Study population

The study was conducted in Nha Trang, a coastal city in south-central Vietnam with a total population of just over 426,958 and under-five population of 28,495 in the 2018 census (personal communication with Dr. Minh Xuan Bui, Khanh Hoa Health Service, October 2, 2021—see S3 File). Nha Trang city consists of 27 communes of similar population size. Each commune has a commune health centre, providing a range of basic health services. Similarly, educational services including nurseries and schools are largely provided on a commune basis.

In October 2016, a cluster randomised controlled trial was initiated to evaluate alternative PCV dosing schedules [13]. In the 24 communes that were allocated to receive PCV, routine infant vaccination was initiated according to the allocated schedule: 2 priming doses and 1 booster dose (2p+1), 3p+0, 1p+1, or 0p+1, and a catch-up campaign including children under 3 years old was conducted. The primary endpoint of the trial is dependent on pneumococcal carriage, and, as such, this is monitored in each commune through annual cross-sectional nasopharyngeal carriage surveys in 60 infants (4 months to 11 months), 60 toddlers (14 months to 23 months), randomly selected from an administrative population list, and their respective main carers. Completion rate of routine PCV vaccination in the community was 81.2%, 88.5%, 97.4%, and 87.0%, in the 2p+1, 1p+1, 0p+1, and 3p+0 arms, respectively, using the 18 months population as denominator for the 2p+1, 1p+1, and 0p+1 arms and the 12 months population for the 3p+0 arm, in 2018 when the infant contact survey was conducted. Due to serotype replacement, the trial did not affect overall carriage rates in the Nha Trang population.

Study design

In October 2018, 2 years after the start of the trial, an infant contact survey was nested into the cross-sectional carriage study. All infants enrolled for the carriage study were also eligible for enrolment into the contact survey.

For each commune health centre, a staff member was identified and trained to conduct the contact survey in their local commune. During an initial home visit, written informed consent was obtained for both the infant contact and carriage studies from parents or guardians before any collection of data or specimen.

Subsequently, a background questionnaire was filled in together with the parent or guardian on behalf of the infant, collecting information including sex, age, household composition, and the infant’s mobility (see S1 File for the full questionnaire). The parent or guardian was asked to monitor their infant’s contacts the following day, and another home visit was scheduled for the day after. For that visit, the interviewer would aid the carer’s memory by discussing their day and noting initials of the infant’s contacts. Subsequently, information on those contacts’ age, sex, duration, and location was filled in (see S2 File for the full questionnaire). A contact was defined as skin-to-skin contact with the infant, because the common conversational contact definition would be inappropriate to apply to infants and because physical contacts, as a proxy for close contacts, have been shown to better correlate with pneumococcal carriage risk [5,6].

Consent to participate in the contact study and the nasopharyngeal carriage study was sought from the carers by home visit. The date of keeping a contact diary was assigned, and field workers conducted a second visit 1 or 2 days after to record the contact information based on the diary. Participants were also given an appointment with the commune health centre during which a nasopharyngeal swab from the infant was obtained by a study nurse. Sample collection, handling and storage were performed according to the World Health Organization guidelines [14]. At the Pasteur Institute of Nha Trang, DNA extractions were performed on a QIAcube HT instrument using the QIAmp 96 DNA QIAcube HT Kit (QIAGEN, Hilden, Germany) with a lysis buffer (20 mg/ml lysozyme, 20 mM Tris/HCl, 2 mM EDTA, 1% v/v Triton) and RNase A treatment, as previously described [15]. Real-time quantitative PCR (qPCR) to detect the autolysin-encoding gene (lytA) of Streptococcus pneumoniae was performed. Final reaction volumes of 25 μl were run on an Applied Biosystems 7500 Fast real-time PCR instrument with 5 μl of template DNA, TaqMan GeneExpression Mastermix (Applied Biosystems, Massachusetts, USA), and 200 nM of primers and probe (forward primer: 5′-ACGCAATCTAGCAGATGAAGCA-3′, reverse primer: 5′-TCGTGCGTTTTAATTCCAGCT-3′, and lytA probe: 5′ FAM -TGCCGAAAACGCTTGATACAGGGAG- 3′ BHQ1) [16]. lytA positive (cycle threshold (Ct) value < 35) or equivocal (Ct value 35 to 40) samples were cultured on sheep blood agar plates with 5 μg/ml of gentamicin. Samples with alpha-hemolytic growth were harvested and DNA was extracted from the bacterial pellet on a QIAcube HT instrument as described previously [17]. Pneumococcal serotyping was performed by DNA microarray at the Murdoch Children’s Research Institute using Senti-SPv1.5 microarrays (BUGS Bioscience, London, UK) [18].

Pneumococcal carriage was defined as samples that were both lytA positive/equivocal and culture positive and subsequently confirmed by microarray. A small number of samples that were lytA qPCR positive but were nonculturable (n = 5) or had technical issues (n = 4) were also deemed pneumococcal positive.

This study is reported as per the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline (S1 STROBE Checklist).

Statistical analyses

The infant contact study is a cross-sectional study, and no prospective analysis plan exists. The plan for the current study combining contact and carriage data was conceived during a collaborative meeting that happened after the infant contact and pneumococcal carriage surveys were conducted, but without being informed by the data. We used negative binomial regression to estimate crude rate ratios of the mean number of contacts of participants for each characteristic of the participants, as defined in Table 1. We also calculated rate ratios adjusted for age (categorical: 4 to 7 months and 8 to 11 months of age) and for clustering of participants into communes.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Mean number of contacts reported for infants and effect of each infant’s characteristic on incidence of contacts, estimated using a negative binomial regression model.

https://doi.org/10.1371/journal.pmed.1004016.t001

After completion of the nested contact survey, we also examined the association of infant infection and their contact behaviour but, in an extension to previous studies, also accounted for the risk of infection in each contact. We used logistic regression to estimate the crude ratio of odds for a contact being made outside the commune of residence, for each characteristic of the participants. In an adjusted analysis, we included the age group (4 to 7 and 8 to 11 months of age: the first 4 months and the second 4 months of the target age range) and day care attendance as covariates and accounted for clustering of participants into communes.

To explore whether an infant’s contact patterns are correlated with their probability to carry pneumococcus, we created the potential exposure index (PEI). For each infant participating in the study, PEI estimates the probability that at least one of the reported contacts was carrying pneumococcus and hence that the infant was exposed to the pneumococcus on that day at least once. We define PEI as (1) where p_i represents the probability that the ith contact of the respective study participant is a pneumococcal carrier. We assume that p_i is solely dependent on the contact’s age. To estimate age-dependent probabilities for pneumococcal carriage in the study setting, we performed Bayesian penalised B-spline regression on logit-transformed carriage probability, obtained from previously reported low-granularity age-stratified pneumococcal carriage across age groups from Nha Trang [19]. These data were collected in October 2006 from 519 individuals in a cross-sectional survey. The age group stratifications were yearly from ages until 5 years old and then 6 to 10 years old, 11 to 17 years old, 18 to 30 years old, 31 to 49 years old, and over 50 years old. We visually validated our estimates against carriage prevalence observed in young children in the trial. The PEI estimate is based on the contact patterns recorded for a single day but were interpreted as a proxy for the general contact behaviour of the respective infant. The Mann-Whitney U test was used to assess whether there was a significant difference in average PEI value in carriers and noncarriers.

We used a hierarchical Bayesian logistic regression model (see Table B in S4 File) to estimate the correlation of PEI (covariate) and pneumococcal infection (outcome). Based on selection via a Bayesian belief network (see Fig A in S4 File), we included age in months as a covariate and the resident commune as a random effect. Hence, our logistic regression model was (2) where

• p_i is the probability that the ith infant in the study is a pneumococcal carrier;
• b₀ is the model intercept under a single-level regression model (fixed effects only);
• C_i is the random effect on the intercept due to the influence of the commune to which the ith individual belongs;
• b₁ is the coefficient corresponding to the effect of PEI;
• E_i is the value of PEI, for the ith individual;
• b₂ is the coefficient corresponding to the effect of the infant’s age; and
• A_i is the age of the ith individual in months.

The model was implemented in JAGS [20]. The uncertainty in the individual level estimates for PEI in this analysis was retained from the posterior estimates in the previous regression (namely, the cubic spline on age) by assuming a normal distribution for each year of age and drawing samples from this distribution when calculating the posterior samples of PEI. We chose to assume a normal distribution for simplicity and since, by visual inspection, the distribution of carriage rates across all ages appeared to follow a Gaussian (see Fig C in S4 File). Subsequently, this individual-level uncertainty in PEI is further propagated by assuming a beta distribution for these PEI estimates, since PEI is bounded between 0 and 1, and each individual’s PEI distribution is right skewed (see Fig B in S4 File); this distribution then serves as an input to the logistic regression predicting carriage status. Potential correlations among the predictor variables were checked before model fitting using the variance inflation factor (VIF) [21].

The total pneumococcal exposure to an infant on the day that contacts were reported was approximated by the sum of each reported contact’s age specific probability of pneumococcal carriage. The contribution of a specific age group to that total was calculated as the pneumococcal exposure from contacts of that age group divided by the total exposure. This can be generalised to calculate the contribution of specific age groups to the total pneumococcal exposure across all infants by including all contacts in the study.

The descriptive statistics were done in STATA version 14.0 [22]. The analyses on the association of contact patterns and pneumococcal carriage were conducted in R 4.0.2 [23] and RJags [24]. The code is available on https://github.com/GeorgeYQian/Nha-Trang-Contact-Study-Data-and-Code.

Ethical approval

Ethical approval for the study was granted by the respective ethics boards of the National Institute of Hygiene and Epidemiology in Hanoi, Vietnam (4405/QD-BYT) and the London School of Hygiene & Tropical Medicine (15881).

Characteristics of study participants and their contacts

The contact study was conducted between October 19 and November 7, 2018. Overall, 1,583 infants aged between 4 and 13 months (median: 9 and interquartile range: 7 to 11 months) were enrolled in both the contact survey and the carriage survey (Table 1). All children were aged 11 months or younger at the time of enrolment, although 122 children were 12 or 13 months by the time of data collection. An additional 11 infants were enrolled in the carriage survey but did not consent to the contact survey because it was not convenient for the carer; these infants were not included in any of the following analyses (response rate was 99%). There were no missing data in the contact survey except that S. pneumoniae carriage status could not be determined in one infant due to technical reasons. This infant was excluded from the analysis of pneumococcal carriage. Over half (917, 58%) of infants had siblings living in the same household and the average household size was 4.9 (SD 1.0). The majority of infants were able to sit on their own (84%), many could crawl (55%), and some could walk (12%). Most families (98%) had a motorbike; other means of transport were rarely used. Moreover, 50% of infants did not leave their commune of residence during the preceding week. Few infants (5%) attended day care. Pneumococcal carriage prevalence was 22% and increased from 17% at 4 to 7 months to 26% at 8 to 11 months.

A total of 7,428 contacts aged between 0 and 100 years (median: 33 years) were reported (Fig 1). Contacts mostly occurred at home (89%) and with family members (76%). They were mostly with other children, parents, and grandparents and lasted for more than 1 hour (55%). Few contacts (5%) were reported to occur outside the commune of residence or with other infants (0.4%) (Table 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. The number of infant contacts.

Reported skin-to-skin contacts are shown disaggregated into annual age groups and stratified by contact duration.

https://doi.org/10.1371/journal.pmed.1004016.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Characteristics of infants’ contacts.

https://doi.org/10.1371/journal.pmed.1004016.t002

Factors affecting the number of contacts

The number of skin-to-skin contacts an infant made on the day before the survey ranged from 1 to 15, and the mean was 4.7 (SD 1.8). Infants who lived in households with more than 4 people were reported to have 25% (95% confidence interval (CI): 19% to 30%) more contacts than infants in smaller households (adjusted incidence rate ratio 1.25, 95% CI: 1.19 to 1.30, p-value <0.001). The infant’s ability to crawl or sit was not found to be associated with their number of contacts; however, infants who had travelled beyond commune borders in the week preceding the survey were reported to have 6% (95% CI: 2% to 10%) more contacts (adjusted incidence rate ratio 1.06, 95% CI: 1.02 to 1.10, p-value 0.004). Infants who were found to carry pneumococci were reported to have slightly more contacts; however, this was not statistically significant (adjusted incidence rate ratio 1.03, 95% CI: 0.98 to 1.08, p-value 0.284) (Table 1).

Factors affecting the location of contacts

About 10% of infants had at least one contact outside of their commune of residence (Table A in S4 File). Infants at least 8 months old and those with caretakers in current employment were 67% (95% CI: 16 to 140) and 67% (95% CI: 19 to 134) more likely (according to the crude odds ratios 1.67, 95% CI: 1.16 to 2.40, p-value 0.006 and 1.67, 95% CI: 1.19 to 2.34, p-value 0.003) to have a contact outside their commune, respectively. Also, pneumococcal carriers were more likely to have had contact outside the commune; however, this was no longer the case if adjusted for day care centre attendance (adjusted odds ratio 1.25, 95% CI: 0.92 to 1.71, p-value 0.159), which was strongly associated with contacts outside the commune (adjusted odds ratio 6.56, 95% CI: 3.92 to 10.98, p-value <0.001). Infant-to-infant contact was rare; only 2% of infants were reported to have contacted another infant in another commune. Of such infant-to-infant contacts, the majority (19/31, 61%) happened in day care.

Validating contacts as a risk factor for carriage

Modelled carriage prevalence in Nha Trang based on carriage observations from a previous study [19] increased rapidly in infancy and peaked at about 4 years of age. Secondary school–age children and particularly adults were rarely carrying pneumococcus; about 10% and less than 5% respectively were infected. Carriage rates observed in 2008 and in 2018 were similar in the age groups observed in both surveys (Fig 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Age-stratified carriage prevalence in Nha Trang.

Grey dots indicate the positions of knots for the spline and the open circles and their corresponding 95% binomial CIs the carriage prevalence data that the model was fitted to. The black line and the region highlighted in blue represents the model’s median and 95% quantile estimates. The “x” symbols are added for visual comparison and indicate the carriage prevalence observed in the infant survey. To aid visualisation, we used quadratic scaling of the x-axis. Inset: the distribution of estimated PEI values in infants without (left) and with (right) pneumococcal carriage. CI, confidence interval; PEI, potential exposure index.

https://doi.org/10.1371/journal.pmed.1004016.g002

We estimate that most (61%) infants had less than a 25% probability of having had close contact with a pneumococcal carrier on the surveyed day; however, for 22% of infants, the probability of exposure was much higher at 50% to 70%. The estimated PEI for each infant was associated with substantial uncertainty (Fig B in S4 File). The median PEI value for pneumococcal carriers was 27% (95% CI: 5.1% to 70%), compared with 15% (95% CI: 4.9% to 61%) for noncarriers, and, from the Mann–Whitney U test, there was a statistically significant difference between PEI values from the 2 groups (p < 0.001).

When adjusting for age and locality, we find that the probability of exposure to pneumococci on the surveyed day and the risk of pneumococcal carriage were highly correlated: for every 10 percentage points increase in PEI, the odds of pneumococcal carriage increased by 22% (95% CI: 15 to 29), i.e., the odds of pneumococcal carriage were 7.1 (95% CI: 4.1 to 12.3) times higher in infants who had at least one effective contact with a pneumococcal carrier (PEI = 1), compared to those who had not (Table C in S4 File).

Age-specific infant exposure to pneumococci

Children between 1 and 4 years of age contributed 38% (95% CI: 22% to 56%) of the total pneumococcal exposure to infants; children 5 to 9 and adults 21 to 40 year olds contributed 22% (95% CI: 16 to 29) and 20% (95% CI: 9.2 to 32), respectively. We estimate that 2- to 6-year-old children are key pneumococcal transmitters to infants, contributing 51% (95% CI: 39 to 63) of the total exposure and that infection risk from other infants or older children and adults is likely low in this setting (Fig 3). The majority of potential pneumococcal exposures (80%) originated from household contacts.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. The contribution of different age groups towards the total exposure of infants to pneumococcus.

Bars show the relative contribution of an age group to the total pneumococcal exposure (number of close contacts and their propensity to carry pneumococci) of infants in Nha Trang. Error bars indicate 95% credible intervals. Inset: the contribution of up to 15 year olds to total pneumococcal exposure of infants in Nha Trang.

https://doi.org/10.1371/journal.pmed.1004016.g003