Interaction of Vaccination and Reduction of Antibiotic Use Drives Unexpected Increase of Pneumococcal Meningitis

Abstract

Antibiotic-use policies may affect pneumococcal conjugate-vaccine effectiveness. The reported increase of pneumococcal meningitis from 2001 to 2009 in France, where a national campaign to reduce antibiotic use was implemented in parallel to the introduction of the 7-valent conjugate vaccine, provides unique data to assess these effects. We constructed a mechanistic pneumococcal transmission model and used likelihood to assess the ability of competing hypotheses to explain that increase. We find that a model integrating a fitness cost of penicillin resistance successfully explains the overall and age-stratified pattern of serotype replacement. By simulating counterfactual scenarios of public health interventions in France, we propose that this fitness cost caused a gradual and pernicious interaction between the two interventions by increasing the spread of nonvaccine, penicillin-susceptible strains. More generally, our results indicate that reductions of antibiotic use may counteract the benefits of conjugate vaccines introduced into countries with low vaccine-serotype coverages and high-resistance frequencies. Our findings highlight the key role of antibiotic use in vaccine-induced serotype replacement and suggest the need for more integrated approaches to control pneumococcal infections.

Figure 1. Public health interventions and dynamics of pneumococcal meningitis (PM) in France.

(A) Monthly number of antibiotic prescriptions (beta-lactams and macrolides) in France (black curve), and 7-valent pneumococcal conjugate-vaccine (PCV7) 2-dose coverage in children <2 years old (red curve). Antibiotic data are from the NHI system; vaccine coverage data are yearly estimates from a permanent 1/97 NHI sample (red squares, text S2 and Table S2) (B) Monthly and (C) annual PM cases (per 100,000 population) reported to the National Reference Center for Pneumococci. The biennial pattern apparent in C is caused by different notification rates between odd and even years; we corrected for this effect in our analyses (Supplementary Text S1 and Text S2).

Figure 2. General model structure and derivation of the vaccine/nonvaccine (VNV) and susceptible/resistant (SR) models.

(Top) Simplified model diagram. The population was divided according to carriage status: S, susceptibles; C_VS, carriers of a vaccine-serotype, penicillin susceptible (VS) strain; C_VR, carriers of a vaccine-serotype, penicillin-resistant (VR) strain; C_NVS, carriers of a nonvaccine-serotype, penicillin-susceptible (NVS) strain; C_NVR, carriers of a nonvaccine-serotype, penicillin-resistant (NVR) strain. The per-susceptible rate of pneumococcal acquisition, λ_X, is serotype-dependent. Serotype-specific transmission rates β_X (where λ_X = β_XC_X/N) are assumed constant over time. Carriers of a given serotype can acquire another serotype (super-acquisition) at a rate reduced by a factor 1 − θ compared with noncarriers. Carriers develop PM at a serotype-dependent rate ρ_X, which is assumed to be seasonal. In the full model, the population is further stratified according to vaccination and antibiotic exposure status, so that this simplified diagram is repeated 4 times (individuals unexposed to antibiotics and unvaccinated, exposed and unvaccinated, unexposed and vaccinated, exposed and vaccinated). All model details are given in Supplementary Text S1. (Bottom) The table summarizes the simplifying transmission and invasiveness hypotheses that define the VNV and SR models. In the VNV model, strain’s transmissibility and invasiveness differ between vaccine and nonvaccine serotypes; in the SR model, strain’s transmissibility and invasiveness differ between susceptible and resistant serotypes.

Figure 3. Model fit to data.

For each strain, the observed numbers of PM (red curve), and the results of 10 stochastic runs of the SR model (thin grey lines) are shown. For completeness, the total number of PM (panel TOTAL) is also given; this series was not used for statistical inference. Note the different y-axis for each graph. VS: vaccine-serotype penicillin-susceptible; VR: vaccine-serotype, penicillin-resistant; NVS: nonvaccine-serotype penicillin-susceptible; NVR: nonvaccine-serotype penicillin-resistant.

Figure 4. Simulated counterfactual scenarios of interventions in France and expected benefits of vaccination in various settings.

(A–C) Model predictions according to intervention scenarios in France, 2001–2009: (A) Reference scenario with two interventions, (B) no PCV7 and antibiotic reduction, (C) PCV7 and no antibiotic reduction. The curves represent mean predicted annual cases (per 100,000 population) and the 95% prediction intervals for each type (green: VS, blue: VR, red: NVS, yellow: NVR) and for the total number of PM (black), obtained from 1,000 stochastic simulations of the best (SR) model with parameters fixed at their maximum likelihood values. In A–C, to avoid the biennial notification pattern, the simulated number of PM is represented before applying the observation model. In (A), the triangles represent the observed annual PM-incidence data from Fig. 1C, corrected for this biennial notification pattern (that is, divided by the reporting probability for each year). (D) Effects of vaccination and reduced antibiotic use. We simulated the effects of introducing a conjugate vaccine and reducing antibiotic use, while varying the initial vaccine-serotype coverage (proportion of PM caused by vaccine serotypes, x-axis) and the initial resistance frequency (y-axis). Here, we assume that, after a 2-year period without intervention, the two interventions are implemented simultaneously, and result in 100% vaccine coverage (all scenarios) and 0% (top left), 10% (top right), 20% (bottom left), or 30% (bottom right) reduction of antibiotic use. The colors represent the mean predicted variations of PM incidence (in %) 5 years after the implementation of the two interventions. Negative values correspond to decreases, while positive values correspond to increases.