Global variation in force-of-infection trends for human Taenia solium taeniasis/cysticercosis

  1. Matthew A Dixon
  2. Peter Winskill
  3. Wendy E Harrison
  4. Charles Whittaker
  5. Veronika Schmidt
  6. Astrid Carolina Flórez Sánchez
  7. Zulma M Cucunuba
  8. Agnes U Edia-Asuke
  9. Martin Walker
  10. María-Gloria Basáñez

  • Curated by eLife

    eLife logo

    Evaluation Summary:

    Basic epidemiological parameters such as the force of infection (rate at which susceptible individuals acquire the infection) remain undetermined for human infection with the neglected food-borne zoonotic cestode Taenia solium, which may cause taeniasis and cysticercosis. Dixon and colleagues address this major gap by fitting simple mathematical models to datasets that describe the prevalence of taeniasis and cysticercosis in several countries. Importantly, they found that infection acquisition rates per year vary widely (up to two orders of magnitude) across endemic settings and provide an approach for mapping the global public health impact of taeniasis and cysticercosis.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #1 agreed to share their name with the authors.)

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

Infection by Taenia solium poses a major burden across endemic countries. The World Health Organization (WHO) 2021–2030 Neglected Tropical Diseases roadmap has proposed that 30% of endemic countries achieve intensified T. solium control in hyperendemic areas by 2030. Understanding geographical variation in age-prevalence profiles and force-of-infection (FoI) estimates will inform intervention designs across settings. Human taeniasis (HTT) and human cysticercosis (HCC) age-prevalence data from 16 studies in Latin America, Africa, and Asia were extracted through a systematic review. Catalytic models, incorporating diagnostic performance uncertainty, were fitted to the data using Bayesian methods, to estimate rates of antibody (Ab)-seroconversion, infection acquisition and Ab-seroreversion or infection loss. HCC FoI and Ab-seroreversion rates were also estimated across 23 departments in Colombia from 28,100 individuals. Across settings, there was extensive variation in all-ages seroprevalence. Evidence for Ab-seroreversion or infection loss was found in most settings for both HTT and HCC and for HCC Ab-seroreversion in Colombia. The average duration until humans became Ab-seropositive/infected decreased as all-age (sero)prevalence increased. There was no clear relationship between the average duration humans remain Ab-seropositive and all-age seroprevalence. Marked geographical heterogeneity in T. solium transmission rates indicate the need for setting-specific intervention strategies to achieve the WHO goals.

Article activity feed

  1. Version published to 10.7554/elife.76988 on eLife
  2. eLife

    Author Response

    Reviewer #1 (Public Review):

    1.1) Dixon and colleagues aim to fill a major gap in our understanding of the epidemiology of human disease caused by Taenia solium (taeniasis and cysticercosis), a major food-borne zoonotic cestode. They use a rather heterogeneous dataset comprising "age-prevalence" data to fit catalytic models and infer a key epidemiological parameter, the force of infection of taeniasis and cysticercosis.

    The authors are to be commended for exploring the scarce information regarding the prevalence of Taenia antigens and antibodies in different endemic settings. It remains unclear to me why were the much more numerous studies relying on fecal egg detection for diagnosing taeniasis not included in their analysis. One reason might be that their main focus is on T. solium infection and classic parasitological diagnosis cannot distinguish between T. solium and the even more common human pathogen, T. saginata - but the most common coproantigen detection method used worldwide (Allan et al., 1990) also fails to reliably distinguish between these two species.

    The first clear limitation of the primary datasets analyzed for addressing the prevalence of taeniasis is that they combine infections with different species of Taenia - T. solium, T. saginata, and perhaps T. asiatica.

    We thank the reviewer for this summary and for acknowledging how our study attempts to fill a key research gap, using scarce and heterogeneous data sources. We agree that the challenges around measuring T.solium taeniasis infection in humans are based around a lack of species-specific diagnostics. The reviewer rightly points out that our main justification for not including faecal egg detection is the inability to distinguish at the species level, with this currently remaining a challenge for the copro-antigen-based diagnostics.

    We have included copro-antigen-based surveys (but not included surveys based on faecal egg detection, see below) because, for measuring T. solium-specific infection, the copro-antigen diagnostic characteristics (sensitivity and specificity), including for adapted protocols (e.g., Allan et al. 1990), are readily available to inform the necessary diagnostic adjustment process in the model. This is not the case for faecal egg detection approaches. It is our contention that there would be wide variation in protocols for faecal egg detection methods (without associated sensitivity and specify estimates), which would make inclusion of the diagnostic adjustment step very challenging and uninformative. Including the diagnostic adjustment for copro-antigenbased surveys therefore provides a more robust approach to tackling the issue associated with cross-reactivity (suboptimal specificity), which is then reflected by further uncertainty in the force-of-infection (FoI) estimates.

    1.2) Second, they combine diagnostic data based on coproantigen and antibody detection for modeling the force of infection of taeniasis. These are data of a completely different nature. Although the authors use reverse catalytic models to account for "infection loss", they are coping with different biologic processes classified under "infection loss" - the slow decline in antibody responses vs. the sudden clearance of coproantigens following treatment or spontaneous worm elimination. In areas of high endemicity, people may be often reinfected ("infection acquisition") but antibody seroconversion rates will grossly underestimate reinfection rates if many individuals remain seropositive at the time they are reinfected.

    We thank the reviewer for raising this important distinction to shed light on the interpretation of models fitted to either antigen-based surveys (e.g., copro-antigen), which we use as a marker of infection acquisition (λinf), described on lines 530-532, and models fitted to antibody-based surveys (e.g., rES33-immunoblot for taeniasis), indicating seroconversion (λsero), as described on lines 529-530. The reviewer is correct that we assume these underpin different biological processes. In Figure 1 (note, now Figure 2), we qualify how we interpret these two quantities, λinf and λsero, in conjunction with the structure of the catalytic model configurations. We agree that the additional parameter in the reversible model, seroreversion (ρsero) or infection loss (ρinf), refer to different mechanisms, and we define these two parameters differently depending on whether the reversible model is fitted to antibody-based or antigen-based datasets. However, we agree that we could make this clearer in Table 1, by reinforcing in the Table title that models fitted to antibody data aim to estimate seroconversion and seroreversion (exposure), while models fitted to antigen data aim to estimate infection acquisition and infection loss (active infection). The title for Table 1 title has therefore been amended as follows:

    “Table 1. Parameter posterior estimates for the best-fit catalytic models fitted to human taeniasis age-(sero)prevalence datasets (ordered by decreasing all-age (sero)prevalence). Parameters estimated from antibody-based datasets measure exposure dynamics, with seroconversion λsero and seroreversion ρsero rates. Parameters estimated from antigen-based datasets measure active infection dynamics, with infection acquisition λinf and infection loss ρinf rates.”

    We do not currently include a model capable of estimating seroreversion and infection loss rates from a single dataset, so we keep our interpretations separately. The reviewer points that seroconversion rates may underestimate infection loss rates, which is true where seroconversion rates are slow compared to resolution of natural infection, although we are not equating infection loss rates with seroconversion rates and keep these interpretations distinctly (see Figure 2). This is, however, an important consideration to include in the Discussion, so we have added the following text:

    “In addition, reinfection of individuals with the adult tapeworm is also likely to occur, particularly in high-endemicity settings; therefore, the persistence of antibodies against the adult worm is likely to complicate the measurement of reinfection rates (where antibody seroconversion is equated to infection, although we take care to differentiate between these two processes when interpreting the λsero and λinf parameters). However, with the limited number of HTT-based surveys available to estimate antibody seroconversion and duration of antibody parameters, it is difficult to determine to what extent this is an issue.” (lines 386 – 395)

    1.3) The "human cysticercosis" component of the study also relies on antigen and antibody detection. The diagnostic methods are assumed to be both species-specific (i.e., they distinguish between T. solium and T. saginata) and, even more critically, to be stage-specific (i.e., they distinguish between antibodies elicited by exposure to T. solium cysticerci and those elicited by adult worms). This appears to be the case of the classic EITB assay, but it remains unclear whether the diagnostic method (López et al.) used in the large, nationwide Colombian dataset is sufficiently species- and stage-specific.

    We thank the reviewer for these comments. More generally, Taenia saginata does not cause cysticercosis in humans, but cross-reactivity with other parasites is an important consideration (see below). The assays on which the data we analysed are based, are generally highly (species- and stage-) specific, with values >90%, identified in the literature, being broadly consistent with our posterior estimates of diagnostic specificity. The sensitivity values (>80%, reported in the literature) are also in agreement with our posterior estimates of diagnostic sensitivity. However, we thank the reviewer for requesting more information on the López et al. (1988) human cysticercosis antibody diagnostic.

    The Flórez Sánchez et al. 2013 paper provides further background on the López et al. test, indicating: “diagnostic tool to determine exposure to parasitic infection through the ELISA test for the detection of anti-cysticercus immunoglobulin G (IgG) antibodies” (translated from Spanish). In addition, the test was “standardized and evaluated in Colombian patients with parasitologically proven neurocysticercosis, with a sensitivity of 100% and a specificity of 97.6% in serum samples and 100% in both values with CSF samples”. The authors also state that “in its standardization, cross-reactions with different infectious agents such as Taenia saginata, Hymenolepis nana, Echinococcus sp., Fasciola hepatica, Entamoeba histolytica, Ascaris lumbricoides, Mansonella ozzardi, Treponema pallidum, Cryptococcus neoformans and HIV were evaluated, which were discarded”.

    We therefore conclude that the López et al. assay is both species- (T. solium) and stage- (to cysticerci) specific.

    1.4) Finally, the brief description of the source studies overlooks basic information. Were study participants randomly sampled in each study site? What about sampling units - individuals or households? Are study sites representative of the countries?

    We agree with the reviewer that this additional information should be included, which we now introduce into Supplementary Table S1 (under the new column “Study design, sampling strategy and representativeness”).

    To summarise, study participants were randomly selected in 8 studies, and in 4 studies all eligible participants in study sites (e.g., specific village) were selected. In 3 studies (Moro et al. 2003; Nguekam et al. 2003; Weka et al. 2013), non-random sampling was performed or information was not available to assess the methodology adequately. The unit of randomisation was the household in 3 studies (as a first sampling stage) followed by all eligible household members being sampled (Gomes et al. 2002; Conlan et al. 2012; Wardrop et al. 2015). In 2 studies, households were randomised first, then one household member was randomly selected (Holt et al. 2016; Sahlu et al. 2019). In the study by Flórez Sánchez et al. (2013), three-stage sampling was conducted. In 1 study individuals were the units of randomisation (Edia-Asuke et al. 2015), and in 1 study the sampling information was not sufficiently clear (Theis et al. 1994).

    Although there was somewhat limited information to determine how representative of the countries the studies were, in several studies the authors indicated that the study sites, selected from different areas, were representative of specific socio-economic factors across a region (e.g., Conlan et al. 2012; Jayaraman et al. 2011). Other study sites were selected based on prior knowledge of the presence of high-risk factors for T. solium or prevalence of porcine cysticercosis (Sahlu et al. 2019; Kanobana et al. 2011; Edia-Asuke et al. 2015).

    1.5) Given the potential limitations inherent to the datasets analyzed, it remains uncertain whether the authors can provide "global force-of-infection trends" derived from a small number of studies with different diagnostic approaches - although they can surely describe productive ways of interrogating available data, point to their limitations and suggest standardized study designs that might generate better data for future pooled analyses.

    We agree and thank the reviewer for this feedback. We have addressed the issue of representativeness of each study (where information is available) under Authors’ response 1.4. In several studies, the authors indicated that the study sites, selected from different areas, were representative of specific socioeconomic factors across a region (e.g., Conlan et al. 2012; Jayaraman et al. 2011). We acknowledge that it was difficult to determine how representative some of the other studies were at country level. Having said that, we strongly contend that our analyses, taken in their entirety, do indicate substantial variation in FoI/seroreversion or infection loss estimates across a range of different epidemiological settings representing the major global endemic areas (e.g., South America, sub-Saharan Africa and Asia). The distribution of study sites is visually presented in Figure S2, which is now Figure 1– figure supplement 1: Geographical distribution of studies with human taeniasis (HTT) and human cysticercosis (HCC) age-(sero) prevalence data included in the final analysis (n = 16) by diagnostic method. For this reason, we believe this study reflects the variation in global trends, and therefore propose modifying the manuscript title as follows: “Global variation in Force-of-Infection trends for Human Taenia solium Taeniasis/Cysticercosis”.

  3. eLife

    Evaluation Summary:

    Basic epidemiological parameters such as the force of infection (rate at which susceptible individuals acquire the infection) remain undetermined for human infection with the neglected food-borne zoonotic cestode Taenia solium, which may cause taeniasis and cysticercosis. Dixon and colleagues address this major gap by fitting simple mathematical models to datasets that describe the prevalence of taeniasis and cysticercosis in several countries. Importantly, they found that infection acquisition rates per year vary widely (up to two orders of magnitude) across endemic settings and provide an approach for mapping the global public health impact of taeniasis and cysticercosis.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #1 agreed to share their name with the authors.)

  4. eLife

    Reviewer #1 (Public Review):

    Dixon and colleagues aim to fill a major gap in our understanding of the epidemiology of human disease caused by Taenia solium (taeniasis and cysticercosis), a major food-borne zoonotic cestode. They use a rather heterogeneous dataset comprising "age-prevalence" data to fit catalytic models and infer a key epidemiological parameter, the force of infection of taeniasis and cysticercosis.

    The authors are to be commended for exploring the scarce information regarding the prevalence of Taenia antigens and antibodies in different endemic settings. It remains unclear to me why were the much more numerous studies relying on fecal egg detection for diagnosing taeniasis not included in their analysis. One reason might be that their main focus is on T. solium infection and classic parasitological diagnosis cannot distinguish between T. solium and the even more common human pathogen, T. saginata - but the most common coproantigen detection method used worldwide (Allan et al., 1990) also fails to reliably distinguish between these two species.

    The first clear limitation of the primary datasets analyzed for addressing the prevalence of taeniasis is that they combine infections with different species of Taenia - T. solium, T. saginata, and perhaps T. asiatica.

    Second, they combine diagnostic data based on coproantigen and antibody detection for modeling the force of infection of taeniasis. These are data of a completely different nature. Although the authors use reverse catalytic models to account for "infection loss", they are coping with different biologic processes classified under "infection loss" - the slow decline in antibody responses vs. the sudden clearance of coproantigens following treatment or spontaneous worm elimination. In areas of high endemicity, people may be often reinfected ("infection acquisition") but antibody seroconversion rates will grossly underestimate reinfection rates if many individuals remain seropositive at the time they are reinfected.

    The "human cysticercosis" component of the study also relies on antigen and antibody detection. The diagnostic methods are assumed to be both species-specific (i.e., they distinguish between T. solium and T. saginata) and, even more critically, to be stage-specific (i.e., they distinguish between antibodies elicited by exposure to T. solium cysticerci and those elicited by adult worms). This appears to be the case of the classic EITB assay, but it remains unclear whether the diagnostic method (López et al.) used in the large, nationwide Colombian dataset is sufficiently species- and stage-specific.

    Finally, the brief description of the source studies overlooks basic information. Were study participants randomly sampled in each study site? What about sampling units - individuals or households? Are study sites representative of the countries?

    Given the potential limitations that are inherent to the datasets analyzed, it remains uncertain whether the authors can provide "global force-of-infection trends" derived from a small number of studies with different diagnostic approaches - although they can surely describe productive ways of interrogating available data, point to their limitations and suggest standardized study designs that might generate better data for future pooled analyses.

  5. eLife

    Reviewer #2 (Public Review):

    In this work the authors study human taeniasis and cysticercosis based on previously published data from South America, Africa and Asia. They use parsimonious catalytic models incorporating sensitivity and specificity of the diagnostics to estimate rates of infection and infection loss.

    Strengths:

    The work is based on multiple datasets extracted from 16 epidemiological studies from 3 continents. The authors account for diagnostic performance uncertainty (beta distribution priors were used to capture literature estimates) and consider parsimonious models to describe antigen and antibody age-prevalence profile of human taeniasis and human cysticercosis. Model parameters were estimated with Bayesian methods assuming uniform or weakly informative prior distributions.

    Weakness:

    Although parsimonious models are always desirable, they sometimes lack mechanisms and only provide an overview of systems. The authors considered simple and reversible catalytic models to describe datasets from distinct setting. By doing so, site specific drivers, heterogeneity in exposure and susceptibility or age-related immunity mechanisms were disregarded.

  6. eLife

    Reviewer #3 (Public Review):

    In this manuscript, the authors systematically review the literature on the epidemiology of Taenia solium in humans to identify possible patterns in transmission dynamics across different regions. The authors' review extends on the work of a prior systematic review (Coral-Almeida et al 2014). The calculation of region-specific force-of-infections (FoIs) is new to this manuscript and answers an important gap in the field. The authors also calculate region-specific FoIs across different departments of Colombia, demonstrating the usefulness of their method in a more real-world setting.

    The authors clearly describe their goals and results, and their claims are supported by their data. While there are limitations to these results, the authors appropriately acknowledge these and their potential impacts in the manuscript.

  7. Version published to 10.1101/2022.02.11.22270710v1 on medRxiv