Relationship between annual influenza vaccination and winter mortality in diabetic people over 65 years

Abstract

Background

Influenza is an important cause of morbidity and mortality in older people, especially in those with some high-risk conditions such as diabetes mellitus. This study assessed the relationship between influenza vaccination status and winter mortality among diabetics 65 y and over during four consecutive influenza seasons.

Methods

Population-based cohort study including 2,650 community-dwelling individuals 65 y or older with diabetes mellitus followed between January 2002 and April 2005 in Tarragona, Spain. Influenza vaccination status was evaluated every year of the study and the primary endpoint was considered all-cause death during the study period. Deaths were classified as occurring within influenza periods (January–April) or non-influenza periods. The relationship between vaccination and winter mortality was evaluated by multivariable discrete-time hazard models.

Results

Influenza immunization was associated with a reduction of 33% (95% confidence interval: 4–53) in the adjusted risk of all-cause mortality throughout the overall influenza periods 2002–2005. The attributable risk to vaccination in reducing mortality was 13.5 per 100,000 person-weeks within influenza periods, estimating that one death was prevented for every 435 annual vaccinations.

Conclusion

Our data confirm the benefit of influenza vaccination in reducing mortality and supports the strategy of annual vaccination in diabetics aged at least 65 y.

Keywords: :

• Diabetes Mellitus
• Influenza vaccine
• effectiveness
• elderly
• mortality

Introduction

The influenza viruses are a major cause of morbidity and mortality. The incidence of influenza is higher in children and young adults, but influenza-associated morbidity and mortality increase with age, especially in individuals with underlying medical conditions such as diabetes mellitus.^1-3 The impact of annual epidemics of influenza on morbidity and mortality in older people and the effectiveness of influenza vaccine have been the basis for the implementation of vaccination programs in older people worldwide.⁴ The immunization against influenza is important in preventive medicine for chronic diseases like diabetes, in which the primary care service has an important role. According to the Advisory Committee on Immunization Practices of the American Academy of Family Physicians (AAFP), annual vaccination of high-risk individuals before the influenza season is the most effective measure to reduce the impact of influenza. In relation to diabetes mellitus, the goal in primary care should be the immunization of all people with diabetes, especially if they have risk factors such as renal or cardiac disease or those who have been hospitalized recently.⁵

Although the effectiveness of influenza vaccine in reducing mortality has been studied extensively in hospitalized and institutionalized patients in times of high influenza activity, there are few studies on the clinical benefit of vaccination in the medium to long-term non-institutionalized patients,^6-14 and the effectiveness of annual influenza vaccination programs in individuals at risk is controversial.¹⁵

To assess the possible effectiveness of influenza vaccination in preventing mortality, we conducted a cohort study of 11,240 Spanish community-dwelling elderly individuals followed between 2002–2005.¹⁶ The analysis on vaccine effectiveness covering patients with chronic pulmonary and heart disease has been published previously.^17,18 In the present study, we have assessed the relationship between the annual influenza vaccine status and all-cause winter mortality in the subgroup of 2,650 individuals with diabetes mellitus. Moreover we have studied the risk of mortality in each influenza and non-influenza periods throughout the study period.

Patients and Methods

Design, setting and study population

The 2650 cohort members represented all subjects 65 y or older assigned to the 8 Primary Health Care Centres (PHCCs) in the region of Tarragona (Catalonia, Spain) participating in the study who had a diagnosis of diabetes mellitus coded in their electronic clinical record at the beginning on the study (ICD-9: 250). Details on design and setting have been described extensively elsewhere.¹⁶

All cohort members were followed from the beginning of the study (January 1, 2002) until enrolment from the PHCC ceased, the occurrence of death or until the end of the study (April 30, 2005). The study was approved by the Ethical Committee of the Catalonian Health Institute and it was conducted in accordance with the general principles for observational studies.

Sources of data

All participating PHCCs have an institutional computerized clinical record system which contains registries of immunizations, laboratory tests, medication prescription, diagnoses associated with outpatient visits and chronic diseases coded according to the International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9). The electronic records of each cohort member were used to identify whether the individual had received or not the influenza vaccine in each influenza vaccination campaign, and it was also used to identify the presence of diabetes mellitus (ICD-9 code 250), co-morbidities and other medical conditions.

Outcome measure and definitions

Primary outcome was all-cause death. Deaths were identified initially in the Institutional Demographic Database (which is updated monthly with administrative data about deaths, patients who have moved or new patients assigned to a PHCC). Afterwards, a review of the reference Civil Registry Offices of the 8 PHCCs was used to identify those deaths that had occurred in cohort patients who had not been registered in the Institutional Database. This review was also used to validate the exact date of death in all cases. Deaths were classified as occurring during influenza period (January–April) or during the rest of the year. The influenza period was defined as the period during which influenza-like illnesses were frequently reported in the study area, from January 1 to April 30 for each year of the study.¹⁶ As a reference non-influenza period we considered deaths occurring from July 1 to August 31 (reference Summer period).

Among the 167 patients who died within the influenza periods, cause-specific death was registered in the primary care clinical record in only 89 cases (53.3%). Among these patients, the specific cause of death was a cardiovascular disorder in 31 cases (34.8%), a neoplasia in 21 cases (23.6%), an infectious cause in 7 cases (7.9%), a respiratory cause in 6 cases (6.7%) and other or unspecific causes in 24 cases (27%). Although cause-specific death was the ideal outcome measure, we considered that the degree of missing data on cause-specific death justified using all-cause mortality as the primary outcome measure.

Exposure to influenza vaccination

For each year, information on the influenza vaccination status of the subjects was determined by a review of the PHCCs’ clinical records, which contain specially designated fields for annual influenza vaccinations. We assumed that information in clinical records was complete, so a subject was considered as non-vaccinated when data on vaccination were not recorded.

Influenza vaccine status was considered as a dichotomous (vaccinated or non-vaccinated) time-varying condition throughout the study period (i.e., in the analysis covering the overall study period, the same person could be considered non-vaccinated in 2002, vaccinated in 2003 and non-vaccinated in 2004 according to the reception or not of a dose of influenza vaccine in the prior autumn).

Covariates

Covariates included dichotomous variables for sex, chronic heart disease (including heart failure and coronary artery disease), chronic lung disease (including asthma, emphysema or chronic bronchitis), chronic severe nephropathy (nephrotic syndrome, renal failure, dialysis or transplantation), chronic severe liver disease (cirrhosis), cancer (solid organ or hematological neoplasia), hypertension, obesity, current smoking and immunocompromised status. Age and the number of outpatient visits in the previous two years were considered as continuous covariates. The presence of co-morbid conditions was determined by a review of the diagnosis codes in the electronic clinical record of each cohort member.

Statistical analysis

Incidence rates (IR) of death were calculated as person-years and person-weeks. For the numerator we used number of deaths and for denominator the total number of person-years or person-weeks of observation for each study period considered. So, for each individual we determine the amount of observation time contributed to that period and to add up those contributions for all cohort members. Attributable risk (AR) was the difference between IR among vaccinated and non-vaccinated subjects (AR = IR exposed-IR non-exposed). Number needed to be vaccinated (NNV) to save one death was estimated for influenza periods (January–April = 17.1 weeks) and was calculated as the inverse of the AR (NNV = 1/AR) with its confidence interval.¹⁹

We performed stratified analysis by influenza period and reference summer period and by year.

To adjust for potential selection bias, we calculated a propensity score by using a logit model with the dependent variable influenza vaccine status. The above covariables and the influenza vaccine status the previous year were considered potential candidates for the calculation of the propensity score. Main effects of variables were selected for inclusion in the first logit model using a stepwise logit regression. The covariates were entered into the model if they were significant at the 0.05 level. A second step analysis added needed interactions and quadratics of those variables whose main effects were selected by the first stepwise procedure. The propensity score was then divided into quintiles and this score was included as a covariate in all the multivariate models.^20,21

As we were interested on time recorded on a discrete scale, each year of influenza and non-influenza periods, we fit a discrete-time survival model.

To calculate the relationship between the risk of death in each time period and individuals’ characteristics, a discrete-time hazard model was calculated, adjusting for propensity quintiles score and using the methodology described by Singer et al.²² and Hosmer et al.²³ Time was defined as each influenza or reference summer period of each year and the event as occurrence of death. The data set was structured as person-period. The person-period data set has a separate record for each time period when an individual is at risk of event occurrence. The values of the variables in the person-period data set reflect the status of person i on that variable in the jth period. This model that assumes proportional odds, can be expressed as a linear relationship between logit hazard and predictors and introduce the discrete-time periods as dummy variables, D_j (Equation 1 (1) ).

(1)

We quantified the relative magnitude of risk in each time period, h(t_j) by taking the antilogit of α_j: for individuals in the baseline group. When we antilog the coefficient β_k we assess the effect of its associated predictor in the odds of event occurrence in every time period while controlling for the other predictors in the model

The cumulative survival at period j,, can be calculated as the survival at period j-1, S_j-1, by 1 minus hazard at period j, h_j : S_j = S_j-1 *(1-h_j).²²

We evaluated the proportional odds assumptions adding the covariate by time periods interactions and by a linear interaction with time. All the models have been compared through the partial likelihood ratio test or the Akaike’s information criterion (AIC).²¹

Except for the annual influenza vaccine status, the other covariates were time-invariant and defined at study entry. All statistical tests were two-sided at the 5% significance level. The analyses were performed using Stata/SE version 10.1 for Windows (StataCorp. LP).

Results

During the total 40-mo study period, the 2650 cohort members were observed for an amount of 8167 person-years (426,143 person-weeks). The mean age of the subjects when the study started was 74.4 y (standard deviation: 6.7) and 41.2% were men. At baseline, 2180 (82%) of the cohort members had some other form of co-morbidity, mostly hypertension (66.6%), chronic heart disease (16.3%) or chronic lung disease (12.3%). Regarding treatment of diabetes 21% patients get insulin, 71% take oral antidiabetic and 8% only dietetic treatment. Table 1 shows the characteristics of the study population when the study started (Jan 1, 2002) according to the reception or non-reception of the influenza vaccine in Autumn 2001. Vaccinated subjects were slightly older than non-vaccinated subjects, and they had more frequency of attendance and co-morbidity than non-vaccinated subjects.

Table 1. Baseline characteristics of the Study Population according to their influenza vaccination status when the study started (Jan 1, 2002)

		Non-vaccinated	Vaccinated
		(N=1064)	(N=1586)	p value^(a)
		No of persons (%)
Age group^(b)	65-74 yrs	665 (62.5)	350 (51.9)	<0.001
	75-84 yrs	317 (29.8)	622 (39.2)
	≥ 85 yrs	82 (7.7)	141 (8.9)
Sex	Male	423 (39.8)	670 (42.2)	0.202
	Female	641 (60.2)	916 (57.8)
Number of outpatient visits during previous 2 yrs^(c)	0-14 visits	287 (27.0)	115 (7.3)	<0.001
	15-29 visits	366 (34.4)	518 (32.7)
	30 or more visits	411 (38.6)	953 (60.1)
Pneumococcal vaccination in previous 5 yrs		244 (22.9)	1112 (70.1)	<0.001
Medical conditions	Chronic heart disease	157 (14.8)	276 (17.4)	0.071
	Chronic lung disease	98 (9.2)	229 (14.4)	<0.001
	Hypertension	665 (62.5)	1100 (69.4)	<0.001
	Obesity	256 (24.1)	422 (26.6)	0.141
	Smokers	99 (9.3)	111 (7.0)	0.031
	Chronic liver disease	27 (2.5)	41 (2.6)	0.94
	Chronic nephropathy	40 (3.8)	79 (5.0)	0.137
	Active neoplasia	20 (1.9)	49 (3.1)	0.055
	Immunosuppressive medication	38 (3.6)	118 (7.4)	<0.001
Immunocompromised status ^(d)		114 (10.7)	257 (16.2)	<0.001

^(a) p-values were calculated with chi-square test or t-test;

^(b) Mean age was 73.1 y (SD: 6.9) for non vaccinated subjects and 75.2 y (SD: 6.5) for vaccinated subjects; p < 0.001;

^(c) The mean number of outpatient visits was 28.1 (SD: 23.6) for non vaccinated subjects and 39.0 (SD: 25.5) for vaccinated subjects; p < 0.001;

^(d) Immunocompromise was a composite variable defined by the presence of any one of the following: cancer (solid organ or hematological neoplasia), chronic severe nephropathy (nephrotic syndrome, renal failure, dialysis or transplantation), chronic severe liver disease (cirrhosis), anatomical or functional asplenia, AIDS, and long-term corticosteroid therapy (20 mg/day of prednisone) or another immunosuppressive medication.

Of the 2,650 cohort members, 38 (1.4% moved from PCC and 384 (14.6%) died during the 40-mo study period. A total of 167 deaths occurred within the influenza period, whereas 217 occurred between May–December (54 of which occurred within the reference summer period).

If we consider those cohort members who remained in the closed cohort at the beginning of each year (excluding patients who died or moved during the prior year), the annual vaccination coverage reached 59.8% in autumn 2001, 65.8% in autumn 2002, 70.8% in autumn 2003 and 71.3% in autumn 2004. In total, 168,512 person-weeks were observed in all the influenza periods between 2002 and 2005, of which 112,416 person-weeks (66.8%) were vaccinated against influenza in the respective previous autumn.

Table 2 describes incidences of winter death in vaccinated and non-vaccinated subjects for each of the influenza seasons during the study period and shows unadjusted relative risk ratios calculated for vaccinated people vs. non- vaccinated. The incidence of winter death was lower among vaccinated than non-vaccinated subjects in all four influenza seasons. However, although vaccination pointed to a reduction in mortality risk in each influenza period (relative risk ratio for vaccinated people ranging from 0.67 in Winter 2002 to 0.97 in Winter 2005), the differences did not reach statistical significance.

Table 2. Incidence and risk of all-cause mortality among study population within the influenza period (January–April) between 2002 and 2005 according to the reception of the influenza vaccine in prior autumn

Parameter	2002	2003	2004	2005
Total Number of eligible subjects	2650	2558	2422	2286
Non-vaccinated	1064	876	707	657
Vaccinated	1586	1682	1715	1629
Influenza vaccine coverage	59.80%	65.80%	70.80%	71.30%
Number of deaths
Non-vaccinated	12	19	13	17
Vaccinated	12	28	25	41
Incidence of all-cause mortality per 100,000 person-week
Non-vaccinated	66.2	127.9	108.27	153.06
(95% CI)^(a)	(34.4-114.9)	(77.4-198.7)	(57.9-184.1)	(87.5-245.1)
Vaccinated	44.3	97.9	85.67	148.85
(95% CI)	(22.9-77.2)	(65.3-140.9)	(55.6-126.0)	(105.7-202.3)
Incidence of all-cause mortality per 1,000 person-influenza period
Non-vaccinated	11.3	21.7	18.4	25.9
(95% CI)	(5.9-19.7)	(13.3-34.1)	(9.9-31.5)	(15.1-42.1)
Vaccinated	7.6	16.7	14.6	25.2
(95% CI)	(3.9-13.2)	(11.2-24.2)	(9.5-21.6)	(18.2-34.8)
Attributable risk per 1000 person and influenza period	3.7	5	3.8	0.7
(95% CI)	-3.9 to 11.4	-6.4 to 16.5	-7.6 to 15.2	-18.3 to 19.7
Number needed to vaccinate to prevent one death (NNV)^(b)	270	198	263	1429
95% CI)	(88 to ∞)^(c)	(61 to ∞)	(66 to ∞)	(51 to ∞)
Relative Risk Ratio^(d)	0.67	0.77	0.79	0.97
(95% CI)	0.29-1.48	0.43-1.39	0.41-1.56	0.48-1.82
p value	0.32	0.389	0.51	0.732

^(a) CI denotes confidence interval; ^(b)NNV is the number needed to prevent one death within an influenza period (January–April = 17.1 weeks) and is estimated as 1/([mortality rate difference/100,000]x17.1); ^(c)Confidence interval calculated according to Altman,1998. The sign ∞ means that the upper limit of the confidence interval tends to infinite; ^(d)Relative risk ratio in vaccinated people vs. non-vaccinated

Table 3 shows the discrete-time estimations of conditional hazard and cumulative survival probabilities by periods and year, unadjusted and adjusted for influenza vaccine status. As can be seen in the table, the hazard of death in the influenza periods were greater than in the reference summer periods, regardless if we adjusted for influenza vaccine or not. Indeed, if we adjusted for the influenza vaccine, the hazard of death for vaccinated subjects was lower than unvaccinated at each time period. The risk of death increased over time except for the period 2004, reaching the highest risk in 2005.

Table 3. Discrete-Time Estimation of Conditional Hazard and cumulative Survival rate by periods and year, unadjusted and adjusted for influenza vaccine status. Influenza Period (from January 1 to April 30), reference Summer Period (July 1-August 31)

Period	No of deaths			Hazard^#		Cumulative survival^$
				Adjusted for influenza vaccine status
		Unadjusted		Hazard*(%)		Cumulative survival$ (%)(%)
		Hazard*(%)	Cumulative survival^$ (%)	Unvaccinated subjects	Vaccinated subjects	Unvaccinated subjects	Vaccinated subjects
Influenza period
2002	24	0.91	99.09	1.01	0.84	98.99	99.16
2003	47	1.84	97.27	2.07	1.72	96.95	97.46
2004	38	1.57	95.75	1.78	1.48	95.33	96.02
2005	58	2.54	93.32	2.88	2.4	92.48	93.72
Reference summer period
2002	15	0.57	99.43	0.63	0.52	99.37	99.48
2003	22	0.86	98.58	0.97	0.8	98.41	98.68
2004	17	0.7	97.89	0.8	0.66	97.62	98.03

Note: The discrete-time hazard is the conditional probability that an individual will experience the event in a period, given that he or she did not experience it in any earlier time period; *Fitted value of hazard calculated by taking the antilogit of the model Logit[h(t_j)] = [α₁D₁ + α₂D₂ + α₃D₃ + α₄D₄]; ; ^#Fitted value of hazard calculated by taking the antilogit of the model Logit [h(tj)] = [α₁D₁ + α₂D₂ + α₃D₃ +α₄D₄] + β* influenza_vaccine; ^$The cumulative survival at period j,S_j, has been calculated as the survival at period j-1, S_j-1, by 1 minus hazard at period j, h_j : S_j = S_j-1 *(1-h_j).

Considering the sum of deaths of the four influenza periods, 106 deaths were observed among persons who had received the influenza vaccine in the prior autumn whereas 61 deaths occurred among people who had not received it. The attributable risk for non-vaccinated was 13.5 deaths per 100,000 person-weeks, which meant 2.3 deaths per 1000 person-influenza period (95% CI: -1.9 to 6.6) and a NNV of approximately 435 annual vaccinations to save one winter death (95% CI: 152 to infinite). In the multivariable adjusted models the hazard of death was nearly zero in each time period although it remained higher in the influenza periods than in the reference summer period (data not shown).

When considering all of the influenza periods between 2002 and 2005, the reception of the influenza vaccine was a significant predictor for lower mortality, with a significant 33% protective effect (adjusted OR for propensity score quintiles = 0.67, 95% CI: 0.47–0.96), whereas the reception of the vaccine was not significantly associated with lower mortality (adjusted OR for propensity score quintiles = 0.70, 95% CI: 0.37–1.31) in the reference summer periods (Table 4) .

Table 4. Incidence and risk of all-cause mortality among study subjects by all of the influenza and reference summer periods, according to their influenza vaccine status throughout the study period (from January 1, 2002 to April 30, 2005)

	All influenza period (Jan-April 2002-2005)	All reference summer period (July-August 2002-2004)
Number of deaths	167	54
Non-vaccinated	61	21
Vaccinated	106	33
Incidence of all-cause mortality per 100,000 person-week
Non-vaccinated	110.5	91.5
(95% CI)^(a)	85.3-140.5	56.8-138.9
Vaccinated	97	75.8
(95% CI)	80.1-116.4	52.3-105.9
Incidence of all-cause mortality rate per 1,000 person-influenza period^(b)
Non-vaccinated	18.9
(95% CI)	14.6-24.1	-
Vaccinated	16.6
(95% CI)	13.7-20.0
Attributable risk per 1,000 person and influenza period^(b)	2.3
(95% CI)	-1.9 to 6.6	-
Unadjusted Odds Ratio	0.83	0.83
(95% CI)	0.60-1.14	0.48-1.44
p value	0.247	0.503
Multivariable-adjusted Odds ratio^(c)	0.61^(d)	0.71^(e)
(95% CI)	0.44-0.85	0.41-1.24
p value	0.003	0.231
Multivariable-adjusted Odds ratio^(f), additionally adjusted for the propensity score quintiles	0.67^(d)	0.70^(e)
(95% CI)	0.47-0.96	0.37-1.31
p value	0.031	0.263

NOTE: Discrete-time hazard analyses adjusted for age, sex, chronic heart disease, immunocompromise, number of outpatient visits, hypertension and smoking showed point estimates and confidence intervals that hardly varied compared with the reported multivariable adjusted odds ratios (data available on request); ^(a)CI denotes confidence interval; ^(b)An influenza period includes deaths from January 1 to April 30 (17.1 weeks); ^(c)Adjusted odds ratios are based on discrete-time hazard model; ^(d)Adjusted for age (p < 0.001), sex (p = 0.015), chronic heart disease (p < 0.001), immunocompromise (p < 0.001), number of outpatient visits (p < 0.001) and the interaction chronic heart disease x immunocompromise (p < 0.001); ^(e)Adjusted for age (p < 0.001), sex (p = 0.546), hypertension (p = 0.078, confounder), immunocompromise (p = 0.015), smoking (p = 0.436, confounder); ^(f)Previous models adjusted, as well as for the propensity score quintiles.

Discussion

The efficacy of influenza vaccination and the estimated impact of annual influenza epidemics on morbid-mortality have been the basis for implementing influenza vaccination programs for elderly and high-risk individuals.^2,3 However, the effectiveness of vaccination has been reported to decrease in older age-groups and high-risk persons, and the magnitude of clinical effectiveness of annual vaccination campaigns is controversial.^14,15,24,25 Nowadays, in this field, the gold standard of a large randomized controlled trial could create ethical difficulties and non-experimental studies evaluating influenza vaccination effectiveness must be applied.

In the present study, we have assessed the effects of the annual vaccination on winter mortality in diabetics aged 65 y or more throughout a relatively long time period including four consecutive influenza seasons with different intensity in the epidemic activity. Although the study was not randomized, the relatively large sample size of the study cohort, together with the logistic discrete time-hazard regression modeling and the adjustment for relevant covariates and the propensity score of the influenza vaccine in the multivariable analysis, provides an adequate basis for assessing the effects of the influenza vaccine on winter mortality in these persons. The discrete time-hazard methodological approach has allowed us to estimate the joint effect of the vaccine in each period and its association with individual characteristics, an aspect which so far has not been evaluated by other studies.

In the present study influenza vaccination was associated with a statistically significant reduction of 33%, adjusting for the propensity score, in the risk of all-cause winter mortality throughout the overall study period. Our result fits with those reported evaluating vaccination effectiveness among the general elderly population in the study area,¹⁶ and it also agrees with data reported by Voordouw et al. in a large cohort study focused on older people in the Netherlands, who found that the annual influenza vaccination was associated with an all-cause mortality risk reduction of approximately 28% during epidemic influenza periods.¹⁰

Although the benefits of influenza vaccination in preventing hospitalization and death have been largely reported, the effectiveness of the influenza immunization is not well understood for major cause-specific mortality, except for pneumonia. Recent studies have shown an association between influenza, respiratory infections and the onset of acute vascular events such as myocardial infarction, sudden death and stroke. Thus, influenza vaccination could reduce not only deaths from influenza-specific pneumonia, but it also could have an effect by reducing cardiovascular mortality.²⁶ Recently, Wang et al. have reported that influenza vaccination was significantly associated with a 44% lower risk of all-cause mortality among general elderly people in Southern Taiwan, observing a significant 55% reduction in the risk of death from diabetes among vaccinated subjects.⁷

In this study, cause-specific mortality was not available for almost half of the cases who died during the study period. Furthermore, in some patients the specific cause of death was not specific enough to classify as influenza-related mortality or not, which would result in the possibility of misclassification bias and lack of statistical power if analysis of cause-specific mortality is assessed. In general, when the event of interest is death, all-cause mortality is considered a more robust event than cause specific mortality.²⁷ Consequently, given the low frequency of serological confirmation for flu cases in clinical practice, all-cause mortality has been considered an acceptable outcome to assess the effectiveness of influenza vaccine in many observational studies.^28,29 Nevertheless, given that specific mortality was not evaluated, a residual confounding in the estimates of vaccine effectiveness cannot be excluded completely .

Due to the difficulties involved in primary care clinical practice for a serological diagnosis of influenza illness, the variability of immunological response to vaccination in older people, and the ethical problems arising from a randomized design, the effectiveness of the influenza vaccine at the population level has generally been assessed by observational studies using non-specific indexes and/or indirect effect (impact of influenza-like illness, hospitalizations and death from respiratory illness or any cause). If we consider all-cause mortality, the effectiveness of influenza vaccine in reducing deaths from all causes has ranged from 45–56% in cohort studies and 17–30% in case-control studies.^28,29 However, this concern is controversial. Simonsen et al. (who analyzed vaccine coverage and estimated the influenza-related mortality and deaths from any cause from 1968 to 2001 in the US elderly population) found no correlation between increasing vaccination coverage after 1980 with decreasing rates of mortality in any age group and concluded that many studies substantially overestimate the benefits of the influenza vaccine.²⁴

In this study, annual influenza vaccine coverages ranged from 59.8% to 71.3%, which is consistent with other data reported in older people with diabetes in Spain and other developed countries, where approximately 60–70% of these subjects are immunized annually against influenza.^1,2,30,31 Although vaccination coverage increased throughout the study period, the risk of death also increased, reaching the maximum value at 2005. This finding is difficult to interpret due to the fact that the closed cohort was growing older. Nevertheless, it has been observed that the cumulative survival rate was higher in vaccinated than in unvaccinated subjects in the four analyzed winter periods. In our study, the difference between all-cause mortality in non-vaccinated and vaccinated (AR) was 13.5 deaths per 100,000 person-weeks considering the total influenza periods between 2002 and 2005, and we estimated that in the total population one winter death was prevented for every 435 annual vaccinations, although this estimation does not exclude the possibility of a greater number since the value of the upper limits in the confidence interval are infinite.

Important aspects that determine vaccine effectiveness are the intensity of viruses circulating during the study periods and the similarity between vaccine strains and circulating strains.³² It must be noted that, during our study period, influenza activity in the study area as well as in northern hemisphere was mild-to-moderate in most countries, and was associated with a mixed circulation of Virus A and Virus B. In this period, the vaccine strains and predominant circulating strain [mainly A (H3N2)] were generally well matched.^33-36

Some methodological characteristics of this study need to be addressed. The analysis of discrete-time event occurrence, influenza period and summer period, has allowed us to assess not only the association between the influenza vaccine and death, but also the risk of death and cumulative survival in each influenza period from 2002 to 2005 and its relationship with individual characteristics such as age, comorbidity and immunological status. In this study, influenza vaccine status was considered as a time-varying dichotomous variable (“vaccinated” or “not vaccinated “), but other categories of influenza vaccine status (such as “first vaccination,” “revaccination,” or “interrupted vaccination”) that can influence vaccine effects, were not evaluated. Considering that diagnosis of diabetes was considered only according to ICD-9 diagnosis codes registered in clinical records and it was not laboratory-confirmed, a classification bias in some cohort members cannot be completely excluded.

As a major limitation of observational designs is a possible selection bias. In our study, vaccinated subjects were older and had more co-morbidity than non-vaccinated subjects. Those patients who had a higher number of underlying conditions had more visits than those patients who did not, and this meant a higher probability of vaccination. We account for these differences between vaccinated and unvaccinated subjects in the analysis by adjusting for these variables and for the propensity score of influenza vaccination in the logistic discrete time-hazard model. The use of propensity scores, in addition to covariance adjustment, has reduced this bias and increased the precision in the estimation of the true vaccination effect.²¹ However, the propensity score is only designed to balance measured covariates. It makes no claim to balance unmeasured covariates, so the possible influence of residual confounding due to unknown confounding factors on the estimates of vaccine effectiveness cannot be completely excluded.³⁷ In this way, when the present authors attempted to address this issue by comparing vaccine effects in winter with the summer period when influenza does not circulate, a very similar point estimate of adjusted vaccine effect against all-cause death in the winter (OR: 0.67; p = 0.03) compared with the summer (OR: 0.70; p = 0.26) was found, so, regardless of statistical significance, the closeness of these point estimates points to the possibility of unmeasured residual confounding.

In addition, considering that only data contained in the electronic clinical records (working since 1999) were used to establish baseline characteristics of cohort members at study start, important data which can influence mortality in diabetic people were not available (type of diabetes, HbA1c, presence of diabetic nephropathy and the lack of such data may be a limitation in this study.

There is ongoing controversy about the benefit of influenza vaccine in the elderly. Only a small fraction of winter deaths in the general elderly population are thought to be associated with influenza, and it has been argued that the results of cohort studies could be unreliable due to unadjusted selection biases.³⁸ However, given the demonstrated effectiveness of the influenza vaccine in protecting individuals, commencing new RCTs in populations at risk would create difficulties. Thus, well designed cohort studies are an acceptable alternative to estimate vaccine effectiveness against different clinically relevant outcomes among different populations at risk.

In conclusion, our data show that the reception of the annual conventional inactivated influenza vaccine was associated with a significant reduction in the risk of winter mortality among community-dwelling elderly persons with diabetes mellitus. This study adds to the body of knowledge on the effectiveness of influenza vaccination in elderly diabetics. Current clinical guidelines already call for vaccinating all elderly people (with or without diabetes). According to our data, the effects of vaccinating elderly diabetics on winter mortality is similar to reported vaccination effects of the general elderly population or sub-populations.^16-18 Our result shows a benefit of the annual influenza vaccination, even considering mild- or moderate severity of influenza seasons, and it supports an annual vaccination strategy for elderly diabetics. It must not be forgotten that approximately one-third of elderly patients with diabetes remain annually non-vaccinated and the increase in vaccination uptakes should be a major goal in the care of these patients.

Acknowledgments

This study was supported by a grant from the Instituto de Salud Carlos III from the Spanish Health Ministry (Grant ID FIS PI-021117). The authors would like to thank all the family physicians and nurses of the Primary Care Centres of Tarragona-Valls who collaborated in this study.

Disclosure of Potential Conflicts of Interest

No potential conflicts of interest were disclosed.

Financial support

This study was supported by a grant from the Instituto de Salud Carlos III of the Spanish Health Ministry (Grant ID FIS PI-021117).

Contributors

T. Rodriguez-Blanco, A. Vila-Córcoles, C. de Diego and O. Ochoa-Gondar designed the study, assessed outcomes, and wrote and edited the paper; F. Bobé, A. Morro, N. Herńndez, A. Martín, F. Calamote, L. Clotas, N Saún, E Valdivieso and MI. Herreros obtained the data; T. Rodríguez-Blanco did statistical analysis; A. Vila-Corcoles coordinated the study.