^{1}

^{2}

^{*}

^{1}

^{3}

^{4}

^{5}

^{5}

Conceived and designed the experiments: MRT. Performed the experiments: HC STM EB RO. Analyzed the data: MRT HC LJ. Wrote the paper: MRT. Interpreted the data: MRT HC LJ. Contributed to all phases of the field study: STM. Supervised the acquisition of data: EB RO.

The authors have declared that no competing interests exist.

Polyparasitism can lead to severe disability in endemic populations. Yet, the association between soil-transmitted helminth (STH) and the cumulative incidence of

Longitudinal data from 2276 participants in 50 villages in Samar province, Philippines treated at baseline for

The misclassification bias increased with decreasing test accuracy. Hookworm infection was found to be associated with increased 12-month cumulative incidence of

Hookworm, roundworm, and whipworm are collectively known as soil-transmitted helminths. These worms are prevalent in most of the developing countries along with another parasitic infection called schistosomiasis. The tests commonly used to detect infection with these worms are less than 100% accurate. This leads to misclassification of infection status since these tests cannot always correctly indentify infection. We conducted an epidemiological study where such a test, the Kato-Katz technique, was used. In our study we tried to show how misclassification error can influence the association between soil-transmitted helminth infection and schistosomiasis in humans. We used a statistical technique to calculate epidemiological measures of association after correcting for the inaccuracy of the test. Our results show that there is a major difference between epidemiological measures of association before and after the correction of the inaccuracy of the test. After correction of the inaccuracy of the test, soil-transmitted helminth infection was found to be associated with increased risk of acquiring schistosomiasis. This has major public health implications since effective control of one worm can lead to reduction in the occurrence of another and help to reduce the overall burden of worm infection in affected regions.

Polyparasitism is a common feature in parasite endemic regions, which includes most developing countries

Laboratory studies suggest that infection with one helminth may influence the outcome of infection with another helminth

The purpose of this study was to show the impact of adjusting for misclassification error in estimating the effect of STH infections on the 12-months cumulative incidence of

The research was approved by the institutional review board (IRB) of the Brown University in the United States and the IRB of the Research Institute for Tropical Medicine in the Philippines. The data analysis component of the study was reviewed and approved by the University of Oklahoma Health Sciences Center IRB. The chiefs of all villages were asked permission for the village to be included in the study. In addition, all eligible participants were asked for their consent to participate. Only those individuals who provided written informed consent were included. Written informed consent for individuals below 18 years old was obtained and provided by parents or legal guardians.

We used data from a longitudinal study conducted between January 2004 and December 2005 in the province of Samar, the Philippines. The main purpose of the original study was to assess the effect of water and animal management systems on the transmission of

Seventy-five out of 134 villages endemic for

Eligible households were those of at least five members and where at least one member was working full time in a rain-fed farm in “rain-fed” villages and at least 50% of the time in a man-made irrigated farm in “irrigated” villages. A maximum of 35 eligible households were randomly selected from each village using the following procedure. A list of 50 random numbers was created (one list per village). Eligible households were allocated consecutive numbers and visited in the order chosen at random. If a household refused to participate, the next available household was asked to participate. When 35 or fewer households were eligible in a village, they were all invited to participate in the study. At most six individuals including at least one full-time rice farmer were selected at random from each household.

An individual-level interview included questions on age, gender, and occupation. Participants were asked to provide one stool sample (morning or first) per day for three consecutive days. Each participant provided between one and three stool samples. If a participant provided a stool sample on one of the three days but was unable for any reason to provide stool samples on other days, that person was still considered as a stool sample provider. Stool envelopes (of wax paper and book paper) with popsicle sticks were distributed to participants a day before the actual stool collection. At least thumb-size stool samples were submitted. Portions from different parts of the stool were taken to fill up the template. Although consistency of the stool sample was not recorded, only pasty to formed stool could be accommodated in the stool envelopes. Stool samples were processed 2–3 h after collection. Two slides were prepared from each stool sample. All slides were placed in a styrofoam box with cold packs inside at the end of each collection day. At the end of each collection week all slides were brought to a designated laboratory and transferred to a refrigerator. The time delay between stool sample processing and microscopic reading associated with day one stool collection (provided by 99.45% of participants) ranged from less than 24 hours to as long as 20 days with a median of 4 days (inter-quartile range: 2–6 days). Stool samples were examined for the presence of eggs of

Details about the mass treatment have been published elsewhere

All of the study participants were asked to provide three stool samples over three consecutive days 12 months after the mass treatment. All individuals who provided at least one stool sample were considered as follow-up stool sample providers. Stool samples were processed and examined in the same manner and by the same people as at baseline.

Some of the participants who provided the baseline stool samples did not participate in the mass treatment program. Moreover, not all participants provided stool samples during the follow-up survey. The 12-month cumulative incidence of

As mentioned earlier, we obtained between one and three stool samples on consecutive days from each participant at baseline and follow-up. This introduces individual variations in the sensitivity and specificity of the Kato-Katz to detect infection. To take this variation into account, and to adjust for the village-level clustering of infection, we used a Bayesian latent class hierarchical cumulative-logit regression model based on a method described by Joseph and others (1995) and adapted to our problem (1, 2, or 3 days of sampling) for

The probability of any single test being positive is the sum of the probability of a true positive result and the probability of a false positive result. If

When there is more than one test per person, the properties of multiple tests can be modeled using probability

The main outcome of interest here is the probability distribution of the true

Each hierarchical model consists of three levels, as follows: the first level includes one intercept parameter for each village and independent variables for age, sex, occupation, and one of the STHs under study. At the second level of the hierarchical model, the intercept parameters from each of the 50 villages are modeled as a linear regression to account for the clustering of infection within village. At the third level, prior distributions were specified for all parameters. Uniform (uninformative) prior distributions on the range from 0 to 1 (parameters of the beta distribution: α = 1, β = 1) were used for sensitivity and specificity of all three STH infections. For

The above model was modified to construct three additional models: one model accounted for misclassification error in outcome but not in exposure, one accounted for misclassification error in exposure but not in outcome, and another one did not account for any misclassification error. For models where misclassification error was not accounted for, an individual with any stool sample positive for a particular STH was considered as infection positive for that STH. For

We assumed conditional independence between subsequent tests in our model, meaning in practice that when more than one sample was available from a subject, the test results are independent from each other, conditional on the person's true infection status. In other words, the probability of a positive (or negative) test depends only on the true status, and once this true staus is known, does not depend on any test results from other days. This assumption seemed reasonable, and simplifies the statistical model compared to a model that might account for any between-day dependencies.

WinBUGS software (version 1.4.3, MRC Biostatistics Unit, Cambridge, UK) was used to implement the Gibbs sampler algorithm. Posterior medians of random samples derived from marginal posterior densities were used as point estimates, reported with 95% Bayesian credible intervals (BCI). The programs written in WinBUGS are available upon request to the authors.

Of the 5624 individuals who agreed to participate in the study at baseline, 2276 (40.5%) constitute the group “at-risk”. The “at-risk” group and those who were not treated with praziquantel or did not provided any stool sample during the follow-up (“not at-risk” group) are compared in

Characteristic | At-risk group, no. (%) | Not at-risk group, no. (%) |

2276 (40.5) | 3348 (59.5) | |

<10 | 658 (28.9) | 1193 (35.6) |

11–16 | 399 (17.5) | 458 (13.7) |

17–40 | 618 (27.2) | 1047 (31.3) |

>40 | 601 (26.4) | 650 (19.4) |

1274 (56.0) | 1692 (50.5) | |

1142 (50.2) | 1368 (40.9) | |

534 (23.5) | 350 (10.5) |

Exposure variable is respective soil-transmitted helminth infection; All odds ratio estimates are adjusted for age, sex, and occupation; BCI: Bayesian credible intervals; ME: misclassification error; ^{a} Correction of misclassification error in exposure (respective STH infection) and outcome (

All ME correction |
Outcome ME correction | Exposure ME correction | No ME correction | |||||

Covariates | OR | 95% BCI | OR | 95% BCI | OR | 95% BCI | OR | 95% BCI |

Reference: ≤10 yrs | ||||||||

11–16 yrs | 1.05 | 0.51, 2.11 | 1.05 | 0.50, 2.08 | 1.03 | 0.62, 1.67 | 1.04 | 0.63, 1.68 |

17–40 yrs | 0.49 | 0.22, 1.07 | 0.50 | 0.22, 1.04 | 0.60 | 0.34, 1.01 | 0.61 | 0.35, 1.03 |

>40 yrs | 0.21 | 0.09, 0.49 | 0.22 | 0.09, 0.49 | 0.36 | 0.20, 0.63 | 0.37 | 0.20, 0.65 |

Reference: female | ||||||||

Male | 2.33 | 1.50, 3.75 | 2.33 | 1.48, 3.70 | 1.83 | 1.34, 2.50 | 1.84 | 1.35, 2.52 |

Reference: rice farming | ||||||||

Non-rice farming | 0.42 | 0.20, 0.86 | 0.42 | 0.20, 0.82 | 0.48 | 0.29, 0.78 | 0.49 | 0.30, 0.79 |

Reference: ≤10 yrs | ||||||||

11–16 yrs | 1.03 | 0.49, 2.07 | 1.07 | 0.52, 2.12 | 1.02 | 0.61, 1.66 | 1.03 | 0.62, 1.67 |

17–40 yrs | 0.44 | 0.19, 0.94 | 0.48 | 0.22, 1.00 | 0.56 | 0.32, 0.96 | 0.58 | 0.33, 1.01 |

>40 yrs | 0.20 | 0.08, 0.45 | 0.21 | 0.09, 0.47 | 0.34 | 0.18, 0.60 | 0.35 | 0.19, 0.63 |

Reference: female | ||||||||

Male | 2.12 | 1.35, 3.34 | 2.15 | 1.38, 3.48 | 1.72 | 1.26, 2.37 | 1.75 | 1.28, 2.40 |

Reference: rice farming | ||||||||

Non-rice farming | 0.44 | 0.21, 0.87 | 0.45 | 0.22, 0.88 | 0.49 | 0.30, 0.79 | 0.50 | 0.30, 0.82 |

Reference: ≤10 yrs | ||||||||

11–16 yrs | 1.09 | 0.52, 2.16 | 1.08 | 0.52, 2.14 | 1.03 | 0.62, 1.68 | 1.03 | 0.63, 1.68 |

17–40 yrs | 0.54 | 0.25, 1.15 | 0.54 | 0.24, 1.13 | 0.63 | 0.36, 1.07 | 0.62 | 0.36, 1.05 |

>40 yrs | 0.24 | 0.10, 0.53 | 0.24 | 0.10, 0.53 | 0.37 | 0.21, 0.66 | 0.37 | 0.21, 0.66 |

Reference: female | ||||||||

Male | 2.30 | 1.48, 3.68 | 2.31 | 1.48, 3.66 | 1.82 | 1.34, 2.50 | 1.83 | 1.34, 2.50 |

Reference: rice farming | ||||||||

Non-rice farming | 0.43 | 0.21, 0.85 | 0.43 | 0.20, 0.84 | 0.49 | 0.30, 0.79 | 0.49 | 0.30, 0.79 |

Correction of misclassification error in exposure (respective STH infection) and outcome (

To our knowledge, this is the first longitudinal study to estimate the effect of STH infection on the 12-month risk of

Although our analysis included only about one third of the baseline participants from 50 villages, the longitudinal sample size was large enough for this analysis. When comparing individuals included in and excluded from the analysis, we found more rice farmers in the ‘at-risk’ group than in the ‘not at-risk’ group. This is because more males were treated than females (56.4% vs. 43.6%), and because more rice farmers were infected with

Our results show that OR estimates for all three STHs are pulled away from the null value when the OR estimates are adjusted for misclassification error. This effect of non-differential misclassification has long been recognized, although this is not always the case when exposure and outcome variables are dependent, a discrete variable assumes more than two values, or there is misclassification error in the confounding variable

The effect of misclassification on the OR estimates of the association between STH and the risk of

Two studies have reported estimates of cross-sectional association between hookworm infection and infection by another schistosome species (

Important changes in OR estimates for other covariates were also observed. The OR estimates for covariates when only

The largest impact of misclassification error was observed for the association between hookworm and

The efficacy of praziquantel for the treatment of schistosomiasis has been reported to range between 71% and 99% in published literature

Our data suggest that hookworm infection is associated with increased 12-month cumulative incidence of

STROBE Checklist.

(0.09 MB DOC)