Streptococcus pneumoniae (pneumococcus) is one of the most important bacterial causes of respiratory tract infections worldwide [1]. While it can cause serious illnesses, pneumococcus carriage in its ecological niche, the human nasopharynx, is generally asymptomatic. More than 90 serotypes have been described based on differences in the polysaccharide capsule [2]. Colonisation by pneumococcus can occur soon after birth and remains common in the first years of life [3], with virtually every child experiencing a sequence of colonisation events by alternating serotypes [4]. Previous statistical epidemiology studies have found that serotypes compete to some extent for colonisation, as carriage of a certain serotype interferes with subsequent colonisation by pneumococcus of a different serotype [5–10].

Serotype specificities in transmission and competition parameters are subject to ongoing research and debate [7, 9, 11, 12]. Cauchemez et al. [11] concluded that there were no significant differences in rates of acquisition and clearance between vaccine and non-vaccine serotypes within school classes in France. Using the same dataset, de Cellès et al. [12] reported significant differences in the rates of acquisition estimated independently for different serotypes while assuming shared clearance rates. Erästö et al. [7] estimated serotype-specific parameters simultaneously in Bangladeshi families and concluded that, while more common serotypes in the study area were more often acquired from the community, these differences loose significance when conditioning on exposure within the family. More recently, Lipsitch et al. [9], focusing on the competitive ability of specific serotypes in Kenya, reported significant differences in susceptibility to competition, which appears positively correlated with the rate of clearance.

Studies that estimate pneumococcal transmission parameters have either used different models, statistical frameworks, parametrisations or transmission contexts, for example families, hospitals or day-care centres (DCCs). Our aim was to compare two different pneumococcal settings in Europe, one with high carriage prevalence (Portugal) and one with low carriage prevalence (Finland) by estimating pneumococcal transmission parameters using a single statistical method with data from a common context, in this case, DCCs. We apply a Bayesian data augmentation approach [6, 13] to estimate acquisition and clearance rates in day-care centres in Portugal [14] and Finland [15]. We assume shared parameters among serotypes in each country and focus on the comparison between the two settings.

In this article we explored what determines the prevalence of pneumococcal carriage in two European day-care settings. Parameters related to pneumococcal transmission were estimated from longitudinal studies conducted in Portugal and Finland. Differences in rates of transmission and clearance of carriage were identified as the main determinants of the observed differences in prevalence (61% in Portugal vs. 26% in Finland). Although the two studies differed in the ages of enrolled children (mean age of 2 years in Portugal vs. 4 years in Finland), this does not explain the whole extent of the prevalence difference since other studies have reported carriage levels higher than 60% among Portuguese children of age 4 years [17] and less than 40% in Finnish children of age 2 years [4]. In particular, pneumococcal transmission appears stronger in the Portuguese setting irrespective of age.

Both the rate of within-room transmission (1.05 vs. 0.63 per month) and community acquisition (0.26 vs. 0.12 per month) were twice as high in Portugal than in Finland. Cultural differences could explain the higher within-room intensity of transmission in Portugal, for example a longer time spent indoors or at the DCC. In both settings, children from the same DCC spent time together in the playground. Since we considered each of the DCC rooms separately, transmission between rooms of the same DCC would appear as community acquisition. Thus, the higher community acquisition in Portugal was probably a consequence of the Portuguese study subjects attending a larger DCC than children in any of the Finnish DCCs in this study. The credible intervals for the within-room transmission and outside acquisition were wide due to the strong (posterior) correlation between these two parameters: high within-room transmission together with low community exposure was as likely as vice-versa.

From the same dataset of Finnish DCCs, but complemented with samples from the children’s families, Hoti et al. [6] estimated a slightly lower within-DCC transmission rate of 0.53 per month and a within-family transmission rate of 0.36 per month. A study of pneumococcal transmission in U.K. families, however, estimated a higher within-family transmission rate of 1.41 per month [18]. Three previous studies have reported only small differences in transmission rates across different serotypes [7, 11, 12]. The transmission rates within families varied between 0.64 and 0.84 per month (Bangladesh) [13] and within DCCs between 1.04 and 1.18 [12] and between 1.38 and 1.53 [11] (France). The two French studies may have over-estimated the transmission rates, since the first [12] assumed a fixed duration of carriage of 28 days, and the second [11] assumed that carriage needs to be cleared before colonisation by another serotype.

The duration of carriage (here defined as time until immune clearance, cf. [9]) was found to be longer in Portugal (estimated mean 47 days) than in Finland (37 days). The shorter duration in older children (mean age 4 years in Finland vs. 2 years in Portugal) is in agreement with previous analyses [9, 19]. Serotype 19 F [9, 13] and serotype 6B [9] are typically found to be carried much longer. The fact that these serotypes were common in the Portuguese dataset but almost non-existent in the Finnish may have contributed to the estimated difference in the rates of clearance.

Similar durations of carriage were found in other studies with children under 3 years old, 48 days in Danish DCCs [8] and 51 days in English households [18]. Serotype-specific durations of carriage in Bangladeshi infants up to 1 year of age were estimated between 43 and 48 days for all serotypes except for 19 F, which was estimated to be carried on average 62 days [13]. Lipsitch et al. [9], who also estimated serotype-specific parameters, estimated the duration of carriage between 28 and 123 days. The mean duration decreased with age, 105 days in children less than 2 years old and 29 days in children between 3.5 and 5 years old.

Current carriage was found to reduce the subsequent acquisition rate by a factor of 0.5 in Portugal and 0.7 in Finland. The estimation of this competition parameter (ϕ), however, was sensitive to the choice of the prior distribution. In particular, the analysis warranted for a small prior probability for values very close to zero. In the current analysis the competition parameter was included mainly to adjust for possible confounding effects on acquisition by concurrent carriage. More data on the frequency of multiple carriage and more frequent sampling would have been necessary to learn adequately about competition [20].

The overall prevalence of pneumococcal carriage in the Finnish dataset was low, even when a more sensitive method of detecting pneumococcal carriage was applied on samples from one of the day care centres. Moreover, the proportion of multiple carriage in these samples (13%) was comparable to that in another study (10%) using the same detection method in day care children [16], although the prevalence in our study was clearly smaller (29% vs. 58%). This suggests that a low level of multiple carriage does not explain the lower prevalence and transmission rates in the Finnish data, although we did not have similar data from the Portuguese setting in this study. Moreover, it appears natural to assume that the dominant serotype is the one most likely transmitted as well as detected by sampling.

Although the prevalence of carriage and the total number of isolates were lower in Finland, the number of different serotypes was clearly larger in any of the Finnish DCCs than in the Portuguese DCC. The Portuguese children were on average younger than their Finnish counterparts (2 vs. 4 years), which could have affected the diversity of carriage: younger children carry the common paediatric types while older children and adults carry a wide variety of rarer serotypes. Cobey et al. [10] found an increase in the diversity of carriage with age. The Finnish DCCs were also more scattered in the outskirts of an urban area, such that there was probably no transmission between the DCCs and clear serotype clustering occurred, with rare serotypes being exceptionally prevalent in samples from individual DCCs [6]. The Portuguese DCC was more connected to the community and other DCCs and, as such, more mixing could occur within and across DCCs resulting in less clustering of serotypes.

Not all children in the studied DCCs were sampled. This may have caused some bias, especially in one of Finnish DCCs in which only 26% of children were sampled. In the Portuguese DCC, only three rooms were included, corresponding to 22% of the children in the DCC. Although this ensured homogeneous ages among the study subjects, the sample may not be representative of the DCC as a whole. On the other hand, one month may be too long a sampling interval, especially for the Portuguese setting, in which the prevalence and the rates of transmission and community acquisition were higher. The optimal spacing between observations has been determined for a binary model, i.e. considering only carrier and non-carriers as possible states, and varied between 1 month for very low prevalence settings and 1 week in very high prevalence settings [21]. Nevertheless, the prevalence of carriage implied by the estimated parameters was close to the observed, and may be a good representation of the transmission dynamics. It would be interesting to replicate the study reported here to a larger set of comparable longitudinal studies from different prevalence contexts or countries so as to gain a better understanding of pneumococcal transmission differences.

We were successful at estimating realistic parameters for pneumococcal transmission, which were comparable among themselves and compatible with observed prevalences in two European settings (61% in Portuguese vs. 26% in Finnish day care children). The force of transmission in Portuguese children was found to be significantly higher than in the Finnish children. The community rate of acquisition, affected by the community structure, was also found higher in the Portuguese setting. The difference in carriage was explained by the higher rates of transmission and community acquisition as well as a lower rate of clearing carriage in younger DCC attendees in the Portuguese data.

Here we derive crude estimates for the four model parameters. Let n _g be the number of groups, $n_v^{\left(g\right)}$ the number of visits in group g ∈ {1, …, n _g }, and $n_c^{\left(g\right)}$ the number of children in the group. For visit v, let $N_v^{r,s}$ denote the observed number of children who were in state r at visit v and in state s at visit v + 1. The total number of transitions from r to s is then the sum over all visits in all groups: $N^{r,s}={\displaystyle \sum _{g=1}^{n_g}{\displaystyle \sum _{v=1}^{n_v^{\left(g\right)}-1}{N}_v^{r,s}}}$. Let us also consider these numbers stratified by exposure to serotype q: N ^r,s (q+) and N ^r,s (q−), where exposure is defined as presence of the serotype q in the room at the time of transition. The total number of samples of serotype r is $N^r={\displaystyle \sum _{x=0}^{n_s}{N}^{r,x}}$. Stratified by exposure to serotype q, the total numbers of samples of serotype r are denoted by N ^r (q+) and N ^r (q−).

So the estimate for the competition parameter is $\widehat{\mathit{\phi }}=\frac{{\widehat{\kappa }}^{\mathit{\text{carriers}}}}{\widehat{\kappa }}$.