Seasonal Variations in Clostridium difficile Infections Are Associated with Influenza and Respiratory Syncytial Virus Activity Independently of Antibiotic Prescriptions: a Time Series Analysis in Québec, Canada

The incidence and severity of Clostridium difficile-associated diarrhea (CDAD) are increasing in North America and Europe (5, 26, 28, 30). The rising CDAD rates have largely been attributed to the emergence of an epidemic hypervirulent strain known as restriction enzyme analysis (REA) type BI, North American pulsed-field gel electrophoresis type 1 (NAP1), and type ribotype 027 (toxinotype III) (30). This strain is more resistant to fluoroquinolones (FQs) than historical isolates (30) and exhibits increased sporulation capacity compared to other strains (16). A large number of C. difficile isolates now show an alarming pattern of resistance to the majority of FQs currently used in hospitals and outpatient settings (39). In addition to FQs, an increased prevalence of resistance to macrolide-lincosamide-streptogramin B (MLS_B) and its association with FQ resistance have been reported (1).

Seasonal variations in CDAD with a higher incidence occurring during winter months, have been described (4, 18, 22). The mechanisms implicated in this seasonality remain poorly understood. A recent study found an association between influenza virus activity and CDAD incidence derived from an administrative database (35). This association may be due to increased antibiotic use or to cocirculating respiratory viruses during winter months. Originally considered a pediatric pathogen, respiratory syncytial virus (RSV) has been shown to also be a significant cause of respiratory illness among adults. RSV is the second leading cause of winter morbidity in the elderly, after influenza virus (15). Moreover, influenza virus and RSV are frequent contributors to acute exacerbations of chronic obstructive pulmonary disease (COPD), for which the elderly are treated with antibiotics and hospitalized, especially during the winter season (13, 25, 31). Advanced age and comorbidities are also established risk factors for CDAD.

The province of Québec, Canada, experienced a province-wide CDAD epidemic from 2002 to 2005 which was driven by the spread of the NAP1 strain. Increased resistance to FQs and macrolides (MLs) was observed among the strains from Québec hospitals (9, 21). FQs and MLs are increasingly prescribed for respiratory infections occurring mainly during winter months. Two components are essential for CDAD to occur: first, exposure to antimicrobials and, second, exposure to toxigenic C. difficile (23). While exposure to C. difficile occurs most frequently in the hospital, exposure to antimicrobials as long as 6 months before the start of a CDAD episode has been associated with an increased risk of developing symptomatic CDAD (12). Indeed, almost half of elderly patients admitted to hospitals with community-acquired CDAD in Québec received antibiotics 3 months prior to their admission (14). In a recent cohort study in more than 4,000 patients admitted to Québec hospitals, an association between antibiotic use in the 8 weeks prior to hospitalization and increased risk of CDAD and C. difficile colonization was detected (27). As such, patients receiving antibiotics prior to hospital admission are at increased risk of developing a symptomatic CDAD disease when exposed to the hospital environment.

Therefore, we hypothesized that seasonal variations in antibiotic prescription in the community may be associated with the seasonality of CDAD in hospitalized patients.

The objective of this study was to examine temporal relationships between the two most frequent respiratory viruses (influenza virus and RSV), the antibiotics prescribed for respiratory infection (FQs and MLs) in Québec's community pharmacies, and CDAD incidence in Québec hospitals by using multivariable time series analysis. We hypothesized that (i) both influenza virus and RSV are temporally associated with CDAD incidence and (ii) this association is explained by seasonal variations in antibiotic prescriptions. To circumvent two problems of traditional regression models, i.e., (i) the assumption that consecutive respiratory virus counts, antibiotic use data, or hospitalizations are independent and (ii) their inability to detect time-delayed associations, which often exist between antibiotic use/increasing respiratory virus activity and hospitalization, a Box-Jenkins or autoregressive integrated moving average (ARIMA) transfer function model has been used.

Time series of CDAD incidence, influenza virus and RSV circulation, and antibiotic prescriptions are presented in Fig. 1. CDAD incidence seems to follow closely influenza virus and RSV seasonal activity, as well as FQ and ML prescriptions. Some specific drugs also showed obvious seasonal variations (among FQs, gatifloxacin-moxifloxacin and levofloxacin; among MLs, clarithromycin and azithromycin) (Fig. 2A and B).

By applying the Spearman coefficient, both respiratory viruses were positively and significantly correlated with CDAD incidence (R = 0.47 for influenza virus; R = 0.37 for RSV) (Fig. 3A). FQs (R = 0.31) and MLs (R = 0.40), as well as all specific drug prescriptions, with the exception of ciprofloxacin, were positively and significantly correlated with CDAD incidence (Fig. 3B and C). Ciprofloxacin was negatively correlated with CDAD incidence (R = −0.50). The magnitude of the correlation was greater for levofloxacin (R = 0.60), ofloxacin (R = 0.62), and norfloxacin (R = 0.60) among the FQs (Fig. 3B) and for erythromycin (R = 0.73) and azithromycin (R = 0.45) among the MLs (Fig. 3C).

For influenza virus and RSV time series, as well as for most antibiotic time series, significant terms of the order of 1 (corresponding to the previous month) and 12 (corresponding to the similar month of the previous year and indicating seasonality) were detected (Table 1). For CDAD incidence time series, an autoregressive term of order 12 presented a trend (P = 0.06).

In bivariate analysis, transfer function models showed significant relationships between almost all explanatory time series (respiratory viruses and antibiotic prescriptions) and CDAD incidence (Table 2), with the exception of ciprofloxacin, ofloxacin, and norfloxacin, for which no significant parameter was detected.

Multivariate time series analysis showed that temporal variations in CDAD incidence followed temporal variations in FQ use with a lag of 2 months (P = 0.01) (Table 3). This means that, on average, an increase (or decrease) of FQ use by 1,000 DDDs/100,000 inhabitants resulted in an increase (or decrease) of the CDAD incidence by 0.38/1,000 discharges 2 months later. In addition, temporal variations in CDAD incidence followed temporal variations in influenza virus and RSV time series with a lag of 1 month (P = 0.043 and P = 0.004, respectively) (Table 3). A trend for the relationship between CDAD incidence and ML use with a lag of 1 month was detected (P = 0.05). Globally, this model explained 95.64% of the CDAD incidence variance over the study period.

In sensitivity analyses restricted to the period without an epidemic peak (August 2005 to December 2008), similar results were obtained for Spearman correlation coefficients and time series analyses (data not shown). Results of multivariate sensitivity analysis are presented in Table 3.

To our knowledge, this is the first study to assess the impact of two respiratory viruses (influenza virus and RSV) on CDAD while controlling for antibiotic prescription. Circulation of influenza virus was previously shown to be associated with changes in CDAD rates (35). Our study suggests that in addition to influenza virus, RSV is also associated with seasonal variations in CDAD incidence. RSV is a common wintertime respiratory virus that affects persons of all ages and is the major cause of serious lower respiratory tract infections in young children. Detection of RSV in young children contributes importantly to the epidemic curve describing RSV circulation. However, epidemic cycles of respiratory viruses consistently occur 1 to 2 weeks later in adults than in children (24), and as such, a time series of aggregated 1-month RSV circulation is representative of overall RSV seasonal variations. Also, influenza virus and RSV surveillance data precede the peak in hospitalizations associated with influenza virus and RSV infections by 1 to 2 weeks (17, 36). Thus, monthly aggregated data capture variations in both community and hospitalized influenza virus and RSV infections.

Unexpectedly, influenza virus and RSV time series had a significant impact on CDAD independently of antibiotic prescription. The force of impact of influenza virus and RSV time series on CDAD rates decreased in multivariate models controlling for the impact of antibiotic prescription compared to models with only respiratory viruses. This means that the effect of viruses may be partially explained by antibiotics. However, there is an independent role for viruses. This may be related to close sequestering of individuals, coupled with likely changes in diet and outdoor activity, hospital overcrowding, shortages of medical staff, deficiencies in applying infection control measures during winter months, or other unknown factors.

Other factors which could potentially influence CDAD rates include seasonality in C. difficile contamination of the food supply, such as ground beef (37), and increased C. difficile carriage by calves and swine in winter (33, 38). A relationship between CDAD and norovirus outbreaks (which tend to occur during winter months) has also been reported (6, 41). Additional studies are needed to assess the degree to which these candidate determinants contribute to CDAD seasonality.

It is of note that significant correlations with CDAD were observed for both examined classes of drugs and specific drugs, as well as for influenza virus and RSV. In the bivariate transfer function models, some of the examined time series (ciprofloxacin, ofloxacin, and norfloxacin) did not show a significant impact on CDAD time series. This suggests that some of the significant correlations between time series detected may be due only to similar patterns or a correlation of consecutive observations in the examined time series and not necessarily to a real association. Correlation of consecutive observations in time series is especially true when these are cases of infection. In contrast to the positive association between ciprofloxacin in outpatients and CDAD detected in Germany by using Spearman's correlation coefficient (8), we detected a negative significant correlation between ciprofloxacin prescription and CDAD when applying the same statistical method. However, this association was no longer significant when using the transfer function model. The absence of an association between ciprofloxacin and CDAD in our study could be due to the poor activity of ciprofloxacin against respiratory pathogens, which makes it a less than ideal candidate to treat respiratory infections. The divergence in results obtained in our study and the study from Germany may also be explained by methodology differences, differences in antibiotic prescription practices, or differences in the genotypes and antibiotic susceptibility profiles of circulating C. difficile strains in different geographical regions. For example, the NAP1 strain is the predominating strain in Québec, while other strains predominate in European countries (7, 21).

The temporal relationships between FQ and ML prescriptions in the community and CDAD incidence in hospitals detected in our study are congruent with those observed in other ecologic studies using similar statistical methods relating antibiotic use within hospitals and CDAD incidence in hospitals (2, 40). In our study, an increase in FQ and ML prescriptions in the community was associated with an increase in the CDAD incidence with delays of 2 months and 1 month, respectively. This is in line with the findings of other studies that detected an association between health care-acquired or community-acquired CDAD and the antibiotics used prior to hospitalization (14, 27). Our results add to the evidence that when assessing factors associated with an increased risk of CDAD in hospitalized patients, exposure to antibiotics before admission should be considered in addition to the antibiotics received during hospitalization. In addition to FQ and ML classes, we explored the effects of specific drugs on CDAD. Our results suggest that differences in the effects of different specific drugs may be observed and that an aggregated class effect may be observed but the effect does not allow generalization to all its representatives. This observation is of particular importance for setting priorities in settings with different antibiotic prescribing practices.

The strengths of our study include the use of a robust multivariable analysis method allowing assessment of independent temporal relationships between circulating respiratory viruses, antibiotic prescriptions, and CDAD incidence. In addition, the Box-Jenkins transfer function method assumes that each additional percentage of virus positivity and each additional DDD of antibiotic prescription contribute additively to CDAD incidence (17). As such, detection of significant associations in our study is indicative of a dose-response relationship between respiratory virus activity, antibiotic prescriptions, and CDAD. Also, we used CDAD incidence from a surveillance program including the great majority of provincial acute care hospitals and using a standardized definition.

Our study has several limitations. First, an ecologic design with aggregated group-level data may not represent individual-level associations. However, the association between FQs and MLs as a class and their representatives with CDAD is in line with what is observed in individual-level case-control and cohort studies (28, 34). Furthermore, temporal relationships with circulating viruses at a regional level may be indicative of other phenomena at the local (hospital) or individual level and merit further exploration. Second, we did not consider the antibiotics prescribed in hospitals. One can expect that seasonal variations in prescribing FQs and MLs in the community and in hospitals are similar; however, this may be different for specific drugs. Although the strength of association that we found may not necessarily quantify exactly the relation between antibiotics and hospitalized CDAD cases, we think that the significance and delays in the temporal relationships with FQs and MLs detected are probably true. Third, we cannot exclude the role of other antibiotic classes and other respiratory viruses which we did not consider in our study. For example, we did not assess the impact of beta-lactam antibiotics on CDAD incidence, as local clinical practice guidelines do not advise their use in lower respiratory tract infections in adults (11). Finally, the number of observations in time (48 monthly observations) limited the power of the transfer function models to detect significant associations.

In conclusion, our study attempted to clarify the temporal association of influenza virus, RSV, FQs, and MLs and the incidence of CDAD. While the initial hypothesis was that the role of winter viruses is no longer significant after controlling for the antibiotics prescribed against respiratory infections, we detected an independent impact of influenza virus and RSV on CDAD time series. This suggests that measures aimed at preventing influenza virus and RSV infection may help decrease the risk of CDAD. However, this association was observed at an aggregated level (ecologic study) and may be indicative of other antibiotic classes or other phenomena occurring during wintertime. More research including longer periods of time and more antibiotic classes is needed to confirm these findings.

We appreciate the help of Louis Dumont, pharmacologist at the Institut Universitaire de Cardiologie et de Pneumologie de l'Université Laval, Québec, Canada, for useful comments during the preparation of the manuscript.