Global Typhoid Fever Incidence: A Systematic Review and Meta-analysis

Abstract Background Contemporary incidence estimates of typhoid fever are needed to guide policy decisions and control measures and to improve future epidemiological studies. Methods We systematically reviewed 3 databases (Ovid Medline, PubMed, and Scopus) without restriction on age, country, language, or time for studies reporting the incidence of blood culture–confirmed typhoid fever. Outbreak, travel-associated, and passive government surveillance reports were excluded. We performed a meta-analysis using a random-effects model to calculate estimates of pooled incidence, stratifying by studies that reported the incidence of typhoid fever and those that estimated incidence by using multipliers. Results Thirty-three studies were included in the analysis. There were 26 study sites from 16 countries reporting typhoid cases from population-based incidence studies, and 17 sites in 9 countries used multipliers to account for underascertainment in sentinel surveillance data. We identified Africa and Asia as regions with studies showing high typhoid incidence while noting considerable variation of typhoid incidence in time and place, including in consecutive years at the same location. Overall, more recent studies reported lower typhoid incidence compared to years prior to 2000. We identified variation in the criteria for collecting a blood culture, and among multiplier studies we identified a lack of a standardization for the types of multipliers being used to estimate incidence. Conclusions Typhoid fever incidence remains high at many sites. Additional and more accurate typhoid incidence studies are needed to support country decisions about typhoid conjugate vaccine adoption. Standardization of multiplier types applied in multiplier studies is recommended.

Salmonella enterica subspecies enterica serovar Typhi (Salmonella Typhi) is the cause of typhoid fever. Typhoid fever is a systemic infection that is an important source of illness and death in low-resource areas [1,2]. Persons living in areas without access to improved sanitation facilities who are exposed to fecally contaminated water and food are at greatest risk for infection [3][4][5]. Timely access to effective antimicrobial therapy is central to preventing complications such as intestinal perforation and death. Consequently, the alarming spread of antimicrobial resistance in Salmonella Typhi is likely to contribute further to poor clinical outcomes [6]. Although there has been considerable progress with expanding coverage of access to improved water and sanitation facilities in most regions except Oceania [7] and typhoid incidence has declined in some countries, typhoid remains a major problem worldwide [8]. The emergence and spread of antimicrobial resistance in Salmonella Typhi and the long-term efforts needed to deliver water and sanitation improvements has focused increasing attention on the use of typhoid vaccines.
Two vaccines, Vi polysaccharide (Vi-PS) and live attenuated oral vaccine (Ty21a), have been widely licensed for typhoid prevention [9]. However, their use is restricted to those aged ≥2 years for Vi-PS and ≥6 years for Ty21a [10]. This, combined with waning immunity after 2 years of vaccination [11][12][13], means that Vi-PS and Ty21a are not widely used in typhoid-endemic areas. Typhoid conjugate vaccines (TCVs) overcome a number of limitations of Vi-PS and Ty21a, having been shown to be safe, immunogenic, and effective in infants and young children [14][15][16]. In October 2017, the World Health Organization (WHO) Strategic Advisory Group of Experts on Immunization recommended TCV for routine use in children >6 months of age in typhoid-endemic countries. In December 2017, the first TCV was prequalified by WHO [17], enabling typhoid-endemic, low-income countries priority access and funding for the vaccine [18].
Contemporary estimates of the incidence of typhoid fever are helpful in supporting countries' decision making about typhoid prevention, including vaccine use. While there have been improvements in the sophistication of efforts to model typhoid fever incidence [2,19,20], primary data have not always been at the forefront. Recognizing a recent increase in the number of typhoid fever incidence studies [8], we performed a systematic review of the literature and a meta-analysis. Our goal was to describe, summarize, and analyze high-quality primary incidence data on typhoid fever. We focused primarily on prospective studies reporting incidence either through population-based surveillance or studies estimating incidence through sentinel site surveillance with a healthcare utilization survey to establish multipliers to account for underascertainment.

Search Strategy
We performed a systematic review of articles published in 3 databases: Ovid, Scopus, and PubMed. We searched Ovid Medline from 1946 to 19 January 2018 with Daily Update, Embase Classic + Embase, and EBM Reviews-Cochrane Central Register of Controlled Trials. Scopus and PubMed databases were searched from inception to 22 January 2018 (search strategies in Supplementary Appendix A). Additionally, reference lists of any article included after the full-text review were searched.
Duplicates from the 3 searches were removed and collated using an online systematic review tool [21]. The resulting list of titles and abstracts were independently screened by 2 authors (J. A. C. and C. Y. H.) for relevance. Any article selected by at least 1 reviewer was moved to full-text review. All subsequent processes of the systematic review were performed in parallel by 2 authors (C. S. M. and J. A. C.), with discrepancies resolved by discussion. The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) checklist was used to document the search process [22]. Full texts of the articles were screened for inclusion. After the final list of included articles was established, data for study characteristics and incidence were abstracted.

Inclusion and Exclusion Criteria
Epidemiologic studies of any design reporting the incidence of typhoid fever were included. No restrictions were placed on age of population, country, language, or date. For intervention studies such as controlled field trials, we used only the control and unvaccinated arms. Only cases confirmed through blood or bone marrow culture were included. We excluded studies where we were unable to separate blood or bone marrow culture-confirmed typhoid from the results of serology (eg, Widal test) or other culture sources (eg, stool, urine).
We excluded studies that were outbreak-associated, case reports, or of travel-associated typhoid fever. In addition, editorials, commentaries, and conference presentations or abstracts were excluded. While government surveillance reports (eg, US FoodNet) have valuable data and likely provide very accurate estimates, countries with robust national surveillance systems experience low typhoid incidence, and typhoid fever in these countries is often predominantly travel-associated. The combination of low incidence with the problem of distinguishing travel-associated from non-travel-associated disease underpinned the decision to exclude such reports from this systematic review. Finally, we excluded studies reporting typhoid incidence in special subsets of populations such as only human immunodeficiency virus-infected persons, and healthcare-associated or hospital-acquired infections.

Data Abstraction and Analysis
Study characteristics that were abstracted included PubMed identifier if available, first author, publication year, study design, city or region and country of study, sample size (total population of surveillance or size of control/unvaccinated group), number of cases of fever, number of blood cultures collected, inclusion age and range, and criteria for blood culture. We subsequently classified study sites by their United Nations geographic areas and regions [23].
The currently available tools for assessing the quality and risk of bias did not translate to disease prevalence or incidence reviews [24,25]. Therefore, we developed criteria specifically for this review, using established quality assessment tools as guidelines (Supplementary Appendix B). Our goal was to evaluate the overall quality of the study in providing an accurate estimate of typhoid incidence in the study setting. We applied greater weight to questions about study design, patient selection, and criteria for collecting a blood culture. Our form evaluated selection and performance and the reference standard, assigning "low, " "moderate, " "high, " "uncertain, " "inapplicable, " "yes, " or "no" to each question. We then assigned an overall score of low, moderate, or high quality to the study.
Data for incidence were abstracted in Excel 2016 (Microsoft, Redmond, Washington). We stratified studies into 2 main types. Studies that estimated incidence through sentinel site surveillance with multipliers to account for underascertainment were categorized as multiplier studies [26]. All other studies were categorized as population-based, which included active household or population-based surveillance, prospective observational studies, and randomized vaccine field trials with control arms. We noted the year the observation period began and how many months the surveillance followed participants.
We recorded the number of typhoid cases that occurred during the study period and then divided by the number of months the study conducted surveillance and multiplied by 12 to report incidence as a rate of cases per 100 000 per year. For multiplier studies, we recorded the estimated incidence separately and also noted the multipliers that were used.
Pooled incidence estimates were calculated with MetaXL version 5.3 software (Epigear International) using a random-effects model [27]. As a secondary analysis of published data this study was exempt from institutional review board approval.
Among included studies, data collection was conducted from 1954 through 2014, with 17 (51.5%) of the studies collecting  Through our quality assessment, we determined that 5 of the included studies were of high quality [44,45,55,57,64]. These studies did not place any restrictions on inclusion age, gave clear and explicit temperature and duration for collecting blood culture, and actively searched the community for cases. We rated 12 studies as moderate quality [46, 48, 49, 56, 59-61, 65-68, 71], and the remaining 16 as lower quality. Multiplier studies were discounted compared with population-based surveillance studies on the basis of lack of validation of the "multiplier method, " and assumed shortcomings of accuracy.

Typhoid Incidence Among Population-based Studies
Population-based typhoid incidence studies were available in 26 sites from 16 countries (Supplementary Table 1). There were a total of 50 separate estimates of typhoid incidence; 7 studies provided the number of cases for each separate year of surveillance during the study period [12,13,50,51,53,69,72]. The majority of estimates of typhoid incidence at all sites were older; data for 34 (68.0%) were collected prior to the year 2000. The overall pooled estimate of incidence was 154.0 (95% confidence interval [CI], 115.1-198.6) typhoid cases per 100 000 per year.

Typhoid Incidence Among Multiplier Studies
Studies that used a multiplier to estimate typhoid incidence were reported in 17 sites in 9 countries (Table 2). Among these studies, 4 types of multipliers were implemented:  In contrast to incidence studies without multipliers, the majority (11 [64.7%]) reporting an estimated incidence with a multiplier were recent, collecting data from 2010 through 2014. The overall pooled incidence estimate from multiplier studies was 141.8 (95% CI, 85.3-212.2) typhoid cases per 100 000 per year. All but 1 of the multiplier study sites were located in Africa [57]. Removing the site located in Asia, the pooled estimate of multiplier studies located in Africa was 134.1 (95% CI, 77.9-204.8) per 100 000 per year (Figure 4). Typhoid incidence varied between sites and countries, with no cases reported in East Wad Medani, Sudan [56], and 882.0 typhoid cases per 100 000 per year in Kibera, Kenya [45]. One study was excluded from the multiplier pooled estimate calculations because it did not report the population under surveillance, only the incidence by age group [49]. Heterogeneity was significant in each pooled estimate (I 2 > 98%; P < .001, Cochran Q test).

DISCUSSION
We performed a systematic review and meta-analysis of blood culture-confirmed typhoid fever incidence. Although an encouraging number of typhoid incidence studies meeting criteria for inclusion have been published since 2000, a substantial proportion had data collection periods much earlier and as long ago as 1954. When examined by time and region, there appears to be a secular trend of lower incidence in more recent studies. This is consistent with other work suggesting declining typhoid fever incidence worldwide [8].
A number of studies included in this review collected typhoid incidence data from the same location over multiple years of surveillance. While some locations such as Alexandria, Egypt [69], recorded stable typhoid incidence between years of surveillance, significant changes in incidence were recorded at others such as Eastern Transvaal, South Africa [52,53], and Kolkata, India [66,67]. As none of the studies included in our systematic review were done during recognized typhoid fever epidemics, the multiyear studies illustrate the considerable variation in typhoid incidence that may occur between years at the same site considered to have endemic disease.
In addition to variation of endemic typhoid fever in time at the same location, our systematic review illustrates the considerable variation in endemic typhoid incidence that may occur between sites within the same country, between countries within a region, and between regions. This serves as a reminder that limited typhoid surveillance data can be misleading and that decisions for typhoid vaccine introduction should be based not only on surveillance data but also be informed by the occurrence of risk factors for typhoid transmission such as unsafe water and unimproved sanitation facilities.
Sites in the Asia area have long been recognized as having high typhoid incidence [73]. However, some considered typhoid fever to be uncommon in the Africa area as recently as 2008 [74]. The substantial number of new typhoid fever incidence estimates from sites in Africa since earlier reviews [2,8,19] confirms that some African sites record high typhoid fever incidence [56] and provides evidence that TCV use should be considered beyond Asia.
We demonstrate that eligible contemporary typhoid fever incidence studies were not available for some areas and regions. Notably, no eligible studies were identified from Oceania despite the major typhoid fever problem that is recognized in some Pacific islands [75]. Studies were also lacking from high-income countries, that rely on robust national surveillance systems to monitor typhoid fever. However, typhoid fever incidence is almost universally low in high-income countries, is often  travel-associated, and likely contributes little to total global cases [76]. While increasingly sophisticated approaches to modeling typhoid fever incidence provide a means of estimating typhoid fever incidence in locations where data are lacking [2,19,20], we sought to focus on systematic presentation of primary data and summary values across eligible studies.
Our quality assessment identified a number of concerns with published typhoid fever incidence studies. We rated 5 studies as high quality, thus classifying the majority as of moderate or low quality. Our data were also limited by the significant heterogeneity, but given the variation in location, inclusion ages, study design, and dates, this was to be expected. Our search strategy may not have identified negative studies from sites with low typhoid incidence.
A number of factors drove downgrading of study quality. Multiplier studies have been increasingly used to estimate typhoid fever incidence for cost and logistical reasons in recent    [48]. Such studies are less expensive and less logistically complex than population-based active surveillance studies and underpin considerable recent progress in available typhoid fever incidence data. However, the multiplier method has not been validated in the context of population-based surveillance and is associated with greater uncertainty introduced at each stage of correction for underascertainment. Furthermore, we identified considerable variation in the types of multipliers that were used. We recommend that consensus be reached on standardization of multiplier selection for future work to ensure that incidence estimates derived from multiplier studies are directly comparable. We also found that patient selection may have introduced bias in a large number of studies. The control arms of vaccine trials remain an important source of data on typhoid fever incidence. However, it must be recognized that vaccine trials are likely to be preferentially done in areas of high typhoid incidence to increase statistical power. Furthermore, a number of studies were done in specific age groups or communities, and resulting incidence estimates may not reflect disease in all age groups or the wider population. We also identified variation in who received blood or bone marrow cultures in studies, which may have introduced bias, and blood culture volume adequacy and the proportion of blood cultures contaminated were rarely reported.
While there are concerns for bias, we describe an overall pattern of declining incidence among included studies, with variation in typhoid incidence between surveillance years at the same site and also variation in place within countries and regions. These observations support the recognition of heterogeneity in endemic typhoid fever in both time and place. They also underscore the importance of considering both surveillance data and typhoid fever risk factors or "receptivity" when making vaccine introduction decisions.
We encourage ongoing efforts to generate more contemporary estimates of typhoid incidence, focusing on areas and regions where data are lacking, including Oceania and many parts of Africa. We also identify the need to develop a consensus standard on the types of adjustments that should be made in multiplier studies. This systematic review serves as a resource for site selection for future incidence studies, as a data source for typhoid fever modeling efforts, and in policy decisions for typhoid fever control.

Supplementary Data
Supplementary materials are available at Clinical Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.

Notes
Author contributions. J. A. C. conceived the study. J. A. C. and C. Y. H. developed the protocol and search strategy, and performed the title and abstract review. J. A. C. and C. S. M. reviewed full-text articles and abstracted the data. C. S. M. performed the analysis, created the tables and figures, and wrote the first draft of the manuscript. J. A. C. provided major revisions and comments to the manuscript. All authors reviewed the manuscript for final revisions before submission.