Age-related trends in the timeliness and prediction of medical visits, hospitalizations and deaths due to pneumonia and influenza, British Columbia, Canada, 1998–2004

Summary

The influenza immunization program in North America has been primarily designed to provide direct benefit to vaccinated individuals at highest risk of serious influenza outcomes. Some evidence suggests that immunization of certain age groups may also extend indirect protective benefit to vulnerable populations. Our goal was to identify age groups associated earliest with seasonal influenza activity and who may have the greatest indirect impact at the population level. We examined age-based associations between influenza medical visits and population-wide hospitalization/mortality due to pneumonia & influenza (P&I) using administrative datasets in British Columbia, Canada. A peak week was identified for each age group based on the highest rates observed in a given week for that study year. Mean rates at the peak week were averaged over the study years per age group. Timeliness (T) was defined as the mean difference in days between the first peak in influenza medical visits and population-wide P&I hospitalizations/deaths. Poisson regression was applied to calculate prediction (Pr) as the average proportion of deviance in P&I explained by influenza medical visits. T and Pr were derived by age group, and the product (T × Pr) was used as a summary measure to rank potential indirect effects of influenza by age group. Young children (0–23 months) and the elderly (≥65 years) had the highest peak rates of P&I hospitalization. Children <6 m and the elderly had the highest peak rates of P&I mortality. We found no significant differences by age for influenza medical visits in predicting population-wide P&I hospitalizations or deaths. School-aged children (5–19 years) showed the best relative combination of T × Pr, followed by preschool-aged children (2–4 years). We conclude that the very young and old suffer the greatest morbidity due to P&I, and an indirect role for school-aged children in anticipating the risk to others warrants further evaluation.

Introduction

Influenza is a significant cause of morbidity and mortality each year [1–5]. While children suffer amongst the highest influenza attack rates (30–50%), with high rates of associated health care use, the elderly suffer the greatest mortality, comprising nearly 90% of influenza-related deaths each year [6–8]. Annual vaccination is the primary method recommended to reduce the impact of influenza. The National Advisory Committee on Immunization (NACI) in Canada and the Advisory Committee on Immunization Practice (ACIP) in the United States have long recommended influenza vaccination for those at high risk of serious complications from infection, including the elderly and those with chronic conditions. Since the 2004–2005 season, children aged 6–23 months have also been recommended recipients of the publicly funded influenza vaccine in North America because of their high rates of influenza-related hospitalization [9,10]. In 2006, ACIP further expanded the recommendation for the publicly funded influenza immunization program to include children aged 2–5 years [11]. Their increased risk for influenza-associated clinic, emergency department, and hospital visits, was cited as a primary rationale for this inclusion.

The publicly funded influenza immunization program in North America has historically been risk-based and intended to provide direct benefit to vaccinated individuals. Some evidence suggests that immunization of other age groups may also extend indirect protective benefit to these vulnerable populations by reducing influenza circulation and its overall community burden [12–16]. Children are thought to be key players in the pathway of influenza transmission, especially within households [12,17,18]. Experience from Japan [16] and the United States [14,17] suggests that influenza immunization programs for school-aged children may significantly reduce morbidity and mortality due to influenza in the wider community. In a recent study, medical visits for respiratory illness among preschool-aged children (3–4 years) were the first and strongest predictors of population-wide peaks in mortality due to pneumonia and influenza (P&I) [19]. Vaccination of preschool-aged children has been shown to reduce transmission among household contacts [20] although effects beyond that into the community have not yet been established epidemiologically. Another study has emphasized the role of working-aged adults in spreading infection through commuting to and from work [21].

In this study, we use administrative data from British Columbia (BC), Canada, to explore the potential for indirect benefit of expanded influenza immunization policies to protect those at high risk of serious complications from influenza. Specifically, we examine age-based differences in health care use and adverse outcomes due to P&I to identify age groups that present earliest and predict the greatest proportion of population-wide P&I hospitalizations and mortality.

Methods

Seasonal profiles and inclusion criteria

Provincial influenza surveillance in BC is managed by the British Columbia Centre for Disease Control (BCCDC) with input to the Public Health Agency of Canada’s FluWatch Program (www.phac-aspc.gc.ca/fluwatch/index.html). Annual online reports of influenza activity in Canada are published in the Canada Communicable Disease Report (CCDR). Provincial and national influenza surveillance reports were reviewed to characterize each influenza season in BC from 1998–1999 to 2003–2004. Based on the reported percentages of influenza type, subtype and characterization data, predominant strains were identified in each season.

Distributions of influenza activity from two data sources were examined for each season to assess fit for the timeliness and prediction analysis. The first indicator was influenza medical visits rates; the second was the percent of all visits due to influenza-like illness (ILI) from the sentinel physician network participating in the provincial influenza surveillance system. Seasons were included in the analysis if clear seasonal peaks were present and if the same influenza types or subtypes were observed across all age groups, subsequently allowing for the analysis of age-dependent associations. Other studies have also conducted similar restricted analysis on A/H3N2 seasons alone [22] or stratified analyses based on influenza subtype [5,21,23].

Defining influenza medical visits, P&I hospitalizations and P&I deaths

Influenza medical visit, P&I hospitalization, and P&I mortality data in BC from 1 September 1998 to 31 August 2004 were obtained from three separate administrative databases managed by the BC Ministry of Health using the International Classification of Diseases (ICD) coding system. Aggregate data were divided by week (with the first Sunday of the year defined as day 1 of week 1) and the following age groups: <6 months, 6–23 months, 2–4 years, 5–9 years, 10–19 years, 20–49 years, 50–64 years, and ≥65 years. Medical visit data in BC are captured by the Medical Service Plan billing system. Influenza medical visits are monitored through ICD-9 codes for influenza (487) for the general practice services of regional examinations, consultations, complete examinations, home visits and emergency visits. Service counts may be counted more than once if a patient receives a service on more than one occasion. The data exclude lab services and institutional visits. A P&I hospitalization was defined as one for which an ICD-9 code for influenza (487) or pneumonia (480–486) or ICD-10 code for influenza (J108, J100, J101, J110, J111, J118) or pneumonia (J13-4, J120-2, J128-9, J150-60, J168, J180-1, J188-9, A403, A491, G001) was listed as the primary diagnosis on the hospital discharge record. Week of diagnosis was based on the hospital admission date. A P&I death was defined as one in which an ICD-10 code for influenza or pneumonia was listed as the direct cause of death. Only primary diagnoses and direct causes of death were included to reduce the likelihood that outcomes were due to other causes, including underlying medical conditions.

Age-specific rates for influenza medical visits, P&I hospitalizations, and P&I deaths were calculated by week. Age-specific crude and adjusted percentages of all influenza medical visits were also calculated by week. Percentages were adjusted for population size in each age group as well as possible age-related differences in health-seeking behaviour using the influenza medical visit rate for 20–49 year olds as the reference for standardization; this is calculated as

$$\begin{array}{l}\left[\frac{\text{number of influenza visits in age group}}{\text{number of all visits in age group}}\right]\hfill \\ \times \left[\frac{\text{age-specific medical visit rate for all causes}}{\text{medical visit rate for all causes for}20–49\text{years}}\right]\hfill \end{array}$$

A peak week was identified for each age group based on the highest rates observed in a given week for that study year. Mean rates at the peak week were averaged over the study years per age group.

Analysis of timeliness and prediction

Data were aggregated by weeks in each study year (week 34 to week 33 of the following year) and by age group. Timeliness (lead time) of influenza medical visits was derived in order to give an indication of the ordering of onset of influenza-like illness by age group. The lead time between influenza medical visits in each age group and population-wide P&I hospitalizations or mortality was calculated as the difference, converted to days, between the first peaks of each distribution for each study year. The mean lead time for each age group over the four included study years was calculated.

Prediction was calculated as a measure of the association between influenza medical visits in each age group and population-wide P&I hospitalizations and mortality. The predictive value of influenza medical visits to P&I hospitalization or mortality was assessed by fitting a generalized linear model. Assuming a Poisson distribution for P&I hospitalization and mortality counts, Poisson regression models for each age group and study year were performed. The predictor was the influenza medical visits from the lead/lag week(s) suggested by the timeliness analysis. The overall model fit for each Poisson regression model was determined by the significance of its deviance statistics. The predictive value for P&I hospitalization or mortality was determined by calculating the proportion of deviance that was explained, similar to the R². The average of the predictive values for each age combination from the four study years was calculated.

The product of timeliness and prediction (T × Pr) was calculated in order to rank age groups by their combined indirect effect on overall population outcomes. This measure assumes equal weight for timeliness and prediction and is useful for relative comparison only; as such, individual T × Pr values have no literal meaning.

Results

Characterizing influenza seasons

In BC, and across Canada, four of the six influenza seasons between 1998 and 2004 were predominantly due to influenza A, during which an influenza A (H3N2) virus accounted for 94–100% of influenza A viruses characterized (Table 1). Based on influenza surveillance in BC, including sentinel physician data, the 2001–2002 season was characterized by prolonged low-level activity, as indicated by a relatively flat distribution in sentinel influenza medical visits rather than a clear peak in activity as seen in the other seasons (Fig. 1a). This is reflected to a lesser extent in the population-wide medical visits data (Fig. 1b).

Table 1.

Influenza seasons in BC by predominant influenza type, subtypes and strains, 1998–1999 to 2003–2004

Season	Influenza types		Predominant subtype(s)	Predominant strain(s)	%a
	%A	%B
1998–1999	92	8	A/H3N2	A/Sydney/5/97-like	100
1999–2000	98	2	A/H3N2	A/Sydney/5/97-like	94
2000–2001	33	67	B	B/Yamanashi/166/98-like	100
			A/H1N1	A/New Caledonia/20/99-like	100
2001–2002	90	10	A/H3N2	A/Panama/2007/99-like	96
2002–2003	46	54	B	B/Hong Kong/330/01-like	100b
			A (H1N2)	A (H1N2)	67 b,c
2003–2004	100	<1	A/H3N2	A/Fujian/411/02-liked	95

^aCalculated as percent of each influenza type (A or B) characterized as predominant strain.

^bNational characterization data, provincial data not available.

^cOther circulating strains include A/New Caledonia/20/99-like, 20% and A/Panama/2007/99-like, 13%.

^dVaccine mismatched to strain.

Figure 1.

(a) Seasonal distribution of percent of all visits due to influenza-like illness (ILI) among sentinel physicians, 1998–1999 to 2003–2004 and (b) seasonal distribution of influenza medical visits in British Columbia, 1998–1999 to 2003–2004.

The 2000–2001 and 2002–2003 seasons were characterized by a mix of influenza A and B activity, with different strains predominating. The predominant influenza B virus in 2000–2001 belonged to the B/Yamagata lineage; viruses of this lineage had been circulating worldwide since 1990. However, in 2002–2003, B/Hong Kong/330/01-like virus circulated; this virus belongs to the B/Victoria lineage, which had not circulated for more than 10 years in North America. B/Hong Kong/330/01-like virus drove a substantial peak late in the season, primarily in children who had less prior exposure to this lineage. Based on surveillance data from 2002 to 2003, there were clear demarcations in influenza activity by influenza type, timing and age group. Influenza A had substantial impact in the elderly early in the season while influenza B was detected primarily in children later in the season.

Methods could not be applied in the absence of clear seasonal peaks (2001–2002) or when distinctive age-stratified peaks occurred based on different influenza A versus B types (2002–2003). For these reasons, the 2001–2002 and 2002–2003 seasons were excluded from the timeliness and prediction analysis.

Direct effects: peak rates of medical visits, hospitalizations and deaths by age group

Older school-aged children (10–19 years) had the lowest peak rates of influenza medical visits compared to other age groups, especially younger children (Table 2). Children <6 m, 6–23 m, 2–4 years and the elderly (≥65 years) had the highest peak rates of P&I hospitalization compared to other age groups; the peak rate among children 2–4 years of age, however, was less than half that of their younger counterparts. Children <6m and the elderly had the highest peak rates of P&I mortality. Compared to all other age groups, older school-aged children (10–19 years) had the lowest peak rates of P&I hospitalization and death (Table 2).

Table 2.

Age profile of influenza medical visits, and P&I hospitalizations and mortality for 1998–1999, 1999–2000, 2000–2001, and 2003–2004, expressed as one week peak values averaged over the four study years [95% confidence interval]

Age group	Influenza medical visits			Adverse outcomes due to P&I
	% of all visits (unadjusted)	% of all visits (adjusted a)	Rate per 1000 people	Hospitalization rate per 1000 people	Mortality rate per 1000 people
<6 months	0.6 [0.3, 0.8]	1.8 [1.2, 2.4]	1.4 [0.8, 1.9]	0.57 [0.25, 0.88]	0.048 [0.046, 0.051]
6–23 months	1.4 [0.9, 1.9]	3.6 [2.3, 4.8]	2.1 [1.6, 2.6]	0.49 [0.32, 0.65]	0.012 [0.000, 0.025]
2–4 years	2.0 [1.5, 2.4]	2.8 [2.2, 3.3]	1.8 [1.3, 2.4]	0.20 [0.19, 0.22]	0.004 [0.000, 0.011]
5–9 years	2.1 [2.0, 2.3]	2.0 [0.9, 3.1]	1.3 [0.9, 1.7]	0.06 [0.03, 0.08]	0.002 [0.000, 0.006]
10–19 years	2.0 [1.7, 2.2]	0.4 [0.3, 0.5]	0.3 [0.2, 0.3]	0.006 [0.004, 0.008]	0.00024 [0.000, 0.0007]
20–49 years	1.5 [0.6, 2.3]	1.5 [0.6, 2.3]	1.1 [0.5, 1.7]	0.03 [0.02, 0.05]	0.001 [0.001, 0.001]
50–64 years	1.1 [0.2, 2.0]	1.3 [0.3, 2.4]	1.1 [0.2, 1.9]	0.09 [0.03, 0.14]	0.005 [0.004, 0.007]
≥65 years	0.8 [0.1, 1.4]	1.2 [0.4, 2.1]	0.9 [0.3, 1.6]	0.48 [0.19, 0.78]	0.097 [0.041, 0.152]

^aAdjusted for population size of age category and health-seeking behavior.

Indirect effects: age-related trends in the timeliness and prediction of influenza medical visits in determining P&I hospitalization and mortality

We found no significant differences by a age for influenza medical visits in predicting population-wide P&I hospitalizations or deaths (Fig. 2 and Table 3). School-aged children showed the best relative combination of timeliness and prediction (T × Pr) for P&I hospitalizations, followed closely by preschool-aged children (2–4 years). Infants and toddlers (6–23 months) as well as younger (20–49 years) and older (50–64 years) working-aged adults showed less of a combined contribution in terms of T × Pr, although they contributed comparably to preschool-aged children in terms of hospitalization prediction alone (70–75%). School-aged children (5–19 years) also showed the best combination of T × Pr in anticipating population-wide P&I mortality. All other age groups demonstrated no lead time advantage of medical visits in anticipating P&I mortality (Table 3). Elderly persons and infants less than 6 months of age who themselves experience the greatest direct impact in terms of serious influenza morbidity (hospitalization and death) showed the lowest T × Pr combination for medical visits in anticipating overall P&I hospitalizations and mortality.

Figure 2.

Prediction and timeliness of influenza medical visits by age group in signaling overall P&I hospitalizations, 1998–1999, 1999–2000, 2000–2001, and 2003–2004.

Table 3.

Predictive ability of influenza medical visits by age group in signaling P&I hospitalizations and mortality for 1998–1999, 1999–2000, 2000–2001, and 2003–2004

Age group	P&I hospitalization			P&I mortality
	Timeliness a,b (mean)	Prediction a, c (mean [95% CI])	T × Pr	Timelinessa,b (mean)	Predictiona,c (mean [95% CI])	T × Pr
<6 months	0.0	0.57 [0.38, 0.75]	0.0	−3.5	0.31 [0.09, 0.52]	−1.1
6–23 months	1.8	0.74 [0.63, 0.86]	1.3	−1.8	0.45 [0.22, 0.68]	−0.8
2–4 years	3.5	0.73 [0.55, 0.91]	2.6	0.0	0.45 [0.14, 0.77]	0.0
5–9 years	5.3	0.66 [0.53, 0.79]	3.5	1.8	0.41 [0.22, 0.60]	0.7
10–19 years	5.3	0.65 [0.53, 0.77]	3.4	1.8	0.43 [0.25, 0.62]	0.8
20–49 years	1.8	0.73 [0.59, 0.87]	1.3	−1.8	0.47 [0.12, 0.83]	−0.8
50–64 years	1.8	0.71 [0.57, 0.86]	1.3	−1.8	0.51 [0.14, 0.88]	−0.9
≥65 years	−1.8	0.64 [0.45, 0.82]	−1.2	−5.3	0.43 [0.40, 0.82]	−2.3

^aInfluenza medical visits by age group in relation to P&I hospitalization or P&I mortality.

^bLead time in days.

^cProportion of deviance explained.

Discussion

Using readily accessible health care and vital statistics databases, we were able to compare the peak P&I health care burden by age and explore potential indirect impacts at the population level. As in other studies, our results underscore that the very young and the very old suffer the greatest burden due to P&I hospitalization and death. We found limited evidence based on influenza medical visits for intermediate age groups in anticipating that risk.

By comparing peak rates by age group, we found expected patterns: young children had the highest rates of influenza medical visits; young children and the elderly had the highest rates of P&I hospitalizations and the very young and elderly had the highest rates of P&I mortality [18,24–27]. Young school-aged children and working-aged adults showed an overall intermediate burden at peak. Older school-aged children showed the lowest peak rates. It should be noted that peak rates may be useful for comparison, but they do not reflect total influenza-attributable disease burden over the course of an entire season. The contribution of non-influenza causes to the peak is also expected to vary by age. Hospitalizations and deaths included in our analysis were only those for which pneumonia or influenza was listed as the primary reason for admission and death, respectively. Since influenza can also exacerbate underlying conditions [24,28], our rates are expected to underestimate the true level of P&I-related morbidity and mortality in the population, especially among older adults. In comparing the direct burden of illness this way, it should also be noted that, as in other studies, we reported rates based on case counts without attempting to further quantify severity of illness. So, for instance, infants and toddlers may be hospitalized at the same rate as elderly persons, but the number of days of hospitalization may be longer among the very old and frail. A review of days per influenza-related hospitalization and comparison between the old and young may be warranted. Similarly, we did not calculate potential years of life lost in comparing deaths in the elderly to deaths in younger age groups. Nevertheless, consistency with other studies in terms of relative impact by age based on these direct measures of disease burden offers some reassurance in terms of the overall validity of our findings [29].

Transmission patterns and indirect effects cannot be proven through this ecological study using administrative data; we found some variation but no significant differences by age. Relative to other age groups, and based on influenza medical visit data, school-aged children showed the best combination of timeliness and prediction related to P&I burden in the community. Medical visits alone, however, may not be the best indicator of influenza circulation or indirect effects; this is particularly emphasized by the negative lead times we identified in some instances. The negative values we report for timeliness reflect an absence of lead time between medical visits in one group and serious outcomes in the community overall. In particular, those groups themselves experiencing the highest rates of hospitalization or death (namely infants and the elderly) provide less advance indication of this through physician visit data. This may reflect more severe illness and patterns of health care utilization that bypass physician visit and lead directly to hospitalization or death. Alternatively, it may reflect less social interaction with each other so that primary patterns of transmission are from a reservoir other than their own peer group. Both explanations likely contribute to this negative trend.

School-aged children are postulated to amplify the indirect risk at the community level through a combination of susceptibility and more extensive and complex social networks compared to other age cohorts. Although not proven by this study, it is not difficult to conceive of a pathway that includes initial acquisition and spread of infection by school-aged children, exchange and amplification through their complex peer networks and branching connections from there to the most vulnerable–the very young and the very old. Simulations based on virus transmission properties and community structures have explored this concept in detail [30].

An earlier US study has instead suggested a pre-eminent role for preschool-aged children in predicting population burden due to influenza [19]; this age group is expected to have higher rates of susceptibility to influenza but may have less complex social connectedness than school-aged children. We also identified children aged 2–4 years as a group of interest, although our results may suggest a lesser indirect role for them compared to their older school-aged counterparts. A recent publication describing correlation between emergency department visits by age and hospitalizations and deaths in New York City also emphasized the greater lead time associated with school-aged children [23]. In the US, preschool-aged children have been included in the annual publicly funded influenza immunization program on the basis of direct benefits in reducing health care use. In Canada, the recommendation has not been made to expand pediatric influenza immunization beyond infants and toddlers 6–23 months of age–how the direct benefit in vaccinating preschool-aged children should be weighed against the potential indirect benefit through vaccination of school-aged children requires further study and debate.

In contrast with our findings and that of others emphasizing the role of children, another recent study examining spatiotemporal trends in the United States found that the spread of influenza correlates best with workflows (i.e. the movement of adults to and from the workplace) [21]. Such variation between studies in age-related findings may be due to differences in data sources, diagnostic code validity, health care access and utilization patterns, remuneration and insurance schemes, analysis methods as well as chosen influenza seasons, circulating strains and their differential spread and surveillance across geographic and administrative borders. Variation may also reflect true differences in transmission networks based on social structure and mixing patterns within and between age groups, notably in relation to the proportion of elders living within or distant from the family household or care of children in daycare or other organized settings. In that context, the potential for indirect benefit may be regionally dependent and may best be assessed and determined at that level to account for relevant demographic differences.

Our study, using administrative databases, requires a number of caveats in the interpretation of results. Misclassification and lack of precision in diagnostic codes and their timing can affect internal validity and diminish or enhance the likelihood of observing significant associations in administrative data. Loss of resolution and dilution of patterns can be a particular problem when pooling regional data and interpreting at a provincial or national level. We used clinically diagnosed influenza-like illness and P&I outcomes rather than confirmed influenza infection; such non-specific case definitions will tend to overestimate influenza-attributable morbidity. Some visits may have been due to other respiratory viruses such as respiratory syncytial virus, human parainfluenza viruses or human metapneumovirus [15,16,31–33] for which influenza immunization programs would not be expected to provide any benefit. In interpreting our results, it should also be remembered that medical visits and hospitalizations are determined by many factors, including severity of illness and patient or physician-related concern. Timeliness by age related to influenza medical visits may not reflect earliest or most efficient circulation of influenza virus by age in the community or the potential to intervene upon that. Our decision to exclude specific influenza seasons may also have affected the generalizability of our results. Thus, although administrative data have the advantage of providing broad and powerful representation of influenza-like illness, they should be used only cautiously in guiding age-based recommendations for vaccination. Specific values we have presented in this paper cannot be interpreted literally or definitively but are useful in highlighting trends and areas for further assessment. We recommend other epidemiologic designs to quantify lead time and the potential for vaccination programs to take advantage of that. Our results should be considered in the context of other evidence relevant to determining immunization programs including further studies of disease burden, vaccine safety and efficacy, direct and indirect effects, economic and ethical implications as well as assessment of feasibility and acceptance in relation to program goals and objectives.

In summary, this study is consistent with others emphasizing that the very young and the very old suffer the greatest morbidity due to P&I, and the indirect role of school-aged children in anticipating the risk to others warrants further evaluation.