Modeling the initial impact and predicted future benefits of TCV from two Pakistani provinces

Summary

Background

While trials have demonstrated high efficacy of typhoid conjugate vaccine (TCV), data on effectiveness are limited. We report initial impacts and predict future benefits of TCV from two provinces in Pakistan.

Methods

We used blood culture-confirmed typhoid cases from the Surveillance for Enteric Fever in Asia Project (SEAP) and Impact assessment of Typhoid conjugate vaccine following a catch-up campaign and introduction in Routine Immunization Program of Pakistan (ITRIPP) to estimate the population-level impact of vaccination (2018–2023). We used regression models to estimate initial impacts and an agent-based model to predict future benefits.

Findings

In Sindh, typhoid incidence was higher and cases occurred in younger children compared with Punjab. TCV reduced incidence by 48.9% in Sindh (95% CI: 47.3–50.3%) and 66.2% in Punjab (95% CI: 64.7%, 67.6%) over the first 2 years after vaccine rollout but declined each year. In Sindh, waning was quicker and models predicted that population incidence would stabilize near pre-vaccine levels in 2024. An additional campaign could provide short-term, but not long-term, benefits. In contrast, in Punjab, incidence is projected to remain low for several years, and the catch-up campaign with routine immunization at 9 months of age may be sufficient. However, follow up data from Punjab are needed to better characterize waning immunity.

Interpretation

TCV has reduced incidence in Pakistan, but protection varies by site. Routine immunization at 9 months of age along with a catch-up campaign may be sufficient to control incidence in settings with moderate transmission. However, in settings with particularly high incidence and/or short duration of protection, alternative strategies to reduce the force of infection may be needed.

Funding

Bill & Melinda Gates Foundation.

Research in context.

Evidence before this study

As typhoid conjugate vaccines were only introduced into routine immunization programs in a few countries relatively recently, there is a need to move beyond initial efficacy studies to quantifying real-world effectiveness, particularly. Long-term data on both efficacy and effectiveness are also lacking. We searched PubMed from January 1, 2018 to August 22, 2024 for the search terms (“typhoid conjugate vaccine”) AND (“efficac∗” OR “effectiv∗”), returning a total of 66 results. We focus on summarizing the results here from primary research articles that reported efficacy or effectiveness estimates for Tybar-TCV vaccination. A total of 12 articles provided efficacy or effectiveness estimates, 7 of which were efficacy estimates from randomized trials. Only one published study from Malawi reported efficacy greater than 2 years after vaccination, with no decrease in protection reported. All other studies, including all effectiveness studies, reported outcomes over two years or less after vaccination. Among individually randomized studies, efficacy was approximately 80% across all sites but effectiveness was more variable, ranging from 56 to 98%. Across these non-randomized studies, effectiveness was higher among older children and shortly after vaccination. One cluster-randomized study suggested stronger impacts among households with access to improved water, sanitation, and hygiene.

Added value of this study

This study provides longer term follow up data at the population level up to four years post-vaccine introduction on population-level vaccine effectiveness for Pakistan. We find substantial differences in vaccine impact, clinical waning, and optimal dosing strategy between two provinces in Pakistan, with the higher burden province of Sindh having lower impact, shorter duration of protection, and showing a need for more frequent vaccine doses and/or complementary interventions to reduce transmission. This study provides insight into heterogeneity in vaccine impacts across sites and potential drivers of those differences. As World Health Organization (WHO) prepares to revise their typhoid vaccine recommendations, understanding the magnitude and drivers of heterogeneity in vaccine impacts is important for maximizing public health impact.

Implications of all the available evidence

TCV vaccination has the potential for substantial impact, but effectiveness may vary by site. In high burden settings, complementary interventions to reduce the force of infection may be needed to eliminate typhoid as a public health concern. Given this heterogeneity in impact, the optimal typhoid vaccination policy may be context-specific. Specifically, additional typhoid vaccine doses may be useful to control large outbreaks, and expanding the routine immunization program to include school age doses may also provide more sustainable benefits in certain sites.

Introduction

Typhoid is an enteric infection with high disease burden in South Asia. In 2017, there were an estimated 14.3 million infections and 136,000 deaths, with the vast majority of these deaths occurring among young children in South Asia.¹ While incidence has been declining in many countries, Pakistan continues to suffer from large outbreaks due to a variety of factors, including poor access to adequate water, sanitation, and hygiene infrastructure and increasing prevalence of antibiotic resistance, which makes treating cases more difficult.² Most recently, extensively drug resistant (XDR) typhi has emerged in Pakistan, which indicates resistance to co-trimoxaxzole, chloramphenicol, ampicillin, fluoroquinolones, and third generation cephalosporins, limiting effective oral antibiotics to azithromycin.³ Indeed, Pakistan is now considered to be the main global source of antibiotic resistant cases,⁴ with 70% of cases being extensively drug resistant in 2020.⁵

One tool to reduce typhoid is the typhoid conjugate vaccine. In clinical trials, TCV reduced blood confirmed typhoid incidence by 79% in Nepal⁶ and 81% in Malawi.⁷ These initial trial data were published on the basis of 2 years of follow-up data and formed the evidence base that led to TCV approval and endorsement by the WHO.⁸ In Pakistan, a surveillance study found strong efficacy of up to 97% in an outbreak setting,^9,10 suggesting similar efficacy was also achievable in Pakistan. More recently, data on individual-level protection after 4 years has been published from Malawi, with a recent paper showing no waning by 4 years after vaccination.^7,11 However, longer term population level field data on vaccine impacts after large scale rollout are needed to understand differences between vaccine efficacy and real world effectiveness across different epidemiologic settings.

For many vaccines, obtained protection and waning of that protection is known to vary by site.¹² One reason for this heterogeneity includes differences in initial vaccine response,^13,14 which may interact with the age distribution of cases. Some studies have shown that immunogenicity for typhoid vaccination is age-specific, with older ages showing a potentially stronger response.¹⁵ While an imperfect proxy due to difficulties with typhoid case ascertainment among children, a lower age of first infection generally indicates a higher transmission intensity.¹⁶ Thus, in settings with more cases among children, the overall population averaged protection might be lower due to interactions between vaccine response and age-specific incidence. High transmission intensity can also overwhelm expected protection due to high frequency and/or level of exposure. More specifically, the ultimate difference in cumulative risk over follow up narrows over time even if immunological waning does not change in high transmission settings. Given the wide variation in incidence across countries targeted for TCV vaccination,¹⁷ studies are needed to assess duration of protection at the population level across sites.

Large scale vaccine roll-out for typhoid fever occurred in Sindh and Punjab. In this study, we show population level impacts of typhoid vaccination for Sindh and Punjab provinces in Pakistan covering 4 years of vaccine rollout in Sindh and 3 years in Punjab. We provide estimates by age group and over time to show the dynamics of protection and differences by site. We then use a transmission model to explore how incidence is expected to change in the future and how vaccine strategy could shape these patterns.

Methods

Data

Data are from the laboratory network in Sindh and Punjab provinces from the Surveillance for Enteric Fever in Asia (SEAP) study¹⁷ and ITRIPP (unpublished) include the number of blood culture confirmed typhoid cases by month and age group (<2, 2–4, 5–9, 10–14, and ≥15 years). Additional details are included in the Supplementary Material. Based on a comparison of incidence and test positivity over the surveillance period (see Supplementary Material Text and Supplementary Figure S1), we focus on data from 2019 to 2023 for calculating impacts in the main text for both provinces. We also considered running Sindh with 2018–2023 data to allow for two years of pre-vaccine data as a sensitivity analysis (Supplementary Figure S2), but focus on 2019–2023 in the main text to match Punjab.

Statistical analysis

Pre-post Poisson regression model

We used Poisson regression models to estimate the impact of vaccination comparing the pre-vaccine era, the reference period used in the model, with the post-vaccine introduction era. For both provinces, we excluded data collected from February to June 2020 during a national COVID lockdown. For Sindh, we used data from January 2019 to November 2019 as the pre-vaccine period and January 2020 and July 2020–December 2023 as the post-vaccine introduction period. In addition to the lockdown months, we excluded the month of December 2019 because the campaign was being conducted during that month, and seroconversion after vaccination takes about 28 days.¹⁸ For Punjab, we used January 2019–January 2021 (excluding lockdown months) as the reference and March 2021–December 2023 as the vaccine period, excluding February 2021, the month of vaccine rollout and the vaccine campaign in Punjab.

Population size was included as an offset for all models, based on estimates from the Pakistan Bureau of Statistics from the 2017 census for the urban areas of each province.¹³ We assumed population growth rates were applied evenly to all age groups. We ran these models for the entire follow up period and all age groups combined, but also ran separate models for each year and age group to better characterize differences in impact by age group and over time. All Poisson regression models were conducted in R (version 4.3.1).

Simulation model

We used a previously published individual-based typhoid transmission model built in EMOD 2.11.¹⁹ Briefly, susceptible individuals can become infected through transmission through environmental reservoirs. We also considered fitting a direct contact route, but the two modes of transmission in the model were not identifiable and caused difficulties comparing transmission across the two sites. Once infected, the infection is initially latent after which it progresses to acute or subclinical infection. Both acute and subclinical infections can become chronic carriers and may continue to transmit at a reduced rate. We also track the number of infections each agent has experienced with repeated infections being less common (Supplementary Tables S1 and S2).

The model uses a baseline number of agents of 10,000, with growth rates set to match the regression model. Unlike the regression model, population growth is assumed to be resultant from births outweighing deaths. We simulated this model from 1990 to 2040, with 1990–2015 being treated as a burn in period.

Vaccination was implemented in December 2019 for Sindh and February 2021 for Punjab and is modeled as reducing the risk of infection by an amount proportional to the calibrated vaccine efficacy (see Supplementary Material for more details). For each province, vaccine introduction is implemented as a one-time campaign among children 9 months to 15 years of age followed by continued routine immunization for 9 month olds. Incidence during lockdown months was excluded from both models for calibration.

Model calibration and validation

The model was calibrated to the age distribution pre-vaccine as well as overall incidence for each province during the reference and vaccine period (2019–2023 for Sindh and 2019–2023 for Punjab) using a gradient ascent algorithm. We assumed a constant force of infection during the pre-vaccine period. Model parameters are shown in Supplementary Tables S1 and S2 and more details of the calibration are shown in the Supplementary Material.

To validate model predictions, we compared incidence from vaccine simulations with a counterfactual, no vaccine simulation designed to predict what would have happened if the vaccine had never been implemented. We ran 100 stochastic simulations for each parameter set/scenario to capture uncertainty inherent in the transmission process and population dynamics. To facilitate comparison of transmission intensity between the two provinces, we also computed the average age of first infections in 2015–2019, when the model had reached equilibrium, across the 100 stochastic no-vaccination simulations. The age distribution is expected to be at its peak prior to vaccination, and inversely related to the force of infection.^16,20

Forward simulations

To explore how transmission dynamics might depend on future vaccine strategies we implemented a booster campaign in June 2024, aligning with action considered by GAVI to control the ongoing disease outbreak. We considered: a control scenario with no additional vaccination, a default campaign using the same strategy as the initial campaign, an expanded age campaign, where adults over 15 were vaccinated, and an additional school age dose. We also explored whether combining these interventions led to additional impacts.

Ethics approval

Ethical approval was granted by the AKU Ethics Committee on December 15, 2019 (#2019-2136-6929), and by the Pakistan National Bioethics Committee on February 23, 2021 (4-87/NBC-439/19/1816) and December 13, 2024 (#4-87/NBC-439/21/821). Written informed consent was obtained from all hospital enrolled cases and telephonic informed consent from the lab enrolled cases.

Role of the funding source

The funders had no role in the analysis, or interpretation of the data; the preparation, review, or approval of the report; or the decision to submit the manuscript for publication.

Results

In the population overall, reported cases in Sindh were roughly 4× as high as in Punjab pre-vaccine, with a younger age distribution (Fig. 1, Fig. 2, Fig. 3). In Sindh, incidence was highest in children under 2, but was also high among 2–4 and 5–9 year olds. Incidence among ≥15 year olds was low. In contrast, in Punjab, incidence was lowest in children under 2 years of age and peaked among 5–9 year olds, with over 30% of cases 15 years of age and older.

Fig. 1.

Time series of blood confirmed typhoid cases over time by age for A) Sindh and B) Punjab. The months of February 2020–June 2020 are excluded from the time series due to a national lockdown that occurred in response to surging cases of COVID-19. Timing of vaccine campaigns in each province are shown with the dotted lines. See Supplementary Figure S1 for pooled time series collapsed across all ages.

Fig. 2.

Poisson regression derived vaccine effectiveness for each year post-vaccine introduction period, compared to the pre-vaccine reference period for A) Sindh, all ages, B) Sindh, stratified by age, C) Punjab, all ages, D) Punjab, stratified by age. Note that y-axes are different for each panel. A version with the same y-axis for panels A and C and panels B and D is shown in Supplementary Figure S3.

Fig. 3.

Model calibration figure. A) Simulated and measured age distribution of cases in Sindh pre vaccination and B) Simulated and measured age distribution of cases in Punjab pre vaccination. Pooled time series of blood confirmed reported typhoid cases from data (solid line) vs. model (dashed line and grey ribbon for C) Sindh and D) Punjab. Sindh panels show results with the 2019 reference period.

For both provinces, incidence declined after vaccination. However, in Sindh, incidence began to rebound 1.5 years later, largely driven by increased incidence in 2–4 year olds. In Punjab, incidence did not begin to rebound until 2.5 years after vaccine introduction, and the resultant peak was not as pronounced as in Sindh (Fig. 1). While the rebound in Punjab was highest among the 2–4 year olds, the overall reduction in vaccine effectiveness was greatest among the oldest age group (≥15 years) (Fig. 2).

The regression models revealed that the pattern in protection was heterogeneous by province, age, and time. For both provinces, efficacy was highest in the first year after vaccine introduction, with an initial estimate of 56.1% (95% CI: 54.1–57.9%) in Sindh and 73.3% (95% CI: 71.5, 74.9%) in Punjab (Fig. 2). During the second year after introduction (January–December 2022 in Sindh and March–December 2023 in Punjab), estimates of vaccine impact had fallen in both provinces, with a VE of 44.7% (95% CI: 42.9–46.5%) in Sindh and 59.3% (95% CI: 57.1–61.4%) in Punjab. Thus, average protection over the first 2 years after rollout was 48.9% in Sindh (95% CI: 47.3–50.3%) and 66.2% in Punjab (95% CI: 64.7–67.6%). In Sindh, protection remained high for the youngest age group for all four years, with stable protection of about 70%. However, older age groups appeared to have a decline in protection. For Punjab, protection was most stable for all but the ≥15 age group, who experienced a relatively large decline in efficacy in 2023. While protection was moderate for children <2 years in Punjab, it was lower than in Sindh, with an average protection of 40–60% and a stronger decay over time. Both provinces experienced a large spike in cases in 2023 and had lower vaccine efficacy in that year. If 2018 was included in the reference period for Sindh, the estimates of vaccine efficacy were lower, but the trend over time was similar (Supplementary Figure S2).

The transmission model for both provinces fit the age distribution pre-vaccine well (Fig. 3A and B). In Punjab, the model tended to overestimate the number of cases among children 10–14 years of age, but all other age groups were consistent between the model and the data. The model also performed well in capturing the decline and then rebound in incidence after vaccine rollout (Fig. 3C and D). Fitted parameters suggested longer duration of protection in Punjab compared with Sindh. In Sindh, the estimated duration of immunity among children was much lower, with half of all vaccinated children being fully susceptible before age 5. In contrast, in Punjab, half of young children were expected to be fully susceptible by age 10. Fitted infectiousness was also over four times as high in Sindh compared with Punjab, consistent with higher baseline transmission intensity (Supplementary Table S2).

In Punjab, incidence began to fall late in 2020 after the COVID-19 lockdown but before the vaccine campaign began, which was not captured in our model. The vaccine efficacy obtained from the model closely matched what was seen in the surveillance data when averaged over the first 2 years after vaccine introduction. For Sindh, initial estimates of VE were not as closely matched to the data, but the trajectory was better whereas for Punjab the initial VE estimates matched closely but the rate of rebound was slightly offset. Specifically, for Punjab modeled VE was 68.9% (63.6–73.7%) compared with 66.2% (95% CI: 64.7–67.6%) in the data for 2021–2022. For Sindh, modeled VE was 41.2% (95% CI: 34.3–46.7%) over the first two years compared with 48.9% (95% CI: 47.3–50.3%) in the data for 2020–2021. The differences in rebound of population incidence were strongly patterned by age, with modeled protection being more stable in <2 year olds in Sindh and showing a much faster rebound in older age groups, similar to the primary data. Fits to age-specific incidence are shown in Supplementary Figures S4 and S5.

In the absence of any campaigns or shift in vaccine strategy, incidence was predicted to rebound in both provinces, with incidence being about 25% below pre-vaccine levels in Punjab by 2030 or near pre-vaccine levels in Sindh (Fig. 4). For Punjab, the decline in vaccine efficacy was expected to be slower and smaller. In both provinces, a 2024 campaign conferred a substantial short-term benefit. However, in Sindh, this benefit rapidly declined and was followed by a rebound in 2025–2027. In contrast, in Punjab, the rebound in incidence after the campaign was smaller, but no long-term benefit was expected. Augmenting planned campaigns with an expanded age range and/or adding a routine school-age dose did not influence this pattern in Sindh. In Punjab, a school age dose had a greater benefit, appearing to slow the rate of rebound in incidence. Thus, repeated campaigns with similar reach would likely be needed to sustain population level benefits in Sindh. While augmenting the base schedule in Punjab could be helpful, the timeline for such interventions could be slower and could be developed as more data on long-term protection in the province become available.

Fig. 4.

Impact of future vaccine strategies on incidence of blood-confirmed typhoid in A) Sindh and B) Punjab.

As fitted initial campaign coverage was lower in Sindh than Punjab, we also conducted a sensitivity analysis where we simulated the potential impact of vaccination if the follow-up campaign in Sindh were able to achieve 100% coverage, similar to the fitted initial campaign coverage in Punjab. In those simulations, the initial benefit of the campaign was greater (bringing incidence to as low as 2.1 per 10,000 person-years compared with 2.9 per 10,000 for the simulations at the original coverage level), but there was no long-term benefit (Supplementary Figure S6).

Discussion

Overall, TCV has substantially reduced typhoid incidence in Pakistan. We found strong impacts shortly after vaccination, particularly for young children. However, protection varied by province and over time. The status quo vaccine policy of routine immunization at 9 months of age may be sufficient to control incidence in settings with moderate transmission intensity and/or slow clinical waning like Punjab. However, in settings with particularly high incidence and/or faster clinical waning, such as Sindh, alternative strategies to reduce the force of infection and tailor vaccine campaigns may be needed. Moreover, complementary strategies will likely be needed to eliminate typhoid.

The differences in overall clinical effectiveness and rebound rates over time were primarily driven by the lower fitted duration of immunity and secondarily by higher transmission intensity in Sindh compared with Punjab. Unlike the other age groups, protection for the youngest group was slightly lower in Punjab compared with Sindh. Effectiveness estimates may have been limited by the extremely low baseline incidence in children <2 years, which reached zero in some of the pre-vaccine years. In Sindh, the estimated duration of immunity among children was much lower. For older age groups, the discrepancy was even greater. All differences in immunity are based on observed clinical protection may be due to multiple factors, including the interplay between antibody decline and force of infection, potential differences in circulating strain diversity, antibiotic resistance, or other unidentified factors. Given that the dynamics of antibody decline have been shown to have similar patterns across countries, even where there are large differences in incidence²¹ and appear to be similar in Pakistan,²² it is unlikely that faster antibody titer waning in Pakistan can explain our findings. However, higher antibody titer may be needed to confer protection in high force of infection sites compared with lower transmission intensity settings.

Our fitted estimates of vaccine coverage were high, with coverage of about 95% for routine immunization estimated in both provinces. These high fitted point estimates were likely driven by consistently low incidence in children <2 years of age throughout the study follow up, which indicate high vaccine efficacy, high vaccine coverage, or both. We additionally estimated high campaign coverage in Punjab, which was driven by consistently lower incidence in older age groups.

Based on our fitted parameters and the age distribution of infections, the expected force of infection was higher in Sindh compared with Punjab, which may partially contribute to these trends (R₀ values of 8.0 and 5.6). This transmission intensity for Sindh is higher than has previously been estimated for any location. Bangladesh has been estimated to have an R₀ of 7, with India having a much lower estimate of 2.8.^23,24 This high transmission intensity may represent a break point above which additional interventions are needed.

While the faster decline in initial benefits in Sindh may also reflect the lower fitted coverage of the initial campaign (Supplementary Table S2), these differences cannot explain the gap in long term protection (Supplementary Figure S5). For all forward simulations, we assumed that coverage would likely be similar to the initial campaign to be conservative. However, our sensitivity analysis suggests that a campaign in Sindh with broader reach would likely have a stronger initial impact, but would not impact long term trends. Rather than uniformly increasing coverage, future campaigns targeting high risk subgroups who are commonly infected and more likely to spread disease, such as food handlers, might be beneficial.²⁵

While our simulations suggest the potential for continued waning in Punjab and thus the possible need for follow up campaigns and/or additional doses in the future, we note that the parameters related to waning immunity in Punjab were particularly uncertain, given only ∼2.5 years of post-vaccine introduction data in Punjab compared with 4 years in Sindh. Thus, additional data in Punjab would be useful to determine if population level protection will continue to decline or will stabilize. These results have implications for future vaccine policy recommendations, which are being considered by WHO to better control typhoid in endemic settings based on data from large-scale rollout that is now becoming available.²⁶

While beyond the scope of this paper, we note that alternative strategies beyond vaccination may also be able to reduce transmission intensity and will likely be necessary to eliminate typhoid as a public health problem in Pakistan. For example, increasing access to safe water and improved sanitation may also be helpful, particularly for reducing long-cycle transmission through environmental sources.^27,28 Treating chronic carriers might also be useful.²⁷ Particularly in settings such as Pakistan with high rates of AMR and ceftriaxone resistance, effective treatment to stop continued shedding of the bacteria and onward transmission may also play a key role in elimination. Our results suggest that elimination is unlikely to occur with vaccine interventions alone, similar to the findings of other modeling studies.²⁹

Our study has a number of limitations. First, we note that surveillance sites were initially limited, particularly in Punjab, and we were thus limited to focusing on 6 years of data in Sindh and 5 years in Punjab, which only included 2 years of pre-vaccine data for each province. The limited length of pre-vaccine data made our comparisons of the pre and post-vaccine introduction periods more sensitive to stochastic effects of high transmission years shortly before vaccination. In 2019, incidence was exceptionally high throughout Pakistan. Although the specific drivers are unknown, the increase in incidence may have been due to extreme flooding through the region, leading to more exposure to contaminated water systems or population displacement. Additionally, AMR increased substantially between years 2017 and 2018 for many antibiotics including ceftriaxone and fluroquinolones.³⁰ AMR may have impacted disease trends due increased transmission rates of XDR strains, or increased healthcare seeking for effective treatment.

If the increase in incidence in 2019 was due to increased transmission and not the surveillance system, it is possible that incidence would have naturally declined in 2020 even if vaccination had not been implemented, and our effect estimates are biased high. However, in our sensitivity analysis for Sindh that also included 2018, a lower burden year, in the pre-vaccine reference period, we still find significant protection in Sindh with a similar pattern of decline in vaccine efficacy over time. Moreover, in Punjab, which also included 2020, an especially low burden year, in the pre-vaccine period, we estimate even higher levels of protection than in Sindh. Thus, while transient changes in the force of infection could influence exact values of vaccine efficacy obtained in our study, it is unlikely that these changes can explain the large declines in incidence after vaccine introduction that we have shown here.

The elevated incidence in 2019 may also have been attributable to improving surveillance quality or case ascertainment due to AMR-related healthcare seeking. As shown in Supplementary Figure S1, blood culture positivity in Sindh appeared to be higher than would have been expected given incidence in 2017 and 2018, suggesting that surveillance was not initially strong enough to capture most cases. However, from 2019 onwards there is good concordance between these two time series. Further, a study comparing AMR rates between 2019 and 2021 show high, yet stable, rates of MDR,³¹ compared to changes in 2017–2018.³⁰ We do not use any data earlier than 2019 in our analyses to minimize this possible confounding. As other countries prepare to launch their own vaccine campaigns, careful quantification of the baseline burden is important to understand the ultimate impact of vaccination.

From 2020 onwards, our estimates may have also been impacted by external events. At the end of the time series (in 2022 and 2023) there was a large flood coincident with the surge in incidence in both provinces.³² While vaccination did not prevent this large resurgence, this may not be reflective of what we can expect from vaccine impacts going forward. Nevertheless, large storms are expected to become more frequent globally and in Pakistan as climate change progresses,³² so understanding the impacts of vaccination in the context of extreme weather events is important.

Impacts may also have been influenced by the COVID-19 outbreak. COVID-19 may have reduced reported incidence due to reduced social contact, reducing eating out,³³ and reduced health seeking behavior/reporting.³⁴ While we removed lockdown months from model calibration and the regression models, we cannot completely exclude the impact of COVID-19 on our impact estimates, as the impact on population level immunity may take years to stabilize, in addition to returning to normal healthcare-seeking. This bias differentially impacts estimates by province. If the reduction in incidence in Sindh post-vaccine in 2020 in non-lockdown months was due to COVID-19 and not the vaccine, we have overestimated estimates of vaccine impact. In Punjab, the same bias would lead us to underestimate initial impact estimates, as 2020 was included in the pre-vaccine period. Due to the agreement of substantial impact between sites, we can conclude that COVID-19 is not the sole driver of incidence changes during this period.

In this study, we provide population-level estimates of effectiveness of typhoid vaccination in Pakistan. Our results suggest that TCV can provide substantial benefits, but that long-term benefits may be context-specific and require additional efforts to sustain in high burden settings. As more data become available, the expected impacts of typhoid vaccination may become clearer.

Contributors

ANMK, MTY, JG, and FQ conceptualized the study and co-developed the methodology. FQ and MTY acquired funding. MTY and SQ curated data. ANMK and JG conducted formal analysis. ANMK validated the model, made all visualizations, and wrote the first draft of the manuscript. ANMK and MTY have directly accessed and verified the underlying data. All authors reviewed and edited the manuscript for important intellectual content and were responsible for the decision to submit the manuscript.

Data sharing statement

Data is available upon request and submission of proposal to FQ after publication of the primary study.

Declaration of interests

This work is based on research conducted by Institute for Disease Modeling, a research group within, and solely funded by, the Gates Foundation. JG and AK are employees of the Gates Foundation, however, this study does not necessarily represent the views of the Bill and Melinda Gates Foundation.