Introduction

Streptococcus pneumoniae is a major cause of morbidity and mortality among children and adults worldwide. The incidence of invasive pneumococcal disease (IPD) shows significant seasonal variations, with a peak incidence during the cold winter season [1–3]. This phenomenon has been well documented, but the reasons are poorly understood. Different environmental factors have been proposed as potential explanations for the seasonal variation in IPD occurrence [1–4]. Moreover, it is widely assumed that the higher number of cases of pneumococcal disease that occur during the winter is closely related to the increased activity of the influenza virus [1, 3, 5, 6]. Some epidemiological studies have provided strong evidence for this relationship during both pandemic and inter-pandemic periods and, recently, this association has been strengthened by experimental findings [1, 3, 5–9].

Although these observations support the synergism among influenza virus, cold seasons, and pneumococcus, important questions remain unanswered [10]. Few studies have explored this association, controlling for other potential confounders such as climatic factors, environmental pollution or pollen and spores concentration. It is also unknown if the interaction of the influenza virus and environmental factors with pneumococcus facilitates the development of more severe clinical presentations of pneumococcal disease. Finally, it would be interesting to know if the synergism between influenza virus and pneumococcus is common to all pneumococcal serotypes. Thus, it has recently been suggested that the effect of influenza on pneumococcal disease varies according to pneumococcal serotype and host comorbidities [11]. To further understand the relationship between pneumococcal disease and these factors in an urban setting we designed an ecological study to explore the association among IPD, fluctuation in influenza infections, and environmental conditions, over the last 17 years. The study is mainly aimed at exploring the potential association between the seasonal variations and the causal serotype and clinical presentation of IPD.

Materials and methods

Identification of patients with IPD

Patients were enrolled as part of an ongoing observational study initiated in 1996 of all adults (aged ≥ 18 years) hospitalized with IPD in a teaching hospital from Barcelona, Spain (Hospital Universitari Vall d’Hebron). In the hospital, all microbiological strains isolated in sterile samples are collected systematically.

Data collection

From each patient we recorded the following variables:

  1. 1.

    Sociodemographic data

  2. 2.

    Underlying diseases

  3. 3.

    Immunosuppressive diseases

  4. 4.

    Clinical syndrome

  5. 5.

    Variables related to respiratory status (respiratory failure, need for mechanical ventilation and chest radiograph pattern)

  6. 6.

    Other variables related to clinical presentation and outcome (septic shock, intensive care unit [ICU] admission, suppurative lung complications, length of hospital stay, and mortality)

  7. 7.

    Antimicrobial therapy

  8. 8.

    Microbiological data

Definitions

Invasive pneumococcal disease was defined as the isolation of S. pneumoniae from a normally sterile site. Invasive pneumococcal pneumonia (IPP) was diagnosed when a patient had consistent clinical findings plus a new pulmonary infiltrate on chest radiography and isolation of S. pneumoniae in blood and/or pleural fluid cultures.

Patients were considered to have no comorbidities when they had neither underlying diseases nor immunosuppressive conditions. Definitions of septic shock and respiratory failure have been described previously [12].

Microbiological data

Streptococcus pneumoniae strains were identified using Gram staining, optochin susceptibility testing, bile solubility testing, and latex agglutination testing. Serotyping was performed by Quellung reaction and/or dot–blot assay at the Spanish Reference Laboratory for Pneumococci (Instituto de Salud Carlos III, Madrid, Spain). Serotypes were classified into highly invasive serotypes (1, 4, 5, 7 F, 9 V, 14, 18C, and 19A) and nonhighly invasive serotypes (all others) according to the classification of Brueggemann et al. and Sleeman et al. [13, 14]

Meteorological data

The Barcelona metropolitan area is situated at latitude 41.23 N and has a temperate climate, with the colder months occurring between November and March. Minimum temperatures rarely fall below freezing point, but there is marked seasonal variation. Meteorological data from Barcelona were obtained from the National Meteorology Service [15].

The following data were recorded: mean daily temperature (°C), maximum and minimum daily temperature (°C), number of days per month with rainfall, monthly accumulated rainfall (mm), daily sunshine irradiation (MJ/m2), mean daily wind speed (m/s), mean daily relative humidity (%), and mean daily atmospheric level pressure (hPa).

Air pollution data

Air pollution data were obtained from the General Direction for Environmental Quality [16]. The variables recorded were: mean daily concentrations of ozone (O3), sulfur dioxide (SO2), carbon monoxide (CO), nitric oxide (NO), and nitrogen dioxide (NO2) in ambient air, mean average daily concentration of metals (lead and benzene), mean daily concentrations of particulate matter up to 10 mm (PM10) and 2.5 mm (PM2.5) in diameter (μg/m3), and number of days per month with a 24-h average PM10 concentrations >50 mg/m3 and PM2.5 concentrations >25 mg/m3.

Aerobiological data

Data on airborne pollen and spores were obtained from the Point of Information of Aerobiology [17]. The following information was recorded: monthly index (sum of the mean daily concentrations in the month) of Cupressus (cypress), Platanus (plane tree), Parietaria (wall pellitory) and Chenopodium sp. (goosefoot), total pollen (pollen grains) and Alternaria, and total spores (spores).

Respiratory virus infection surveillance

In Barcelona, since 1988, there has been an active community surveillance system to monitor influenza infection (PIDIRAC program) [18]. A total of 30 general practitioners and 28 pediatricians participate in the viral sentinel surveillance system, covering 0.91 % of the Catalonian population. These sentinel physicians systematically collect throat swab specimens and nasal wash specimens from patients who present with febrile illnesses accompanied by upper respiratory symptoms. Viruses are detected either by immunofluorescence or polymerase chain reaction. The influenza epidemic period was considerable when the weekly incidence was > 50 cases/100,000 inhabitants in the reference area

Statistical methods

For the purposes of this study, monthly average values were calculated from the daily data. The relationships between the number of episodes of IPD diagnosed per month and the incidence rates of influenza, meteorological data, air pollution, and aerobiological variables were assessed using Spearman’s rank correlation coefficient. The analysis was repeated with variables from the previous month (a 1-month lag).

A negative binominal regression analysis was used to further assess the relationship between rates of IPD and influenza virus controlled for potential confounder environmental variables. In all models, the monthly number of episodes of IPD was the outcome variable. The explanatory variables used in the exploratory models were that the factors correlated significantly in the first analysis with a greater Spearman’s rank correlation coefficient. We excluded from the regression analysis variables with high co-linearity.

To assess differences in the disease characteristics and serotype distribution between the cases of IPD during the non-epidemic and epidemic influenza period, we compared patients from the two periods. We also repeated the analysis stratified by the invasiveness of the serotype and the comorbidities of patients, and by cold and warm seasons. All statistical analyses were performed using the statistical software package SPSS for Windows, version 19.0.

Results

During the 17 years of the study, 1,150 episodes of IPD were diagnosed in adults. The mean age of patients was 59 (±19.3) years and 62.5 % of the episodes occurred in men. A comorbid condition was present in 61 % of cases. Most episodes were IPP (n = 934, 81.2 %), followed by meningitis (n = 105, 9.1 %), and primary bacteremia (n = 65, 5.7 %).

Correlation between episodes of IPD, influenza virus and environmental factors

A seasonal incidence of IPD was observed such that 67.5 % of the cases were diagnosed during the 6-month period from October to March (OR 2.07, 95 % CI 1.75–2.46) and 40.5 % during the 3-month period from December to February (OR 2.05, 95 % CI 1.71–2.45). Most striking was the decline in the incidence of IPD in the summer time; only 114 cases (9.9 %) occurred in the 3-month period from June to August (OR 0.33, 95 % CI 0.26–0.42; Fig. 1). The monthly frequency of episodes of IPD correlated significantly with the incidence rates of influenza infection (r = 0.642, p <0.001). When a time lag of 1 month was applied, the correlation decreased slightly. Inverse correlations between pneumococcal infection and ambient temperature (r = −0.671, p <0.001), sunshine irradiation (r = −0.501, p < 0.001) and relative humidity (r = −0.147, p =0.037) were also observed (Table 1). Moreover, regarding ambient pollution variables, the number of episodes of IPD correlated positively with the air concentrations of nitric oxide, lead, and particles of PM2.5, and correlated negatively with the concentrations of ozone (Table 1). Concerning aerobiological data, an inverse correlation was found with concentrations of pollen and spores, with the exception of Cupressus pollen (Table 1). After adjusting for confounding environmental variables, the only factors that correlated with the rates of IPD were the incidence of influenza infection and the average ambient temperature (Table 2). IPD rates increased about 25 % in the influenza epidemic periods, and decreased by 8 % for each degree of ambient temperature increase. Similar findings were observed when only respiratory infections were analyzed (data not shown).

Fig. 1
figure 1

Monthly proportion of cases of invasive pneumococcal disease (IPD). Rates of IPD are expressed in percentage (%) of monthly case diagnoses throughout the 17 years of the study period

Full size image
Table 1 Correlation of monthly episodes of invasive pneumococcal disease (IPD) with rates of influenza infection, meteorological data, air pollution, and aerobiological data
Full size table
Table 2 Multivariate regression analysis of factors associated with rates of IPD
Full size table

Figures 2 and 3 show the seasonal variation in the number of episodes of IPD and the variation in influenza rates and average temperature respectively.

Fig. 2
figure 2

Monthly rates of IPD and influenza infection over the 17 years of the study period. Rates of IPD (dashed line) are expressed as the number of cases and rates of influenza (solid line) as the incidence (cases/100,000 population)

Full size image
Fig. 3
figure 3

Monthly rates of IPD and monthly average temperature over the 17 years of the study period. Rates of IPD (dashed line) are expressed as the number of cases and temperature (solid line) as the mean

Full size image

Comparison of IPD cases during influenza epidemic and non-epidemic periods

The number of patients with IPD during the flu epidemic weeks does not differ significantly from the number of those with IPD during the non-epidemic weeks with regard to sex, age, and comorbid conditions (Table 3). However, we observed some differences in respect of clinical presentation. Patients with IPD during the flu epidemic period had a worst respiratory status, with a greater proportion of patients presenting with respiratory failure (45.6 % vs 52 %, p =0.032). During the influenza period patients with IPD required ICU admission (19.3 % vs 24.7 %, p =0.018) and mechanical ventilation (11 % vs 15.1 %, p =0.038) more often. Nevertheless, we did not observe any significant differences in mortality. Similar findings were observed when only respiratory infections were analyzed (data not shown).

Table 3 Comparison of IPD during the influenza and non-influenza epidemic periods
Full size table

When we repeated the analysis stratified by invasiveness of pneumococcal serotypes and the presence of comorbid conditions we found that the increase in the severity of clinical presentation observed during the influenza period was focused on healthy adults with IPD caused by nonhighly invasive serotypes (Table 4).

Table 4 Clinical presentation and outcome of IPD during influenza and non-influenza epidemic periods, stratified by invasiveness of serotype and presence of comorbidities
Full size table

There were no significant differences in the distribution of specific serotypes during the influenza period compared with the non-influenza period. However, there was a trend toward an increase in the proportion of cases of IPD caused by serotypes 1, 3, and 23 F and serogroup 19 during the influenza season. During the influenza period highly invasive serotypes were more likely to cause pneumococcal infections in patients with comorbidities.

Comparison of IPD cases during warm and cold seasons

Episodes of IPD occurring in the colder months did not differ in comparison to the episodes diagnosed in the warmer months with regard to baseline characteristics of the patients, clinical presentation, and prognosis of the disease. Only a higher proportion of infections caused by serotype 5 occurred in warm months than in colder months (6.5 % vs 2.4 %, p = 0.01).

Discussion

Although several epidemiological studies have reported that influenza infections could play a role in the burden of pneumococcal disease, the exact contribution of the influenza virus has been difficult to demonstrate owing to the lack of control of other underlying factors [1, 3, 5, 6]. To our knowledge, only three previous studies have explored this association controlling for potential seasonal confounders [4, 19, 20]. Murdoch and Jennings found that the incidence rates of IPD were associated with the increased activity of some respiratory viruses, after adjusting for virological, meteorological, and air pollution variables in a multivariate model [4]. More recently, two larger studies have confirmed this correlation [19, 20].

Our observations are consistent with the results of these studies. We have also found that influenza infection seems to be the main factor associated with fluctuations in the incidence of pneumococcal disease after adjusting for other multiple variables. It has been estimated that at least 11–14 % episodes of invasive pneumococcal pneumonia can be attributed to influenza infection during the epidemic period [21].

However, not only viral but also other factors beyond influenza infection should probably be taken into account as being responsible of the peak of IPD cases in the winter season. The largest study that attempted to examine seasonal meteorological factors and their association with IPD found that pneumococcal disease correlated inversely with the mean ambient temperature and variations in the photoperiod [2]. Our data are in accordance with these observations, since we have also found an inverse correlation of the incidence of IPD and ambient temperature, sunshine irradiation, and relative humidity. Interestingly, when we adjusted the model for other environmental factors, average temperature is the only meteorological variable associated with an increased risk of pneumococcal disease. This correlation has not been found in other studies that have used a similar multivariate approach [4, 19]. Despite experimental evidence that associates changes in ambient temperature with variation in host susceptibility to pneumococcal infection having been published, the exact mechanism is not clearly explained [2, 22]. More research is needed in order to understand the exact role of these environmental factors, as well as other variables, such as ambient pollution and aerobiological data.

Another interesting point is the impact of viral co-infection on the clinical characteristics and outcome of patients with IPD. Mouse and squirrel monkey models suggest that the co-infection of pneumococcus and influenza virus result in a greater severity of the disease than infections caused by either microorganism alone [8, 9]. From a clinical point of view, data supporting the association between clinical severity and pneumococcal–viral co-infection are controversial. While O’Brien et al. observed an increase in the severity of pneumococcal pneumonia in children with an influenza-like illness in the preceding days [23], two other studies that evaluated children with microbiological confirmation of pneumococcus and influenza virus co-infection did not find any clinical differences in the severity of the disease [24, 25].

A recent published surveillance study performed in adults observed an increase in the mortality of pneumococcal disease caused by nonhighly invasive serotypes in patients without comorbidities during the influenza period (20.2 % vs 14.8 %) [11, 26]. In our study, a worse respiratory clinical profile was observed in patients with IPD during the influenza season. In this period, patients had respiratory failure more frequently and a greater proportion of them required ICU admission and mechanical ventilation. When we stratified by causal serotype and host comorbidities, this greater severity was observed mainly in healthy patients with IPD caused by low invasiveness serotypes. Our data support the hypothesis that the influenza co-infection affects the severity of IPD, although it probably depends on the specific serotype. The underlying pathogenic mechanism to describe the interaction between the two microorganisms should be focused on a local action that causes lung damage but with low systemic consequences. This would explain why respiratory failure is the main clinical complication observed over other systemic complications, such as septic shock [9, 27]. Differences between adult and pediatric patients may be due to a more intense interaction between pneumococus and influenza virus in adults than in children [1, 3, 4].

Another finding of our study is that the pattern of age and comorbidities of patients with IPD was similar during the influenza and non-influenza periods. These results contrast with a recently published article that analyzes the effect of the 2009 Influenza A (H1N1) pandemic on invasive pneumococcal pneumonia [28]. In this study, patients with IPP during the influenza pandemic were younger and were less likely to have traditional pneumococcal disease risk factors (56.2 vs 75.6 %, p < 0.001) than patients with IPP in the same months of the previous years. This discrepancy may be explained by the special tropism of influenza 2009 A virus (H1N1) in that it affected young and healthy people.

Beyond the impact on clinical presentation, it has been suggested that influenza epidemics might be associated with changes in the distribution of pneumococcal serotypes causing disease [7]. A study performed in children observed that influenza co-infection was more frequent in children with IPD caused by nonhighly invasive serotypes, suggesting that a viral synergism might help certain serotypes to make invasiveness more likely [29]. In the same way, a study by Weinberger et al. found that among healthy adults, influenza was associated with IPP caused by low-invasive serotypes. In contrast, among individuals with comorbidities the opposite occurs and the influenza virus had a greater effect on IPD caused by highly invasive serotypes [11]. Our results confirm this observation, since we have also found a greater proportion of IPD caused by invasive serotypes in patients with comorbidities during the influenza season. Interestingly, serotypes covered by the PCV-13 formulations were more frequently isolated during influenza periods. Thus, the spread of PCV-13 could be an important strategy to prevent the excess of pneumococcal disease observed during the influenza period. Further studies are needed to understand the relationship, now more evident, between influenza virus and specific pneumococcal serotypes.

Our analysis is subject to some limitations. First, it is ecological in design, using data from independent and unlinked surveillance systems. An optimal design should probably be a prospective follow-up of a large population cohort and monitor each individual for infection by influenza or S. pneumonia. However, the relative rarity of IPD in addition to the fact that the influenza rapid diagnostic test is not widely available for routine use makes this study unfeasible at present. Thus, it is generally accepted that the burden of influenza is usually estimated through labor-intensive active prospective surveillance [10]. Second, although our model incorporates an considerable number of covariables, we did not examine viruses other than influenza that might also cause seasonal fluctuations in pneumococcal disease. Epidemiological data, however, indicate that many of these pathogens circulate throughout the year (e.g., rhinoviruses and adenoviruses) [30] or peak in other seasons (e.g., parainfluenza and metapneumovirus) [30, 31], suggesting that the role of these pathogens in the seasonal fluctuation of pneumococcal disease might be less important than that of viruses that peak in the winter season. Third, the influenza vaccine status of the patients was unknown. Different vaccine proportions may have biased some results. Fourth, the co-linearity of some variables could make it difficult to evaluate the real effect on the incidence of IPD. Finally, the results might not translate to other geographical areas where meteorological conditions and influenza infection differ.

In conclusion, this study adds new evidence regarding the close relationship between IPD and influenza virus infection. It is remarkable that beyond the increase in the burden of IPD associated with the interaction between the two pathogens, a more severe clinical pattern of pneumococcal disease was observed during the seasonal influenza period. In any case, the effect of influenza on pneumococcal disease is dependent on some additional factors: pneumococcal serotype, host comorbidities, and age. The development of new and novel strategies to prevent pneumococcal disease should unquestionably include measures to control the spread of influenza.