High residual vaccine-serotype Streptococcus pneumoniae carriage 4 to 6 years after the introduction of 13-valent pneumococcal conjugate vaccine in Malawi: a prospective serial cross-sectional study

Background There are concerns that current pneumococcal conjugate vaccine (PCV) schedules in sub-Saharan Africa sub-optimally interrupt vaccine-serotype (VT) carriage and transmission, thus limiting indirect protection. We assessed pneumococcal carriage in vaccinated children and unvaccinated populations targeted for indirect vaccine protection, between 4 and 6 years after the 2011 introduction of a 13-valent PCV (PCV13) 3+0 schedule in Malawi. Methods We conducted four sequential prospective nasopharyngeal carriage surveys in urban Blantyre, from June, 2015, to April, 2017. We recruited healthy PCV13-vaccinated children 3-6 years old, children 5-10 years old born before PCV13 introduction, and HIV-infected adults 18-40 years old on antiretroviral therapy. Carriage risk by age was analysed by non-linear regression. Findings We sampled 1382 PCV13-vaccinated children, 889 PCV13-unvaccinated children, and 985 adults. VT carriage prevalence declined from 23% to 17% among vaccinated children (adjusted prevalence ratio [aPR] 0.75, 95% CI 0.56-1.01; p=0.062) and 27% to 15% among unvaccinated children (aPR 0.65, 95% CI 0.44-0.98; p=0.039). Adult prevalence remained 14% (aPR 0.92, 95% CI 0.59-1.44; p=0.72). VT carriage probability declined with age, with a decay half-life of 5.3 years (95% CI 3.2-9.0). Interpretation The PCV13 3+0 schedule in Malawi has not achieved optimal reduction in pneumococcal carriage prevalence, compared to high-income settings. This is likely due to recolonisation of vaccinated children with waning vaccine-induced immunity and suboptimal indirect protection of unvaccinated populations. Rigorous evaluation of strategies to augment vaccine-induced control, including alternative schedules and catch-up campaigns among children under 5 years old is required. Funding Bill & Melinda Gates Foundation, Wellcome Trust UK, Medical Research Council.


INTRODUCTION
Streptococcus pneumoniae is estimated to be responsible for over 500 000 deaths every year in children aged 1 to 59 months worldwide, with the highest burden among African children. 1 S. pneumoniae has over 90 immunological serotypes and is a common coloniser of the human nasopharynx, particularly in young children, resource-poor and HIV-affected populations. 1 Although most carriers are asymptomatic, pneumococcal colonisation is a necessary precursor for transmission and the development of pneumonia, meningitis, and bacteraemia. 2 In Europe and North America, routine infant administration of pneumococcal conjugate vaccine (PCV) has rapidly reduced vaccine-serotype (VT) invasive pneumococcal disease (IPD) and carriage. [3][4][5][6] Importantly, this has occurred in vaccinated and unvaccinated age groups. Thus, indirect protection resulting from diminished carriage and transmission amplifies PCV impact and cost-effectiveness. 7 Pneumococcal epidemiology in sub-Saharan Africa is characterised by high rates of carriage and transmission, differing markedly from high-income settings. 8,9 Carriage studies pre-dating PCV introduction in Kenya, 8 Mozambique, 10 Malawi, 11 The Gambia, 12 and South Africa 13 reported VT carriage prevalences ranging from 49.7% to 28.2% in under 5s with colonisation occurring early in life, 14 consistent with higher transmission rates than those found in resource rich settings.
Vaccine trials and post-routine-introduction studies in Africa have demonstrated substantial direct effects of PCV against IPD, pneumonia, and all-cause mortality among young children. [15][16][17][18] Although Kenya, 19 The Gambia, 18 Mozambique 20 , and South Africa 21 have reported VT carriage reductions, these prevalences are still higher than in industrialised countries, [22][23][24] and NVT replacement is emerging. 25 Thus, it is uncertain whether PCV introduction in sub-Saharan Africa will achieve the sustained direct or indirect protection necessary to reduce pneumococcal carriage to levels sufficient to interrupt transmission and disease. This is of particular concern in many sub-Saharan African countries where the 3+0 schedule has been implemented into infant expanded programmes on immunisation (EPI). 26 In November 2011, Malawi (previously PCV-naïve) introduced 13-valent PCV as part of the national EPI using a 3+0 schedule (6, 10 and 14 weeks of age). A three-dose catch-up vaccination campaign included infants <1 year of age. Field studies among age-eligible children have reported an even higher PCV13 uptake (90-95%) 27,28 than the 83% previously reported by WHO/UNICEF. 29 We hypothesised that despite evidence of PCV13 impact on IPD and pneumonia in Malawi, 30,31 there would be persistent VT carriage and that this would maintain transmission in both childhood and adult reservoirs. We have investigated this among PCV13-vaccinated children (in whom vaccine-induced immunity wanes after the first year of life 32 ); children too old to have received PCV13; and HIV-infected adults on antiretroviral therapy (ART) who do not routinely receive pneumococcal vaccination (previously demonstrated to have a high carriage prevalence). 33,34

Study Design
This was a prospective cross-sectional observational study using stratified random sampling to measure pneumococcal nasopharyngeal carriage in Blantyre, Malawi. Sampling consisted of a time series profile from twice-annual surveys over 2 years.

Study population and recruitment
Blantyre is located in Southern Malawi with a population of approximately 1·3 million.
Recruitment included three groups: i) randomly sampled healthy children 3-6 years old (3-6yrs) who received PCV13 as part of EPI or the catch-up campaign, ii) randomly sampled healthy children 5-10 years old (5-10yrs) who were age-ineligible (born on or before 11 November 2010 and therefore too old to receive PCV13 as part of EPI or the catch-up campaign), recruited from households and Blantyre schools; and iii) HIV-infected adults (18-

Site selection
Households and schools were selected from within three non-administrative zones representative of urban Blantyre's socioeconomic spectrum in high-density townships. These zones were further divided into clusters, allowing for approximately 25 000 adults per zone and 1 200 adults per cluster. Within each zone, two to three clusters were randomly selected per survey. Clusters were not purposely resampled but eligible if randomly selected in subsequent surveys. Within each cluster, a handheld GPS device guided study teams to a randomly selected starting point. After the first house was chosen randomly, teams moved systematically, recruiting one eligible child per household until the required number of children were recruited from each cluster. Individual schoolgoers were randomly selected from school registers, and letters sent home inviting parents or legal guardians to discuss the study and consider consenting to their child's participation.

Determining PCV13 vaccination status
A child was considered "PCV13-vaccinated" if s/he had received at least one dose of PCV13 prior to screening. Vaccination status and inclusion/exclusion criteria were further assessed from subject-held medical records (known as health passports). If a child was reported by the parent/guardian to be PCV13-vaccinated but no health passport was available, a questionnaire was applied. The questionnaire was developed by identifying in a subset of 60 participants, the four questions most commonly answered correctly by parents/guardians of children with proof of PCV vaccination. The questions included child's age when vaccinated, vaccine type (oral or injectable), anatomical site of vaccination, and which other (if any) vaccines were received at time of PCV13 vaccination. If all four questions were answered correctly, the child was recruited as "PCV13-vaccinated."

Ethics
The study protocol was approved by the College of Medicine Research and Ethics Committee,

Sample size
The sample size strategy was a pragmatic approach to allow for adequate precision of the carriage prevalence estimates. Using VT carriage as the primary endpoint, the sample size was calculated based on the precision of the prevalence estimation, assuming an infinite sampling population. Among children 3-6yrs (vaccinated), an absolute VT prevalence up to 10% was expected, with a sample of 300/survey providing a 95% confidence interval (CI) of 6·6-13·4%. Among children 5-10yrs (unvaccinated) and HIV-infected adults, an absolute VT prevalence of 20% was expected, with a sample of 200/survey providing a 95% CI of 14·5-25·5%.

Nasopharyngeal swab collection
A nasopharyngeal swab (NPS) was collected from each participant using a nylon flocked swab (FLOQSwabs TM , Copan Diagnostics, Murrieta, CA, USA) and then placed into 1·5mL skim milk-tryptone-glucose-glycerol (STGG) medium and processed at the Malawi-Liverpool-Wellcome Trust (MLW) laboratory in Blantyre, according to WHO recommendations. 35 Samples were frozen on the same day at −80°C.

Pneumococcal identification and latex serotyping
After being thawed and vortexed, 30 µL NPS-STGG was plated directly on gentamicin-sheep blood agar (SBG; 7% sheep blood agar, 5 µl gentamicin/mL) and incubated overnight at 37°C in 5% CO2. Plates showing no S. pneumoniae growth were incubated overnight a second time before being reported as negative. S. pneumoniae was identified by colony morphology and optochin disc (Oxoid, Basingstoke, UK) susceptibility. The bile solubility test was used on isolates with no or intermediate (zone diameter <14mm) optochin susceptibility. A single colony of confirmed pneumococcus was selected and grown on a new SBG plate as before.
Growth from these secondary plates was used for serotyping by latex agglutination (ImmuLex™ 7-10-13-valent Pneumotest; Statens Serum Institute, Denmark). This kit allows for differential identification of each PCV13 VT but not for differential identification of NVT serotypes; NVT and non-typeable isolates are therefore reported as NVT. Samples were batch tested on a weekly basis, blinded to the sample source. Latex serotyping results showed good concordance with whole genome sequence and DNA microarray serotyping. 36 There were no changes to protocols over the duration of the study.

Statistical analysis
Participant demographic characteristics were summarised using means, standard deviations, medians, and ranges for continuous variables and frequency distributions for categorical variables. Non-ordinal categorical variables were assessed as indicators. Carriage prevalence ratios (PR) were calculated over the study duration by log-binomial regression using months (30.4 days) between study start and participant recruitment, coded as a single time variable, allowing an estimate of prevalence ratio per month. Comparisons between surveys included estimates of PR per survey. Potential confounders were identified by testing the association between variables and included in the multivariable models when p<0.1. Adjusted prevalence ratios (aPR) were calculated using log-binomial regression. Confidence intervals are binomial exact. Statistical significance was inferred from two-sided p<0·05. Statistical analyses were completed using Stata 13.1 (StataCorp, College Station, TX, USA).

Development of non-linear regression analysis for decay rate in VT carriage
To better understand the rate at which VT carriage prevalence was decreasing, we performed non-linear regression analysis. Using empirical study data from children 3-10 years of age, a non-linear model was developed to describe the variation in risk of VT carriage with age, adjusted for baseline characteristics (crowding, number of children <5 years old in the household, gender, socioeconomic status [SES], time since vaccination [i.e. time between first dose PCV13 and date of recruitment], and date of recruitment). The analysis was left-censored at age 6 months. The analysis did not include a seasonality effect because no significant effect was detected. The population-level half-life (i.e. time in years for the carriage in the sampled cohort to reduce to one-half of its peak) was log(2)/δ), where δ = rate of decay of VT carriage prevalence with age. Model parameters were estimated by maximum likelihood, and hypothesis tests were conducted using generalized likelihood ratio tests. This analysis used R open-source software (www.r-project.org). Details of the analysis framework are in appendix 1.

Role of the funding source
The funders had no role in study design, collection, analysis, data interpretation, writing of the report or in the decision to submit the paper for publication. The corresponding author had full access to the study data and, together with the senior authors, had final responsibility for the decision to submit for publication.

Demographics and vaccination history
The three surveyed groups had similar demographics (Table 1) 4 Possessions index, mean (SD) 6·8 (3·2) 8·2 (3·3) 8.2 (3·3) ART=antiretroviral therapy. SD=standard deviation. 1 The gender distribution among adults recruited from ART Clinic is representative of the gender distribution among those attending the clinic. 2 Crowding index: Calculated as number of persons residing in main house divided by number of bedrooms in main house 3 Smoker in household: reports the percentage of households with at least one household member who smokes tobacco 4 Possession index: calculated as a sum of positive responses for household ownership of each of fifteen different functioning items: watch, radio, bank account, iron (charcoal), sewing machine (electric), mobile phone, CD player, fan (electric), bednet, mattress, bed, bicycle, motorcycle, car, television

Reduction in risk of VT carriage with age (VT carriage decay rate)
Using non-linear regression analysis to investigate the risk of VT carriage by age among children 0·5-11 years old, the probability of VT carriage was found to decline with age ( figure   4). The population-averaged immediate effect of vaccination reduced VT carriage prevalence to an estimated fraction, β=0·52 (95% CI 0·33-0·71) at 6 months of age, compared to the prevaccination baseline. Thereafter, the rate of decay in VT carriage translates to an estimated

Discussion
In this community-based assessment of pneumococcal carriage we surveyed potential reservoir populations between 4 and 6 years after the routine introduction of PCV13 in Malawi.
We found only a modest decline in VT carriage in vaccinated children 3-6yrs (26·7%). The residual VT carriage prevalence (17%) among these children was lower than that previously observed among 1-4 year olds in northern Malawi before vaccine introduction (28·2%), 11 but did not reach the levels reported in high-income low carriage prevalence settings (<5%) that have been associated with control of carriage and transmission. [22][23][24] We found a more marked decline among unvaccinated age-ineligible children 5-10yrs (44·4%), and no significant change among HIV-infected adults on ART. All 13 VTs were found among the three study groups, despite high vaccine uptake and good adherence to the three-dose schedule among vaccine-eligible children. In the light of the recent WHO Technical Expert Consultation Report on Optimization of PCV Impact, 39 these data start to address the paucity of information on the long-term impact of the widely implemented 3+0 vaccine schedules on serotype-specific disease and carriage. These findings also highlight the critical need for surveillance postvaccine introduction in high-burden resource-poor countries.
To achieve herd protection in settings with high carriage prevalences, such as Malawi, we naturally acquired immunity to subcapsular protein antigens. 41,42 Thus, the impact of vaccine predicted by transmission models from countries with low carriage prevalences may not translate to settings with high carriage prevalences. Although it has previously been assumed that PCVs would eliminate VT carriage in mature PCV programmes, 43 our data bring into question the potential for either a sustained direct or indirect effect on carriage using the current 3+0 strategy. This schedule has been widely rolled out across high-pneumococcalcarriage-prevalence and high-disease-burden sub-Saharan African countries.
In Malawi, the vaccine impact on carriage prevalence has been less than that observed in Kenya, The Gambia and South Africa which have used different vaccination strategies. Kenya reported a reduction from 34% to 9% VT carriage among PCV-vaccinated children under 5 years of age, 6 years after introduction of 10-valent PCV. 19 The Gambia reported a reduction from 50% to 13% VT carriage among children 2-5 years old, 20 months after introducing the 7-valent PCV. 44 Likewise, a study from South Africa showed a decrease of PCV13-serotype colonisation from 37% to 13% within 1 year of transitioning from PCV7 to PCV13. 45 However, none of these countries has achieved the low carriage rates seen in Europe and North America within 2 to 3 years of vaccine introduction. 3,46 We propose that the high force of infection (FOI) in Malawi and other similar settings limits a 3+0 schedule to achieving only a short duration of VT carriage control in infants. While a 2+1 schedule, as deployed in South Africa, may improve colonisation control, this is as yet unproven in other parts of Africa. Given the likely importance of an early reduction in transmission intensity to maintain a reduced carriage prevalence, a catch-up-campaign with a broader age range (ie, <2 years or <5 years of age) may also be required. Although the Global Alliance for Vaccines and Immunization (GAVI) has considerably reduced PCV costs for low-income countries, 47,48 it is also important for financial sustainability that vaccine impact be optimised (particularly the indirect effects). Indeed, the FOI and the determinants of transmission between and within age groups need to be considered as new approaches to improving vaccine-induced carriage reduction are proposed and tested.
Unlike what has been observed in low-transmission settings, 49 as well as The Gambia 25 and South Africa, 45 we observed an unexplained decrease in NVT carriage among children in Malawi. We implemented a rigorous quality assurance programme, including routine onsite supportive supervision and repeated rigorous retraining of field staff, to avoid surveillance fatigue. It is possible that serotype replacement and redistribution had already occurred before the start of this study, and that as part of a stochastic secular trend, we are now observing an overall decrease in pneumococcal carriage prevalence which may be sustained or may reverse. It is plausible that overall improvement in living conditions (improved nutrition, sanitation and disease control efforts, improved socioeconomic status) and improvements in health care (antiretroviral roll-out, improved measles and rotavirus vaccination) have resulted in a sustained overall drop in pneumococcal carriage as a result of improved health, evidenced by falling under 5 mortality in recent years. 50 However, this decrease in overall prevalence has not been observed in countries with transitioning economies such as Kenya and South Africa.
Either way, the importance of these trends in NVT carriage will become clearer as the trends in NVT invasive disease become available from these different settings.
We have previously shown incomplete pneumococcal protein antigen-specific reconstitution of natural immunity and high levels of pneumococcal colonisation in HIV-infected Malawian adults on ART. 33

Limitations
This work provides a robust community-based estimate of VT and NVT pneumococcal carriage in Blantyre. The study was conducted over a relatively short timeframe for understanding long-term temporal trends. For this reason, the statistical analysis is limited in its ability to disentangle the effects of calendar time and age-since-vaccination, given the small overlap in ages of vaccinated and unvaccinated children in our data. The analysis is also

CONCLUSION
Despite success in achieving direct protection of infants against disease, a 3+0 PCV13 schedule in Malawi has not achieved the low universal VT carriage prevalence reported in high-income settings that is required to control carriage and transmission. We propose that although vaccine-induced immunity reduces the risk of VT carriage in children up to approximately 6 months of age, in the context of a high residual FOI, this impact is limited by rapid waning of vaccine-induced mucosal immunity and pneumococcal recolonisation ( figure   5). Furthermore, we suggest that carriage reduction observed after 6 months of age largely relies on indirect vaccine protection and naturally-acquired immunity. Therefore, alternative schedules and vaccine introduction approaches in high pneumococcal carriage, high-diseaseburden countries should be revisited through robust evaluation rather than through programmatic change without supporting evidence.

Authors' contributions
TDS, NBZ, DE, NF, and RSH designed the study. All contributed to the development or design of methodology. TDS, NF and RHS oversaw the study, data collection, and data management.
CF and PD developed the statistical regression analysis. TDS, NF, and RSH conducted the statistical analysis. TDS, NF, and RSH wrote the first draft of the paper, and all authors contributed to subsequent drafts. All read and approved the final version of the report.

Declaration of interests
Dr. Bar-Zeev reports investigator-initiated research grants from GlaxoSmithKline Biologicals and from Takeda Pharmaceuticals outside the submitted work. No other competing interests were reported by authors. The duration of each survey was 3-5 months, with 2-3 weeks between surveys. 95% confidence interval error bars are shown. Prevalence of non-carriers is calculated by 1−(NVT+VT). Among children 3−6yrs old (vaccinated), there was a modest decrease in VT carriage, from 23·2% (95% CI 17·9-27·1) in survey 1 to 17·0% (95% CI 13·4-21·2) in survey 4. When adjusted for age at recruitment, the adjusted prevalence ratio (aPR) over the study duration was 0·75 (95% CI 0·56-1·01; p=0·062). Among children 5−10yrs old (unvaccinated), VT carriage decreased from 27·4% (95% CI 21·1-32·6) to 14·8% (95% CI 10·1-20·6) over the 2 years (aPR 0·65, 95% CI 0·44-0·98; p=0·039). Among HIV-infected adults on ART, VT prevalence remained unchanged, at 14·1% (95% CI 9·6-19·8) after survey 1 and 13·7% (95% CI 10·0-18·0) after survey-4 (aPR 0·92, 95% CI 0·59-1·44; p=0·72). Refer to appendix 3 for aPR and VT & NVT prevalence stratified by survey. Estimated individual probabilities (solid lines) and pointwise 95% confidence intervals (shaded regions) of VT carriage as a function of a child's age (years), for an unvaccinated child whose baseline characteristics translate to a VT carriage prevalence of 0·4 at age 6 months (red line) and for a vaccinated child with the same baseline characteristics (blue line). Both fitted lines include extrapolations beyond the range of the empiric data. The population-averaged effect of vaccination reduces VT prevalence to an estimated fraction β=0·52 (95% CI 0·33-0·71) by 6 months of age, compared to the pre-vaccination baseline. This translates to an estimated half-life of 5·3 years (95% CI 3·2-9·0), irrespective of vaccination status. For the non-linear modelling framework, see appendix 2. , with an optimal immunogenic response for vaccine-induced mucosal immunity 4 weeks later (C). Among PCV-vaccinated children, 18-weeks is the approximate vaccine-induced set point, with a rapid decrease in VT prevalence (C-E) until 6 months of age (E). At 6 months of age, there is an increase in risk of VT carriage (E-F), driven by increased force of infection in the context of waning vaccine-induced immunity, the former due partly to increased contact with other young children in the household and community. VT carriage prevalence increases until naturally acquired immunity starts to impact on colonisation (F), reducing pneumococcal carriage prevalence. Among PCV-unvaccinated children, risk of VT carriage continues to increase largely unchecked (B-D) until naturally acquired immunity starts to impact on colonisation (D), reducing pneumococcal carriage prevalence. Among these unvaccinated children, 12 months is the approximate set-point induced by naturally-acquired immunity. Indirect vaccine effects will impact on the height of C-F and B-D, as well as the rate of decline in VT carriage prevalence.