Associations between common respiratory viruses and invasive group A streptococcal infection: A time‐series analysis

Abstract Background Invasive infections by group A Streptococcus (iGAS, Streptococcus pyogenes) have a winter seasonality which largely coincides with the season for influenza and other respiratory viruses. Influenza superinfections with GAS have been described to occur regularly and to show a severe clinical picture with high mortality. We aimed to study the extent to which influenza A and B viruses (IAV and IBV), respiratory syncytial virus (RSV) and rhinovirus circulation contribute to iGAS incidence and severity. Methods Time‐series regression models were built to explore the temporal associations between weekly laboratory counts of IAV, IBV, RSV and rhinovirus as independent variables and weekly counts of GAS disease notifications or laboratory GAS cultures as dependent variables. Results The weekly number of IAV detections showed a significant temporal association with the number of notifications of streptococcal toxic shock syndrome (STSS), a severe complication of iGAS. Depending on the season, up to 40% of all notified STSS cases was attributable to IAV circulation. Besides STSS, none of the other iGAS manifestations were associated with a respiratory virus. Conclusions Our study found an ecological temporal association between IAV and STSS, the most severe complication of iGAS. Future studies are needed to confirm this association and assess the possible preventability of STSS by influenza vaccination, especially in the age group 60 years and older.

infections. 3 The drivers of iGAS seasonality are thus far unexplained, but GAS superinfections on common respiratory viruses such as influenza have been reported. 4,5 Notably, during the 1918 influenza pandemic, streptococcal superinfections were important causes of death, which besides Streptococcus pneumoniae also included GAS. 6 In 2010-2011, an increased iGAS incidence was observed in the United Kingdom during a high influenza season. 7 Record linkage with influenza surveillance data showed that 9% of iGAS cases also had a laboratory-confirmed influenza infection, which is likely to be an underestimation of the true coinfection rate as testing for influenza infection is not common. Case series of influenza-iGAS co-infections have shown a severe clinical picture with mortality rates around 70% but are based on low numbers. 8,9 Therefore, the extent to which influenza contributes to iGAS seasonality or severity remains unknown. Importantly, in the case of such an association, a proportion of iGAS cases is potentially avertable by influenza prevention. Patients with iGAS disease are not usually tested for influenza infection, which complicates investigation of a link between the two infections. However, comprehensive surveillance systems both of respiratory pathogens and of iGAS disease separately offer an opportunity to explore the association between iGAS and respiratory pathogens through modelling of their time series. In the absence of individual-level data on co-infections, such time-series analyses using population-level data are an important tool for hypothesis generation and directing future in-depth research. While time series have been used to estimate the association between respiratory virus infections and infections with another Streptococcus species (Streptococcus pneumoniae), 10 such populationlevel studies with GAS infections are very scarce. One such study, by Allard et al, 11 reported a specific association between influenza B virus and GAS laboratory detections in Montréal, Canada.
Our research question was whether influenza (A or B) seasonality was associated with iGAS disease in the Netherlands, and if so, whether this association was specific to influenza or whether it existed for any respiratory virus with a winter season. We constructed time-series regression models to estimate the temporal, population-level associations between weekly respiratory virus counts (influenza A and B, RSV and rhinovirus) and weekly counts of iGAS disease notifications or microbiological GAS cultures.  [17][18][19][20][21] laboratories reporting weekly, and the data were shown to be representative of epidemiological trends on the national level in a recent evaluation comparing the VWR to mandatory notification data. 12 The VWR participating laboratories include smaller and larger laboratories plus several academic hospitals. A study conducted in 1999-2000 estimated the national population coverage of the VWR at 73% for influenza, with 17 participating laboratories at the time. 13 For the current study, detections of common respiratory viruses (influenza A virus (IAV), influenza B virus (IBV), respiratory syncytial virus (RSV) and rhinovirus) were included. sepsis. Since June 2016, also puerperal (ie, within 3 weeks post-partum) fever with a GAS culture from a non-sterile site (ie, culture from the female urogenital tract) is notifiable. Puerperal sepsis or puerperal fever will henceforth be referred to as puerperal GAS. STSS and necrotising fasciitis are notifiable if GAS is cultured from a normally sterile site, or when GAS is cultured from a normally non-sterile site, in the absence of another micro-organism that can cause the disease.

| iGAS disease notifications
Since notifiable by law, notifications of iGAS diseases STSS, necrotising fasciitis and puerperal GAS should include all cases occurring throughout the Netherlands, some underreporting notwithstanding.

| GAS cultures
Weekly numbers of GAS-positive cultures (week 1 2008-week 52 2017), by specimen type, were derived from the national Infectious Diseases Surveillance Information System-Antimicrobial Resistance (ISIS-AR). 14 This surveillance system is a combined initiative of the Dutch Ministry of Health, Welfare and Sport and the Netherlands Society for Medical Microbiology, and is maintained by the Centre for Infectious Disease Control at the National Institute for Public Health and the Environment and the participating medical microbiology laboratories (MMLs). All results of antimicrobial susceptibility testing (AST) from participating laboratories are uploaded in this surveillance system. As AST is routinely performed for all GAS isolates, we assumed counts of GAS isolates to be representative for the number of GAS diagnoses. However, to avoid overestimation of diagnoses by including follow-up cultures of the same disease episode, we included only isolates that were not preceded by a GAS isolate from the same patient in the 6 months before. Coverage of ISIS-AR has increased during the study period (2008-2017) from around 50% to 70% of the number of laboratories in the Netherlands; however, the exact catchment areas of the laboratories are unknown. Isolates from blood, upper and lower respiratory tract material, genital tract material and pus or wounds were separately included in the analysis.
The representativeness of the GAS culture data for GAS disease will vary per specimen type: culturing upper respiratory tract specimens for, for example, tonsillitis is likely to occur regularly, but is not a standard procedure. On the other hand, blood cultures are always performed in sepsis cases.

| Statistical models
We used time-series regression to estimate the temporal association between respiratory virus detections and both iGAS notifications and GAS cultures. Negative binomial models were constructed with weekly counts of respiratory viruses as independent variables and weekly counts of iGAS disease notifications (STSS, necrotising fasciitis or puerperal GAS) or GAS cultures (per specimen type) as dependent variables. Building a separate model per each dependent variable, we thus built eight models. For estimating the association between the weekly counts of respiratory virus detections and iGAS disease notifications, the period of week 1 2011 to week 26 2018 was included. The GAS culture data were available from week 1 2008 up to week 52 2017; therefore, this period was included in the GAS culture models. For ease of interpretation, we used an identity link, which assumes the effects of different respiratory viruses on GAS incidence to be additive rather than multiplicative. In model selection steps, P-values were used to determine whether a parameter was associated with the dependent variable. Parameters with P-values <0.1 were included in a stepwise model selection.
To allow a baseline seasonal variation in GAS, independent of respiratory virus activity, first a sine wave was modelled before including the respiratory virus variables. For each GAS outcome, initially a model was constructed with four sets of sine/cosine terms, and the number of significant sine/cosine pairs was determined by backward selection. Linear time trends were added only when significant. For all four respiratory viruses, we also explored whether they were significantly associated with GAS when a time lag was taken into account (ie, whether a current increase in virus detections was predictive of current GAS or GAS in later weeks). Time lags of zero up to 4 weeks were explored in this manner. Negative (inverse) associations between respiratory viruses and GAS were disregarded, as we assumed a protective effect of respiratory viruses on GAS disease to be biologically implausible.
If the number of IAV detections showed an overall significant association with GAS, we explored whether a time-dependent approach improved the model fit, that is separating IAV detections into

| RE SULTS
Weekly numbers of detections of respiratory viruses between January 2008 and June 2018 reflected the typical seasonality of respiratory viruses IAV, IBV and RSV, that is with clear annual peaks around winter and almost absent outside of their season (Figure 1).
Rhinovirus detections remained quite consistently present throughout the year. All GAS disease notifications and cultures also showed marked seasonality (Figures 2, 3), but retained a baseline presence throughout the year. Annual totals of respiratory virus detections, iGAS disease notifications and GAS isolates are shown in Table S1.

| Time-series regression
In all models, one set of sine/cosine terms (of 1-year periodicity) showed the best fit. A secular time trend improved the fit of (and was therefore included in) all models with GAS culture numbers as dependent variables, likely reflecting the increasing number of laboratories contributing to the data over the study period. Detections However, the number of IAV detections was significantly associated with STSS notifications in the same week (P = 0.007, Table   S2). When season-specific IAV variables were used, for three seasons the IAV P-value was below 0.1, and beta coefficients differed notably by season. The model with season-specific IAV variables was selected as final model (Table 1). Figure 4 shows the ob-    (Table 1), most likely reflecting such an IAV strain dependency.

| D ISCUSS I ON
A time-series regression similar to ours found influenza B rather than influenza A to be associated with iGAS detections in Montréal during 1996-2008. 11 Therefore, it was surprising that we did not find an association with influenza B. This might be due to different choices in model construction; the Montréal study incorporated six sine/cosine pairs, which either may be a better approximation of the baseline or may compete away a larger part of the seasonal influenza A association compared to our model. Our model, with one sine/cosine pair, also potentially overestimates the baseline at the expense of IAV. This is illustrated by season 2013-2014, when the number of STSS notifications fell below the baseline sine wave of our model, resulting in a negative beta coefficient for IAV (see Table 1 and Figure 4). tifications. The secular trend added to the puerperal GAS and ISIS-AR models crudely adjusts for these increases but may have left residual surveillance artefacts in the data. Also, the formal iGAS surveillance of notifiable disease presentations is partially limited in the Netherlands as national surveillance data on GAS pneumonia are absent (because not notifiable), while it is strongly associated with STSS. 21 IAV infection is known to predispose for bacterial pneumonia. 22 However, we did not observe an association between respiratory viruses and GAS cultures from lower respiratory tract specimens, which might be a proxy measure for GAS pneumonia. Strengths of this study include the multiple GAS outcomes (three disease presentations and five culture specimen types), which enabled the distinction between diseases with differing severity and of different organ systems. In addition, the inclusion of four respiratory viruses enabled the detection of a specific association with IAV, rather than a general co-seasonality of iGAS disease and respiratory virus infections in winter and early spring.
Our study supports the notion that iGAS burden may be reduced by influenza prevention. Indeed, a case-control study among US military found a strong protective effect of influenza vaccination on GAS disease. 23 Further research is needed to estimate the STSS disease burden preventable by seasonal influenza vaccination, such as a case-control study to ascertain the influenza (vaccination) status of STSS cases and (matched) controls. Such a study would ideally include an elderly F I G U R E 4 Five-week moving average of the weekly number of STSS notifications ("observed") and predictions from the final model with a baseline sine wave, season-specific IAV variables, IBV, RSV and rhinovirus ("predicted") population, as people of age 60 and older are eligible for influenza vaccination in the Netherlands and are also at highest risk of STSS ( Figure S1).
Our finding of a specific association between IAV and STSS seasonality and not other (i)GAS manifestations should be interpreted with caution as it is an ecological association and needs further confirmation in future studies. However, we believe our study adds a population perspective to the evidence from experimental studies and clinical case reports that influenza significantly impacts iGAS severity. Replication of our study in other settings or systematic influenza testing of iGAS patients may provide more insight into the burden attributable to influenza-iGAS superinfection.

ACK N OWLED G EM ENTS
The authors would like to thank all participating microbiological laboratories and support team of ISIS-AR, the participating laboratories and support team of the Virological Weekly Reports (Dutch Working

Group for Clinical Virology) and the Municipal Public Health Services
and Osiris support team. This work was supported by the Dutch Ministry of Health, Welfare and Sports.