Influenza-like illnesses (ILIs) are caused by several respiratory pathogens. These pathogens show weak to strong seasonal activity implying seasonality in ILI consultations. In this paper, the contribution of pathogens to seasonality of ILI consultations was statistically modelled. Virological count data were first smoothed using modulation models for seasonal time series. Second, Poisson regression was used regressing ILI consultation counts on the smoothed time series. Using ratios of the estimated regression parameters, relative measures of the underreporting of pathogens were obtained. Influenza viruses A and B, parainfluenza virus and respiratory syncytial virus (RSV) significantly contributed to explain the seasonal variation in ILI consultations. We also found that RSV was the least and influenza virus A is the most underreported pathogen in Belgian laboratory surveillance. The proposed methods and results are helpful in interpreting the data of clinical and laboratory surveillance, which are the essential parts of influenza surveillance.
Infectious disease epidemiology; influenza; statistics; surveillance system
The European Paediatric Influenza Analysis (EPIA) project is a multi-country project that was created to collect, analyse and present data regarding the paediatric influenza burden in European countries, with the purpose of providing the necessary information to make evidence-based decisions regarding influenza immunisation recommendations for children. The initial approach taken is based on existing weekly virological and age-specific influenza-like illness (ILI) data from surveillance networks across Europe. We use a multiple regression model guided by longitudinal weekly patterns of influenza virus to attribute the weekly ILI consultation incidence pattern to each influenza (sub)type, while controlling for the effect of respiratory syncytial virus (RSV) epidemics. Modelling the ILI consultation incidence during 2002/2003–2008 revealed that influenza infections that presented for medical attention as ILI affected between 0.3% and 9.8% of children aged 0–4 and 5–14 years in England, Italy, The Netherlands and Spain in an average season. With the exception of Spain, these rates were always higher in children aged 0–4 years. Across the six seasons analysed (five seasons were analysed from the Italian data), the model attributed 47–83% of the ILI burden in primary care to influenza virus infection in the various countries, with the A(H3N2) virus playing the most important role, followed by influenza viruses B and A(H1N1). National season averages from the four countries studied indicated that between 0.4% and 18% of children consulted a physician for ILI, with the percentage depending on the country and health care system. Influenza virus infections explained the majority of paediatric ILI consultations in all countries. The next step will be to apply the EPIA modelling approach to severe outcomes indicators (i.e. hospitalisations and mortality data) to generate a complete range of mild and severe influenza burden estimates needed for decision making concerning paediatric influenza vaccination.
Influenza; Children; Paediatric; Burden of disease; Modelling; Europe
We consider inference in randomized longitudinal studies with missing data that is generated by skipped clinic visits and loss to follow-up. In this setting, it is well known that full data estimands are not identified unless unverified assumptions are imposed. We assume a non-future dependence model for the drop-out mechanism and partial ignorability for the intermittent missingness. We posit an exponential tilt model that links non-identifiable distributions and distributions identified under partial ignorability. This exponential tilt model is indexed by non-identified parameters, which are assumed to have an informative prior distribution, elicited from subject-matter experts. Under this model, full data estimands are shown to be expressed as functionals of the distribution of the observed data. To avoid the curse of dimensionality, we model the distribution of the observed data using a Bayesian shrinkage model. In a simulation study, we compare our approach to a fully parametric and a fully saturated model for the distribution of the observed data. Our methodology is motivated by, and applied to, data from the Breast Cancer Prevention Trial.
Informative drop-out; Prior elicitation; Intermittent missingness
The diagnosis of influenza is often missed in older adults and illness presentation may be modified by prior vaccination. We evaluated the symptoms and immunologic markers predicting laboratory-confirmed influenza (LCI) among vaccinated older adults. In subjects with influenza-like illness (ILI), fever distinguished subjects with laboratory-confirmed influenza (LCI) from those with other ILI (39% vs. 12.5%, p=0.009). In LCI subjects who did not seroconvert to influenza infection, pre-infection levels of the cytolytic mediator, granzyme B, correlated with fever (r=1.00; p=0.01) and the IFN-γ:IL-10 ratio (r=0.99; p=0.03), and increased following influenza infection in LCI vs. ILI subjects (p=0.03). The cell-mediated immune response to influenza distinguishes A/H3N2 LCI from other ILI in older adults, and suggests a link between cell-mediated immunity and influenza illness severity in vaccinated older adults.
influenza; granzyme B; interferon-γ; interleukin-10; elderly
An Approximate Bayesian Bootstrap (ABB) offers advantages in incorporating appropriate uncertainty when imputing missing data, but most implementations of the ABB have lacked the ability to handle nonignorable missing data where the probability of missingness depends on unobserved values. This paper outlines a strategy for using an ABB to multiply impute nonignorable missing data. The method allows the user to draw inferences and perform sensitivity analyses when the missing data mechanism cannot automatically be assumed to be ignorable. Results from imputing missing values in a longitudinal depression treatment trial as well as a simulation study are presented to demonstrate the method’s performance. We show that a procedure that uses a different type of ABB for each imputed data set accounts for appropriate uncertainty and provides nominal coverage.
Not Missing at Random; NMAR; Multiple Imputation; Hot-Deck
Limited vaccine availability and the potential for resistance to antiviral medications have led to calls for establishing the efficacy of non-pharmaceutical measures for mitigating pandemic influenza. Our objective was to examine if the use of face masks and hand hygiene reduced rates of influenza-like illness (ILI) and laboratory-confirmed influenza in the natural setting. A cluster-randomized intervention trial was designed involving 1,178 young adults living in 37 residence houses in 5 university residence halls during the 2007–2008 influenza season. Participants were assigned to face mask and hand hygiene, face mask only, or control group during the study. Discrete-time survival models using generalized estimating equations to estimate intervention effects on ILI and confirmed influenza A/B infection over a 6-week study period were examined. A significant reduction in the rate of ILI was observed in weeks 3 through 6 of the study, with a maximum reduction of 75% during the final study week (rate ratio [RR] = 0.25, [95% CI, 0.07 to 0.87]). Both intervention groups compared to the control showed cumulative reductions in rates of influenza over the study period, although results did not reach statistical significance. Generalizability limited to similar settings and age groups. Face masks and hand hygiene combined may reduce the rate of ILI and confirmed influenza in community settings. These non-pharmaceutical measures should be recommended in crowded settings at the start of an influenza pandemic.
To use laboratory data to assess the specificity of syndromes used by the New York City emergency department (ED) syndromic surveillance system to monitor influenza activity.
For the period from October 1, 2009 through March 31, 2010, we examined the correlation between citywide ED syndrome assignment and laboratory-confirmed influenza and respiratory syncytial virus (RSV). In addition, ED syndromic data from five select NYC hospitals were matched at the patient and visit level to corresponding laboratory reports of influenza and RSV. The matched dataset was used to evaluate syndrome assignment by disease and to calculate the sensitivity and specificity of the influenza-like illness (ILI) syndrome.
Citywide ED visits for ILI correlated well with influenza laboratory diagnoses (R=0.92). From October 1, 2009, through March 31, 2010, there were 264,532 ED visits at the five select hospitals, from which the NYC Department of Health and Mental Hygiene (DOHMH) received confirmatory laboratory reports of 655 unique cases of influenza and 1348 cases of RSV. The ED visit of most (56%) influenza cases had been categorized in the fever/flu syndrome; only 15% were labeled ILI. Compared to other influenza-related syndromes, ILI had the lowest sensitivity (15%) but the highest specificity (90%) for laboratory-confirmed influenza. Sensitivity and specificity varied by age group and influenza activity level.
The ILI syndrome in the NYC ED syndromic surveillance system served as a specific but not sensitive indicator for influenza during the 2009-2010 influenza season. Despite its limited sensitivity, the ILI syndrome can be more informative for tracking influenza trends than the fever/flu or respiratory syndromes because it is less likely to capture cases of other respiratory viruses. However, ED ILI among specific age groups should be interpreted alongside laboratory surveillance data. ILI remains a valuable tool for monitoring influenza activity and trends as it facilitates comparisons nationally and across jurisdictions and is easily communicated to the public.
Aims: To estimate excess morbidity during periods of influenza and respiratory syncytial virus (RSV) activity.
Methods: Retrospective analysis of a sentinel practice network database in active and non-active virus periods. Main outcome measures: clinical diagnoses of new episodes of influenza-like illness (ILI), acute bronchitis, asthma, and otitis media.
Results: The clinical diagnosis of ILI was consistent with influenza virus activity and acute bronchitis with RSV. During periods of virus activity, estimates of excess morbidity in children aged 1–4 and 5–14 years diagnosed as having acute otitis media exceeded those diagnosed with each of the other three conditions; in children <1 year estimates for acute bronchitis were highest. Using a broad definition of virus activity and summarising the data for all children diagnosed with ILI, 60% was attributable to influenza (40% RSV) as were 37% of episodes diagnosed as acute bronchitis, 9% of those with asthma and 48% of those with otitis media. Using a narrow definition, corresponding proportions were: for ILI diagnoses 77% (23% RSV), acute bronchitis 32%, asthma zero, and otitis media 45%. Acute bronchitis was diagnosed twice as frequently in association with RSV as with influenza in all age groups: excess asthma episodes were only evident in RSV active periods.
Conclusions: Except in relation to ILI, RSV caused more illness than the influenza virus in the respiratory diagnoses examined, emphasising the need for RSV prevention and treatment. Influenza was not associated with excess asthma episodes.
Influenza is a human pathogen that continues to pose a public health threat. The use of small mammalian models has become indispensable for understanding the virulence of influenza viruses. Among numerous species used in the laboratory setting, only the ferret model is equally well suited for studying both the pathogenicity and transmissibility of human and avian influenza viruses. Here, we compare the advantages and limitations of the mouse, ferret and guinea pig models for research with influenza A viruses, emphasizing the multifarious uses of the ferret in the assessment of influenza viruses with pandemic potential. Research performed in the ferret model has provided information, support and guidance for the public health response to influenza viruses in humans. We highlight the recent and emerging uses of this species in influenza virus research that are advancing our understanding of virus-host interactions.
Syndromic surveillance is increasingly being evaluated for its potential for early warning of increased disease activity in the population. However, interpretation is hampered by the difficulty of attributing a causative pathogen. We described the temporal relationship between laboratory counts of influenza and respiratory syncytial virus (RSV) detection and alternative groupings of Emergency Department (ED) respiratory diagnoses.
ED and laboratory data were obtained for the south-eastern area of Sydney, NSW for the period 1 June 2001 - 1 December 2006. Counts of ED visits and laboratory confirmed positive RSV and influenza cases were aggregated by week. Semi-parametric generalized additive models (GAM) were used to determine the association between the incidence of RSV and influenza and the incidence of respiratory syndrome ED presentations while controlling for temporal confounders.
For every additional RSV laboratory count, ED diagnoses of bronchiolitis increased by 3.1% (95%CI: 2.7%-3.5%) in the same week. For every additional influenza laboratory count, ED diagnoses of influenza-like illness increased by 4.7% (95%CI: 4.2%-5.2%) one week earlier.
In this study, large increases in ED diagnoses of bronchiolitis and influenza-like illness were independent and proxy indicators for RSV and influenza activity, respectively.
A routine challenge is that of making inference on parameters in a statistical model of interest from longitudinal data subject to drop out, which are a special case of the more general setting of monotonely coarsened data. Considerable recent attention has focused on doubly robust estimators, which in this context involve positing models for both the missingness (more generally, coarsening) mechanism and aspects of the distribution of the full data, that have the appealing property of yielding consistent inferences if only one of these models is correctly specified. Doubly robust estimators have been criticized for potentially disastrous performance when both of these models are even only mildly misspecified. We propose a doubly robust estimator applicable in general monotone coarsening problems that achieves comparable or improved performance relative to existing doubly robust methods, which we demonstrate via simulation studies and by application to data from an AIDS clinical trial.
Coarsening at random; Discrete hazard; Dropout; Longitudinal data; Missing at random
Longitudinal studies with binary repeated measures are widespread in biomedical research. Marginal regression approaches for balanced binary data are well developed, while for binary process data, where measurement times are irregular and may differ by individuals, likelihood-based methods for marginal regression analysis are less well developed. In this article, we develop a Bayesian regression model for analyzing longitudinal binary process data, with emphasis on dealing with missingness. We focus on the settings where data are missing at random, which require a correctly specified joint distribution for the repeated measures in order to draw valid likelihood-based inference about the marginal mean. To provide maximum flexibility, the proposed model specifies both the marginal mean and serial dependence structures using nonparametric smooth functions. Serial dependence is allowed to depend on the time lag between adjacent outcomes as well as other relevant covariates. Inference is fully Bayesian. Using simulations, we show that adequate modeling of the serial dependence structure is necessary for valid inference of the marginal mean when the binary process data are missing at random. Longitudinal viral load data from the HIV Epidemiology Research Study (HERS) are analyzed for illustration.
Repeated measures; Marginal model; Nonparametric regression; Penalized splines; HIV/AIDS; Antiviral treatment
Nonparametric Bayesian techniques have been developed recently to extend the sophistication of factor models, allowing one to infer the number of appropriate factors from the observed data. We consider such techniques for sparse factor analysis, with application to gene-expression data from three virus challenge studies. Particular attention is placed on employing the Beta Process (BP), the Indian Buffet Process (IBP), and related sparseness-promoting techniques to infer a proper number of factors. The posterior density function on the model parameters is computed using Gibbs sampling and variational Bayesian (VB) analysis.
Time-evolving gene-expression data are considered for respiratory syncytial virus (RSV), Rhino virus, and influenza, using blood samples from healthy human subjects. These data were acquired in three challenge studies, each executed after receiving institutional review board (IRB) approval from Duke University. Comparisons are made between several alternative means of per-forming nonparametric factor analysis on these data, with comparisons as well to sparse-PCA and Penalized Matrix Decomposition (PMD), closely related non-Bayesian approaches.
Applying the Beta Process to the factor scores, or to the singular values of a pseudo-SVD construction, the proposed algorithms infer the number of factors in gene-expression data. For real data the "true" number of factors is unknown; in our simulations we consider a range of noise variances, and the proposed Bayesian models inferred the number of factors accurately relative to other methods in the literature, such as sparse-PCA and PMD. We have also identified a "pan-viral" factor of importance for each of the three viruses considered in this study. We have identified a set of genes associated with this pan-viral factor, of interest for early detection of such viruses based upon the host response, as quantified via gene-expression data.
Health authorities find thresholds useful to gauge the start and severity of influenza seasons. We explored a method for deriving thresholds proposed in an influenza surveillance manual published by the World Health Organization (WHO).
For 2002-2011, we analysed two routine influenza-like-illness (ILI) datasets, general practice sentinel surveillance and a locum medical service sentinel surveillance, plus laboratory data and hospital admissions for influenza. For each sentinel dataset, we created two composite variables from the product of weekly ILI data and the relevant laboratory data, indicating the proportion of tested specimens that were positive. For all datasets, including the composite datasets, we aligned data on the median week of peak influenza or ILI activity and assigned three threshold levels: seasonal threshold, determined by inspection; and two intensity thresholds termed average and alert thresholds, determined by calculations of means, medians, confidence intervals (CI) and percentiles. From the thresholds, we compared the seasonal onset, end and intensity across all datasets from 2002-2011. Correlation between datasets was assessed using the mean correlation coefficient.
The median week of peak activity was week 34 for all datasets, except hospital data (week 35). Means and medians were comparable and the 90% upper CIs were similar to the 95th percentiles. Comparison of thresholds revealed variations in defining the start of a season but good agreement in describing the end and intensity of influenza seasons, except in hospital admissions data after the pandemic year of 2009. The composite variables improved the agreements between the ILI and other datasets. Datasets were well correlated, with mean correlation coefficients of >0.75 for a range of combinations.
Thresholds for influenza surveillance are easily derived from historical surveillance and laboratory data using the approach proposed by WHO. Use of composite variables is helpful for describing influenza season characteristics.
Despite considerable research efforts, little is yet known about key epidemiological parameters of H5N1 highly pathogenic influenza viruses in their avian hosts. Here we show how these parameters can be estimated using a limited number of birds in experimental transmission studies. Our quantitative estimates, based on Bayesian methods of inference, reveal that (i) the period of latency of H5N1 influenza virus in unvaccinated chickens is short (mean: 0.24 days; 95% credible interval: 0.099–0.48 days); (ii) the infectious period of H5N1 virus in unvaccinated chickens is approximately 2 days (mean: 2.1 days; 95%CI: 1.8–2.3 days); (iii) the reproduction number of H5N1 virus in unvaccinated chickens need not be high (mean: 1.6; 95%CI: 0.90–2.5), although the virus is expected to spread rapidly because it has a short generation interval in unvaccinated chickens (mean: 1.3 days; 95%CI: 1.0–1.5 days); and (iv) vaccination with genetically and antigenically distant H5N2 vaccines can effectively halt transmission. Simulations based on the estimated parameters indicate that herd immunity may be obtained if at least 80% of chickens in a flock are vaccinated. We discuss the implications for the control of H5N1 avian influenza virus in areas where it is endemic.
Outbreaks of highly pathogenic H5N1 avian influenza in poultry first occurred in China in 1996. Since that time, the virus has become endemic in Asia, and has been the cause of outbreaks in Africa and Europe. Although many aspects of H5N1 virus biology have been studied in detail, surprisingly little is known about the key epidemiological parameters of the virus in its avian hosts (the length of time from infection until a bird becomes infectious, the duration of infectiousness, how many birds each infectious bird will infect). In this paper we show, using experimental transmission studies with unvaccinated and vaccinated chickens, that H5N1 avian influenza induces a short duration of infectiousness (∼2 days) and a very short period of time from infection until infectiousness (∼0.25 day) in unvaccinated chickens. Furthermore, while transmission was efficient among unvaccinated birds, no bird-to-bird transmission was observed in vaccinated chickens. Our results indicate that it may be difficult to curb outbreaks by vaccination after an introduction in a flock has been detected. On the other hand, preventive vaccination could be effective in preventing virus introductions and limiting the size of outbreaks.
Influenza viruses cause significant morbidity and mortality in adults each winter. At the same time, other respiratory viruses circulate and cause respiratory illness with flu-like symptoms. Respiratory syncytial virus (RSV), the human parainfluenza viruses (HPIV), and human metapneumovirus (HMPV) have all been associated with morbidity and mortality in adults, including nosocomial infections. We evaluated 154 respiratory specimens collected from adults with influenza-like/acute respiratory illness (ILI) seen at the Edward Hines Veteran’s Affairs (VA) Hospital during two successive winters, 1998-1999 and 1999-2000. The samples were tested for 10 viruses in two nested multiplex RT-PCR reactions. One to three respiratory viruses were detected in 68% of the samples. As expected, influenza A virus (Flu-A) infections were most common (50% of the samples), followed by RSV-A (16%). Surprisingly, HPIV-4 infections (5.8%) were the third most prevalent. Mixed infections were also relatively common (11%). When present, HPIV infections were approximately three times more likely to be included in a mixed infection than Flu-A or RSV. Mixed infections and HPIV-4 are likely to be missed using rapid diagnostic tests. This study confirms that ILI in adults and the elderly can be caused by RSV and the HPIVs, including HPIV-4, which co-circulate with Flu-A.
Semicontinuous data in the form of a mixture of zeros and continuously distributed positive values frequently arise in biomedical research. Two-part mixed models with correlated random effects are an attractive approach to characterize the complex structure of longitudinal semicontinuous data. In practice, however, an independence assumption about random effects in these models may often be made for convenience and computational feasibility. In this article, we show that bias can be induced for regression coefficients when random effects are truly correlated but misspecified as independent in a 2-part mixed model. Paralleling work on bias under nonignorable missingness within a shared parameter model, we derive and investigate the asymptotic bias in selected settings for misspecified 2-part mixed models. The performance of these models in practice is further evaluated using Monte Carlo simulations. Additionally, the potential bias is investigated when artificial zeros, due to left censoring from some detection or measuring limit, are incorporated. To illustrate, we fit different 2-part mixed models to the data from the University of Toronto Psoriatic Arthritis Clinic, the aim being to examine whether there are differential effects of disease activity and damage on physical functioning as measured by the health assessment questionnaire scores over the course of psoriatic arthritis. Some practical issues on variance component estimation revealed through this data analysis are considered.
Correlated random effects; Excess zeros; Outcome-dependent sampling; Repeated measures
Longitudinal data often contain missing observations and error-prone covariates. Extensive attention has been directed to analysis methods to adjust for the bias induced by missing observations. There is relatively little work on investigating the effects of covariate measurement error on estimation of the response parameters, especially on simultaneously accounting for the biases induced by both missing values and mismeasured covariates. It is not clear what the impact of ignoring measurement error is when analyzing longitudinal data with both missing observations and error-prone covariates. In this article, we study the effects of covariate measurement error on estimation of the response parameters for longitudinal studies. We develop an inference method that adjusts for the biases induced by measurement error as well as by missingness. The proposed method does not require the full specification of the distribution of the response vector but only requires modeling its mean and variance structures. Furthermore, the proposed method employs the so-called functional modeling strategy to handle the covariate process, with the distribution of covariates left unspecified. These features, plus the simplicity of implementation, make the proposed method very attractive. In this paper, we establish the asymptotic properties for the resulting estimators. With the proposed method, we conduct sensitivity analyses on a cohort data set arising from the Framingham Heart Study. Simulation studies are carried out to evaluate the impact of ignoring covariate measurement error and to assess the performance of the proposed method.
Estimating equations; Longitudinal data; Measurement error; Missing data; Simulation and extrapolation method
Schools are important foci of influenza transmission and potential targets for surveillance and interventions. We compared several school-based influenza monitoring systems with clinic-based influenza-like illness (ILI) surveillance, and assessed the variation in illness rates between and within schools.
During the initial wave of pandemic H1N1 (pdmH1N1) infections from June to Sept 2009 in Singapore, we collected data on nation-wide laboratory confirmed cases (Sch-LCC) and daily temperature monitoring (Sch-DTM), and teacher-led febrile respiratory illness reporting in 6 sentinel schools (Sch-FRI). Comparisons were made against age-stratified clinic-based influenza-like illness (ILI) data from 23 primary care clinics (GP-ILI) and proportions of ILI testing positive for pdmH1N1 (Lab-ILI) by computing the fraction of cumulative incidence occurring by epidemiological week 30 (when GP-ILI incidence peaked); and cumulative incidence rates between school-based indicators and sero-epidemiological pdmH1N1 incidence (estimated from changes in prevalence of A/California/7/2009 H1N1 hemagglutination inhibition titers ≥ 40 between pre-epidemic and post-epidemic sera). Variation in Sch-FRI rates in the 6 schools was also investigated through a Bayesian hierarchical model.
By week 30, for primary and secondary school children respectively, 63% and 79% of incidence for Sch-LCC had occurred, compared with 50% and 52% for GP-ILI data, and 48% and 53% for Sch-FRI. There were 1,187 notified cases and 7,588 episodes in the Sch-LCC and Sch-DTM systems; given school enrollment of 485,723 children, this represented 0.24 cases and 1.6 episodes per 100 children respectively. Mean Sch-FRI rate was 28.8 per 100 children (95% CI: 27.7 to 29.9) in the 6 schools. We estimate from serology that 41.8% (95% CI: 30.2% to 55.9%) of primary and 43.2% (95% CI: 28.2% to 60.8%) of secondary school-aged children were infected. Sch-FRI rates were similar across the 6 schools (23 to 34 episodes per 100 children), but there was widespread variation by classrooms; in the hierarchical model, omitting age and school effects was inconsequential but neglecting classroom level effects led to highly significant reductions in goodness of fit.
Epidemic curves from Sch-FRI were comparable to GP-ILI data, and Sch-FRI detected substantially more infections than Sch-LCC and Sch-DTM. Variability in classroom attack rates suggests localized class-room transmission.
Respiratory tract infections; Vaccination; Serology
Public health surveillance systems need to be refined. We intend to use a generic approach for early identification of patients with severe influenza-like illness (ILI) by calculating a score that estimates a patient’s disease-severity. Accordingly, we built the Intelligent Severity Score Estimation Model (ISSEM), structured so that the inference process would reflect experts’ decision-making logic. Each patient’s disease-severity score is calculated from numbers of respiratory ICD9 encounters, and laboratory, radiologic, and prescription-therapeutic orders in the EMR. Other ISSEM components include chronic disease evidence, probability of immunodeficiency, and the provider’s general practice-behavior patterns.
Sensitivity was determined from 200 randomly selected patients with upper- and lower-respiratory tract ILI; specificity, from 300 randomly selected patients with URI only. For different age groups, ISSEM sensitivity ranged between 90% and 95%; specificity was 72% to 84%.
Our preliminary assessment of ISSEM performance demonstrated 93.5% sensitivity and 77.3% specificity across all age groups.
We assessed the severity of the 2009 influenza pandemic by comparing pandemic mortality to seasonal influenza mortality. However, reported pandemic deaths were laboratory-confirmed – and thus an underestimation – whereas seasonal influenza mortality is often more inclusively estimated. For a valid comparison, our study used the same statistical methodology and data types to estimate pandemic and seasonal influenza mortality.
Methods and Findings
We used data on all-cause mortality (1999–2010, 100% coverage, 16.5 million Dutch population) and influenza-like-illness (ILI) incidence (0.8% coverage). Data was aggregated by week and age category. Using generalized estimating equation regression models, we attributed mortality to influenza by associating mortality with ILI-incidence, while adjusting for annual shifts in association. We also adjusted for respiratory syncytial virus, hot/cold weather, other seasonal factors and autocorrelation. For the 2009 pandemic season, we estimated 612 (range 266–958) influenza-attributed deaths; for seasonal influenza 1,956 (range 0–3,990). 15,845 years-of-life-lost were estimated for the pandemic; for an average seasonal epidemic 17,908. For 0–4 yrs of age the number of influenza-attributed deaths during the pandemic were higher than in any seasonal epidemic; 77 deaths (range 61–93) compared to 16 deaths (range 0–45). The ≥75 yrs of age showed a far below average number of deaths. Using pneumonia/influenza and respiratory/cardiovascular instead of all-cause deaths consistently resulted in relatively low total pandemic mortality, combined with high impact in the youngest age category.
The pandemic had an overall moderate impact on mortality compared to 10 preceding seasonal epidemics, with higher mortality in young children and low mortality in the elderly. This resulted in a total number of pandemic deaths far below the average for seasonal influenza, and a total number of years-of-life-lost somewhat below average. Comparing pandemic and seasonal influenza mortality as in our study will help assessing the worldwide impact of the 2009 pandemic.
To assess how weekly percent of influenza-like illness (ILI) reported via Early Notification of Community-based Epidemics (ESSENCE) tracked weekly counts of laboratory confirmed influenza cases in five influenza seasons in order to evaluate the early warning potential of ILI in ESSENCE and improve ongoing influenza surveillance efforts in Missouri.
Syndromic surveillance is used routinely to detect outbreaks of disease earlier than traditional methods due to its ability to automatically acquire data in near real-time. Missouri has used emergency department (ED) visits to monitor and track seasonal influenza activity since 2006.
The Missouri ESSENCE system utilizes data from 84 hospitals, which represents up to 90 percent of all ED visits occurring in Missouri statewide each day. The influenza season is defined as starting during Centers for Disease Control and Prevention (CDC) week number 40 (around the first of October) and ending on CDC week 20 of the following year, which is usually at the end of May.
A confirmed influenza case is laboratory confirmed by viral culture, rapid diagnostic tests, or a four-fold rise in antibody titer between acute and convalescent serum samples. Laboratory results are reported on a weekly basis. To assess the severity of influenza activity, all flu seasons were compared with the 2008–09 season, which experienced the lowest influenza activity based on laboratory data. Analysis of variance (ANOVA) was applied for this analysis using Statistical Analysis Software (SAS) (version 9.2).
The standard ESSENCE ILI subsyndrome includes ED chief complaints that contain keywords such as “flu”, “flulike”, “influenza” or “fever plus cough” or “fever plus sore throat”. The ESSENCE ILI weekly percent is the number of ILI visits divided by total ED visits.
Time series of weekly percent of ILI in ESSENCE were compared to weekly counts of laboratory confirmed influenza cases. Spearman correlation coefficients were calculated using SAS. The baseline refers to the mean of three flu seasons with low influenza activity (2006–07, 2008–09 and 2010–11 seasons). The threshold was calculated as this baseline plus three standard deviations.
The early warning potential of the ESSENCE weekly ILI percent was evaluated for five consecutive influenza seasons, beginning in 2006. This was accomplished by calculating the time lag between the first ESSENCE ILI warning versus the first lab confirmed influenza warning. A warning was identified if either lab confirmed case counts or weekly percent of ILI crossed over their respective baselines.
For each influenza season evaluated, weekly ILI rates reported via ESSENCE were significantly correlated with weekly counts of laboratory-confirmed influenza cases (Table 1). The baseline of ILI activity in ESSENCE was 1.8 ILI /100 ED visits/week and the threshold was set at 4.1 ILI visits per 100 ED visits/week. The ESSENCE ILI baseline provided, on average, two weeks of advanced warning for seasonal influenza activity. Figure 1 shows that two influenza seasons (2007–08 and 2009–10) were more severe than others examined based on the ESSENCE percent ILI threshold analysis, this result is consistent with the examination of severity of influenza activity based on lab confirmed influenza data (p<0.05).
The significant correlation between ILI surveillance in ESSENCE and laboratory confirmed influenza cases justifies the use of weekly ILI percent in ESSENCE to describe seasonal influenza activity. The ESSENCE ILI baseline and threshold provided advanced warning of influenza and allowed for the classification of influenza severity in the community.
ESSENCE; syndromic surveillance; influenza-like illness (ILI); baseline; threshold
When an outbreak of a novel pathogen occurs, some of the most pressing questions from a public-health point of view relate to its transmissibility, and the probabilities of different clinical outcomes following infection, to allow an informed response. Estimates of these quantities are often based on household data due to the high potential for transmission in this setting, but typically a rich spectrum of individual-level outcomes (from uninfected to serious illness) are simplified to binary data (infected or not). We address the added benefit from retaining the heterogeneous outcome information in the case of the 2009-10 influenza pandemic, which posed particular problems for estimation of key epidemiological characteristics due to its relatively mild nature and hence low case ascertainment rates.
We use mathematical models of within-household transmission and case ascertainment, together with Bayesian statistics to estimate transmission probabilities stratified by household size, the variability of infectiousness of cases, and a set of probabilities describing case ascertainment. This novel approach was applied to data we collected from the early "containment phase" stage of the epidemic in Birmingham, England. We also conducted a comprehensive review of studies of household transmission of influenza A(H1N1)pdm09.
We find large variability in the published estimates of within-household transmissibility of influenza A(H1N1)pdm09 in both model-based studies and those reporting secondary attack rates, finding that these estimates are very sensitive to how an infected case is defined. In particular, we find that reliance on laboratory confirmation alone underestimates the true number of cases, while utilising the heterogeneous range of outcomes (based on case definitions) for household infections allows a far more comprehensive pattern of transmission to be elucidated.
Differences in household sizes and how cases are defined could account for an appreciable proportion of the reported variability of within-household transmissibility of influenza A(H1N1)pdm09. Retaining and statistically analysing the full spectrum of individual-level outcomes (based on case definitions) rather than taking a potentially arbitrary threshold for infection, provides much-needed additional information. In a future pandemic, our approach could be used as a real-time analysis tool to infer the true number of cases, within-household transmission rates and levels of case ascertainment.
Influenza A(H1N1)pdm09; Household; Case ascertainment; Markov Chain Monte Carlo; Transmission dynamics
Using validation sets for outcomes can greatly improve the estimation of vaccine efficacy (VE) in the field (Halloran and Longini, 2001; Halloran and others, 2003). Most statistical methods for using validation sets rely on the assumption that outcomes on those with no cultures are missing at random (MAR). However, often the validation sets will not be chosen at random. For example, confirmational cultures are often done on people with influenza-like illness as part of routine influenza surveillance. VE estimates based on such non-MAR validation sets could be biased. Here we propose frequentist and Bayesian approaches for estimating VE in the presence of validation bias. Our work builds on the ideas of Rotnitzky and others (1998, 2001), Scharfstein and others (1999, 2003), and Robins and others (2000). Our methods require expert opinion about the nature of the validation selection bias. In a re-analysis of an influenza vaccine study, we found, using the beliefs of a flu expert, that within any plausible range of selection bias the VE estimate based on the validation sets is much higher than the point estimate using just the non-specific case definition. Our approach is generally applicable to studies with missing binary outcomes with categorical covariates.
Bayesian; Expert opinion; Identifiability; Influenza; Missing data; Selection model; Vaccine efficacy
In epidemics of infectious diseases such as influenza, an individual may have one of four possible final states: prior immune, escaped from infection, infected with symptoms, and infected asymptomatically. The exact state is often not observed. In addition, the unobserved transmission times of asymptomatic infections further complicate analysis. Under the assumption of missing at random, data-augmentation techniques can be used to integrate out such uncertainties. We adapt an importance-sampling-based Monte Carlo EM (MCEM) algorithm to the setting of an infectious disease transmitted in close contact groups. Assuming the independence between close contact groups, we propose a hybrid EM-MCEM algorithm that applies the MCEM or the traditional EM algorithms to each close contact group depending on the dimension of missing data in that group, and discuss the variance estimation for this practice. In addition, we propose a bootstrap approach to assess the total Monte Carlo error and factor that error into the variance estimation. The proposed methods are evaluated using simulation studies. We use the hybrid EM-MCEM algorithm to analyze two influenza epidemics in the late 1970s to assess the effects of age and pre-season antibody levels on the transmissibility and pathogenicity of the viruses.
Data augmentation; EM algorithm; Infectious disease; Missing data; Monte Carlo