% of elderly persons were immunized. In the late 1980s, this percentage began to increase. By the 1993-1994 season, 58% of elderly persons were receiving influenza vaccine annually [52]. But this is still inadequate; our target should be all elderly persons.

From the articles reviewed—20 cohort studies, 3 case–control studies, and 1 randomized, double-blind, placebo-controlled study-we can conclude that morbidity and mortality are significantly reduced when influenza vaccine is administered before an epidemic if the vaccine strain is identical or similar to the epidemic strain. Annual universal influenza immunization of elderly persons is a public health imperative that should be carried out by practicing physicians and public health organizations.

Dr. Hermogenes: Sheehan Memorial Hospital, 425 Michigan Avenue, Buffalo, NY 14203.

Dr. Sacks: Mount Sinai Medical Center, Clinical Trials Unit, Box 1042, One Gustave L. Levy Place, New York, NY 10029-6574.

Dr. Lau: Division of Clinical Care Research, Tufts-New England Medical Center, NEMC #63, 750 Washington Street, Boston, MA 02111.

Dr. Levandowski: Division of Viral Products, Center for Biologics Evaluation & Research, 1410 Rockville Pike, Rockville, MD 20892.