- Record: found
- Abstract: found
- Article: not found
The epidemiological and public health research response to 2009 pandemic influenza A(H1N1): experiences from Hong Kong
review-article
Author(s):
Peng Wu
1 ,
Benjamin J. Cowling
1 ,
Joseph T. Wu
1 ,
Eric H. Y. Lau
1 ,
Dennis K. M. Ip
1 ,
Hiroshi Nishiura
1
,
2
Publication date (Electronic):
09 August 2012
Publisher:
Blackwell Publishing Ltd
Read this article at
There is no author summary for this article yet. Authors can add summaries to their articles on ScienceOpen to make them more accessible to a non-specialist audience.
Abstract
In recent years, Hong Kong has invested in research infrastructure to appropriately respond to novel infectious disease epidemics. Research from Hong Kong made a strong contribution to the international response to the 2009 influenza A (H1N1) pandemic (pH1N1). Summarizing, describing, and reviewing Hong Kong’s response to the 2009 pandemic, this article aimed to identify key elements of a real‐time research response. A systematic search in PubMed and EMBASE for research into the infection dynamics and natural history, impact, or control of pH1N1 in Hong Kong. Eligible articles were analyzed according to their scope. Fifty‐five articles were included in the review. Transmissibility of pH1N1 was similar in Hong Kong to elsewhere, and only a small fraction of infections were associated with severe disease. School closures were effective in reducing pH1N1 transmission, oseltamivir was effective for treatment of severe cases while convalescent plasma therapy has the potential to mitigate future pandemics. There was a rapid and comprehensive research response to pH1N1 in Hong Kong, providing important information on the epidemiology of the novel virus with relevance internationally as well as locally. The scientific knowledge gained through these detailed studies of pH1N1 is now being used to revise and update pandemic plans. The experiences of the research response in Hong Kong could provide a template for the research response to future emerging and reemerging disease epidemics.
Related collections
Most cited references61
- Record: found
- Abstract: found
- Article: not found
Comparative epidemiology of pandemic and seasonal influenza A in households.
Benjamin Cowling, Kwok Chan, Vicky J. Fang … (2010)
There are few data on the comparative epidemiology and virology of the pandemic 2009 influenza A (H1N1) virus and cocirculating seasonal influenza A viruses in community settings. We recruited 348 index patients with acute respiratory illness from 14 outpatient clinics in Hong Kong in July and August 2009. We then prospectively followed household members of 99 patients who tested positive for influenza A virus on rapid diagnostic testing. We collected nasal and throat swabs from all household members at three home visits within 7 days for testing by means of quantitative reverse-transcriptase-polymerase-chain-reaction (RT-PCR) assay and viral culture. Using hemagglutination-inhibition and viral-neutralization assays, we tested baseline and convalescent serum samples from a subgroup of patients for antibody responses to the pandemic and seasonal influenza A viruses. Secondary attack rates (as confirmed on RT-PCR assay) among household contacts of index patients were similar for the pandemic influenza virus (8%; 95% confidence interval [CI], 3 to 14) and seasonal influenza viruses (9%; 95% CI, 5 to 15). The patterns of viral shedding and the course of illness among index patients were also similar for the pandemic and seasonal influenza viruses. In a subgroup of patients for whom baseline and convalescent serum samples were available, 36% of household contacts who had serologic evidence of pandemic influenza virus infection did not shed detectable virus or report illness. Pandemic 2009 H1N1 virus has characteristics that are broadly similar to those of seasonal influenza A viruses in terms of rates of viral shedding, clinical illness, and transmissibility in the household setting. Copyright 2010 Massachusetts Medical Society.
- Record: found
- Abstract: found
- Article: not found
The Severity of Pandemic H1N1 Influenza in the United States, from April to July 2009: A Bayesian Analysis
Anne M. Presanis, Daniela De Angelis, Angela Hagy … (2009)
Introduction The H1N1 2009 influenza (pH1N1) pandemic has resulted in over 209,000 laboratory-confirmed cases and over 3,205 deaths worldwide as of 11 September 2009 (http://www.who.int/csr/don/2009_09_11/en/index.html, accessed 14 September 2009), but national and international authorities have acknowledged that these counts are substantial underestimates, reflecting an inability to identify, test, confirm, and report many cases, especially mild cases. Severity of infection may be measured in many ways, the simplest of which is the case-fatality ratio (CFR), the probability that an infection causes death. Other measures of severity, which are most relevant to the burden a pandemic exerts on a health care system, are the case-hospitalization and case-intensive care ratios (CHR and CIR, respectively), the probabilities that an infection leads to hospitalization or intensive care unit (ICU) admission. In the absence of a widely available and validated serologic test for infection, it is impossible to estimate these quantities directly, and in this report we instead focus on the probabilities of fatality, hospitalization, and ICU admission per symptomatic case; we denote these ratios sCFR, sCHR, and sCIR respectively. Although it is difficult to assess these quantities, estimates of their values and associated uncertainty are important for decision-making, planning, and response during the progression of this pandemic. Initially, some national and international pandemic response plans were tied partly to estimates of the CFR, but such plans had to be modified in the early weeks of this pandemic, as it became clear that the CFR could not at that time be reliably estimated [1]. Costly measures to mitigate the pandemic, such as the purchase of medical countermeasures and the use of disruptive social distancing strategies may be acceptable to combat a more severe pandemic but not to slow a milder one. While past experience [2] and mathematical models [3]–[5] suggest that between 40% and 60% of the population will be infected in a pandemic with a reproduction number similar to those seen in previous pandemics, the number of deaths and the burden on the health care system also depend on the age-specific severity of infection, which varies by orders of magnitude between pandemics [6] and even between different waves in the same pandemic [7]. Reports from the Southern Hemisphere suggest that a relatively small fraction of the population experienced symptomatic pH1N1 infection (7.5% in New Zealand, for example [8]), although these numbers are considered highly uncertain [8]. On the other hand, primary care utilization for influenza-like illness (ILI) has been considerably higher than in recent years [8], and anecdotal reports in the Southern Hemisphere have indicated that some intensive care units (ICUs) have been overwhelmed and surgery postponed due to a heavy burden of pH1N1 cases [9],[10]. The problem of estimating severity of pH1N1 infection includes the problem of estimating how many of the infected individuals in a given population and time period subsequently develop symptoms, are medically attended, hospitalized, admitted to ICU, and die due to infection with the virus. No large jurisdiction in the world has been able to maintain an accurate count of total pH1N1 cases once the epidemic grew beyond hundreds of cases, because the effort required to confirm and count such cases is proportionate to the size of the exponentially growing epidemic [11], making it impossible to reliably estimate the frequency of an event (e.g., death) that occurs on the order of 1 in 1,000 patients or fewer. As a result, simple comparisons of the number of deaths to the number of cases suffer from underascertainment of cases (making the estimated ratio too large), and underascertainment of deaths due to inability to identify deaths caused by the illness and due to delays from symptom onset to death (making the estimated ratio too small) [1]. Imperfect ascertainment of both numerator and denominator will lead to biased estimates of the CFR. Estimating the number of persons at these varying levels of severity therefore depends on estimating the proportion of true cases that are recognized and reported by existing surveillance systems. Similar problems affect estimates of key parameters for other diseases, such as HIV. In HIV, a solution to this problem—which now forms the basis for the UK's annual HIV prevalence estimates published by the Health Protection Agency [12],[13]—has been to synthesize evidence from a variety of sources that together provide a clearer picture of incidence, prevalence, and diagnosis probabilities. This synthesis is performed within a Bayesian framework that allows each piece of evidence, with associated uncertainties, to be combined into an estimate of the numbers of greatest interest [14],[15]. Here we use a similar framework to synthesize evidence from two cities in the United States—New York and Milwaukee—together with estimates of important detection probabilities from epidemiologic investigations carried out by the US Centers for Disease Control and Prevention (CDC) and other data from CDC. We estimate the severity of pH1N1 infection from data from spring–summer 2009 wave of infections in the United States. The New York City and Milwaukee health departments pursued differing surveillance strategies that provided high-quality data on complementary aspects of pH1N1 infection severity, with Milwaukee documenting medically attended cases and hospitalizations, and New York documenting hospitalizations, ICU/ventilation use, and fatalities. These are the numerators of the ratios of interest. The denominator for these ratios is the number of symptomatic pH1N1 cases in a population, which cannot be assessed directly. We use two different approaches to estimate this quantity. In the first (Approach 1), we use self-reported rates of patients seeking medical attention for ILI from several CDC investigations to estimate the number of symptomatic cases from the number of medically attended cases, which are estimated from data from Milwaukee. In the second (Approach 2), we use self-reported incidence of ILI in New York City, and making the assumption that these ILI cases represent the true denominator of symptomatic cases, we directly estimate the ratio between hospitalizations, ICU admissions/mechanical ventilation, and deaths (adjusting for ascertainment) in New York City. Each of these two methods provides estimates for the general population, and also for broad age categories 0–4, 5–17, 18–64, and 65+ years. The result of each approach is a tiered severity estimate of the pandemic. Methods Methods Overview The overall goal of this study was to estimate, for each symptomatic pH1N1 case, the probability of hospitalization, ICU admission or mechanical ventilation, or death, overall and by age group. The challenge is that in any population large enough to have a significant number of patients with these severe outcomes, there is no reliable measure of the number of symptomatic pH1N1 cases. This problem was approached in two ways. Approach 1 was to view the severity of infection as a “pyramid” [16], with each successive level representing greater severity; to estimate the ratio of the top level to the base (symptomatic cases), we estimated the ratios of each successive level to the one below it (Figure 1, left side). Thus we broke down (for example) the sCFR (Figure 1, black), i.e., the probability of death per symptomatic case, into components for which data were available – the probability of a case coming to medical attention given symptomatic infection (CDC survey data); the probability of being hospitalized given medical attention (Milwaukee data); and the probability of dying given hospitalization (New York data, including a correction for those who died of pH1N1 but were not hospitalized). Approach 2 was to use the self-reported incidence of ILI from a telephone survey in New York City as the estimate of total symptomatic pH1N1 disease, and the total number of confirmed deaths in New York City as the estimate of the deaths (after accounting for imperfect ascertainment, in this case due to possibly imperfect viral testing sensitivity). In each case, prior distributions were used to quantify information on the probability that cases at each level of severity were detected; these prior distributions reflected the limited data available on detection probabilities and associated uncertainty. 10.1371/journal.pmed.1000207.g001 Figure 1 Diagram of two approaches to estimating the sCFR. Approach 1 used three datasets to estimate successive steps of the severity pyramid. Approach 2 used self-reported ILI for the denominator, and confirmed deaths for the numerator, both from New York City. Both approaches used prior distributions, in some cases informed by additional data, to inform the probability of detecting (confirming and reporting) cases at each level of severity (not shown in the diagram; see Text S1). The Bayesian evidence synthesis framework was used as a formal way to combine information and uncertainty about each level of severity into a single estimate and associated uncertainty that reflected all of the uncertainty in the inputs. All of these estimates were combined within a Bayesian evidence synthesis framework. This framework permits the estimation of probabilities for the quantities of interest (the sCFR, sCIR, and sCHR) and associated uncertainty (expressed as credible intervals [CIs]). These credible intervals appropriately reflect the combined uncertainties associated with each of the inputs to the estimate—mainly, the true numbers of cases at each level of severity, after accounting for imperfect detection—as well as the uncertainties due to sampling error (chance). Study Populations Data were obtained from enhanced pandemic surveillance efforts by the City of Milwaukee Health Department and the New York City Department of Health and Mental Hygiene (DOHMH). Details of testing policies, data acquisition, and analysis are given in Text S1. All data were analyzed first in aggregate and then by age category. Milwaukee Data Between April 6 and July 16, 2009, Milwaukee recorded 3,278 confirmed cases and four deaths due to pH1N1, reflecting sustained efforts to test patients reporting ILI and their household contacts from the start of the epidemic in April until mid-July. On April 27, Milwaukee initiated protocols including recommendations for testing persons with influenza symptoms and travel history to areas reporting novel H1N1 cases, using a reverse transcriptase polymerase chain reaction (RT-PCR) test specific for pH1N1. By May 7, Milwaukee issued testing guidance updated to recommend testing persons with moderate to severe symptoms, except that testing continued to be recommended for health care workers with mild, moderate or severe symptoms. We used a line list dated July 21, and in a preliminary analysis examined the frequency of hospitalization among cases by “episode date” (the earliest date in their case report). The proportion of confirmed cases hospitalized was stable around 3% up to May 20, after which it increased markedly to 6%–8% in the following weeks. We judged that this change reflected reduced testing of mild cases and limited our analysis (used to inform the ratio of hospitalizations to medically attended cases) to the 763 cases with an episode date up to or including May 20. While Milwaukee data were not the main source of estimates of ICU admission or death probabilities, we did employ hospitalized cases up to an episode date of June 14 to contribute to estimates of the ratio of deaths or ICU admissions to hospitalizations, since these should not be affected by failure to test mild cases. New York Case Data New York City maintained a policy from April 26 to July 7, 2009 of testing hospitalized patients with ILI according to various criteria. These criteria evolved up to May 12, from which point they remained as follows: all hospitalized ILI patients received a rapid influenza antigen test. Those patients who tested positive on rapid test (which is known to have low sensitivity for seasonal influenza [17] and for pH1N1 [18]), and any patient in the ICU or on a ventilator, regardless of rapid test result, received RT-PCR tests for pH1N1. We obtained a line list of confirmed or probable hospitalized cases dated July 7, and found in a preliminary analysis that all patients in this line list had a date (onset or admission) in their record no later than June 30, 7 d prior to the date of the line list. Given that >90% of hospitalizations were reported in New York within 7 d, we used this entire line list without accounting for delays in reporting of hospitalizations. Also, given that 98% of admissions occurred after May 12, we did not attempt to account for changes in testing practices before May 12. This line list included a field indicating whether the patient had been admitted to the ICU or ventilated; patients were not followed up after admission to determine if this status changed. However, a chart review of 99 hospitalized cases indicated that none had been admitted to the ICU after admission, so no effort was made to account for this limitation. Separately, we obtained a list of 53 patients whose deaths were attributed to pH1N1, of whom 44 (83%) had been hospitalized before dying. All patients with known influenza or unexplained febrile respiratory illness at the time of death had postmortem samples and/or samples taken before they died sent for PCR testing. New York Telephone Survey Data To estimate levels of ILI in New York City, DOHMH conducted 1,006 surveys between May 20 and May 27, 2009, and 1,010 between June 15 and June 19. Interviews lasted 5 min and were conducted with households in both English and Spanish. The survey used a random-digit dialing (RDD) telephone sampling methodology to obtain data from a random sample of residential households in New York City. A nonrandom individual from each selected household was interviewed and provided information about all household members. Sampled numbers were dialed between five and 15 times to contact and interview a household, or until the sampled number was determined to be nonworking. To account for this design, the data were weighted to the 2007 American Community Survey (ACS); respondents were weighted to householders by borough, age, gender, and race/ethnicity, and the population was weighted by age to the borough of residence. The survey's RDD sampling methodology gave a useful overview of ILI in the community, but it has limitations. The design does not include individuals living in households only reachable by cellular telephone but not by a landline telephone number, and it omitted those living in group or institutional housing. Although households were randomly selected, for the sake of efficiency the interviewed adult was not. Instead, an available adult in the household provided information about all household members and themselves, which may have introduced bias. The results of the survey are being compiled for publication elsewhere. Here, we use summaries of these results by age group (see Text S1) as one means to provide denominators of symptomatic cases. Data on Detection Probabilities from CDC Investigations Sources of data include two community surveys on ILI and health-seeking behavior, and two field investigations conducted during early outbreaks of pH1N1 in the US. These sources are described in further detail elsewhere [19], but are summarized here briefly. In 2007, the Behavioral Risk Factor Surveillance Survey (BRFSS), an RDD telephone survey, included a module on ILI in nine states. This module included questions to assess the incidence of ILI, health-seeking behavior, physician diagnosis of influenza, and treatment of influenza with antiviral medications during the annual 2006–2007 influenza season. In May 2009, following the emergence of pH1N1, an RDD telephone survey sampled similar to the BRFSS was conducted in the same nine states using only the ILI module from the 2007 BRFSS and limited demographic questions. In addition, some data were available from field investigations conducted during large outbreaks of pH1N1 in one community in Chicago and a university campus in Delaware. Investigations of these outbreaks consisted of household interviews in a Chicago neighborhood and an online survey of students and faculty in Delaware. These data were used to inform detection probabilities. In addition, these data were used to inform a prior distribution on the ratio between symptomatic and medically attended cases, cM |S : these surveys estimated that between 42% and 58% of symptomatic ILI patients sought medical attention [19]. Analysis Estimation of the probabilities of primary interest, cH |S , cI |S , and cD |S , respectively the sCHR, sCIR, and sCFR, was undertaken using a Bayesian evidence synthesis framework [14]. Details are given in Text S1, and a schematic illustration of the model is given in Figure 2. Briefly, in this framework, prior information about the quantities of interest (including the uncertainty associated with this prior information) is combined with the information coming from the observed cases at each severity level to derive a posterior distribution on these quantities. This posterior distribution fully reflects all information about the quantities of interest that is contained in the prior distribution and the observed data. Specifically, it was assumed that detected cases O at each level of severity—medically attended (M), hospitalized (H), ICU-admitted (I), and fatal (D)—represented binomially distributed samples from the true number of cases N at the corresponding level of severity, in the given location (New York, abbreviated N or Milwaukee, abbreviated W), with probability equal to the probability of detection at each level (d). The probability d for each level was informed by evidence on the probability of testing at each level of severity (which may have depended on the sensitivity of the rapid test if this was required for PCR testing) and the sensitivity of the PCR test (Table 1). Thus, for example, we defined the probability of detecting a hospitalized case in New York as dHN = dHN1dHN2 , where dHN1 was the probability of performing an RT-PCR–based test and dHN2 was the sensitivity of that test. Hence, the observed number of hospitalized patients in New York, OHN , was assumed to be distributed as Binomial(NHN,dHN ). 10.1371/journal.pmed.1000207.g002 Figure 2 Schematic illustration of the relationship between the observed data (rectangles) and the conditional probabilities (blue circles). The key quantities of interest, sCHR, sCIR, and sCFR, are products of the relevant conditional probabilities. (A) Approach 1, synthesizing data from New York City and Milwaukee. Note that cM |S (double circle) is informed by prior information [19] rather than observed data. (B) Approach 2, using data from New York City only, including the telephone survey. Variables: cD |M : the ratio of non-hospitalized deaths to medically-attended cases; cD |H : the ratio of deaths to hospitalized cases; cI |H : the ratio of cases admitted to intensive care or using mechanical ventilation to hospitalized cases; cH |M : the ratio of hospitalized cases to medically attended cases; cM |S : the ratio of medically attended cases to symptomatic cases; cD |S : the ratio of deaths to symptomatic cases; cH |S : the ratio of hospitalized cases to symptomatic cases. 10.1371/journal.pmed.1000207.t001 Table 1 Detection probabilities and their prior distributions. Detection Probability Components Distributions Rationale dM Medically attended illness dM 1 probability of testing, follow-up, and reporting among medically attended patients Uniform (0.2,0.35) Data from CDC epi-aids in Delaware and Chicago [19] dM = dM 1 dM 2 dM 2 PCR test sensitivity Uniform (0.95,1) Assumption [19] dHW Hospitalization (Milwaukee) dHW1 probability of testing, follow-up, and reporting among hospitalized patients Uniform (0.2,0.4) Assumption [19] dHW = dHW 1 dHW 2 dHW 2 PCR test sensitivity Uniform (0.95,1) Assumption [19] dIW ICU admission (Milwaukee) dIW 1 probability of testing, follow-up and reporting among hospitalized patients Uniform (0.2,0.4) Assumption [19] dIW = dIW 1 dIW 2 PCR test sensitivity Uniform (0.95,1) Assumption [19] dDW Deaths (Milwaukee) PCR test sensitivity and other detection Beta (45,5) Assumption [19] (mean 0.9, standard deviation 0.05) dHW Hospitalization (New York City) dHN 1 probability of performing PCR (rapid A positive or ICU/ventilated) 0.27+0.73 (Uniform (0.2,0.71)) 27% of cases were ICU-admitted so received PCR test; remainder were tested if rapid A positive, which has a sensitivity of 0.2 [17] to 0.71 (sensitivity among ICU patients in NYC) dHN = dHN 1 dHN 2 dHN 2 PCR test sensitivity Uniform (0.95,1) Assumption [19] dIN ICU/ventilation (New York City) PCR test sensitivity Uniform (0.95,1) Assumption [19] dDN Deaths (New York City) PCR test sensitivity and other detection Beta (45,5) Assumption [19] (mean 0.9, standard deviation 0.05) We noted that the ratios cH |S , cI |S , and cD |S can be built up multiplicatively from simpler components: for instance, the ratio of deaths to symptomatic infections may be expressed as cD |S = cD |H cH |M cM |S , the product of the ratios of deaths:hospitalizations, of hospitalizations:medically attended cases, and of medically attended cases:symptomatic cases. These ratios of increasing severity are similar to conditional probabilities but are not strictly so in all cases, since for example some deaths in New York City occurred in persons who were not hospitalized. For this reason we model deaths separately among hospitalized and nonhospitalized patients, i.e., cD |S = cD |H cH |M cM |S + cD |M cM |S . For each observed level of severity (medically attended, hospitalized, ICU, death), the true number of cases was modeled as a binomial sample from the true number of cases at an appropriate lower level, hence where the first subscript indicates severity and the second indicates the population (New York, Milwaukee to May 20, Milwaukee to June 14). In Approach 1 (New York and Milwaukee data combined), for the unobserved level of severity (symptomatic cases) we used a prior distribution of cM |S ∼ Beta(51.5,48.5) to represent uncertainty between 42% and 58% [19]; this distribution has 90% of its mass in this range, with a mean of 0.515. The main analysis of this first approach was performed using prior information to inform the detection probabilities. An additional “naïve” analysis was performed, in which the detection probabilities d were set equal to 1 at all levels of severity. Our prior distributions for the number of symptomatic cases in New York (overall and by age) were taken as ranging uniformly between zero and the proportion reporting ILI in the telephone survey (with the upper bound of that distribution itself having a prior distribution reflecting the confidence bounds of the survey results; details in Text S1). For Milwaukee, the prior distribution on symptomatic cases was taken as uniform between 0 and 25% of the population. In Approach 2 (New York case data and telephone survey data), we made the assumption that self-reported ILI cases represented symptomatic pH1N1 infection, and used the mean and 95% confidence intervals from that survey to define a prior distribution on the number of symptomatic cases overall and by age group. We then used observed hospitalizations, ICU/ventilator use, and fatalities along with prior distributions on detection probabilities as above to inform estimates of true numbers of hospitalizations, ICU/ventilator use, and fatalities, and these in turn were used to estimate sCHR, sCIR, and sCFR. The evidence was synthesized through a full probability model in a Bayesian framework, implemented in the OpenBUGS software [20], which uses Markov chain Monte Carlo to sample from the posterior distribution. Results Table 2 shows the numbers of medically attended cases, hospitalizations, ICU admissions, and deaths in the two cities, with the Milwaukee data separated into the period (to May 20) for which we believe medically attended cases were consistently detected, and the period (to June 14) for which we consider only hospitalized cases, ICU admissions, and deaths. 10.1371/journal.pmed.1000207.t002 Table 2 Cases at each level of severity. Location Age Group Severity Medically Attended Hospitalized ICU-Admitted Dead to May 20 to May 20 to Jun 14 to Jun 14 to Jun 14 Milwaukee 0–4 126 (16%) 7 (28%) 27 (18%) 5 (20%) 0 5–17 470 (60%) 6 (24%) 29 (20%) 7 (26%) 2 (50%) 18–64 189 (24%) 12 (48%) 87 (59%) 14 (52%) 2 (50%) 65+ 3 (0.4%) 0 4 (3%) 1 (4%) 0 Total 788 25 147 25 4 New York Age Group Medically Attended Hospitalized ICU-Admitted Dead (total)/Dead but not hospitalized 0–4 — 225 (23%) 44 (17%) 2 (4%)/2 5–17 — 197 (20%) 51 (20%) 2 (4%)/1 18–64 — 518 (52%) 147 (57%) 46 (87%)/6 65+ — 56 (6%) 15 (6%) 3 (6%)/0 Total — 996 257 53/9 Approach 1 We considered two alternatives to estimate the ratios of interest from the combined New York and Milwaukee data, using self-reported rates of seeking medical attention to establish the denominator. First, we obtained a naïve estimate of the ratios of deaths to hospitalizations, ignoring differences in detection across levels of severity; and second, we obtained an estimate that incorporated evidence and expert opinion on the detection probabilities at each level of severity. The naïve estimate would suggest a median (95% CI) ratio of deaths to hospitalizations (cD |H ) of 4.3% (95% CI 3.2%–5.5%), of ICU admissions to hospitalizations (cI |H ) of 25% (95% CI 22%–27%), and of hospitalizations to medically attended cases (cH |M ) of 3.1% (95% CI 2.0%–4.4%). The ratio of deaths outside of hospitals to medically attended cases (cD |M ) is estimated to be 0.03% (95% CI 0.01%–0.06%). Incorporating the prior evidence that 42%–58% of symptomatic ILI is medically attended, this would imply a naïve estimate of the sCFR (cD |S = cD |H cH |M cM |S + cD |M cM |S ) of 0.081% (95% CI 0.049%–0.131%), a corresponding estimate of the sCIR (cI |S = cI |H cH |M cM |S ) of 0.38% (95% CI 0.24%–0.58%), and an estimate of the sCHR (cH |S = cH |M cM |S ) of 1.55% (95% CI 0.98%–2.32%). If one assumes that detection probabilities are no worse at higher levels of severity than at lower levels, then these figures would be reasonable upper bounds on the symptomatic CFRs and CIRs. Incorporating prior evidence of the detection probabilities at each level of severity, and thus accommodating structural and statistical uncertainties in these probabilities, we estimated that ratio of deaths to hospitalizations (cD |H ) of 2.7% (95% CI 1.8%–3.8%) of ICU admissions to hospitalizations (cI |H ) of 17% (95% CI 12%–21%) and of hospitalizations to medically attended cases (cH |M ) of 2.9% (95% CI 1.6%–5.0%). The ratio of deaths outside of hospitals to medically attended cases (cD |M ) is estimated to be 0.02% (95% CI 0.01%–0.04%). Table 3 shows the estimates for the quantities of primary interest, overall and by age group, in the analysis that incorporated prior evidence of detection probabilities. Here, the posterior median estimate for the sCFR is 0.048% (95% CI 0.026%–0.096%) and for the sCIR is 0.239% (95% CI 0.134%–0.458%). The sCHR is estimated as 1.44% (95% CI 0.83%–2.64%). 10.1371/journal.pmed.1000207.t003 Table 3 Posterior median (95% CI) estimates of the sCFR, sCIR, and sCHR, by age group, based on a combination of data from New York City and Milwaukee, and survey data on the frequency of medical attendance for symptomatic cases. Age sCFR sCIR sCHR 0–4 0.026% (0.006%–0.092%) 0.321% (0.133%–0.776%) 2.45% (1.10%–5.56%) 5–17 0.010% (0.003%–0.031%) 0.106% (0.043%–0.244%) 0.61% (0.27%–1.34%) 18–64 0.159% (0.066%–0.333%) 0.542% (0.230%–1.090%) 3.00% (1.35%–5.92%) 65+ 0.090% (0.008%–1.471%) 0.327% (0.035%–4.711%) 1.84% (0.21%–25.38%) Total 0.048% (0.026%–0.096%) 0.239% (0.134%–0.458%) 1.44% (0.83%–2.64%) Estimates of each of these severity measures vary dramatically by age group, with the lowest severity by each measure in the 5–17 year age group. Comparing the two groups for which we have the most data, the relative risk of death for a symptomatic 18–64-year-old compared to a symptomatic 5- to 17-year-old is 15 (95% CI 5–57). The corresponding relative risks of ICU admission and hospitalization are 5 (95% CI 2–13) and 5 (95% CI 2–12) respectively. The Bayesian framework provides a natural way to estimate confidence (measured as the posterior probability) that one rate is higher than another. The probability that severity is higher in the 18- to 64-y age group than in the 5–17 age group is >99.9%, for each of fatality, ICU admission, and hospitalization respectively. The data are too sparse to say with confidence whether adults over 65 or under 65 have greater severity. For example, among the four age groups, the symptomatic case-fatality ratio is highest in the 18- to 64-y age group with posterior probability 62.%, and in those 65 and over with probability 38%. The symptomatic case-ICU admission ratio is highest in 18- to 64-year-olds with posterior probability 51% and in those over 65 with posterior probability 38%. The sCHR is highest in 18- to 64-year-olds with posterior probability 37% and in those over 65 with posterior probability 37%. Approach 2 Table 4 shows the estimates for the sCFR, sCIR, and sCHR, by age group, when self-reported ILI is used as the denominator for total symptomatic cases. Overall these estimates are: sCFR = 0.007% (95% CI 0.005%–0.009%), sCIR = 0.028% (95% CI 0.022%–0.035%) and sCHR = 0.16% (95% CI 0.12%–0.26%). Compared to Approach 1, these estimates are nearly an order of magnitude smaller, and the age distribution differs. The relative risks for each severity in the 18- to 64-year-old group compared to the 5- to 17-year-old group are 7 (95% CI 3–25) for fatalities, 1.5 (95% CI 0.9–2.5) for ICU admissions, and 1.4 (95% CI 0.9–2.1) for hospitalizations. The CFR is highest in the 18–64 y group with posterior probability 52%. In contrast to Approach 1, the CIR is highest among 0- to 4-year-olds, with posterior probability 79%, and the CHR is highest among 0- to 4-year-olds, with posterior probability 99%. 10.1371/journal.pmed.1000207.t004 Table 4 Posterior median (95% CI) estimates of the sCFR, sCIR, and sCHR, by age group, using self-reported ILI as the denominator of symptomatic cases. Age sCFR sCIR sCHR 0–4 0.004% (0.001%–0.011%) 0.044% (0.026%–0.078%) 0.33% (0.21%–0.63%) 5–17 0.002% (0.000%–0.004%) 0.019% (0.013%–0.027%) 0.11% (0.08%–0.18%) 18–64 0.010% (0.007%–0.016%) 0.029% (0.021%–0.040%) 0.15% (0.11%–0.25%) 65+ 0.010% (0.003%–0.025%) 0.030% (0.016%–0.055%) 0.16% (0.10%–0.30%) Total 0.007% (0.005%–0.009%) 0.028% (0.022%–0.035%) 0.16% (0.12%–0.26%) Discussion We have estimated, using data from two cities on tiered levels of severity and self-reported rates of seeking medical attention, that approximately 1.44% of symptomatic pH1N1 patients during the spring in the US were hospitalized; 0.239% required intensive care or mechanical ventilation; and 0.048% died. Within the assumptions made in our model, these estimates are uncertain up to a factor of about 2 in either direction, as reflected in the 95% credible intervals associated with the estimates. These estimates take into account differences in detection and reporting of cases at different levels of severity, which we believe, based on some evidence, to be more complete at higher levels of severity. Without such corrections for detection and reporting, estimates are approximately two-fold higher for each level of severity. Using a second approach, which uses self-reported rates of influenza-like illness in New York City to estimate symptomatic infections, we have estimated rates approximately an order of magnitude lower, with a symptomatic sCHR of 0.16%, an sCIR of 0.028%, and an sCFR of 0.007%. In both approaches, the sCFR was highest in adults (in Approach 1, 18–64 y, while Approach 2 cannot distinguish whether it is higher in that group or in those 65y and older) and lowest in school-aged children (5–17 y). Data on children 0–4 and adults 65 and older were relatively sparse, making statements about their ordering more difficult. Nonetheless, these findings, along with surveillance data on the age-specific rates of hospitalization and death in this pandemic (http://www.cdc.gov/vaccines/recs/ACIP/downloads/mtg-slides-oct09/12-2-flu-vac.pdf), indicate that the burden of hospitalization and mortality in this pandemic falls on younger individuals than in seasonal influenza [21]. A shift in mortality toward nonelderly persons has been observed in previous pandemics and the years that immediately followed them [22]. These estimates are valuable for attempting to project, in approximate terms, the possible severity of a fall–winter wave of pH1N1, under the assumption that the virus does not change its characteristics. In the 1957 and 1968 pandemics, it appears that perhaps 40%–60% of the population was serologically infected, and that of those, 40%–60% were symptomatic [2],[23]–[25]. Current estimates of the transmission of pH1N1 range between about 1.4 and about 2.2, consistent with estimates of the reproduction numbers from prior pandemics [26]–[30]. To convert our estimates into population impacts, one needs to make an assumption about the attack rate and its age distribution. For each 10% of the US population symptomatically infected (with the same age distribution observed in the spring wave), our Approach 1 estimates suggest that approximately 7,800–29,000 deaths (3–10 per 100,000 population), 40,000–140,000 intensive care admissions (13–46 per 100,000 population), and 250,000–790,000 hospitalizations (170–630 per 100,000 population) will occur. These estimates scale up or down in proportion to the attack rate; for example, they should be doubled if 20% of the population were symptomatic, producing for example 15,000–58,000 deaths (6–20 per 100,000 population). Approach 2 suggests much smaller figures (for each 10% of the population symptomatic) of 1,500–2,700 deaths (0.5–0.9 per 100,000), 6,600–11,000 ICU admissions/uses of mechanical ventilation (22–35 per 100,000), and 36,000–78,000 hospitalizations (12–26 per 100,000). Again, these numbers should be scaled in proportion to the attack rate. To date, symptomatic attack rates seem to be far lower than 25% in both the completed Southern Hemisphere winter epidemic and the autumn epidemic in progress in the US; severe outcomes seem to be considerably less numerous than those described for Approach 1 with a 25% attack rate. In New Zealand, just under 2% of the population consulted a general practitioner (GP) for ILI during the winter wave of the pandemic (http://www.moh.govt.nz/moh.nsf/indexmh/influenza-a-h1n1-update-138-180809), which is consistent with an attack rate significantly lower than 25%, though somewhat higher than the GP consultation rate observed in severe seasonal flu outbreaks such as those in 2003 and 2004 (http://www.surv.esr.cri.nz/PDF_surveillance/Virology/FluWeekRpt/2004/FluWeekRpt200444.pdf). The level of severity estimated for the United States reflects in part the availability of antiviral treatment and other medical interventions that will not be available in all populations. Oseltamivir use was common in Milwaukee (Milwaukee Department of Health, unpublished data), and although the health care system was put under strain in both cities studied, there was no shortage of intensive care or other life-saving medical resources. In a situation of greater stress on the health system, as has been observed in certain locations in the Southern Hemisphere ([9],[10]; http://www.capegateway.gov.za/eng/your_gov/3576/news/2009/aug/185589), or in areas that lack a high-quality health care system, severity might increase in proportion to decreased availability of adequate medical attention. Worryingly, our estimates of the proportion of symptomatic cases requiring mechanical ventilation or ICU care was approximately 4–5× our estimate of the sCFR. It is possible that a substantial proportion of those admitted to ICUs could have died without intensive care. In populations without widespread access to intensive care, our results suggest that the same burden of disease could lead to a death rate 4–5× higher. Likewise, a change in the virus to become more virulent or resistant to existing antiviral drugs, or the emergence of more frequent bacterial coinfections, could increase the severity of infection compared to that observed so far. Estimates of severity for an infection such as influenza are fraught with uncertainties [1]. Our analysis has accounted for many of these uncertainties, including imperfect detection and reporting of cases, bias due to delays between events (such as the delay from illness onset to death), and the statistical uncertainties associated with limited numbers of cases, hospitalizations, and deaths. Another major source of difficulty is the spatial and temporal variation in reporting effort for mild and severe cases; for example, most jurisdictions in the US stopped reporting mild cases on or before the second week of May, but this change varied by jurisdiction. We have attempted to avoid this difficulty by focusing on individual jurisdictions—New York and Milwaukee—for which the approach to reporting was relatively stable over time. One limitation is that Milwaukee changed its guidance during our surveillance period from testing of all symptomatic cases to testing of all symptomatic health care workers but only moderate-to-severe cases in non-health care workers. We believe that testing policies did not change dramatically during this period, because the proportion of hospitalized cases remained fairly constant; however, the sample size before this change in guidance was small. Thus, our estimates should be seen as being the risk of severe outcome among persons with symptoms, possibly biased somewhat toward those with more severe symptoms. Despite our efforts to account for sources of uncertainty, several others remain and have not been accounted for in our analysis. First, we have assumed that for each level of severity (from medically attended up to fatal), case reporting was equal across age groups; for example, we assumed that medically attended cases were as likely to be reported for young children as for adults. It is possible that this is not the case, for example that mild cases were more likely to come to medical attention if they occurred in children than if they occurred in adults. If this were true, our conclusion that severity was higher in adults than children could be partly a result of differential reporting. Second, the overall estimates of severity (not stratified by age group) reflect the age composition of cases in the sample we studied, especially the age composition of the lowest level of severity examined, medically attended illness. Among medically attended cases in Milwaukee, 60% were in the 5–17 y age group, the one in which severe outcomes were the least likely. A preponderance of cases within this age group may be typical of the early part of influenza epidemics, and while it has been argued that there is a shift from younger to older age groups in seasonal influenza [31] as the epidemic progresses, there is evidence, at least from the 1957 pandemic, that attack rates remain higher in children than adults throughout the course of the epidemic [2]. Since severity of pH1N1 influenza appears to be considerably higher in adults, a shift in the burden of disease from children to adults as the epidemic progresses would lead to an increase in average severity. We note that the association between age and severity may also affect observed trends in the characteristics of cases. The World Health Organization has noted worldwide a shift from younger to older mean age among confirmed cases (http://www.who.int/csr/disease/swineflu/notes/h1n1_situation_20090724/en/index.html). If severity is lowest among children, this upward shift in age distribution may partially reflect a shift toward detection of more severe cases, rather than a true shift in the ages of those becoming infected. Third, the symptomatic CFR, CIR, and CHR are dependent upon our estimates of the true number of symptomatic cases, N iSk , and hence are sensitive to the choice of prior distribution for these, as well as to our prior assumptions on the detection probabilities. In particular, if the probability that symptomatic patients seek medical attention and are confirmed is lower than we assume in our prior distributions, then there are more cases than are inferred by our model, and severity is correspondingly lower than our estimates. If the probability of detecting severe outcomes (hospitalizations, deaths, ICU) is lower than our prior distributions reflect, then there are more severe outcomes than our model infers, so severity is correspondingly higher. Finally, the small sample sizes in some age groups, the over-65 year olds in particular, lead to large uncertainty in the age-specific estimates. This level of uncertainty is reflected in the wide 95% credible intervals for the estimates. Our two approaches yield estimates that differ by almost an order of magnitude in the severity of the infection, on each of the three measures considered. How should planners evaluate these contrasting estimates? The lower estimates, using the denominator of self-reported ILI in New York City, may reasonably be considered lower bounds on the true ratios. ILI is thought to be relatively rare in May–June, hence true ILI was probably largely attributable to pH1N1 during this period in New York City. However, self-reported ILI is notoriously prone to various biases, most of which suggest that true rates are probably lower than self-reported rates. A previous telephone survey conducted in New York City found that 18.5% of New Yorkers reported ILI in the 30 d prior to being surveyed in late March 2003 [32], which represented a period of above-baseline but declining influenza activity nationally and no known influenza outbreaks in New York City [32]. The survey was repeated in October–November 2003, prior to the appearance of significant influenza activity, and 20.8% reported ILI in the 30 d prior [32]. If these surveys represent a baseline level of self-reported ILI in the absence of significant influenza activity, then the approximately 12% self-reported ILI in the telephone survey is substantially lower than this out-of-season baseline, suggesting that it likely overstates the total burden of symptomatic pH1N1 disease. The lower estimates are also broadly consistent with estimates from New Zealand, which has experienced a nearly complete influenza season [8], and from Australia (http://www.health.gov.au/internet/main/publishing.nsf/Content/cda-surveil-ozflu-flucurr.htm/FILE/ozflu-no14-2009.pdf). The higher estimates, on the other hand, were obtained using ratios of hospitalizations to confirmed medically attended cases and self-reported rates of seeking medical attention for ILI, which have been consistently measured in the range of about 40%–60%. It is possible that the special efforts of the New York City health department to identify pH1N1-related fatalities (including those not hospitalized) provides a fuller picture of the total number of deaths from this infection. Interestingly, New York City reports about the same number of hospitalizations for our study period (996) as New Zealand reports up to mid-August (972), but 3.5× as many deaths (53 versus 16) [8]. If this discrepancy reflects more complete ascertainment of deaths in New York City, it may account for much of the difference between our higher estimates of case-fatality ratios and those from New Zealand. Given the number of uncertainties cataloged above (which apply also to other jurisdictions within and outside the US), we believe that our two approaches probably bracket the reasonable range of severity for the US spring wave. Age-specific severity patterns as estimated here are largely consistent with those one would obtain by simply comparing the incidence of confirmed cases, hospitalizations, and deaths in the US as a whole for a similar period [19], although the estimates for persons over age 65 are highly uncertain, with 95% credible intervals spanning several orders of magnitude, due to the very small number of individuals in our sample from that age group. The estimates provided here may be compared to those for seasonal influenza. Compared to seasonal influenza, these estimates (assuming a 25% symptomatic attack rate) suggest a number of deaths in the US that could range from about half the number estimated for an average year to nearly twice the number estimated for an average year [33] (Approach 1), or a range about 10-fold lower than that (Approach 2); however, the deaths would be expected to occur in younger age groups, compared to the preponderance of deaths in persons over 65 in seasonal influenza. Such a shift in age distribution is typical for pandemics and the years that follow them [22]. Under Approach 1, and assuming a typical pandemic symptomatic attack rate of 25%, the estimated number of hospitalizations for an autumn–winter pandemic wave is considerably more than the approximately 300,000 estimated for typical seasonal influenza [34], whereas Approach 2 suggests a number between 1/3 and 2/3 of that observed in typical seasonal influenza. It should be noted that most hospitalizations, and about 90% of deaths attributed to seasonal influenza, are categorized as respiratory and circulatory, not including the more specific diagnoses of pneumonia and influenza; that is, they are due to myocardial infarction, stroke, and other proximate causes, but are nonetheless likely initially caused by influenza infection [35]. The deaths included in our study may have reflected more directly influenza-related causes and may not reflect these indirect causes of influenza-related death. Indeed, it is unclear whether the proportion of indirect respiratory and circulatory causes of death and hospitalization will be as high in this pandemic year, given the younger ages involved in most severe cases. Given these differences between the estimates here based on virologically confirmed deaths and the ecological statistical approach to estimating influenza-attributable deaths and hospitalizations for seasonal influenza, it will be difficult to interpret comparisons between the two types of estimates until (after the pandemic has finished) comparisons can be made between the ecological and the confirmed-case approach to estimating burden of hospitalization and deaths. Our estimate of the sCFR is lower than those provided by Garske et al. [16], which ranges from 0.11% to 1.47% overall, and between 0.59% and 0.78% in the US, but which was based on confirmed plus probable (rather than symptomatic) cases. Nishiura et al. [36] estimate that between 0.16% and 4.48% of confirmed cases in the United States and Canada were fatal. Our Approach 1 includes a probability of approximately 1/8 (∼50% probability of symptomatic patients seeking care × ∼28% probability of testing and report for a symptomatic × ∼97% test sensitivity, with associated ranges for each; Table 1) to convert symptomatic into medically attended cases, and this factor accounts for most of the difference between our estimates and the earlier estimates based on confirmed or confirmed plus probable cases. Wilson and Baker [37], on the other hand, use a denominator of infections (rather than symptomatic or confirmed cases) and estimate a range of CFR from 0.0004% up to 0.6%. Our estimates fall in the middle part of this range. More recently, Baker et al. [8] used their estimates of the total incidence of symptomatic disease in New Zealand to estimate an sCFR of 0.005%, equal to the lower end of the credible interval for our Approach 2 estimate, and considerably below our Approach 1 estimate. The generally downward trend in the estimates of severity reflects early ascertainment of more severe cases (e.g., mainly hospitalized cases in the early Mexican outbreak); the authors of each of these earlier reports recognized and discussed the issue of ascertainment and its potential biasing effect on severity estimates. While we have been careful to highlight uncertainties in the estimates of severity, our results are sufficiently well-resolved to have important implications for ongoing pH1N1 pandemic planning. The estimated severity indicates that a reasonable expectation for the autumn–winter pandemic wave in the US is a death toll less than or equal to that which is typical for seasonal influenza, though possibly with considerably more deaths in younger persons. If attack rates in the autumn match those of prior pandemics and hospitalization rates are comparable to our estimates using Approach 1, the surge of ill individuals and subsequent burden on hospitals and intensive care units could be large. However, using Approach 2, estimates of hospitalizations and ICU admissions are considerably lower. Either set of estimates places the epidemic within the lowest category of severity considered in pandemic planning conducted prior to the appearance of pH1N1 in the United States, which considered CFRs up to 0.1% (http://www.flu.gov/professional/community/community_mitigation.pdf). Continued close monitoring of severity of pandemic (H1N1) 2009 influenza is needed to assess how patterns of hospitalization, intensive care utilization, and fatality are varying in space and time and across age groups. Increases in severity might reflect changes in the host population—for example, infection of persons with conditions that predispose them to severe outcomes—or changes in the age distribution of cases—for example a shift toward adults, in whom infection is more severe. Changes in severity might also reflect changes in the virus or variation in the access and quality of care available to infected persons. Supporting Information Text S1 Supplementary methods. (0.43 MB DOC) Click here for additional data file.
- Record: found
- Abstract: found
- Article: not found
Delaying the International Spread of Pandemic Influenza
Ben Cooper, Richard J Pitman, W. John Edmunds … (2006)
Introduction The scale of threat posed by hypervirulent avian influenza subtypes [1,2], and the memory of the 20–100 million who died in the 1918 pandemic [3,4], warrant consideration of large-scale, concerted, and potentially highly disruptive control measures [5]. Were such a virus to acquire the ability to spread efficiently between humans, control would almost certainly be hampered by limited vaccine supplies [6]. Interventions able to substantially impede global spread, by providing time for vaccine stocks to accumulate, could have profound public health benefits. Border controls and World Health Organization travel advisories formed central and sometimes controversial components of the control efforts during the severe acute respiratory syndrome (SARS) epidemic [7,8], and travel restriction is thought likely to occur during an influenza pandemic (although enforcement is currently considered by the World Health Organization to be impractical in most countries) [9]. In the absence of sufficient vaccine stocks, other control measures such as the use of antiviral agents could also be used [10,11]. Ideally, such measures would reduce the average number of secondary cases caused by each primary case (the effective reproduction number, R t) to below one, making sustained transmission impossible. This happened during the SARS epidemic, where isolation, quarantine, and behaviour change were able to bring about control [12]. The much shorter serial interval for influenza makes the chances for early epidemic termination much lower [13]. The main value of interventions is more likely to be in reducing the incidence and slowing the rate of spread of the virus. To evaluate the potential of travel restriction and local control measures to impede global dissemination we developed a stochastic (i.e., probabilistic) model of the international spread of influenza based on extensions of coupled deterministic epidemic transmission models [14–19]. This class of models has been shown to be capable of accurately forecasting local and global spread of epidemic and pandemic influenza [14–19] and accounting for the global distribution of other pathogens [20,21], but has not previously been used to assess the impact of travel restrictions or other control options for pandemic influenza. Methods We used a metapopulation model that consists of a set of coupled dynamic epidemic transmission models (Figure 1). Each component model represents one city and tracks the progression of individuals through four classes: susceptible to infection (S); exposed to the virus but not yet infectious (E); infectious (I); and recovered and no longer susceptible (R). We assumed that infectiousness coincides with disease onset and that infectious cases do not travel. Previous work has used deterministic approximations to study the evolution of this system [14–19]. With this approach the first case in each city (except the originating city) was assumed to occur only when the average number of incubating cases arriving from other cities exceeded one, an approximation that will artificially slow the rate of spread between cities. Our stochastic model, which has a similar underlying structure to its deterministic counterpart, avoids this distortion and uses probabilistic transitions to capture the inherent uncertainty in the course of the epidemic. This is more appropriate than the deterministic approaches previously adopted, because chance effects dominate in the early stages of the epidemic in each city and in the seeding of each city's epidemic. As well as providing greater realism, this approach allowed us to quantify the uncertainty in model predictions due to demographic stochasticity. Thus, rather than assuming that each city's epidemic starts at a determined time, we assumed that the initiation of an epidemic depends on the timing of a sequence of chance events: a person incubating the virus must board a plane; that person must infect others in the destination city; some of those others must cause further transmission, and so on. Each time the model is run, even with identical starting conditions and parameter values, a different answer is obtained. This stochastic model was used to estimate key parameters using data from the 1968–1969 (1968/9) influenza pandemic and to evaluate the impact of interventions using contemporary demographic and transport data. Mathematical details of the model are provided in Protocol S1. Coupling between cities was estimated using data from the International Air Transport Association for 2002 that gives the number of seats on flights between 105 cities, including the 100 with the highest number of international scheduled passengers and all 52 used in the 1968/9 data. City sizes were taken from the United Nations urban agglomeration data (available at: http://unstats.un.org/unsd/demographic/sconcerns/densurb/urban.aspx). When fitting models to 1968/9 data, air transport data, sizes of urban agglomerations, and influenza data were taken from a previous study [14,15]. To select from variants of the basic model, we compared deterministic and stochastic model fits to data from the 1968/9 pandemic by choosing parameters to minimize the sum of squared deviations (SSQ) between times of observed and predicted peaks (this contrasts with previous work that aimed to forecast the pandemic, and therefore based parameter estimates only on data from the first affected city [14]). A major strength of our approach is that the observed time of an epidemic peak should be unaffected even by large between-country variation in influenza reporting rates. The predicted epidemic peak for a given city was defined as the day on which the highest number of infected people first developed symptoms. When a city had more than one observed peak in the 1968/9 data, only the first peak was used. We used the statistic SSQ(m)/(n − 2m) to compare model fits [22], where m is the number of fitted parameters (between one and four), and n the number of fitted data points (the number of cities with observed and predicted epidemic peak times). This formula selects models that are parsimonious and fit the data well. The full stochastic model was then used to estimate parameters for the best-fitting model by choosing parameters that minimised the mean SSQ from ten simulation runs for each combination of parameter values. We evaluated models with sine wave, square wave, and no seasonal variation in transmission parameters for cities outside the tropics. In the sine wave formulation, the peak transmissibility occurred on the shortest day in each hemisphere, while in the square wave formulation, peak transmissibility lasted 6 mo, also centred on the winter solstice. We also considered model formulations where the transmission parameter in the tropics was taken as the maximum, minimum, and mean over 1 y of that outside the tropics. For some parameter values, the model predicted no epidemic peaks in some cities for which an epidemic peak was in fact recorded. When fitting the models we penalized these regions of parameter space by arbitrarily assigning a deviation between model and data of 500 d. Interventions We used the stochastic model to consider the effects of (i) reducing local transmission (this simulates the effects of isolation, behaviour changes, antiviral use, or other measures that may reduce the average number of secondary cases produced by one primary case); and (ii) restricting travel to and from affected cities. We assessed the ability of these measures to delay epidemics in individual cities. We considered only major epidemics, which we defined as those peaking with at least one case per 10,000 people per day. We assumed that measures were introduced only after the first 100 symptomatic cases in each city except the originating city, for which 1,000 cases were required, although we also evaluated the sensitivity of the results to these assumptions. We considered a number of other scenarios to assess the sensitivity of the results to the most important unknowns: patterns of seasonal variation in influenza transmission; variation in transmissibility between tropical and temperate regions; the proportion of individuals initially susceptible to the virus; the basic reproduction number, R 0 (defined as the mean number of secondary cases in a local and susceptible population caused by the introduction of one primary case); the distribution of the infectious period; the city in which the pandemic begins; and the date on which the virus first begins to spread. Results Amongst the model variants considered, the best fit to data from the 1968/9 pandemic was achieved when transmissibility varied sinusoidally in temperate regions and was constant and equal to the north/south maximum in the tropics. We used this model to estimate key parameters using 1968/9 data, and to evaluate the impact of interventions. Models without seasonal forcing terms gave poor fits to data and could not account for the large differences in epidemic timing between cities in the north and south temperate regions. Models in which transmissibility in the tropics was set to the north/south mean also performed surprisingly poorly, with best-fit SSQs approximately three times greater than those obtained when transmissibility in the tropics was set to the north/south maximum (Figure 2A and 2B). Less surprisingly, models that assumed all cities were equally connected by air travel (but with the same total volume of air traffic) also performed poorly, with best-fit SSQs about twice as large as those from the models that used the air travel data. Previous work with the deterministic version of the model has assumed a square wave variation in transmissibility, assigning transmission outside the influenza season to be one-tenth of the value during the season [14]. We found the fit to data under this assumption to be substantially poorer compared with models in which maximum and minimum seasonal transmission parameters were both estimated. An exploration of the parameter space for the best-fit model showed that, assuming 60% of the population to be initially susceptible (the approximate value estimated previously [14]), maximum R 0 values (R 0,max) ranging from about 2.5 to 3.5 gave the best fits to data, while minimum R 0 values (R 0,min) between about 0.5 and 1.5 had the most support (Figure 2C). The maximum R 0 value and the fraction initially susceptible could not be identified simultaneously: A high value of one implied a low value of the other (Figure 2D). However, the initial maximum effective reproduction number, R max (equal to the product of the two and giving the average number of secondary cases produced by one primary case in an actual population, accounting for immunity) was well defined, with only a narrow range of values between about 1.5 and 2.2 supported by the data. This result is consistent with other estimates from influenza pandemics [14,23]. We therefore took as our baseline scenario an R 0,max value of 3 and an R 0,min of 1.2, assumed 60% of the population to be initially susceptible, and used a model in which the R 0 value varied sinusoidally and peaked in midwinter, and in which the pandemic originated in Hong Kong on 1 June. The model showed good agreement with data from the 1968/9 pandemic, with observed epidemic peaks almost always occurring at times when the model predicted a very high probability of influenza activity (Figure 3). Observed and predicted times of epidemic peaks differed, on average, by 31 d. There were, however, some anomalies: the first epidemic peaks occurred much later than might have been expected in London and Tokyo, and somewhat earlier than predicted in Manila and Madras. Despite large variation in the timing of predicted epidemic peaks in individual cities between simulation runs, the overall course of the pandemic was quite predictable (Figure 3A), although there was markedly more between-run variability in the tropics and the south than in the north. The roughly ten-fold increase in air traffic since 1968 causes epidemics in most cities to peak between 1 and 2 mo earlier than they would have done in 1968 (in some southern hemisphere cities the epidemic peaks 1 y earlier) and substantially reduces variation between simulation runs (Figure 3B). The model reproduced another interesting aspect of influenza epidemiology: the tendency for peak periods of influenza activity in the tropics to shift with latitude, so that in the northern tropics they are closer to countries north of the tropics, while the southern tropics tend to be more closely aligned with countries south of the tropics [24]. This occurs despite the fact that the model has no explicit assumptions about seasonality for cities in the tropics; the behaviour arises only as a result of the strength of transport connections between different regions. It is also notable that the pandemic starts early enough to allow some probability of influenza activity in the south during the end of the flu season in 1968. Despite this, predicted epidemic peaks (the weeks with the greatest number of reported cases in each location) still occur in 1969 in the south. When we used the model to evaluate interventions using contemporary air travel and demographic data, we found that travel restrictions to and from affected cities would slow epidemic spread, but unless almost all air travel from affected cities (i.e., greater than 99%) was suspended, the potential for delaying the pandemic was limited (Figures 4–6 and Table 1). Even when 99.9% of air traffic was suspended, most cities had a low probability of ultimately escaping the pandemic (Figure 4), and delays large enough to be of clinical significance (6 mo or more) were common only if interventions were made after the first few cases (Figure 5). Interventions that reduced transmission could typically lead to more pronounced delays (Figures 5 and 6 and Table 1), although only when R t was reduced to slightly above one were these sufficient to delay epidemics until the next influenza season. These findings were not highly sensitive to assumptions about initial susceptibility and transmissibility (Table 1). Decreasing the number initially susceptible (while holding R max constant) has two opposing effects (Table 1). First, within cities the time between seeding with influenza cases and the epidemic peak decreases. This is because the initial epidemic growth rate is unaffected, but each new case represents a greater proportional reduction in the susceptibles and causes a greater reduction in R t (the epidemic peaks when R t = 1). Conversely, between-city dynamics are slowed because there are fewer infectious people to spread the disease. Which effect dominates varies between cities; those affected at the start of the pandemic tend to experience peak activity earlier when there are fewer initial susceptibles; for the rest it usually occurs later. Although travel restriction always reduces the rate of spread between cities, under most scenarios so many people become infected that even near-total restriction has remarkably little effect. However, for a given R max, the smaller the number of susceptibles the greater the impact of this intervention. For example, when 90% of the population are initially immune, the most extreme travel restrictions can be quite effective in preventing international spread. Conversely, reducing transmission has the greatest effect on impeding international spread when (for a given R max) more people are susceptible. The large delays and reductions in the number of affected cities result from two effects acting in the same direction: The reduced R t slows the epidemic within each city (delaying epidemic peaks), and the reduced total number of cases reduces the rate of spread between cities. Larger reductions in transmission led, in extreme cases, to smaller delays in epidemic peaks (Table 1). This happened only when R t was reduced to below one, causing the epidemic decline to begin immediately; the peak therefore occurred at the time of the intervention, earlier than it would have done with a less effective intervention. Under such circumstances the time of the epidemic peak is not a good measure for fully evaluating local control measures. Previous influenza modelling work has used both square and sine wave seasonal forcing terms [14,25]. We found that the outcomes of interventions were not highly sensitive to the precise assumptions made. The delays in the timing of epidemic peaks depended only to a limited extent on the city in which the pandemic started and to a somewhat greater extent on the date of release (Table 2), with larger delays more likely when the first cases occurred towards the end of the influenza season in the place of origin. Results were, however, highly sensitive to the timing of the intervention (Figure 5). Large delays in the timing of epidemic peaks and the prevention of epidemics in a large number of locations could be achieved with the most extreme interventions, but only when they were made sufficiently early. However, making the interventions after fewer than 1,000 cases in the place of origin had minimal additional benefit in slowing pandemic spread. Similarly, preemptive travel restrictions had no advantage over interventions made after one case in affected cities (Figure 5A and 5B). The course of infection with a future pandemic influenza virus might differ in important ways from our baseline assumptions, and could be quite unlike typical interpandemic influenza. We therefore assessed the robustness of our conclusions to the assumed latent and infectious periods. We found that assuming a greater degree of infectiousness early in the course of infection (reducing the serial interval from 4.2 to 2.6 d, as suggested by recent analysis of household influenza transmission data [11,26]) did not substantially alter the conclusions about the value of the interventions (Figure 6A–6C) compared with the baseline scenario (Figure 5C and 5D), although if this assumption was used when fitting the model to the 1968/9 data the estimated value of R max was reduced from about 1.8 to 1.5. Conclusions were also robust to moderate variation in the distribution of the latent period (Figure 6D–6F). If, however, the virus behaved more like the SARS coronavirus, with extended latent and infectious periods (Figure 6G–6I), a greatly delayed rate of global spread could be expected, giving more chance of delaying epidemics until the next influenza season. In this case, smaller reductions in travel and transmission can achieve clinically significant delays (6 mo or more) in epidemic take-off in many cities. Assuming reduced transmission in the tropics (Figure 6J–6L) also led to a substantial reduction in the rate of global dissemination. Under this scenario much smaller reductions in transmission would be sufficient to greatly reduce the chance of a pandemic; this happens because the lower transmission in the tropics (where the virus is assumed to originate) means that a further transmission reduction of just 21% would be sufficient to make sustained spread impossible in this region. Discussion The relative ineffectiveness of travel restrictions for controlling pandemic influenza is a consequence of the rapid initial rate of growth of the epidemic in each city and the large number of people infected. For example, with a serial interval of 3 d, ignoring depletion of susceptibles, an R t of two would cause a 128-fold increase in new cases within 21 d (128 = 221/3). This means that if travel from the first affected city was restricted to 1/128 of its former value on (and after) day 1, there would be approximately the same number of influenza cases leaving the city on day 21 + t as there would have been on day t had there been no intervention; even such an extreme intervention would therefore buy only about 3 wk. The highly connected nature of the air travel network prevents such minor delays between pairs of cities combining into substantial delays over the whole network. Hufnagel et al. [20] used a related model to study the global spread of SARS. Although this model differed in important respects from the one used here (the implicit assumptions that air travel frequency varies with neither infection state nor country would not be tenable in the context of pandemic influenza), the conclusion that “remarkable success [in SARS epidemic control] is guaranteed if the largest cities are isolated in response to an outbreak” might, at first sight, be thought to apply equally to influenza. In fact, pandemic influenza is expected to have a much shorter serial interval than SARS, and delays in international spread that could be achieved by restricting almost all travel would be far more modest. Even if 99.9% of all travel could be stopped, epidemics in most cities would be delayed by no more than 4 mo. Moreover, the conclusion that a policy of isolating only the largest cities would guarantee success implicitly assumes that closing major airports would cause infected individuals who would have travelled through them to abandon their journeys rather than seek alternative routes, and that disease spread by routes other than air travel can be ignored without substantially altering the conclusions. This seems rather implausible, and for these reasons we think that the conclusions of Hufnagel and colleagues, while of undoubted theoretical interest, would be misleading if taken too literally. Large and important uncertainties abound in influenza epidemiology: We do not know whether or not a significant proportion of transmission occurs before the onset of symptoms or whether subclinical infections are an important source of transmission, and we know very little about the determinants of seasonality [24,27,28]. In evaluating the potential to delay the spread of influenza by restricting travel and reducing transmission, we have systematically adopted optimistic assumptions, chosen to give the interventions the greatest chance of success. Thus we have assumed that seasonal effects are important (delaying the rate of spread outside the influenza season), and that asymptomatic cases do not contribute to transmission (minimizing the numbers capable of spreading the virus, and maximising the chance of detecting them); we have ignored travel that is not by air and not between major airports; and we have ignored the possibility of transmission during flights themselves. Despite these optimistic assumptions we found that even large and widely enforced travel restrictions would usually delay epidemic peaks by only a few days; to have a major impact, restrictions would have to be almost total and almost instantaneous. Only if a pandemic strain were considerably less transmissible, or had a considerably longer serial interval than influenza strains seen in the past, or if very few people were initially susceptible, would such measures be likely to have an important impact on the rate of pandemic spread. Local control measures able to reduce influenza transmission were found to have greater potential for reducing the rate of global spread (they could also substantially reduce the total number of cases, although an evaluation of this benefit is beyond the scope of this paper). Under most plausible scenarios, however, delays would still fall far short of those required to produce large quantities of vaccine unless they were implemented early and able to reduce R t to close to one. Elsewhere it has been shown that airport entry screening would be unlikely to detect more than 10% of passengers latently infected with influenza when boarding [29]. The results in this paper show that such an intervention would have a negligible impact on the course of a pandemic once it was underway. The results also raise interesting questions about the importance of seasonality in influenza transmission. The evidence for strong seasonal effects in temperate regions found here with 1968/9 data is supported by a recent analysis of interpandemic influenza [30]. However, it is not clear how important such seasonal effects have been in previous pandemics, nor is it clear why a much better model fit should be obtained when transmission in the tropics is assumed to be the maximum (rather than the mean) of that in temperate regions. Indeed, a fuller understanding of the determinants of seasonal effects and their variation with latitude remains one of the outstanding problems of influenza epidemiology [24,27,28]. Recent models of pandemic influenza have accounted for household and social contact patterns [10,11]. While such details are needed for evaluating the possibility of containment at source, they would not be expected to affect the broad conclusions presented here. However, for a given R 0, assuming nonhomogeneous local mixing patterns would result in a somewhat reduced attack rate and rate of spread within each city, causing a slight decrease in the rate of global spread. For this reason, estimates of R max based on fitting models that assume homogeneous local mixing to pandemic data may underestimate the true value. A new pandemic strain might not show the same pattern of seasonality as in 1968/9 and could potentially have greater transmissibility than strains seen previously. Both SARS and smallpox transmission can be greatly amplified by nosocomial spread [31,32]; a similar amplification effect could occur with an unusually virulent influenza virus that led to many hospitalisations. In these more pessimistic scenarios, even more heroic efforts would be required to have any chance of significantly delaying the virus's spread by restricting travel. The results here suggest that resources might be better directed at reducing transmission locally and at attempting to control outbreaks during the earliest stages of sustained human-to-human spread, when movement restrictions are likely to be a more valuable containment measure [10,11] Supporting Information Protocol S1 Detailed Description of the Model (65 KB DOC) Click here for additional data file. Editors' Summary Background. Most people who get influenza (flu) recover quickly, although it can cause serious illness and death, most often in the elderly. Sometimes a new type of flu virus appears that is much more likely to kill. This happened, for example, in 1918, when a worldwide flu pandemic killed between 20 million and 100 million people. Recently, there have been concerns about a flu virus that affects birds, and often kills them. At present the virus does not pass easily from birds to humans, and it does not seem to pass from one human to another. However, the fear is that this virus might change and that human-to-human infection could then be possible. Should all this happen, the changed virus would be a major threat to human health. With current technology, it would take several months to produce enough vaccine for even a small proportion of the world's population. By that time, it would probably be too late; the virus would already have spread to most parts of the world. It is therefore important for health authorities to consider all the methods that might control the spread of the virus. With the increase in international travel that has taken place, the virus could spread more quickly than in previous worldwide pandemics. Restrictions on international travel might, therefore, be considered necessary, particularly travel by air. Why Was This Study Done? It is important to estimate how useful restrictions on air travel might be in controlling the spread of a flu virus. Travel restrictions are usually unpopular and could themselves be harmful, and, if they are not effective, resources could be wasted on enforcing them. What Did the Researchers Do and Find? This research involved mathematical modelling. In other words, complex calculations were done using information that is already available about how flu viruses spread, particularly information recorded during a worldwide flu outbreak in 1968–1969. Using this information, virtual experiments were carried out by simulating worldwide outbreaks on a computer. The researchers looked at how the virus might spread from one city to another and how travel restrictions might reduce the rate of spread. Their calculations allowed for such factors as the time of the year, the number of air passengers who might travel between the cities, and the fact that some people are more resistant to infection than others. From the use of their mathematical model, the researchers concluded that restrictions on air travel would achieve very little. This is probably because, compared with some other viruses, the flu virus is transmitted from one person to another very quickly and affects many people. Once a major outbreak was under way, banning flights from affected cities would be effective at significantly delaying worldwide spread only if almost all travel between cities could be stopped almost as soon as an outbreak was detected in each city. It would be more effective to take other measures that would control the spread of the virus locally. These measures could include use of vaccines and antiviral drugs if they were available and effective against the virus. Additional Information Please access these Web sites via the online version of this summary at http://dx.doi.org/10.1371/journal.pmed.0030212. • Fact sheets are available about various aspects of flu from the Web site of the World Health Organization, which takes a global overview of the impact of the infection Many health Web sites aimed at patients provide basic information about flu. • US National Institute of Allergy and Infectious Diseases page about flu • National Institute of Allergy and Infectious Diseases fact sheet about cold and flu symptoms • US Centers for Disease Control and Prevention page about flu • The Journal of the American Medical Association's patient page about influenza • Page on flu from BBC Health • Information about pandemic influenza from The Health Protection Agency
Author and article information
Journal
Journal ID (nlm-ta): Influenza Other Respir Viruses
Journal ID (iso-abbrev): Influenza Other Respir Viruses
Journal ID (doi): 10.1111/(ISSN)1750-2659
Journal ID (publisher-id): IRV
Title:
Influenza and Other Respiratory Viruses
Publisher:
Blackwell Publishing Ltd
(Oxford, UK
)
ISSN
(Print):
1750-2640
ISSN
(Electronic):
1750-2659
Publication date
(Electronic):
09
August
2012
Publication date
(Print):
May
2013
Volume: 7
Issue: 3
,
Part 2 A(H1N1)pdm09 Papers
(
doiID:
10.1111/irv.2013.7.issue-3
)
Pages: 367-382
Affiliations
Author notes
[*]Benjamin J. Cowling, School of Public Health, Li Ka Shing Faculty of Medicine, The
University of Hong Kong, Units 624‐7, Cyberport 3, Pokfulam, Hong Kong. E‐mail:
bcowling@
123456hku.hk
Article
Publisher ID: IRV420DOI: 10.1111/j.1750-2659.2012.00420.x
PMC ID: 3705741
PubMed ID: 22883352
SO-VID: 78703911-6f82-41c4-b227-c56825328ac4
Copyright © © 2012 Blackwell Publishing Ltd
History
Page count
Figures: 2,
Tables: 5,
Pages: 16
Categories
Subject:
Part 2 A(H1N1)pdm09 Papers
Subject:
Review Article
Subject:
Review Article
Custom metadata
source-schema-version-number 2.0
cover-date May 2013
details-of-publishers-convertor Converter:WILEY_ML3GV2_TO_NLMPMC version:4.9.1 mode:remove_FC converted:12.07.2016
ScienceOpen disciplines: Infectious disease & Microbiology
Data availability:
ScienceOpen disciplines: Infectious disease & Microbiology