JCK was involved in all aspects of the study. TAS, JL, AJM, REGU, HJ, WWT, FM, LWS, and DGM provided input on study design and interpretation of results. TAS provided statistical expertise. CS, FM, and LWS assisted with data collection. DT assisted with data analyses. DGM provided overall supervision and guidance. JCK, TAS, JL, AJM, REGU, HJ, CS, WWT, DT, FM, LWS, and DGM critically reviewed and approved the manuscript for publication.
AJM reports receiving travel grants from Sanofi Pasteur and Solvay Pharmaceuticals for speaking at meetings, and payment from Sanofi Pasteur for chairing a safety committee for a clinical trial.
In 2000, Ontario, Canada, initiated a universal influenza immunization program (UIIP) to provide free influenza vaccines for the entire population aged 6 mo or older. Influenza immunization increased more rapidly in younger age groups in Ontario compared to other Canadian provinces, which all maintained targeted immunization programs. We evaluated the effect of Ontario's UIIP on influenza-associated mortality, hospitalizations, emergency department (ED) use, and visits to doctors' offices.
Mortality and hospitalization data from 1997 to 2004 for all ten Canadian provinces
were obtained from national datasets. Physician billing claims for visits to EDs and
doctors' offices were obtained from provincial administrative datasets for four
provinces with comprehensive data. Since outcomes coded as influenza are known to
underestimate the true burden of influenza, we studied more broadly defined conditions.
Hospitalizations, ED use, doctors' office visits for pneumonia and influenza, and
all-cause mortality from 1997 to 2004 were modelled using Poisson regression,
controlling for age, sex, province, influenza surveillance data, and temporal trends,
and used to estimate the expected baseline outcome rates in the absence of influenza
activity. The primary outcome was then defined as influenza-associated events, or the
difference between the observed events and the expected baseline events. Changes in
influenza-associated outcome rates before and after UIIP introduction in Ontario were
compared to the corresponding changes in other provinces. After UIIP introduction,
influenza-associated mortality decreased more in Ontario (relative rate
[RR] = 0.26) than in other provinces (RR = 0.43)
(ratio of RRs = 0.61,
Compared to targeted programs in other provinces, introduction of universal vaccination in Ontario in 2000 was associated with relative reductions in influenza-associated mortality and health care use. The results of this large-scale natural experiment suggest that universal vaccination may be an effective public health measure for reducing the annual burden of influenza.
Comparing influenza-related mortality and health care use between Ontario and other Canadian provinces, Jeffrey Kwong and colleagues find evidence that Ontario's universal vaccination program has reduced the burden of influenza.
Seasonal outbreaks (epidemics) of influenza—a viral disease of the nose, throat, and airways—affect millions of people and kill about 500,000 individuals every year. These epidemics occur because of “antigenic drift”: small but frequent changes in the viral proteins to which the human immune system responds mean that an immune response produced one year by exposure to an influenza virus provides only partial protection against influenza the next year. Immunization can boost this natural immunity and reduce a person's chances of catching influenza. That is, an injection of killed influenza viruses can be used to prime the immune system so that it responds quickly and efficiently when exposed to live virus. However, because of antigenic drift, for influenza immunization to be effective, it has to be repeated annually with a vaccine that contains the major circulating strains of the influenza virus.
Public-health organizations recommend targeted vaccination programs, so that elderly people, infants, and chronically ill individuals—the people most likely to die from pneumonia and other complications of influenza—receive annual influenza vaccination. Some experts argue, however, that universal vaccination might provide populations with better protection from influenza, both directly by increasing the number of vaccinated people and indirectly through “herd immunity,” which occurs when a high proportion of the population is immune to an infectious disease, so that even unvaccinated people are unlikely to become infected (because infected people rarely come into contact with susceptible people). In this study, the researchers compare the effects of the world's first free universal influenza immunization program (UIIP), which started in 2000 in the Canadian province of Ontario, on influenza-associated deaths and health care use with the effects of targeted vaccine programs on the same outcomes elsewhere in Canada.
Using national records, the researchers collected data on influenza vaccination, on all deaths, and on hospitalizations for pneumonia and influenza in all Canadian provinces between 1997 and 2004. They also collected data on emergency department and doctors' office visits for pneumonia and influenza for Ontario, Quebec, Alberta, and Manitoba. They then used a mathematical model to estimate the baseline rates for these outcomes in the absence of influenza activity, and from these calculated weekly rates for deaths and health care use specifically resulting from influenza. In 1996–1997, 18% of the population was vaccinated against influenza in Ontario whereas in the other provinces combined the vaccination rate was 13%. On average, since 2000—the year in which UIIP was introduced in Ontario—vaccination rates have risen to 38% and 24% in Ontario and the other provinces, respectively. Since the introduction of UIIP, the researchers report, influenza-associated deaths have decreased by 74% in Ontario but by only 57% in the other provinces combined. Influenza-associated use of health care facilities has also decreased more in Ontario than in the other provinces over the same period.
These findings are limited by some aspects of the study design. For example, they depend on the accuracy of the assumptions made when calculating events due specifically to influenza, and on the availability and accuracy of vaccination and clinical outcome data. In addition, it is possible that influenza-associated deaths and health care use may have decreased more in Ontario than in the other Canadian provinces because of some unrecognized health care changes specific to Ontario but unrelated to the introduction of universal influenza vaccination. Nevertheless, these findings indicate that, compared to the targeted vaccination programs in the other Canadian provinces, the Ontarian UIIP is associated with reductions in influenza-associated deaths and health care use, particularly in people younger than 65 years old. This effect is seen at a level of vaccination unlikely to produce herd immunity so might be more marked if the uptake of vaccination could be further increased. Thus, although it is possible that Canada is a special case, these findings suggest that universal influenza vaccination might be an effective way to reduce the global burden of influenza.
Please access these Web sites via the online version of this summary at
Read the related
A related
The Ontario Ministry of Health provides information on
The World Health Organization provides information on
The US Centers for Disease Control and Prevention provide information for patients
and professionals on all aspects of
MedlinePlus provides a
The UK National Health Service provides information about
Annual epidemics of influenza continue to cause worldwide morbidity, mortality, and
societal disruption. When there is good match to circulating strains, influenza vaccines are
generally efficacious and cost-effective for most age groups [
In October 2000, the Canadian province of Ontario initiated the world's first large-scale
universal influenza immunization program (UIIP) to provide free influenza vaccinations for
the entire population ≥6 mo of age [
Ontario was the lone Canadian province to implement UIIP; other provinces maintained
targeted programs. Canadian vaccination programs traditionally involve centralized vaccine
procurement by provincial governments with publicly insured local delivery to high risk
individuals and their close contacts/care providers. Vaccine is delivered in health care
settings by nurses or physicians, or in community settings through public health
departments. High risk individuals include seniors aged ≥65 y, individuals with
chronic medical conditions, children aged 6–23 mo (since 2004), and pregnant women
(since 2007) [
Our earlier work found that universal vaccination was associated with an overall
incremental increase in vaccine uptake in the household population aged ≥12 y of 9
percentage points (24- versus 15-percentage-point increase in Ontario versus other
provinces) between 1996–1997 and 2005 [
In this study, we evaluated the effect of Ontario's UIIP on mortality, hospitalizations, emergency department (ED) use, and doctors' office visits compared to targeted programs in other Canadian provinces.
This study included the population of the ten Canadian provinces from 1997 to 2004 who were eligible for universal, publicly insured health care services. Ethics approval was obtained from the Sunnybrook Health Sciences Centre Research Ethics Board, Toronto, Ontario.
We used a pre-/post-intervention study design with concurrent controls to assess the
effect of Ontario's UIIP on influenza-associated mortality and health care use. Although
ecological studies do not permit inference to individuals [
Mortality data were obtained from Statistics Canada's Mortality Database, a national vital statistics dataset.
Hospitalization data were obtained from Statistics Canada's Hospital Morbidity Database, a national discharge abstract dataset.
Physician services data were obtained from the following provinces: Ontario, Quebec, Alberta, and Manitoba. These provinces account for approximately 76% of the Canadian population and were selected because of availability of comprehensive data on physician services.
Annual population estimates were obtained from Statistics Canada.
Vaccination rate data were obtained from the 1996–1997 cycle of the National
Population Health Survey (NPHS) and the 2000–2001, 2003, and 2005 cycles of the
Canadian Community Health Survey (CCHS). Conducted by Statistics Canada using telephone
and in-person interviews, these surveys covered the household population aged ≥12 y
but excluded members of the Canadian Forces, Indian reserves, and some remote areas, and
those living in institutions. The surveys had response rates of between 79% and
85%; details have been described previously [
Influenza outcomes are difficult to quantify because influenza infections are typically
not confirmed. Since outcomes coded as influenza are known to underestimate the true
burden of influenza, we studied more broadly defined conditions. These mortality and
health care utilization outcomes vary according to a cyclical
“baseline” function with winter peaks and summer troughs, but
typically exhibit spikes during periods of influenza virus activity [
For our primary mortality outcome, we included all deaths between 24 August 1997 and 14 August 2004 (primary study period comprising seven 52-wk periods).
We included all pneumonia and influenza (P&I) (ICD-9-CM 480–487,
ICD-10-CA J10-J18) hospitalizations to acute care facilities during the study period,
excluding admissions of nonresidents, transfers between institutions, and readmissions
within 1 wk of discharge. Although discharge abstracts list up to 16 diagnoses, we used
the first five codes to capture cases where influenza may have precipitated another
condition requiring hospitalization (e.g., congestive heart failure) as well as nosocomial
spread of influenza, while limiting the potential effects of differential coding practices
between provinces and over time [
To assess ED use and visits to doctors' offices, we included all physician claims in ED and office settings during the study period for P&I. We included only one service claim per patient per physician per day, and where possible, we excluded claims that were associated with vaccination service codes (i.e., a visit to receive influenza vaccination).
A national network of hospital and provincial laboratories submit weekly reports of numbers of tests performed (using viral culture or direct antigen detection) and numbers of positive tests for influenza A, influenza B, and respiratory syncytial virus (RSV) to the Public Health Agency of Canada. The four Atlantic provinces were grouped owing to small numbers of specimens tested.
A subset of laboratories report further case-specific data including subtype. Since
A(H3N2)-predominant seasons have historically been more severe than A(H1N1)- and/or
B-predominant seasons [
A subset of viral isolates is sent to Canada's National Microbiology Laboratory for
strain characterization. The sample was assumed to represent the distribution of strains
for all influenza viruses circulating in the population. The percentage of circulating
strains that did not match vaccine strains was calculated for each province and season
(percent mismatch) (
Periods of peak influenza activity were defined separately for each province and year,
starting when the weekly percentage of tests positive for influenza was greater than
10% and ending when the percentage fell below that threshold for 2 consecutive
wk (mean period duration = 11 wk) (
We estimated influenza-associated events using a two-step procedure. We first ran multivariate regression models to estimate weekly events as a function of population and influenza season factors, and used this model to generate an expected baseline representing the pattern of events occurring in the absence of influenza. We then computed influenza-associated events as the difference between the number of observed events and expected baseline events during periods of influenza activity.
In the first step, we ran separate Poisson regression models for each outcome, according
to province and age group (<5, 5–19, 20–49 [<50
for mortality], 50–64, 65–74, 75–84, 85+
y) [
Weekly influenza-associated events were subsequently computed as the difference between
the number of observed events and expected baseline events during periods of peak
influenza activity, where expected counts were based on the adjusted Poisson models (
Pre- and post-UIIP influenza-associated event rates were compared by dividing the
adjusted postintervention rates by the preintervention rates to produce relative rates
(RR) of UIIP effect separately for Ontario and the other provinces combined. The standard
error of
Model fit was evaluated by examining the standardized Pearson residuals for outlying
points and secular trends. We also evaluated the presence of influential provinces by
removing them individually from the model and reestimating the UIIP effect. Although the
model did not optimally fit the extreme, short-lived spikes that occurred during the peaks
of the influenza season, the fit during the remainder of the season was reasonable. We
used the Durbin-Watson
To confirm the robustness of the findings, several sensitivity analyses were conducted.
To test the consistency of our findings, we used a 5% threshold for defining periods of peak influenza activity, leading to longer periods. We also lagged influenza and RSV activity forward or backward by 1 wk, in case of delays in viral testing or onset of serious illnesses (i.e., reports of positive tests and illness onset consistently occurring in different weeks).
To increase the number of influenza seasons in the study, we included data from the 1993–1994 to 1996–1997 influenza seasons, but this analysis excluded data from Quebec and the Atlantic provinces because prior to 1997–1998, the former did not report the number of tests performed, preventing calculation of weekly percentages of tests positive, and the latter performed too few tests (<25/wk during influenza season periods) to permit reliable calculation of percentages of tests positive.
As another test of consistency, we restricted the analysis to seasons in which A(H3N2) detections accounted for greater than 50% of the total isolates (A[H3N2]-predominant seasons).
For hospitalizations, we repeated the primary analysis using all 16 codes (any-listed) and using only the first code (primary-listed).
For mortality, we examined deaths from respiratory and circulatory (R&C)
conditions as a more specific outcome for influenza. Deaths from P&I were not
assessed because of the small numbers of events at the province- and age-specific level,
and because of a sharp decrease in P&I deaths coinciding with ICD-10
introduction in 2000 that is known to be a coding artifact [
The 2003–2004 influenza season was noteworthy for the emergence of the novel
A/Fujian strain that accounted for 92% of the characterized isolates in
Canada but was not included in that season's vaccine [
For the health care use outcomes, we examined the effect of UIIP on less specific
outcomes. We examined hospitalizations for all respiratory conditions (ICD-9-CM
460–519, ICD-10-CA J00-J99), and for ED use and office visits, we examined
selected respiratory conditions (composite of P&I, acute respiratory diseases
[ARD] [ICD-9 460–466], otitis media
[OM] [ICD-9 381–383] among those
≤50 y, and chronic obstructive pulmonary disease [COPD]
[ICD-9 490–492, 496] among those aged ≥20 y)
[
Data for the 2004–2005 season were available for hospitalizations, so a
sensitivity analysis adding that year to the study period was performed. For that
season, 35% of characterized isolates were A/California and 4%
were B/Hong Kong strains, both of which were not included in that season's vaccine,
although the A/Fujian component of the vaccine was believed to provide partial
protection against A/California [
To assess the specificity of our findings, we examined rates of all observed events, as opposed to influenza-associated events, during the month of July, since influenza viruses are not circulating in the summer and we would not expect any influenza-associated events nor any benefit from influenza vaccination.
As another test of specificity, we examined the effect of the UIIP on selected control conditions, by computing RRs for all events during the month of February, when influenza viruses are almost always circulating. For mortality, we used all non-R&C conditions, for hospitalizations, we examined hernia, and for ED use and office visits, we considered lacerations.
Regions with greater increases in vaccination rates are expected to have greater decreases in influenza-associated outcomes. To assess the presence of such a relationship, we plotted age group-, province-, and outcome-specific RRs for mean influenza-associated events against absolute changes in vaccination rates. We used the mean of the estimates from the three cycles of the Canadian Community Health Survey as the measure of post-UIIP vaccine uptake and the 1996–1997 National Population Health Survey estimate as the measure of pre-UIIP vaccine uptake. To match with the age groupings of the outcomes data, we assumed that the vaccination rates of those aged 12 to 19 y applied to children as young as 5 y. We fitted weighted linear models separately by outcome and age group (<65 y and ≥65 y), weighting by the inverse of the variance of the RRs.
Over the study period, mean annual outcome rates were highest in older people and young
children (
Study Population Demographics and Mean Annual Event Rates over Study Period
Event rates (from top to bottom) for doctors' office visits, ED use, hospitalizations, and mortality (grey lines are for Ontario and black lines are for other provinces combined) are expressed as rates per 100,000 on the upper sections of the vertical axis. Viral surveillance data (grey shaded areas) are expressed as the weekly percentage of tests positive on the lower section of the vertical axis. Vaccination rates for the household population aged ≥12 y (grey vertical bars are for Ontario and black vertical bars are for other provinces combined) are expressed as the percentage of the population vaccinated on the lower section of the vertical axis. The horizontal axis represents time. The black vertical line represents UIIP introduction.
Between the pre-UIIP 1996–1997 estimate to the mean post-UIIP vaccination rate,
influenza vaccination rates for the household population aged ≥12 y increased 20
percentage points (18%–38%) for Ontario, compared to 11
percentage points (13%–24%) for other provinces
(
Influenza Vaccination Rates over Time for Ontario and Other Provinces
After UIIP introduction, influenza-associated mortality for the overall population
decreased 74% in Ontario (RR = 0.26, 95% confidence
interval [CI], 0.20–0.34) compared to 57% in other
provinces (RR = 0.43, 95% CI, 0.37–0.50) (ratio of RRs
= 0.61,
Effect of UIIP on Influenza-Associated Mortality and Health Care Use Rates
Overall, influenza-associated health care use decreased more in Ontario than other
provinces for hospitalizations (RR = 0.25 versus 0.44, ratio of RRs
= 0.58,
There were no patterns detected in the standardized residual plots against time, and no
evidence of influential provinces (
The sensitivity analyses generally produced similar RR ratios as the primary analysis for
tests of consistency, smaller RR ratios for tests of amplification, larger RR ratios for
tests of attenuation, and RR ratios approximately equal to 1 for tests of specificity, as
expected (
Sensitivity Analyses
We found a dose-response relationship where greater increases in vaccine uptake were
associated with greater decreases in influenza-associated outcomes for all health care use
outcomes for age groups <65 y (all slopes negative,
The vertical axis represents the age group-, province-, and outcome-specific post-
versus pre-2000 RRs. The horizontal axis represents the absolute post- versus pre-2000
change in influenza vaccination rates (%). The bubble sizes represent the
inverse of the variances of the post- versus pre-2000 RRs. The
After introduction of the world's first large-scale UIIP, we found that influenza-associated mortality, hospitalizations, ED use, and office visits decreased more overall in Ontario compared with other provinces. Age-specific analyses revealed greater temporal drops in health care use in Ontario compared to other provinces for younger age groups (i.e., ratios of pre-/post-RRs for Ontario versus other provinces being closer to zero), with the gap between the RRs narrowing with increasing age (i.e., ratios of RRs being closer to one). This is consistent with the age-specific pattern for temporal changes in vaccine uptake, with greater increases over time in Ontario compared to other provinces for younger age groups leading to greater expectations of benefits from the UIIP. However, for older age groups, in particular those ≥75 y, greater incremental increases in vaccine uptake over time were observed in other provinces compared to Ontario, yet among all the outcomes, none of the RR ratios were greater than one. This result is contrary to what one would expect based on the assumption that greater increases in vaccine uptake in those populations should have been associated with greater decreases in outcome rates, and therefore suggests that influenza vaccines may not be as effective in reducing mortality and health care use in older people compared to younger age groups.
Sensitivity analyses demonstrated consistency, specificity, and the presence of a
dose-response relationship between temporal changes in vaccination rates and outcomes among
those <65 y. A limited number of inconsistent results warrant discussion. While the
analysis restricted to A(H3N2)-predominant seasons was expected to be a test of consistency,
the results for the three health care use outcomes suggested effect attenuation. This was
likely because that analysis over-weighted the influence of the 2003–2004 poor
match season by excluding the 2000–2001 and 2002–2003
non-A(H3N2)-predominant seasons. Other unexpected results were the lack of attenuation when
examining the less specific secondary outcomes for ED use and office visits. This may have
been because the selected conditions are nearly as specific to influenza as the primary
outcome. Another inconsistency was the lack of amplification for the supposedly more
specific secondary outcome for mortality, respiratory, and circulatory conditions. This lack
of amplification may be because R&C mortality accounts for the bulk of the mortality
from all causes [
A previous study reported that UIIP introduction did not lead to reduced influenza
incidence [
Although many studies have found that for years with good vaccine match, influenza vaccines
are effective in preventing hospitalizations and mortality in the elderly [
Despite universal availability of publicly insured health care services, enhanced access to
free influenza vaccines in Ontario since 2000, and extensive media communications, only an
estimated average of 38% of the overall household population reported receiving
them during the post-UIIP introduction period. Although vaccine uptake was higher in older
age groups (e.g., approximately 80% among those ≥75 y), it is uncertain
that levels required for appreciable herd immunity effects were obtained in the overall
population, particularly in younger age groups. Several studies have claimed the existence
of indirect benefits arising from vaccinating children, including a report on the impact of
vaccinating a community's children on illness rates in the adult population during the 1968
pandemic [
Strengths of the study include the unique opportunity for a “natural
experiment” and the fact that size of the populations of Ontario and the other
provinces combined are comparable. We addressed most of the criteria outlined recently in a
framework for addressing residual bias by Simonsen et al., including: the selection of more
specific outcomes for the primary study outcomes (with the exception of all-cause mortality)
and the use of consistent modeling techniques and laboratory surveillance data to estimate
influenza-associated events to reduce the potential dilution of benefits from universal
vaccination that might arise from employing all influenza season-associated events as the
outcome measure (end-point specificity); the sensitivity analyses with poor match seasons
excluded (vaccine match); and the inclusion of events only during periods of peak influenza
activity and the sensitivity analysis using a summer period (seasonality) [
This study has a number of limitations. Individual-level vaccination and outcome data were
not available, necessitating an ecological study. However, use of this design is appropriate
for assessing the public health impact of a population-wide intervention. As with other
influenza studies using health databases, the selected outcomes are nonspecific and may be
due to causes other than influenza, but the strategies described above partly address this
issue. The quality and reliability of the outcome data over time, for multiple
jurisdictions, and across different classification systems (i.e., ICD-9 versus ICD-10)
remain uncertain. The validity of statistical models to estimate influenza-associated events
is limited by uncertainty of their accuracy, in spite of our best efforts to achieve optimal
model fit. Laboratory viral surveillance data are potentially susceptible to ascertainment
and reporting biases; the weekly proportion of tests positive is felt to be the most robust
measure of viral activity. Another drawback of the study is that no vaccination rate data
are available for those <12 y of age, an age group that experiences particularly high
rates of less severe influenza-associated outcomes, or for institutionalized seniors, a
group that experiences higher rates of more severe outcomes. Relatively high residual
autocorrelation was likely due to the difficulty in modeling a baseline influenza rate that
did not display purely cyclical behaviour despite inclusion of higher order Fourier series
terms; however, it suggests the possibility of residual confounding due to season-specific
factors such as temperature or relative humidity. We were also not able to include other
potential confounders such as strain-specific influenza surveillance data, prevalence of
individual comorbidities, socioeconomic status, smoking rates, polysaccharide or conjugated
pneumococcal vaccination, antiviral medication use, and provincial health care system
capacity, but we have no reason to believe that these factors changed more over time in
Ontario compared to other provinces. For example, pneumococcal vaccination of children has
been shown to reduce rates of pneumonia-related admissions in older age groups
[
Despite these limitations, this study provides suggestive evidence of the population-based effectiveness of universal vaccination programs using inactivated influenza vaccines. It is not possible to definitively declare superiority of universal programs over targeted programs, as the findings from this study may not generalize to other settings. But by reducing financial barriers and increasing awareness and accessibility, universal vaccination may be an effective strategy for increasing a population's protection against influenza. Future studies to develop more immunogenic influenza vaccines, to test novel strategies for further increasing vaccine uptake, and to examine the cost-effectiveness of universal influenza vaccination may be valuable.
The left vertical axis represents hospitalization rate per 100,000. The horizontal axis represents time. The solid black line represents observed hospitalizations, the dashed black line represents baseline hospitalizations in the hypothetical absence of influenza, and the black vertical bars represent influenza-associated hospitalizations (observed hospitalizations minus baseline hospitalizations). Viral surveillance data (grey shaded areas) are expressed as the weekly percentage of tests positive on the right vertical axis. The grey dashed vertical lines denote periods of peak influenza activity.
(539 KB PPT)
(97 KB DOC)
(74 KB DOC)
(43 KB DOC)
We thank the sentinel laboratories participating in the Respiratory Virus Detection Surveillance System and the FluWatch team at the Public Health Agency of Canada for providing the viral surveillance data, and Charles Burchill and Wendy Au for assistance with data preparation and analyses of the Manitoba data. JCK had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
confidence interval
emergency department
pneumonia and influenza
respiratory and circulatory
relative rate
universal influenza immunization program