H1N1pdm in the Americas

Abstract

In late April 2009 the emergence of 2009 pandemic influenza A (H1N1pdm) virus was detected in humans. From its detection through July 18^th, 2009, confirmed cases of H1N1pdm in the Americas were periodically reported to the Pan-American Health Organization (PAHO) by member states. Because the Americas span much of the world’s latitudes, this data provides an excellent opportunity to examine variation in H1N1pdm transmission by season.

Using reports from PAHO member states from April 26^th, 2009 through July 18^th, 2009, we characterize the early spread of the H1N1 pandemic in the Americas. For a geographically representative sample of member states we estimate the reproductive number (R) of H1N1pdm over the reporting period. The association between these estimates and latitude, temperature, humidity and population age structure was estimated.

Estimates of the peak reproductive number of H1N1pdm ranged from 1.3 (for Panama, Colombia) to 2.1 (for Chile). We found that reproductive number estimates were most associated with latitude in both univariate and multivariate analyses. To the extent that latitude is a proxy for seasonal changes in climate and behavior, this association suggests a strong seasonal component to H1N1pdm transmission. However, the reasons for this seasonality remain unclear.

Introduction

In April 2009, a novel influenza A (H1N1) virus capable of infecting humans was detected in North America¹. This influenza virus had not circulated previously in humans and contained a unique combination of avian, human, and swine influenza genes². Based on the available evidence, the Director General of the World Health Organization (WHO) determined that the criteria for an influenza pandemic had been met and declared the first pandemic of the 21^st century.³ Furthermore, the circulation of this novel virus resulted in the declaration on April 25, 2009 of the first “public health emergency of international concern” as defined by International Health Regulations (IHR) ⁴ since its entry into force on 15 June 2007.⁵ The IHR requires countries to notify WHO, by “the most efficient means of communication available, by way of the National IHR Focal Point, and within 24 hours of assessment of public health information, of all events which may constitute a public health emergency of international concern”. ⁵

From its initial focus in North America in early April, the virus has spread worldwide, yielding hundreds of thousands of confirmed cases and over 6,000 deaths as of November 6, 2009.⁶ Countries in the Americas Region notified the Pan American Health Organization (PAHO), Regional Office of the WHO, of their confirmed cases of pandemic (H1N1) 2009 though the Regional IHR point of contact. By July 23^rd, all 35 countries of the Americas had reported confirmed cases.

Influenza virus transmission varies seasonally in most parts of the globe. In temperate regions (both northern and southern hemisphere) there is strong seasonality with the majority of cases occurring in the winter months.^7,8 In tropical regions the seasonal pattern neither as pronounced nor as consistent across regions, but a local influenza season, or seasons, is often still evident.^7,8 The reasons for this seasonality are unknown but current hypotheses include the impact of climate or seasonal changes in social mixing (e.g., the start of school terms).⁷ This novel virus emerged at different times in relation to the typical influenza season across the globe; in the northern hemisphere, it was the end of the usual influenza season, and in the southern, the beginning. Incidence data from the Americas provides and opportunity to estimate the impact of latitude, climate and the lag from usual influenza peak seasons on initial growth rates of H1N1pdm.

Methods

Reporting of Confirmed Cases

Notifications of confirmed cases were made from countries of the Americas to the Regional IHR point of contact. Cases were considered confirmed infections with pandemic (H1N1) 2009 virus if the virus was detected by either polymerase chain reaction (PCR), detected by viral culture, or a four-fold rise in pandemic (H1N1) 2009 virus-specific neutralizing antibodies was measured. The IHR point of contact was notified on a daily basis for the initial weeks of the pandemic, but by June several countries including Canada, Mexico, and the United States, had moved to notifying on specific days of the week. This analysis includes confirmed cases reported to PAHO until July 18, 2009. Ten countries (Argentina, Canada, Chile, Colombia, Costa Rica, Ecuador, Mexico, Panama, Peru, and the United States,) were selected for further analysis based upon adequate numbers of case reports, geographic coverage and a qualitative evaluation of case reporting.

Estimation of Incidence Rates and Reproductive Number

Incidence rates of confirmed cases were estimated by two methods. In the first, primary, method confirmed case reports were grouped into bins of three days in length. Where there was a greater than three day interval between case reports, the empty bin was merged with subsequent three day bins until an interval containing at least one report of confirmed cases was obtained. As a sensitivity analysis bins of four and five days in length were also considered in the same manner. Confirmed cases were assumed to occur at an even rate throughout the interval. In the second, confirmatory, method, a smooth spline with seven degrees of freedom was fit to the reported cumulative confirmed case counts at each report. The first derivative of the smooth spline was then taken as an estimate of the incidence rate of confirmed cases.

The time dependent reproductive number, R_t, was estimated using an extension of methods presented by Nishiura et al. to deal with variable reporting intervals (see supplemental material).⁹ Within each interval k of length Δt_k it is assumed that incidence is growing exponentially at the rate of r_k. Hence, the expected number of cases within each period k is:

$$E(J_k|J_{k-1})=\frac{r_{k-1}}{r_k}\text{exp}(r_{k-1}\mathrm{\Delta }{t}_{k-1})\frac{\text{exp}(r_k\mathrm{\Delta }{t}_k)-1}{\text{exp}(r_{k-1}\mathrm{\Delta }{t}_{k-1})-1}{J}_{k-1}$$

Where J_k is the number of incident cases observed during interval k. We assume that J_k, follows a Poisson distribution with rate λ_k = E(J_k|J_k−1). The rate of growth r_t is assumed to vary smoothly over time, hence r = βX where X is the design matrix for a B-Spline with four degrees of freedom (five degrees of freedom was also considered for comparison), and β is a vector of estimated coefficients.

The βs were then fit using maximum-likelihood techniques giving an estimate for r = {r₁,…,r_n}. The reproductive number, R_k was calculated for each interval k from the exponential growth rates by the estimator:¹⁰

$$R_k={(1+r_k\mu {\nu }^2)}^{\frac{1}{{\nu }^2}}$$

Where μ is the mean generation time, assumed to be 2.7 days, and ν is the coefficient of variation of the generation time, assumed to be 47%.⁹ For each country the first and last two bins were dropped from the final estimates of reproductive number as these estimates are sensitive to both early increases in reporting as each country’s surveillance capacity came on line and edge effects in the estimation procedure.

For comparison a model was fit based on the daily incidences from the smooth spline estimates, dC_t, as r_t = log dC_t+1 − log dC_t. Estimates were converted to reproductive rates by the same equation as above.

Analysis of Factors Influencing Transmission

Assuming that the peak R_t observed approximates the basic reproductive number (R₀) for H1N1pdm in that setting, we regressed Log-basic reproductive numbers on possible factors influencing transmission: latitude, mean temperature May through July 2009, mean relative humidity May through July 2009, median age, proportion of the population under 14 years of age, proportion 15–64 years of age, and proportion over 65 years of age. Each country’s latitude was taken to be that of the capital city. Mean temperature rainfall and relative humidity were calculated based on data from the Weather Underground website (http://www.wunderground.com/) for the capital city of each country. Age distribution estimates for the entire country were obtained from the CIA world fact book (https://www.cia.gov/library/publications/the-world-factbook/). To account for differences in the precision of estimates of R₀, inverse variance weighted regressions were also performed. Differences in mean and median R_t across latitudes were also compared (see supplement).

For countries reporting influenza cases to FLUNET (http://gamapserver.who.int/GlobalAtlas) during the 2008 influenza season (Argentina, Chile, Mexico, Panama, Peru and the United States) the cross-correlation between weekly H1N1 cases reported in 2009 and seasonal influenza cases reported during the corresponding week of the year in 2008 was calculated using the ccf function in the R statistical package (see supplemental material for estimated coefficients). The estimated offset between the 2008 and 2009 H1N1 epidemics was calculated by the number of week shift at which correlation between 2008 and 2009 epidemics was maximized. Univariate regression analyses using the correlation between time series and the estimated offset were performed.

All statistical analysis was done in the R statistical package (version 2.8, www.r-project.org).

Results

There is wide variation between countries in the Americas in the number of reported confirmed cases as of July 18, 2009 (Table 1, Figure 1). The United States, Mexico, Canada and Chile all have reported over 10,000 confirmed cases, while many countries had reported less than 10 cases. Similar variation is seen in the number of reported deaths. This variation is a result of both variation in the actual number of cases in each region and each particular country’s laboratory capacity, so it is not possible to compare these counts between countries or regions. However, it is clear that by July 18, 2009 the H1N1pdm pandemic had spread to all regions of the Americas.

Table 1.

Country level data reported to PAHO as of July 18, 2009

Country	Population	Confirmed Cases	Confirmed Cases per 100,000 population	Deaths
Antigua & Barbuda	69,000	3	4.3	0
Argentina	40,482,000	3,056	7.5	137
Bahamas	330,549	23	7.0	0
Barbados	279,000	23	8.2	0
Belize	320,000	15	4.7	0
Bolivia	9,863,000	589	6.0	2
Brazil	191,241,714	1,175	0.6	11
Canada	33,740,000	10,156	30.1	45
Chile	16,928,873	10,491	62.0	33
Colombia	44,928,970	214	0.5	8
Costa Rica	4,509,290	428	9.5	7
Cuba	11,451,652	144	1.3	0
Dominica	72,660	1	1.4	0
Dominican Republic	9,523,209	108	1.1	2
Ecuador	13,625,000	376	2.8	7
El Salvador	5,744,113	379	6.6	3
Grenada	110,000	0	0.0	0
Guatemala	13,000,000	374	2.9	2
Guyana	772,298	2	0.3	0
Haiti	9,035,536	3	0.0	0
Honduras	7,483,763	123	1.6	1
Jamaica	2,825,928	49	1.7	2
Mexico	111,211,789	13,646	12.3	125
Nicaragua	5,891,199	402	6.8	0
Panama	3,309,679	524	15.8	0
Paraguay	6,349,000	164	2.6	8
Peru	29,132,013	2,557	8.8	11
St. Kitts & Nevis	45,000	1	2.2	0
St. Lucia	160,765	1	0.6	0
St. Vincent & the Grenadines	120,000	1	0.8	0
Suriname	472,000	11	2.3	0
Trinidad & Tobago	1,305,000	65	5.0	0
United States	307,104,000	40,617	13.2	263
Uruguay	3,361,000	550	16.4	19
Venezuela	26,814,843	260	1.0	0
Americas Total	910,720,588	86,531	9.5	686

Figure 1.

Overview of cases of H1N1pdm in the Americas as of July 18, 2009. Orange circles show the cumulative number of cases reported by country. Blue shading represents the number of cases in sub-national regions of each country per 100,000 population. Sub-national regions information is not available for all countries.

Examination of the cumulative case counts of the ten countries selected for further analysis (Canada, the United States, Mexico, Costa Rica, Panama, Colombia, Ecuador, Peru, Chile and Argentina) shows that the epidemic of H1N1pdm began to increase nearly simultaneously in Canada, Mexico, and the United States and then moved south throughout the month of May, with all countries reporting a substantial number of cases by June 1^st (Figure 2). In North America, Canada and the United States show some evidence that H1N1pdm incidence may have begun to decline during July, 2009, whereas case reports from Mexico and Panama were relatively constant throughout the northern hemisphere’s summer months (June–July). The rate of case reports in Costa Rica appears to still be rising as of July 18, though the total number of cases reports as of July 18^th was still relatively low (428). Increases in H1N1pdm in South American countries were delayed compared to North America. Colombia, Ecuador, and Peru still appeared to be in the growth phase of their epidemics as of July 18^th. Chile and Argentina had intense epidemics characterized by a rapid increase in cases during the month of June, which may have peaked by the end of the analysis period.

Figure 2.

Overview of the H1N1pdm epidemic in select countries in the Americas. Column A shows cumulative cases at each report for each country (black points) and smoothed trend (orange line). Column B shows incident cases per day for each 3 day interval (blue bars), intervals in which there were not case reports were merged with the next interval. For comparison the rate of growth from the smoothed cumulative incidence is shown (orange line). Column (C) shows the estimated reproductive number over time (R_t) for each country (blue line) where the rate of growth is assumed to follow a smooth B-spline with 4 degrees of freedom, the results of a spline with 5 degrees of freedom (light-blue line) and analysis of the rate of growth in incident cases from the smoothed curves (orange line) are shown for comparison.

The maximum reproductive number (R_t) estimated in each country ranged from 1.3 in Colombia to 2.1 in Chile (Figure 2). In most locations R_t was highest at the beginning of the epidemic, and declined thereafter. In some (Ecuador, Peru), reproductive numbers peaked a few weeks after initial cases were reported, which may indicate a lag between the introduction of H1N1pdm and efficient transmission of the virus (or possibly variations in reporting). The range of R₀ estimates is consistent with other estimates for H1N1pdm,^10–12 including other estimates using data from countries considered here.^11,12

Examination of R₀ (as approximated by peak reproductive number) by the latitude of each country’s capital city shows a trend of increasing reproductive numbers with decreasing latitude (Figure 3). A similar trend is evident when comparing median and mean reproductive numbers (supplemental figures S1 and S2). Linear regression of log R₀ on latitude shows a 0.4% decrease (95% CI: 0.1, 0.7) in R₀ for each degree increase in latitude (Table 2). Increasing the bin used in our calculations to four or five days yielded similar estimates of R₀ and a slightly reduced (< 0.1% less) rate of reduction in R₀ with increasing latitude (see supplemental material). Univariate analysis did not show significant statistical association between R₀ and the capital city’s mean temperature, mean relative humidity, median age, proportion of the population under 14 years of age, proportion aged 15 to 64, or proportion over 65. The basic reproductive number shows a nearly significant increase of 1.5% (95% CI: −0.4, 3.6) for every degree decrease in mean temperature over May through June.

Figure 3.

Maximum reproductive number by the latitude of the capital city. The thick shows log-linear regression of peak reproductive number on latitude, showing a clear trend towards more intense epidemics in the more southern countries. Thin vertical lines show the latitudes spanned by each country, and the thin horizontal lines show the bootstrap 95% confidence interval on the maximum reproductive number (2000 iterations).

Table 2.

Regression coefficients (and 95% confidence intervals) from log-linear regression of maximum reproductive number on covariates on select countries (Argentina, Canada, Costa Rica, Chile, Colombia, Ecuador, Mexico, Panama, Peru, and the United States).

Covariate	Univariate	Multivariate 1	Multivariate 2	Multivariate 3
Latitude	−0.0043 (−0.0075, −0.0011)	−0.0068 (−0.0108, −0.0028)	−0.0070 (−0.0111, −0.0030)	−0.0060 (−0.0080, −0.0031)
Mean Temperature	−0.015 (−0.035, 0.004)	0.0032 (−0.0129, 0.0193)	0.0044 (−0.0120, 0.0209)
Relative Humidity	0.0032 (−0.0122, 0.0186)	−0.0029 (−0.0120, 0.0063)	−0.0041 (−0.0143, 0.0061)
Median Age	0.0041 (−0.0206, 0.0289)	0.020 (0.004, 0.037)		0.019 (0.007, 0.031)
Proportion < 14 yrs	−0.6366 (−3.24, 1.97)
Proportion 15 – 64 yrs	1.06 (−5.25, 7.37)		3.04 (−1.52, 7.59)
Proportion > 65 yrs	0.93 (−2.72, 4.57)		1.90 (−0.70, 4.50)
Correlation*	0.039 (−0.393, 0.470)
Weeks out of Phase*	−0.0056 (−0.0305, 0.0193)

Significant results are shown in bold.

^*Regression including correlation with the 2008 influenza season only include those countries with adequate data from 2008 (Argentina, Chile, Mexico, Panama, Peru and the United States).

There are large differences in the precision of our estimates of R₀,for instance, the estimate for the United States is very precise, where as Costa Rica had very large variance. To account for this we performed inverse variance weighted linear regression of log R₀ on latitude and other covariates (Table 3). In univariate weighted analysis we found that latitude no longer was a statistically significant predictor of R₀, nor were other covariates. However, in multivariate analysis latitude was a statistically significant predictor of R₀, and that the effect of other covariates was the same as that seen in the univariate analysis. It should be noted that the weights for Canada, Mexico and the United States were disproportionately high, hence some caution should be used in interpreting these results.

Table 3.

Regression coefficients (and 95% confidence intervals) from inverse-variance weighted log-linear regression of maximum reproductive number on covariates on select countries (Argentina, Canada, Costa Rica, Chile, Colombia, Ecuador, Mexico, Panama, Peru, and the United States). Weighting was by the inverse variance of the estimated maximum reproductive rate used in the regression.

Covariate	Univariate	Multivariate 1	Multivariate 2	Multivariate 3
Latitude	−0.0023 (−0.0054, 0.0007)	−0.0079 (−0.0110, −0.0047)	−0.0080 (−0.0125, −0.0036)	−0.0058 (−0.0077, −0.0039)
Mean Temperature	−0.011 (−0.031, −0.010)	0.0056 (−0.0047, 0.0159)	0.011 (−0.006, 0.027)
Relative Humidity	0.0063 (−0.0009, 0.0135)	−0.0034 (−0.0082, 0.0014)	−0.0031 (−0.0097, 0.0034)
Median Age	0.0029 (−0.0100, 0.0157)	0.024 (0.014, 0.034)		0.018 (0.011, 0.024)
Proportion < 14 yrs	−0.36 (−1.83 1.12)
Proportion 15 – 64 yrs	1.91 (−2.79, 6.62)		2.62 (−4.21, 9.45)
Proportion > 65 yrs	0.37 (−1.68, 2.42)		2.66 (−0.64, 5.97)

Significant results are shown in bold.

In multivariate linear regression of log R₀ on capital city latitude, mean temperature, relative humidity and median age significant associations were seen with both latitude and median age (Table 2). R₀ decreased by 0.7% (95% CI: −1.1, −0.3) for each degree increase in latitude. R₀ increased by 2.0% (95% CI: 0.4, 3.8) for each year increase in median age. The relationship between latitude and R₀ is maintained if we instead adjust for the percentage of the population between 15 and 64 years of age and over 65, but no statistically significant relationship between age categories and R₀ is observed.

For those countries where data from the 2008 flu season was available on FLUNET (http://gamapserver.who.int/GlobalAtlas) (Argentina, Chile, Mexico, Panama, Peru and the United States), the correlation with influenza incidence over a corresponding interval in 2008 as well as the shift in weeks to maximize correlation were examined as a possible indicators of whether seasonality is driving the relationship between H1N1pdm transmission and latitude. High correlation and small phase differences with the 2008 epidemic were correlated with increases in R₀ (Table 2), however these relationships were not statistically significant.

Discussion

The observed rates of epidemic growth for H1N1pdm in the Americas indicate an R₀ between 1.3 and 2.1 during April–July of 2009, with a trend towards higher reproductive numbers at lower latitudes. This trend suggests that, as with interpandemic influenza, there is a strong seasonal component to H1N1pdm transmission. This hypothesis is further supported by the resurgence of H1N1 in the northern hemisphere in the fall of 2009 (starting on September), coinciding with the start of the school year.¹³ The United States observed a peak of activity of respiratory disease that exceeded that of previous years, and reported as many Pandemic (H1N1) cases in the first seven weeks of the 2009–2010 influenza season as in the previous seven months (week ending April 18th to week ending October 9).¹⁴ Similarly, Canada reported as many confirmed hospitalized cases in the first ten weeks of the 2009–2010 influenza season (from August 30 to November 7) as they had in the previous five months (April to August 29th).¹⁵ Following the peak of activity that coincided with southern hemisphere’s winter, no outbreaks of respiratory illness due to pandemic (H1N1) 2009 were reported in the temperate countries of the southern hemisphere of the Americas.¹⁶

The sources of influenza seasonality are the subject of some debate, with the leading hypotheses being that it stems from climatic variation in the efficiency of influenza virus transmission, host susceptibility, changes in social mixing (e.g., the seasonality of school terms), or some combination thereof.⁷ We found no clear relationship between temperature or absolute humidity and influenza transmission after adjusting for latitude, however our numbers are low and there is large variation within countries that may mask this effect. Climate variation may indeed mediate the relationship between latitude and reproductive numbers, however, given the temporal and spatial coarseness of our data, latitude may be a proxy for aspects of climate not captured in the climate covariates considered here. The qualitative trend of increased transmission among lower latitude countries, who were experiencing their fall and winter from May to July, was robust to changes in the assumptions and methods used to calculate the reproductive number. If Northern Hemisphere countries experience increases in transmission during the winter of 2010 equivalent to those we would expect for their equivalent southern hemisphere latitudes, we expect to see increases in the reproductive number ranging from 25% (for Mexico) to 72% (for Canada) during the fall and winter months (based on results from model 3 in Table 2). However, the same rates of transmission seen in Chile and Argentina are unlikely to be observed in North America as the susceptible population has been reduced due to ongoing H1N1pdm transmission. Unfortunately, reporting of interpandemic influenza is inconsistent across countries and only available for 2008 for most regions, hence we are not able to assess how correlation between epidemic month in this epidemic and previous seasons has affected transmission.

It is known that children are an important driver of influenza transmission,^17,18 however we did not observe an association between younger populations and increased reproductive numbers. In fact, the opposite trend was observed, with increased R₀ in countries with older populations. From our knowledge of influenza transmission in general, and the age specific attack rates of H1N1pdm in particular,^11,19 it seems unlikely that this result reflects an actual relationship. This may be an artifact of the lower median ages in more equatorial countries, which generally had lower reproductive numbers than more southern counterparts, or some other unmeasured confounder.

In previous pandemics, novel influenzas have emerged at times when influenza is usually not a major source of morbidity and mortality, for example in the summer of 1918 in the United Kingdom, Denmark and other regions of Europe.^20,21 Though a similar comparison of epidemic growth rates has not been performed using data from past pandemics, season has been found to be a important predictor of the sequence of spread globally of pandemic influenza in 1968.²² Influenza seasonality is also hypothesized to be the cause of resurgences in incidence observed in 1918, 1957 and 1968, by limiting transmission after initial emergence in traditional influenza off-season and increasing the transmission of novel influenzas during the traditional influenza season.^23,24,25

Potential sources of variation in the observed R_t other than actual variation in transmission abound. There is wide variation in reporting across countries, as evidenced by difference in the reporting of confirmed cases and deaths. Some countries included in our analysis (e.g., Costa Rica, Colombia and Ecuador) reported substantially fewer cases than others, and this is reflected in greater uncertainty as to the maximum R_t in these areas. There may also be variations in surveillance over time driving apparent variation in R_t both between countries and within countries over time. Sudden improvements in surveillance would appear as increases in the reproductive number, and saturation of the surveillance capacity of a country (likely in poorer regions) would lead to an observed decrease in the reproductive number. Smoothing was required in order to account for variations in reporting, and the type and level of smoothing used influences the exact reproductive numbers estimated and the extent of time dependent variation. However, the qualitative relationship of an increased maximum R_t at lower latitudes was observed consistently regardless of the method of calculation and the amount of smoothing used. Considering cases by the date of report to PAHO, rather than the time of symptom onset, may mask shorter term trends and the exact timing of changes in transmission; but limiting analysis to only those countries for which exact symptom onset dates are available would substantially reduce the number of regions available for analysis.

The assumption that peak R_t approximates R₀ is based on the observation that for most countries (all but Ecuador and Peru) this peak occurs at the beginning of the epidemic. For those where it does not, it seems likely that the initial spread may have been in a group or region where the reproductive number was reduced, and we make the (possibly false) assumption that the peak R_t more accurately reflects R₀ than the initial R_t. The trends in the main analysis were evident when comparing mean and median R_t.

The 2009 H1N1 pandemic provided a rare opportunity to compare simultaneous epidemics of influenza in the Northern and Southern Hemisphere, when patterns of susceptibility could be assumed ot be similar. While we were unable to identify the mechanistic drivers of the variation in the influenza transmission, by considering epidemics occurring both in an out of season we gain some sense of what the range of such variation might be.