Introduction
Typhoid fever, a disease caused by Salmonella enterica serovar Typhi (S. Typhi), causes over 100000 deaths in low- and middle- income countries (LMICs) annually [Reference Meiring1, Reference Antillón2]. S. Typhi is transmitted through the faecal-oral route (usually contaminated food and water) and is almost exclusively found in settings with limited access to clean water, sanitation, and hygiene (WaSH) infrastructure. The burden of typhoid fever is concentrated in Africa and South Asia and is at risk of growing substantially in the future as a result of climate change, rapid urbanization, and the spread of antimicrobial resistance.
Typhoid conjugate vaccines (TCVs), in which the S. Typhi Vi-polysaccharide antigen is conjugated to a carrier protein to boost immunogenicity, are associated with greater protection than previous typhoid vaccines, with a single dose of Typbar-TCV (Bharat Biotech) estimated to provide 83% protection against typhoid fever over 2 years of follow-up [Reference Batool3]. Since TCV can be administered to infants as young as 6 months of age, it can be incorporated into routine childhood immunization programmes to address the high burden of disease among children in endemic settings. TCV was recommended by the World Health Organization (WHO) in 2018 and has been introduced into routine immunization programmes in Bangladesh, Burkina Faso, Kenya, Liberia, Malawi, Nepal, Niger, Pakistan, Samoa, Tuvalu, and Zimbabwe as of July 2026 [Reference Hancuh4–8].
While the protective efficacy of TCV against clinical typhoid fever has been well established through multiple field trials, it is unclear whether the vaccine prevents S. Typhi infection and can induce population-level ‘herd’ protection, or if vaccination simply attenuates symptomatic disease. This is important to clarify, as many of the mathematical models that are used to inform policy decisions assume a transmission-blocking mechanism [Reference Burrows9]. Indirect protection was observed in a cluster-randomized trial of ViPS vaccine in Kolkata, India, where both adults and children were vaccinated (indirect protection: 44%, 95% CI: 2%–69%), but not in Karachi, Pakistan, where vaccination was limited to children (indirect protection: −10%, 95% CI: −116%−44%) [Reference Dipika10, Reference Khan11]. In the TyVAC-Bangladesh cluster-randomized trial, in which children aged 9 months to 15 years in Dhaka were vaccinated with Typbar-TCV, indirect protection was low and not statistically significant (indirect protection: 19%, 95% CI: −12%−41%) [Reference Qadri12]. However, indirect protection was only a secondary endpoint in the trial, and consequently the study was not explicitly powered to measure it.
The reasons underlying the weak indirect protection from TCV in the TyVAC-Bangladesh trial have not been fully explored. Conjugate vaccines against respiratory-transmitted bacterial pathogens, including H. influenzae type b (Hib), pneumococcus, and meningococcus, have all demonstrated a robust impact on transmission [Reference Shiri13–Reference Kristiansen15], although this pattern may not be generalizable to enteric pathogens. In an immunogenicity trial in India, TCV recipients were 63% less likely than ViPS recipients to have serological evidence of S. Typhi infection in the 2 years following vaccination [Reference Voysey and Pollard16]. TCV recipients were also less likely than unvaccinated recipients to shed S. Typhi in stool samples in controlled human challenge studies [Reference Gibani17]. Indirect protection in the TyVAC-Bangladesh trial may have been diluted by outside contamination (transmission across cluster boundaries). However, limiting the analysis to the innermost cluster residents did not reveal significant changes in indirect protection [Reference Khanam18]. Higher vaccine coverage may be needed to achieve measurable herd protection: only 64% of children between 9 months and <16 years of age were vaccinated in this trial, and older age groups were not eligible for vaccination, although recent analysis shows high seroincidence among adults in Dhaka [Reference Walker19]. Ahmmed et al. observed statistically significant indirect protection (47%, 95% CI: 3%–71%) among unvaccinated residents of clusters in the highest quartile of population-wide coverage [Reference Ahmmed20]. Meanwhile, Zhang et al. stratified unvaccinated cluster residents by TCV coverage at the school and household level, but did not observe significant indirect protection in any group [Reference Zhang21]. Results from mathematical modelling suggest that even if TCV is effective at blocking infection, indirect protection may be lower than expected due to the role of chronic carriers in population-level transmission [Reference Pitzer22]. This is because chronic carriers can continue to shed S. Typhi for years or decades after infection, and therefore the number of infectious carriers in the population is expected to decline relatively slowly following vaccine introduction, even as acute infections fall.
A better understanding of the low indirect protection observed in the TyVAC-Bangladesh trial and its causes is needed to determine whether TCV can have a meaningful impact on transmission when implemented at the population level, and if so, how these transmission effects can be maximized to control typhoid fever. In this study, we used transmission dynamic models to simulate the TyVAC-Bangladesh trial, identify profiles of vaccine protection against infection and chronic carrier transmissibility which are consistent with trial observations, and evaluate strategies for increasing the population-level impact of vaccination.
Methods
Study description
The TyVAC-Bangladesh study was a cluster-randomized trial of Typbar-TCV conducted in Dhaka, Bangladesh between April 2018 and October 2019. The study area was divided into 150 contiguous clusters, which were randomly allocated to receive either TCV or a control vaccine (Japanese encephalitis vaccine (JEV)) in a 1:1 ratio. Eligible and consenting participants between 9 months and <16 years of age received a single vaccine dose and were followed for an average of 17.1 months (n = 61657 total vaccinated participants). The average vaccine coverage among eligible residents was 64% during the study period. The primary trial endpoints were total protection against typhoid fever in TCV recipients relative to JEV recipients (29 vs. 192 blood-culture-confirmed cases; total protection: 85%, 95% CI: 78%–90%) and overall protection against typhoid fever among all residents (144 vs. 331 blood-culture-confirmed cases; overall protection: 57%, 95% CI: 45%–67%) [Reference Qadri12]. Indirect protection against typhoid fever among non-vaccinated residents was a secondary trial endpoint (115 vs. 139 blood-culture-confirmed cases; indirect protection: 19%, 95% CI: −12%−41%) [Reference Qadri12].
Prior to the TyVAC-Bangladesh trial, the Bangladesh arm of the Strategic Typhoid Alliance Across Asia and Africa (STRATAA) study was conducted in the same area of Dhaka (Mirpur) to characterize the local burden and epidemiology of enteric fever [Reference Meiring23]. The STRATAA study used data from passive blood-culture surveillance, household census activities, and a healthcare utilization survey to calculate the age-specific adjusted incidence of typhoid fever. The STRATAA study also included a serological survey, in which paired blood samples were used to estimate the age-specific incidence of typhoidal salmonella infections based on several antigen targets [Reference Walker19]. For this study, we focused on seroincidence estimates based on the HlyE antigen, due to its strong and specific association with acute enteric fever cases at the individual and population levels [Reference Ahmmed20, Reference Meiring23]. However, HlyE is expressed by both S. Typhi and S. Paratyphi A; to isolate serological responses to the former and exclude the latter, we multiplied seroincidence estimates by the age-specific fraction of S. Typhi among blood-culture-positive enteric fever cases in the STRATAA-Bangladesh study.
The TyVAC-Bangladesh trial (ISRCTN11643110) was reviewed by and received ethical approval from the ICDDR,B Research and Ethical Review Committees, the Oxford Tropical Health Research Ethics Committee, and the University of Maryland, Baltimore institutional review board, and was overseen by a local data safety and monitoring board (DSMB) convened by ICDDR,B as well as an international DSMB convened by the University of Maryland. The STRATAA study (ISRCTN12131979) received ethical approval from the ICDDR,B Research Review Committee, the Malawi National Health Sciences Research Committee, the Nepal Health Research Council, and the Oxford Tropical Research Ethics Committee.
Model description and calibration
Our primary analysis is based on a transmission dynamic model of typhoid infection and disease fit to clinical incidence data [Reference Pitzer22], which has previously been used to evaluate the impact of typhoid vaccination and investigate the drivers of typhoid fever outbreaks [Reference Burrows9, Reference Mylona24–Reference Pitzer27]. In this model, individuals are born into a susceptible and previously unexposed class S1 and are infected and moved into the I1 class at rate λ (the force of infection). This force of infection represents the total risk of transmission across person-to-person, environmental, and water-borne pathways. After an average of 4 weeks, an age-specific fraction of primary infections become lifelong chronic carriers (C), while the remainder recover into the R1 class. Recovered individuals remain immune to reinfection for an average of 1 year before waning into the S 2 class. Reinfections (I2) are assumed to be subclinical.
Newly available serological data from the STRATAA-Bangladesh study suggest that the overall incidence of S. Typhi infection (a combination of symptomatic typhoid fever cases and subclinical infections) may be highest within the first few years of life, even though the incidence of symptomatic typhoid fever peaks later in childhood [Reference Walker28]. In order to reconcile these observations, we developed a more flexible alternative model in which 1) additional compartments were added in order to independently track second, third, and subsequent infections (I2, I3, and I4, respectively, Supplementary Figure S1); 2) reinfections are not necessarily subclinical, and instead we assume the symptomatic fraction fi decreases with each subsequent infection i (f1 ≥ f2 ≥ f3 ≥ f4); and 3) the fraction of infections which progress to detectable typhoid fever cases may be lower among those aged <5 years than among older age groups (defined by the ratio rf) to account for potential differences in disease presentation and barriers to case detection among young children. For both models, we assume that the mean duration of infection (in the I classes) and immunity (in the R classes) is independent of infection history and that reinfections do not result in chronic carriage. Finally, we assumed that acute primary infections (I1) and reinfections (I2–4) transmit to susceptible individuals at the same rate (β), while the transmissibility of lifelong chronic carriers is a fraction rC of acute infections.
Within the model, we stratified the population into eight age groups: <9 months, 9 months to <5 years, 5 to <10 years, 10 to <15 years, 15 to <16 years, 16 to <30 years, 30 to <50 years, and 50+ years. These age groups were selected to align with the aggregated age-specific incidence in the STRATAA study and the ages that were eligible for vaccination in the TyVAC-Bangladesh trial (9 months to <16 years) [Reference Qadri12]. For susceptible populations aged <5 and 5 to <10 years, we estimated a relative risk of infection (b1 and b2, respectively) compared with susceptible individuals >10 years old, to account for potential differences in mixing and exposure between age groups. For instance, as children age, the transition from breastfeeding to other water sources and greater social mixing outside the household may increase the risk of exposure to S. Typhi [Reference Meiring1]. Models were run for an initial 100-year burn-in period, during which population demographics and transmission dynamics converged to a steady state, followed by a 2-year observation period.
In our primary model, we simultaneously estimated the S. Typhi transmission rate (β), the relative risk of infection in the <5 and 5 to <10-year-old age groups (b1 and b2, respectively), and the fraction of primary infections which progress to symptomatic disease (f1) by fitting the modelled symptomatic incidence to the adjusted clinical typhoid incidence from the STRATAA-Bangladesh study (Figure 1a,b). In this model, we assume that all reinfections are subclinical (f2–4 = 0) and that the fraction of primary infections which progress to detectable symptomatic disease is independent of age (rf = 1). We fit the alternative model in two stages: first, we estimated the values of the transmission parameters (β, b1, and b2) by fitting the modelled incidence of infection to the adjusted STRATAA HlyE seroincidence (Figure 1c). We then used this calibrated incidence of infection to estimate the symptomatic fraction parameters (f1–4, rf) by fitting the modelled symptomatic incidence to the adjusted STRATAA typhoid incidence (Figure 1d). In both the primary and alternative models, demographic rates (births, deaths, and migration), the duration of acute infection and post-infection immunity, and the fraction of primary infections which become lifelong chronic carriers are fixed based on values from the literature, census data, or reasonable assumptions. The remaining parameters were estimated in a maximum likelihood framework with age group-specific, Poisson-distributed case counts. Models were refit independently for values of rC (the relative transmissibility of chronic carriers) between 0% and 100% in 5% increments. A description of all model parameters and their values is provided in Table 1.
Estimated incidence of clinical typhoid fever and seroincidence of Salmonella Typhi infection for the primary and alternative models. (a) Fitted typhoid fever incidence in the primary model (lines) against adjusted typhoid incidence in the STRATAA study (bars). (b) Fitted typhoid fever incidence in the alternative model (lines) against adjusted typhoid incidence in the STRATAA study (bars). (c) Estimated incidence of S. Typhi infection in the primary model. (d) Fitted incidence of S. Typhi infection in the alternative model and adjusted HlyE seroincidence in the STRATAA study.
Figure 1. Long description
A multi-panel figure with four graphs labeled A through D. All graphs share an X-axis of Age in years with five categories: [0-5), [5-10), [10-15), [15-30), and 30 plus.
Panel A: Typhoid Fever Incidence, Primary Model (lines) vs Clinical Data (bars). The Y-axis is Typhoid Fever Cases per 100 Person-Years ranging from 0 to 4. Dark grey bars show a peak in the [5-10) age group at approximately 3.2. A blue line with dots follows the bar heights closely. A color scale for r C ranges from 0.00 (dark) to 1.00 (light blue).
Panel B: Typhoid Fever Incidence, Alternative Model (lines) vs Clinical Data (bars). Similar to Panel A, but the blue line shows a sharper peak at the [5-10) age group, exceeding the bar height, and a steeper decline in older age groups.
Panel C: Infection Incidence, Primary Model. The Y-axis is Infections per 100 Person-Years ranging from 0.0 to 10.0. Multiple blue lines represent different r C values. All lines show a sharp increase from age [0-5) to a peak at [5-10) of approximately 9.0, followed by a drop to a plateau of approximately 7.0 for all older age groups.
Panel D: Infection Incidence, Alternative Model (lines) vs Seroincidence Data (bars). The Y-axis is Infections per 100 Person-Years ranging from 0 to 80. Dark grey bars show the highest incidence in the [0-5) group at over 70, decreasing steadily with age. The blue line closely tracks the top of these bars, showing a downward trend that levels off around 30 infections per 100 person-years for the 15-30 and 30 plus age groups.
Parameter values for the transmission dynamic model
Table 1. Long description
The table is divided into four horizontal sections.
1. Fixed parameters: Both models.
- Births per 1000 person-years: 18.23 per 1000 person-years (Source: S T R A T A A census data).
- Net mortality plus migration rate per 1000 person-years: Age 0 to 5 is minus 3.72; Age 5 to 10 is 1.83; Age 10 to 15 is minus 2.01; Age 15 to 30 is minus 1.95; Age 30 to 50 is 2.83; Age greater than or equal to 50 is 8.24 (Source: Fitted to S T R A T A A census data).
- Mean duration of acute infection: 4 weeks (Source: Hornick et al.).
- Mean duration of post-infection immunity: 1 year (Source: Assumption).
- Fraction of primary infections which become lifelong chronic carriers: Age 0 to 15y is 0.3 percent; Age 15 to 30y is 1.5 percent; Age 30 to 50y is 6.6 percent; Age greater than or equal to 50y is 9 percent (Source: Ames et al.).
2. Primary model: Fitted and fixed parameters.
- Transmission rate beta: 0.13 to 0.27.
- Relative risk of infection versus age greater than or equal to 10 years: 0 to 5 year olds is 0.70 to 0.71; 5 to 10 year olds is 1.34 to 1.35.
- Symptomatic fraction primary infection: 0.59 to 0.61.
- Symptomatic fraction reinfections: 0 (Source: Assumption).
3. Alternative model: Fitted parameters.
- Transmission rate beta: 0.26 to 0.30.
- Relative risk of infection versus age greater than or equal to 10 years: 0 to 5 year olds b sub 1 equals 5.24; 5 to 10 year olds b sub 2 equals 2.62.
- Symptomatic fraction primary infection age greater than or equal to 5 years old: 0.39.
- Relative symptomatic fraction: 2nd versus 1st infection is 1; 3rd versus 2nd infection is 1; 4th and later versus 3rd infection is 0.04; Age 0 to 5 years versus older r sub f is 0.14.
4. Varied parameters.
- Infectiousness of chronic carriers relative to acute infections r C: Varied from 0 percent to 100 percent in increments of 5 percent.
- Vaccine protection against infection V E sub infection: Varied from 0 percent to 100 percent in increments of 5 percent.
a Range of parameter estimates across rC = 0% to 100%.
b Parameter estimate constrained between 0 and 1 during fitting.
Simulating vaccination in the TyVAC-Bangladesh trial
To model vaccination, we created a vaccinated counterpart for each compartment (S1 and Sv1, I1 and Iv1, etc.). At the end of the burn-in period, a proportion of individuals in the eligible age groups were moved from each compartment to its vaccinated counterpart. Unless otherwise stated, we assumed a target vaccine coverage of 64% among those aged 9 months to <16 years, based on the average coverage level over the duration of the trial [Reference Qadri12]. To keep vaccine coverage constant over time in the eligible age groups, we vaccinated the same proportion of infants at 9 months of age. A similar effect was achieved in the TyVAC-Bangladesh trial by performing catch-up vaccine campaigns every 6 months. For vaccinated individuals, the force of infection is reduced by a fraction VEinfection, while the symptomatic fraction is reduced by a fraction VEprogression, resulting in ‘leaky’ vaccine protection [Reference Halloran, Longini and Struchiner31]. Given the limited duration of the TyVAC-Bangladesh trial (17.1 months of follow-up on average), we assumed no vaccine waning over time in our primary analysis. As a sensitivity analysis, we considered a scenario in which the duration of vaccine protection followed an exponential distribution with an average duration of 5 years.
Using the fitted models, we simulated the TyVAC-Bangladesh trial for 441 unique combinations of rC and VEinfection, with values of each parameter ranging from 0% to 100% in increments of 5%. For each of these parameter combinations, we ran two models, one corresponding to the TCV-receiving treatment clusters and one corresponding to the control clusters. To account for the impact of contamination (transmission between those residing in TCV and control clusters, which is expected to dilute the indirect protection), we reduced vaccine coverage in the treatment cluster model by 10% from the target coverage level, and increased vaccine coverage in the control cluster model from 0% to 10%.
Estimating population-level vaccine effects
To calculate total vaccine protection from the model output, we compared the expected incidence of typhoid fever between eligible TCV and JEV recipients in the treatment and control clusters, respectively. We calculated overall vaccine protection by comparing the expected incidence of typhoid fever between all individuals in the treatment and control clusters. Finally, to calculate indirect vaccine protection, we compared the expected incidence of typhoid fever between the unvaccinated populations of the treatment and control clusters. We standardized modelled incidence by the target vaccine coverage and the age distribution of the vaccinated, unvaccinated, and overall population (see Supplementary Methods). We classified pairs of rC and VEinfection values as inconsistent with the TyVAC-Bangladesh trial results if the projected total, overall, or indirect effects of vaccination did not fall within the 95% confidence intervals of the trial estimates (total VE: 78%–90%, overall VE: 45%–67%, and indirect VE: −12%−41%) for any value of VEprogression between 0% and 100%.
Simulating counterfactual vaccine trials
To evaluate the sensitivity of the trial results to vaccine coverage and age eligibility, we modelled vaccine effects under three scenarios:
-
• Counterfactual #1: increasing vaccine coverage to 90% of eligible children aged ≥9 months to <16 years without extending vaccination to older age groups
-
• Counterfactual #2: extending vaccination to all children and adults aged ≥9 months at 64% coverage
-
• Counterfactual #3: extending vaccination and increasing coverage to 90% of all children and adults aged ≥9 months
We evaluated each scenario for three parameter sets which reproduce the 19% indirect protection estimate from the trial, corresponding to low, medium, and high values of VEinfection and rC, for both the primary and alternative models. For each parameter set, we calibrated VEprogression to reproduce the 85% total protection observed in the trial (see Supplementary Methods).
To compare the efficiency of increasing vaccine coverage among children and extending vaccination to adults, we calculated the incremental number of averted cases per additional dose for each of the scenarios described above, relative to a corresponding strategy in which vaccination is more limited. Specifically, we evaluated the efficiency of increasing vaccine coverage from 0% to 64% and from 64% to 90% of children aged ≥9 months to <16 years, and of expanding vaccination from children only to both children and adults at the same coverage level (64% and 90%).
All modelling and analysis were performed in R (v4.1.1). Code and data to reproduce this analysis are available at github.com/pitzerlab/indirectprotection.
Results
Both the primary and alternative models captured the magnitude and age pattern of typhoid fever incidence observed at the Bangladesh STRATAA site, with incidence peaking among children aged 5 to <10 years and then declining through adulthood, although the primary model reproduced these trends slightly more accurately than the alternative model (Figure 1). In the primary model, which was fit exclusively to the clinical surveillance data, the initial increase in typhoid incidence with age during childhood reflects a higher risk of infection in children aged 5 to <10 years (Figure 1a,b). In the alternative model, where the incidence of infection was fit to the seroincidence data, the risk of infection decreases with age in early childhood, even as typhoid fever incidence increases (Figure 1c,d). In both models, the overall risk of infection (a combination of symptomatic typhoid fever cases and subclinical infections) plateaus through adolescence and adulthood, while the incidence of symptomatic typhoid fever continues to decline with age as immunity from prior infection reduces the risk of symptomatic disease. The age-specific incidence of infection was 4- to 15-fold higher in the alternative model than the primary model (27–73 vs. 5–9 per 100 person-years).
In treatment clusters, the overall modelled incidence of typhoid fever ranged from 160 to 1100 per 100000 person-years in the primary model, and from 290 to 1100 per 100000 person-years in the alternative model, depending on the value of rC, VEinfection, and VEprogression. In control clusters, the overall modelled incidence of typhoid fever ranged from 800 to 1100 per 100000 person-years in the primary model, and from 1000 to 1100 per 100000 person-years in the alternative model. Indirect protection ranged from 0% to 67% for the primary model, and from 0% to 54% for the alternative model, depending on the value of VEinfection and rC (Figures 2 and 3, Table 2). Indirect protection increased with VEinfection and decreased with rC, when holding the other parameter constant. For both the primary and alternative models, we identified a set of VEinfection and rC values which were consistent with the TyVAC-Bangladesh trial results (Figure 3, Supplementary Figures S3–S6), in that the modelled total, overall, and indirect effects of vaccination all overlapped with the 95% confidence interval of the trial estimates for some range of non-negative VEprogression values. The values of VEinfection that were consistent with the trial results ranged from 15% to 85% for the primary model, and from 60% to 90% for the alternative model, although this range was narrower for specific values of rC. Of these qualifying parameters, the values of VEinfection that most closely reproduced the 19% indirect protection observed in the trial ranged from 15% to 85% in the primary model (for rC = 0% and 35%, respectively), and from 60% to 75% in the alternative model (for rC = 0% and 100%, respectively). In sensitivity analyses, modelled indirect protection was up to 16% higher if there was no contamination across cluster boundaries, and up to 15% lower if TCV-induced immunity waned after an average of 5 years (see Supplementary Findings).
Model-predicted indirect protection for different levels of direct protection against infection and relative transmissibility of chronic carriers. Each line represents the model-predicted indirect protection (y-axis) under simulated cluster-randomized trial conditions mimicking the TyVAC-Bangladesh trial as a function of TCV efficacy against infection, with lighter coloured lines corresponding to higher transmissibility of lifelong chronic carriers. The top (a) and bottom (b) panels correspond to the primary and alternative models, respectively. The solid and dashed horizontal lines represent the point estimate (19%) and 95% confidence interval upper bound (41%) of indirect protection in the TyVAC-Bangladesh trial.
Figure 2. Long description
A multi-panel figure with two stacked line graphs labeled A and B. Both graphs share the same axes. The x-axis represents Direct V E Against Infection from 0% to 100%. The y-axis represents Projected Indirect V E Against Typhoid Fever from 0% to 100%. A legend indicates that line colors represent Relative Transmissibility of Chronic Carriers r C, ranging from dark blue at 0% to light blue at 100%.
Panel A, titled Primary Model, Transmission Dynamics Fit to Clinical Surveillance Data, shows a series of curves originating at the 0,0 origin. The curves show a logarithmic-style increase. The darkest blue line at the top reaches approximately 68% indirect protection at 100% direct efficacy. As the r C percentage increases and the lines lighten, the curves flatten significantly, with the lightest blue line reaching only about 10% indirect protection.
Panel B, titled Alternative Model, Transmission Dynamics Fit to Serologic Data, shows curves that follow an exponential-style increase. All lines remain low and tightly clustered until approximately 40% direct efficacy, after which they diverge. The darkest blue line reaches approximately 55% indirect protection at 100% direct efficacy, while the lightest blue line reaches approximately 28%.
In both panels, two horizontal reference lines are present. A solid gray line is positioned at 19% on the y-axis, labeled TyVac Trial point estimate. A dashed gray line is positioned at 41% on the y-axis, labeled TyVac Trial 95% C I Upper Bound.
Each pixel represents the projected indirect protection in the TyVAC-Bangladesh trial for a given level of TCV efficacy against S. Typhi infection (x-axis) and relative transmissibility of chronic carriers compared to acute cases (y-axis). The region bounded by the solid black line contains parameter combinations for which model outputs are consistent with the 95% confidence intervals of the total, overall, and indirect protection in the TyVAC-Bangladesh trial. The dashed line interpolates parameter sets which reproduce the 19% trial estimate of indirect protection. The top (a) and bottom (b) panels correspond to the primary and alternative models, respectively.
Figure 3. Long description
A two-panel attribute space diagram.
Both panels share identical axes. The x-axis represents Direct V E Against Infection from 0% to 100%. The y-axis represents Relative Transmissibility of Chronic Carriers (r C) from 0% to 100%. A color scale to the right indicates Indirect V E, ranging from dark red (0%) through white (19%) to dark blue (70%).
Panel A, titled Primary Model, Transmission Dynamics Fit to Clinical Surveillance Data:
- The heatmap shows a gradient where Indirect V E increases as Direct V E increases and relative transmissibility decreases.
- A solid black line outlines a jagged, roughly triangular region in the center-right, representing parameter combinations consistent with the 95% confidence intervals of the TyVAC-Bangladesh trial.
- A dashed line starts at approximately 15% Direct V E on the x-axis and slopes upward to the right, interpolating the 19% trial estimate of indirect protection.
Panel B, titled Alternative Model, Transmission Dynamics Fit to Serologic Data:
- The heatmap shows a sharper transition to blue (higher Indirect V E) concentrated on the far right of the x-axis.
- The solid black line outlines a tall, narrow rectangular region on the right side of the plot, between 60% and 90% Direct V E.
- A dashed line rises steeply within this bounded region, starting at 60% Direct V E and reaching 100% relative transmissibility at roughly 75% Direct V E.
Modelled indirect protection by vaccine efficacy against infection and chronic carrier transmissibility
Table 2. Long description
The table is divided into two main horizontal sections: Primary model and Alternative model. Both sections use V E sub infection as the first column with values 0 percent, 25 percent, 50 percent, 75 percent, and 100 percent. The subsequent three columns represent Modelled indirect protection at different levels of r sub C.
Primary model data:
* At r sub C equals 0 percent: values are 0 percent, 29 percent, 48 percent, 60 percent, and 68 percent.
* At r sub C equals 50 percent: values are 0 percent, 7 percent, 11 percent, 16 percent, and 20 percent.
* At r sub C equals 100 percent: values are 0 percent, 3 percent, 6 percent, 8 percent, and 10 percent.
Alternative model data:
* At r sub C equals 0 percent: values are 0 percent, 6 percent, 15 percent, 31 percent, and 54 percent.
* At r sub C equals 50 percent: values are 0 percent, 5 percent, 12 percent, 23 percent, and 36 percent.
* At r sub C equals 100 percent: values are 0 percent, 5 percent, 10 percent, 19 percent, and 28 percent.
If the 85% total protection and 19% indirect protection estimate from the TyVAC-Bangladesh trial represent the ‘true’ effects under the observed conditions, we estimate that increasing vaccine coverage to 90% of eligible children would have resulted in 28%–29% indirect protection and 62% overall protection in the primary model, and 33%–38% indirect protection and 62%–64% overall protection in the alternative model, depending on the values of VEinfection and rC (Figures 4 and 5). Expanding vaccination to adults at 64% coverage had an even greater impact, resulting in 40%–47% indirect protection and 70%–73% overall protection in the primary model, and 42%–57% indirect protection and 71%–79% in the alternative model. Combining these effects by increasing vaccine coverage to 90% in children and adults amplified their impact, with a projected 52%–61% indirect protection and 86%–89% overall protection in the primary model, and 59%–80% indirect protection and 89%–94% overall protection in the alternative model. However, increasing vaccine coverage in children was more efficient at preventing cases than expanding vaccination to adults (44–59 vs. 9–17 averted cases per 1000 doses over 2 years of follow-up; Table 3). The impact of expanding vaccine eligibility and increasing coverage beyond that observed in the TyVAC-Bangladesh trial was greatest for lower values of rC.
Projections of indirect protection under alternative vaccine coverage scenarios. Bar heights represent model-based projections of indirect protection under TyVAC-Bangladesh trial conditions (baseline) and in alternative scenarios in which vaccine eligibility is expanded to adults, vaccine coverage in the eligible population is increased from 64% to 90%, or both. Parameter sets are selected to reproduce the 19% indirect protection observed in the TyVAC-Bangladesh trial under baseline conditions (64% vaccine coverage among eligible children).
Figure 4. Long description
The image consists of two vertical panels, A and B. Both panels share the same y-axis, Projected Indirect V E Against Typhoid Fever, ranging from 0% to 100%, and the same x-axis categories representing four vaccine coverage scenarios.
Panel A is titled Primary Model, Transmission Dynamics Fit to Clinical Surveillance Data. The legend defines three categories:
- Green: High V E against infection 76% and r C 30%.
- Orange: Medium V E against infection 46% and r C 15%.
- Purple: Low V E against infection 15% and r C 0%.
Data clusters in Panel A from left to right:
1. 64% Vaccine Coverage, Children Only (Baseline): All three bars are at 19%.
2. 90% Vaccine Coverage, Children Only: Green 28%, Orange 28%, Purple 29%.
3. 64% Vaccine Coverage, Children and Adults: Green 40%, Orange 41%, Purple 47%.
4. 90% Vaccine Coverage, Children and Adults: Green 52%, Orange 53%, Purple 61%.
Panel B is titled Alternative Model, Transmission Dynamics Fit to Serologic Data. The legend defines three categories:
- Green: High V E against infection 77% and r C 100%.
- Orange: Medium V E against infection 68% and r C 50%.
- Purple: Low V E against infection 60% and r C 10%.
Data clusters in Panel B from left to right:
1. 64% Vaccine Coverage, Children Only (Baseline): All three bars are at 19%.
2. 90% Vaccine Coverage, Children Only: Green 33%, Orange 35%, Purple 38%.
3. 64% Vaccine Coverage, Children and Adults: Green 42%, Orange 47%, Purple 57%.
4. 90% Vaccine Coverage, Children and Adults: Green 59%, Orange 67%, Purple 80%.
Projections of overall protection under alternative vaccine coverage conditions. Bar heights represent model-based projections of overall protection under TyVAC-Bangladesh trial conditions (baseline) and in alternative scenarios in which vaccine eligibility is expanded to adults, vaccine coverage in the eligible population is increased from 64% to 90%, or both. Parameter sets are selected to reproduce the 19% indirect protection and 85% total protection observed in the TyVAC-Bangladesh trial under baseline conditions (64% vaccine coverage among eligible children).
Figure 5. Long description
The image consists of two vertically stacked grouped bar charts, Panel A and Panel B. Both charts share the same y-axis, Projected Overall V E Against Typhoid Fever, ranging from 0% to 100% in 10% increments. The x-axis for both charts lists four scenarios from left to right: 64% Vaccine Coverage, Children Only (Baseline); 90% Vaccine Coverage, Children Only; 64% Vaccine Coverage, Children and Adults; and 90% Vaccine Coverage, Children and Adults.
Each scenario contains three bars color-coded by V E against infection and r C: green (High), orange (Medium), and purple (Low).
Panel A: Primary Model, Transmission Dynamics Fit to Clinical Surveillance Data.
- Baseline: All three bars are at 46%.
- 90% Children Only: All three bars are at 62%.
- 64% Children and Adults: Green 70%, Orange 71%, Purple 73%.
- 90% Children and Adults: Green 86%, Orange 86%, Purple 89%.
Panel B: Alternative Model, Transmission Dynamics Fit to Serologic Data.
- Baseline: All three bars are at 44%.
- 90% Children Only: Green 62%, Orange 63%, Purple 64%.
- 64% Children and Adults: Green 71%, Orange 74%, Purple 79%.
- 90% Children and Adults: Green 89%, Orange 91%, Purple 94%.
In both panels, expanding coverage to adults and increasing coverage to 90% results in a step-wise increase in projected overall protection.
Efficiency of increasing vaccine coverage in children and extending vaccination to adults
Table 3. Long description
The table is structured with six columns. The first two columns define the Model (Primary or Alternative) and the Parameter set (Low, Medium, or High V E sub infection and r sub c). The remaining four columns represent the incremental averted cases per additional 1000 doses administered under four scenarios.
Primary Model data:
* Low V E sub infection 15% and r sub c 0%: 57 (0% to 64% children), 45 (64% to 90% children), 14 (64% coverage including adults), 9 (90% coverage including adults).
* Medium V E sub infection 46% and r sub c 15%: 57 (0% to 64% children), 44 (64% to 90% children), 13 (64% coverage including adults), 9 (90% coverage including adults).
* High V E sub infection 76% and r sub c 30%: 57 (0% to 64% children), 44 (64% to 90% children), 13 (64% coverage including adults), 9 (90% coverage including adults).
Alternative Model data:
* Low V E sub infection 60% and r sub c 10%: 54 (0% to 64% children), 59 (64% to 90% children), 17 (64% coverage including adults), 10 (90% coverage including adults).
* Medium V E sub infection 68% and r sub c 50%: 54 (0% to 64% children), 54 (64% to 90% children), 15 (64% coverage including adults), 10 (90% coverage including adults).
* High V E sub infection 77% and r sub c 100%: 54 (0% to 64% children), 52 (64% to 90% children), 14 (64% coverage including adults), 9 (90% coverage including adults).
Discussion
TCVs confer strong direct protection against typhoid fever, as observed in multiple randomized controlled trials [Reference Batool3, [Reference Khanam18], [Reference Shakya32], Reference Patel33], human challenge studies [Reference Jin34], and observational studies [Reference Yousafzai35–Reference Hoffman37]. However, the absence of significant indirect protection in the TyVAC-Bangladesh cluster-randomized trial raises the question of whether TCV is effective at preventing transmission and can induce meaningful herd immunity. By simulating the trial with a mathematical model, we found that the low indirect protection observed in the TyVAC-Bangladesh trial does not necessarily imply low TCV efficacy against transmission. In both our primary and alternative models, vaccines which did not confer any protection against S. Typhi infection and transmission were inconsistent with the observed trial results, because overall protection would have been too low. On the other hand, scenarios in which TCV confers strong direct protection against infection were plausible in both models for a wide range of carrier transmission assumptions.
Our analysis suggests that the ineligibility of adults for vaccination was the primary explanation for low indirect protection in the TyVAC-Bangladesh trial, and consequently for the limited statistical power to detect this effect. Moderate vaccine coverage among eligible children and contamination across cluster boundaries were secondary drivers. It is important to emphasize that these factors have practical explanations and do not represent failures of trial design or execution. Vaccination in the TyVAC-Bangladesh trial was restricted to children since they account for most of the burden of disease in Bangladesh and other typhoid-endemic settings [38]. Sample size calculations for the trial were based on detecting significant total and overall protection, assuming a vaccine coverage level of 60% of the age-eligible population [Reference Qadri12]. For a clinical trial of a relatively new vaccine in a setting with high levels of migration, achieving 64% coverage in eligible children required extensive planning, community engagement, and multiple rounds of immunization.
TCV introduction in the general population could induce stronger indirect protection than that observed in the TyVAC-Bangladesh trial, at least temporarily. Administrative data suggest that over 90% of eligible children were vaccinated with TCV in the 2025 national campaign [8]. Furthermore, the effect of migration and transmission from non-vaccinated areas would be minimized if TCV were introduced over a large geographic region rather than in small neighbourhood-level clusters. Our analysis suggests stronger herd effects could be achieved by vaccinating adults. Specifically, if adults had been vaccinated in the TyVAC-Bangladesh trial at the same coverage level as children, we estimate that indirect protection would have increased to between 40% and 64%, depending on modelling assumptions. This effect is comparable to or greater than the 44% indirect protection in the Kolkata cluster-randomized trial of ViPS, in which children and adults were vaccinated and total protection was only 61% [Reference Dipika10]. However, vaccination of adults in endemic areas would not necessarily be cost-effective or a public health priority, since the burden of clinical disease is concentrated in children. In our analysis, vaccinating children was more than three times as efficient at preventing typhoid fever cases as vaccinating adults on a per-dose basis. In addition, the waning of TCV-induced immunity could erode the strength of indirect protection over time. In a test-negative study performed in the years following the initial TyVAC-Bangladesh trial, TCV protection against typhoid fever declined from 93% (95% CI: 86%–96%) to 39% (95% CI: −23%−70%) between the first and fifth years post-vaccination [Reference Qadri39]. The durability of direct and indirect protection from TCV and the cost-effectiveness of including adults in TCV campaigns should be evaluated in future modelling analyses.
A key observation from previous modelling analyses is that indirect vaccine protection is expected to be lower when chronic carriers drive a greater share of transmission [Reference Pitzer22]. While acute infections generally recover after a few weeks, chronic carriers may continue to shed for years. As a result, the size of the chronic carrier population is expected to decline much more slowly than that of acute circulating infections following vaccine introduction. Models of typhoid fever transmission should account for uncertainty in the importance of chronic carriers to endemic transmission. We accomplish this by varying the relative transmissibility of lifelong chronic carriers while holding their prevalence fixed at a low level predicted by the model (0.6% and 0.3% in the primary and alternative models, respectively, in the absence of vaccination; see Supplementary Figure S7). These modelled estimates of the prevalence of lifelong chronic carriers are consistent with the finding of one stool-culture-positive S. Typhi carrier among 192 healthy serosurvey participants with high IgG titres against the Vi antigen (0.52%), a potential marker of chronic carriage [Reference Qadri39–Reference Charles41]. Previous modelling analyses of TCV impact in Gavi-eligible countries sampled rC from an uncertainty distribution with an expected value of 25%, which is consistent with the TyVAC-Bangladesh trial results [Reference Mylona24, Reference Nolan42].
Our analysis also demonstrates that models of TCV impact are sensitive to the mechanism of direct protection. In transmission dynamic models, vaccine effects often derive entirely from immunity to infection, potentially resulting in overly optimistic levels of herd protection, while static cohort models generally do not account for indirect effects at all. Model comparison studies which include both approaches can be a useful way to evaluate uncertainty in the potential impact of vaccine policies [Reference Burrows9]. Transmission dynamic models should be validated against (or calibrated to) observed data on vaccine impact prior to making predictions.
In our model, we assumed vaccine uptake was uniform across the eligible population and therefore independent of the baseline risk of typhoid fever. However, this may not have been the case in the TyVAC-Bangladesh trial, where typhoid incidence in the control clusters was higher among control vaccine recipients than vaccine-eligible non-recipients (616 vs. 406 cases per 100000 person-years in those aged 2 to <16 years) [Reference Qadri12]. Residents who participate in a vaccine trial may also be more likely to seek medical care when ill. Regardless, vaccine efficacy estimates from the trial should be robust to these differences and comparable to our model-based estimates for several reasons. First, the TyVAC-Bangladesh trial design inherently controls for any association between vaccine acceptance and the baseline typhoid risk when estimating total and indirect effects, because the analysis is restricted to those who either did or did not accept the vaccine. Second, in the statistical analysis of the TyVAC-Bangladesh trial data, estimates of vaccine efficacy were adjusted for several factors that could influence care-seeking and the baseline risk of typhoid fever, including age, sex, reported handwashing behaviour, the source and treatment of household drinking water, and household toilet type.
The compartmental modelling framework used in this study has been extensively validated and used to reproduce patterns of typhoid fever transmission across a variety of settings [Reference Pitzer22, Reference Bilcke25–Reference Pitzer27, Reference Saad43]. While our models do account for population age structure, mixing is otherwise assumed to be homogeneous, with susceptible and infected individuals coming into contact at random. In reality, transmission-relevant contacts may be shaped by household structure and patterns of food consumption outside of the home, although indirect transmission through contaminated water may have a more generalized exposure pattern. In theory, these more complex mixing patterns could be captured with an agent-based or network model. However, the data that would be needed to accurately parameterize such a model is not currently available and would be challenging to acquire given our study site’s size, density, and location within a larger city. Our results are specific to the Dhaka study site and should not be over-interpreted or extrapolated to other settings.
We selected a 10% contamination effect in our primary model as a plausible simplifying assumption. We made the same assumption in model simulations used for sample size calculations when designing the TyVAC-Bangladesh trial [Reference Qadri12]. Higher levels of contamination would be expected to further dilute the model-predicted indirect protection. Our analysis shows that even under this relatively low level of contamination, scenarios with a vaccine efficacy against infection as high as 90% are consistent with the trial results.
One limitation of the HlyE seroincidence data we used to fit the alternative model is that it is not serovar-specific, as both S. Typhi and Paratyphi A infections can induce a strong HlyE seroresponse. To estimate the incidence of S. Typhi infection, we multiplied HlyE seroincidence by the ratio of typhoid fever to all enteric fever cases at the Bangladesh site in the STRATAA study. The validity of this calculation is premised on S. Typhi and S. Paratyphi A infections being equally likely to progress to clinical disease. Given the similarity in their physical structure and clinical presentation, we believe that this is a reasonable assumption. However, if S. Paratyphi A infections are less likely to progress to diagnosed clinical disease, our approach would overestimate the incidence of S. Typhi infection and therefore the force of infection in the alternative model.
Our findings suggest that TCV protects against S. Typhi infection and transmission, not just symptomatic disease, and that the low indirect protection observed in the TyVAC-Bangladesh study can be explained by transmission from unvaccinated adults. Mathematical models of typhoid fever vaccination should explicitly account for indirect protection and the observed uncertainty in its magnitude. In 2025, Bangladesh introduced routine TCV vaccination at 9 months of age and vaccinated over 40 million children up to 15 years of age in a nationwide catch-up campaign [8]. Our analysis suggests that this strategy could at least temporarily induce moderate indirect protection, depending on coverage and the duration of vaccine protection. Routine surveillance programmes are needed to evaluate the population-level impact of TCVs and inform policy decisions in Bangladesh and other endemic settings.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/S0950268826102337.
Data availability statement
Code and data to reproduce this analysis is available at https://github.com/pitzerlab/Typhoid-IndirectProtection.
Acknowledgements
We acknowledge the contributions of the participants who took part in the TyVAC-Bangladesh and STRATAA studies and the large field, clinical, and laboratory teams who supported these studies.
Author contributions
Conceptualization: J.W., M.V., A.P., F.Q., V.P.; Formal analysis: J.W.; Investigation: J.W., F.K., X.L., M.V., M.A.A., M.G.F., N.H.R., P.K.B., A.P., K.Z., F.Q., J.D.C., V.P.; Visualization: J.W.; Writing - original draft: J.W.; Writing - review & editing: J.W., F.K., X.L., M.V., M.A.A., A.P., F.Q., J.D.C., V.P.; Data curation: F.K., M.A.A., M.G.F., P.K.B.; Project administration: F.K., F.Q.; Supervision: F.K., F.Q., V.P.; Methodology: X.L., M.A.A., V.P.; Funding acquisition: V.P.
Funding statement
This analysis was supported by the Bill and Melinda Gates Foundation (INV-030857). The funder had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. The conclusions and opinions expressed in this work are those of the authors alone and shall not be attributed to the Foundation.
Open access
