Evaluation of statistical models of carriage to predict the impact of the 10-valent pneumococcal conjugate vaccine on invasive pneumococcal disease in Nigeria

A substantial fraction of the population-level impact of Pneumococcal Conjugate Vaccines (PCVs) on Invasive Pneumococcal Disease (IPD) is mediated through indirect effects, i.e., their capacity to protect against carriage acquisition of vaccine serotypes (VTs) among vaccinees, thereby proportionately reducing transmission and indirectly averting invasive disease in the whole population. Therefore, by relying on the consequent near elimination of VT carriage, early carriage-based models successfully captured the impact of seven-valent PCV (PCV7) in high-income settings. We sought to determine the applicability of three published statistical carriage-based models for the evaluation of PCV10 impact in Nigeria, where carriage prevalence data are available from urban and rural sites. We applied external data, with assumptions, to empirical carriage prevalence data to predict IPD incidence rate ratios (IRRs). The models assume PCV has no effect on serotype invasiveness among carriers because VT carriage is eliminated. Model 1 uses estimates of relative proportions of pre-PCV VT-IPD to predict IRRs. Model 2 uses pre-PCV serotype IPD incidence, while Model 3 uses measures of serotype invasiveness, the case-carrier ratio (CCR). Model 1 estimates the largest PCV10 impact on overall IPD (IRR:0.38 and 0.50) in the urban and rural sites, respectively. Whereas estimates from Model 2 (IRR:0.69 and 0.78) and Model 3 (IRR:0.63 and 0.70) were more conservative. VT carriage was not eliminated in our setting, so Model 1 estimates the hypothetical maximum impact. Relying entirely on indirect effects, Models 2 and 3 represent the minimum impact of PCV. Predictions would be more accurate if they accounted for direct effects among vaccinated VT carriers. This study illustrates the importance of capturing vaccination data on individuals sampled in carriage prevalence surveys designed to estimate IPD burden at population level.