The authors have declared that no competing interests exist.
Large-scale deworming programs have, to date, mostly targeted preschool- and school-age children. As community-based deworming programs become more common, deworming will be offered to women of reproductive age. The World Health Organization recommends preventive chemotherapy be administered to pregnant women only after the first trimester. It is therefore important for deworming programs to be able to identify women in early pregnancy. Our objective was to validate a short questionnaire which could be used by deworming program managers to identify and screen out women in early pregnancy.
In May and June 2018, interviewers administered a questionnaire, followed by a pregnancy test, to 1,203 adult women living in the Peruvian Amazon. Regression analyses were performed to identify questions with high predictive properties (using the pregnancy test as the gold standard). Test parameters were computed at different decision tree nodes (where nodes represented questions). With 106 women confirmed to be pregnant, the positive predictive value of asking the single question
To identify women in early pregnancy when deworming programs are community-based, both the number and order of questions are important. The local context and cultural acceptability of different questions should inform this decision. When numbers are manageable and resources are available, pregnancy tests can be considered at different decision tree nodes to confirm pregnancy status. Trade-offs in terms of efficiency and misclassification rates will need to be considered to optimize deworming coverage in women of reproductive age.
To date, large-scale deworming programs have strategically targeted the high risk groups of school-age and preschool-age children in worm-endemic areas using the highly cost-effective existing infrastructure of schools. To achieve elimination of worm-attributable morbidity, however, adult populations will also need to be treated. The World Health Organization considers women of reproductive age to also be a high risk group for worm-attributable morbidity and they will be increasingly included in large-scale community-based deworming programs. Although deworming treatment is considered safe and effective at any time, it is recommended that pregnant women in the first trimester be excluded from deworming treatment. Therefore, program managers need to have a screening tool in order to rule out early trimester pregnant women during deworming program implementation. To respond to this need, we evaluated the predictive properties of a parsimonious set of questions in a study population of adult women of reproductive age in a worm-endemic region of the Peruvian Amazon. We present several question scenarios to assist program managers in using questions to rule out early trimester pregnant women. Adapted to local cultural settings, such a screening tool can optimize deworming coverage in women of reproductive age.
Women of reproductive age (WRA) in low- and middle-income countries are particularly vulnerable to morbidity resulting from soil-transmitted helminth infection (STH) [
Most previous research on STH infections and deworming has focused on children. Even though it is well known that worm infections cause and exacerbate anemia during the different stages of a woman’s reproductive life span, WRA have been under-studied in this context [
The deworming intervention currently used in large-scale public health intervention programs consists of a single dose of deworming medicine (either albendazole or mebendazole), repeated either once or twice a year, depending on the prevalence of STH in the area [
In resource-limited settings, determining pregnancy status by ultrasound, blood testing, or a urine-based pregnancy test is impractical, unfeasible and prohibitively costly, so the common practice has been to use the start date of the last menstrual period (LMP) [
Only a small number of studies have examined the performance of a multi-item questionnaire in determining pregnancy status. In a clinic-based study sample of 283 women who had missed their periods and who had requested a pregnancy test, the authors concluded that neither a woman’s self-assessment of pregnancy nor the presence of pregnancy symptoms had a high degree of accuracy in predicting pregnancy status [
To our knowledge, only one previous study has examined the performance of such a questionnaire in the context of women eligible for a mass treatment program (ivermectin in onchocerciasis control) in a limited resource setting [
The objective of the present study, therefore, was to determine the most parsimonious set of questions to identify women in early pregnancy.
This study was approved by the Oficina del Comité Institucional de Ētica en Investigación of the Hospital Regional de Loreto "Felipe Arriola Iglesias" in Iqutios, Peru (ID 019-CIEI-2018) and by the Research Ethics Board of the McGill University Health Centre (CT1 panel; 23-04-2018), in Montreal, Canada. All participants were adults and provided written informed consent.
Data collection was completed in the 3-week period from May 29 to June 15, 2018, in Belén, an impoverished district of the capital city of Iquitos (in the department of Loreto), situated in the floodplain of the Itaya and Amazon rivers. This district has a high prevalence of STH infections in all three high risk groups (i.e. school-age children, preschool-age children and women of reproductive age, including pregnant women) [
Each study visit consisted of the sequential administration of the study questionnaire, followed immediately by a pregnancy test. The cross-sectional study design was chosen over that of a cohort study (where it might be perceived that a better ‘gold standard’ could be used (i.e. confirmed clinical pregnancy), for the following reasons: i) the loss to follow-up between the baseline assessments and the confirmed clinical pregnancy would likely exceed the measurement error of the currently proposed gold standard (i.e. the rapid pregnancy test); ii) some number of new pregnancies and miscarriages occurring between the baseline assessments and follow-up confirmation would be missed, contributing to measurement error; and iii) cost and feasibility issues were of concern.
The questionnaire and pregnancy tests were administered by a group of 11 research assistants (RAs) who were midwives or midwife-trainees. The RAs were trained on the administration of informed consent, the questionnaire and the pregnancy test. They worked in groups of two or three, visiting women in their own households according to a pre-set daily assignment. While obtaining informed consent, the women were asked where in their home they would be most comfortable completing the questionnaire and pregnancy test. Women were included in the study if they were between the ages of 18 and 49 years and consented to participate.
The one-page study questionnaire had five basic identifier questions (e.g. manzana number, household number), 17 questions on socio-demographic and pregnancy-related characteristics (e.g. age, birth date, marital status, number of children breastfeeding status, known or suspected pregnancy status (including gestational age and method of confirmation for women declaring they were pregnant), start date of the last menstrual cycle, and the presence/absence of breast tenderness, darkened areolas, fatigue, nausea, and vomiting) and one last question on whether or not the woman wished to receive the results of the pregnancy test (
The pregnancy test (ABON hCG, Abon Biopharm (Hangzhou) Co., Ltd, China) was performed on a urine sample provided by the woman after the questionnaire was complete. The test was a rapid chromatographic immunoassay designed to detect the presence of hCG. The manufacturer reports test sensitivity to 25mIU/ML hCG with results available within 3 minutes. Elevated hCG levels can be reliably detected by this test after a minimum interval of one day following the first missed menstrual period in a woman’s cycle, with both a sensitivity and specificity of 100% (95% CI [95–100%]) and a precision level of 100% (95% CI [98%-100%]) [
The results from both the questionnaire and the pregnancy test were entered into an electronic database by two independent research staff. Data quality was monitored daily by research personnel through field supervision and internal checks for consistency.
General socio-demographic information was summarized for the population as a whole, and stratified by pregnancy test result, to present an initial unadjusted view of the distribution of baseline variables in the study population. Comparisons were performed using independent
Answers to the questionnaire and the pregnancy test results were summarized using univariate analyses to assess the predictive capacity of each question (answers to each question were considered as variables in the analyses) with respect to the outcome of pregnancy status. A multivariate analysis was then performed in order to assess the stability of the estimates and to identify the variables that independently had the highest predictive capacity. Odds ratios (ORs) and 95% confidence intervals (CIs) for the association between the independent predictor variables and the dependent binary variable (i.e. the pregnancy test result) were computed using logistic regression for both the univariate and multivariate analyses. Variables were removed from modeling when co-linear or sparse. Self-awareness of pregnancy status was categorized as either
Diagnostic measures of sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were computed for each multi-item question response against the gold standard of the binary pregnancy test result. Youden’s J statistic was then calculated to identify questions with the highest accuracy in terms of sensitivity and specificity. (The Youden’s J statistic takes into consideration both false positives (women incorrectly classified as pregnant when asked one or more questions) and false negatives (women incorrectly classified as not pregnant when asked one or more questions), and ranges from 0 (the test is completely uninformative) to 1 (the test is perfect) [
All statistical analyses were performed using R (version 3.4.3, R Core Team 2017, Vienna, Austria).
A total of 1,203 adult women consented to participate in the study. The women were recruited from 1,074 households of a total of 1,258 households (85.4%) canvassed from 182
Study population characteristics are summarized in
Pregnancy test result | ||||
---|---|---|---|---|
Overall | Negative | Positive | p |
|
n | 1 097 | 106 | ||
Age, mean ± SD |
31.12 ± 8.74 | 31.46 ± 8.83 | 27.55 ± 6.79 | <0.001 |
Marital status, n (%) | <0.001 | |||
Single | 258 (21.4) | 252 (23.0) | 6 (5.7) | |
Married | 154 (12.8) | 143 (13.0) | 11 (10.4) | |
Partner in shared household | 766 (63.7) | 680 (62.0) | 86 (81.1) | |
Partner in separate household | 24 (2.0) | 21 (1.9) | 3 (2.8) | |
Other | 1 (0.1) | 1 (0.1) | 0 (0) | |
Has children, n (%) | 1,039 (86.4) | 949 (86.5) | 90 (84.9) | 0.756 |
Number of children, mean ± SD |
2.39 ± 1.79 | 2.42 ± 1.80 | 2.06 ± 1.65 | 0.048 |
The results presented in
The results of the univariate and multivariate logistic regression analyses are presented in
Univariate | Multivariate | |||
---|---|---|---|---|
Variable | OR [95% CI] | p |
aOR [95% CI] | p |
0.94 [0.92, 0.97] | <0.001 | 1.00 [0.94, 1.07] | 0.961 | |
Single | ||||
Married | 3.24 [1.21, 9.59] | 0.023 | NA |
NA |
Partner in shared household | 5.33 [2.50, 13.83] | <0.001 | 5.86 [1.07, 110.69] | 0.099 |
Partner in separate household | 6.02 [1.21, 24.63] | 0.016 | NA |
NA |
0.88 [0.51, 1.59] | 0.646 | 1.45 [0.28, 10.48] | 0.679 | |
0.21 [0.07, 0.48] | <0.001 | 0.64 [0.10, 2.93] | 0.591 | |
Knows she is pregnant | NA |
NA |
NA |
NA |
Thinks she is pregnant: Yes | 4.64 [0.95, 17.64] | 0.034 | 0.84 [0.08, 7.30] | 0.878 |
Thinks she is pregnant: No | 0.13 [0.05, 0.34] | <0.001 | 0.16 [0.05, 0.47] | 0.001 |
Thinks she is pregnant: Does not know | ||||
Approximately 1 month ago | ||||
Less than 1 month ago | 0.20 [0.05, 0.70] | 0.013 | 0.15 [0.03, 0.60] | 0.008 |
1–3 months ago | 11.94 [5.04, 33.09] | <0.001 | 0.73 [0.15, 3.12] | 0.680 |
4–6 months ago | 37.29 [15.36, 105.92] | <0.001 | 1.10 [0.18, 5.51] | 0.911 |
7–9 months ago | 27.74 [10.77, 82.09] | <0.001 | NA |
NA |
In menopause | NA |
NA |
NA |
NA |
Does not know/remember | 1.38 [0.50, 4.13] | 0.540 | 0.16 [0.02, 0.89] | 0.051 |
Soreness of breasts | 4.61 [3.03, 6.99] | <0.001 | 0.75 [0.18, 2.49] | 0.666 |
Darkened areolas | 65.84 [39.26, 114.44] | <0.001 | 9.42 [2.58, 35.34] | <0.001 |
Increasing fatigue | 7.15 [4.64, 11.29] | <0.001 | 2.49 [0.84, 7.46] | 0.098 |
Nausea | 5.69 [3.55, 9.02] | <0.001 | 1.87 [0.39, 7.19] | 0.393 |
Vomiting | 8.66 [4.83, 15.32] | <0.001 | 1.38 [0.20, 7.74] | 0.725 |
OR [95%CI], odds ratio [95% confidence interval]; aOR, adjusted odds ratio (adjusted for all other variables in the model);
a p-values reported are estimated using logistic regression analysis. A significance cut-off of p < 0.05 was applied. p-values inferior to 0.001 are reported as <0.001.
b NA: Estimates unreliable due to having too few observations (in the category as a whole or when controlling for other variables).
c NA: Estimates unreliable due to near perfect separation of the data when stratified by pregnancy test result.
* the question asking about the number of children was dichotomized (having children versus not having children) to better reflect the question which would be asked in a deworming program (rather than asking about a specific number of children)
Questions | Sensitivity (%) | Specificity (%) | PPV (%) | NPV (%) | Youden’s J |
---|---|---|---|---|---|
Single | 5.7 | 77 | 2.3 | 89.4 | < 0 |
Married | 10.4 | 87 | 7.1 | 90.9 | < 0 |
Partner in shared household | 81.1 | 38 | 11.2 | 95.4 | 0.191 |
Partner in separate household | 2.8 | 98.1 | 12.5 | 91.3 | 0.009 |
84.9 | 13.5 | 8.7 | 90.2 | < 0 | |
4.7 | 81 | 2.3 | 89.8 | < 0 | |
Knows she is pregnant | 80.2 | 100 | 100 | 98.1 | 0.802 |
Thinks she is pregnant: Yes/does not know | 66.7 | 82.0 | 6.6 | 99.2 | 0.487 |
Thinks she is pregnant: No | 33.3 | 18 | 0.8 | 93.4 | < 0 |
Less than 1 month ago | 3.8 | 46.9 | 0.7 | 83.4 | < 0 |
Not less than 1 month ago | 96.2 | 53.1 | 16.6 | 99.3 | 0.493 |
Soreness of breasts | 45.3 | 84.8 | 22.3 | 94.1 | 0.301 |
Darkened areolas | 78.3 | 94.8 | 59.3 | 97.8 | 0.731 |
Increasing fatigue | 71.7 | 73.8 | 20.9 | 96.4 | 0.455 |
Nausea | 32.1 | 92.3 | 28.8 | 93.4 | 0.244 |
Vomiting | 21.7 | 96.9 | 40.4 | 92.8 | 0.186 |
PPV, positive predictive value; NPV, negative predictive value.
Pregnancy test result | |||
---|---|---|---|
Positive | Negative | Totals | |
Yes | 85 | ||
No | 1,118 | ||
Totals | 106 | 1,097 | 1,203 |
Test parameters: Sensitivity: 80.2%; Specificity: 100%; Positive predictive value: 100%; Negative predictive value: 98.1%; False positive rate: 0%; False negative rate: 1.9%.
The most accurate second question (Q2), based on test parameters of the 1,118 women and the re-calculated Youden’s J statistic, would be
Questions | Sensitivity (%) | Specificity (%) | PPV (%) | NPV (%) | Youden’s J |
---|---|---|---|---|---|
Single | 4.8 | 77.0 | 0.4 | 97.6 | < 0 |
Married | 0 | 87.0 | 0 | 97.8 | < 0 |
Partner in shared household | 95.2 | 38.0 | 2.9 | 99.8 | 0.332 |
Partner in separate household | 0 | 98.1 | 0 | 98.1 | < 0 |
85.7 | 13.5 | 1.9 | 98.0 | < 0 | |
14.3 | 81.0 | 1.4 | 98.0 | < 0 | |
Thinks she is pregnant: Yes/does not know | 66.7 | 82.0 | 6.6 | 99.2 | 0.487 |
Thinks she is pregnant: No | 33.3 | 18.0 | 0.77 | 93.4 | < 0 |
Less than 1 month ago | 19.0 | 46.9 | 0.7 | 96.7 | < 0 |
Not less than 1 month ago | 81.0 | 53.1 | 3.2 | 99.3 | 0.341 |
Soreness of breasts | 23.8 | 84.8 | 2.9 | 98.3 | 0.096 |
Darkened areolas | 38.1 | 94.8 | 12.3 | 98.8 | 0.329 |
Increasing fatigue | 57.1 | 73.8 | 4.0 | 98.9 | 0.309 |
Nausea | 28.6 | 92.3 | 6.7 | 98.5 | 0.209 |
Vomiting | 14.3 | 96.9 | 8.1 | 98.3 | 0.112 |
PPV, positive predictive value; NPV, negative predictive value.
To reduce the relatively large number of women who would not receive treatment (the number of false positives), while minimizing the risk of misclassifying women who were likely to be pregnant (the number of false negatives), a third question (Q3) (i.e.
Questions | Sensitivity (%) | Specificity (%) | PPV (%) | NPV (%) | Youden’s J |
---|---|---|---|---|---|
Single | 7.1 | 80.3 | 2.5 | 92.4 | < 0 |
Married | 0 | 83.8 | 0 | 92.2 | < 0 |
Partner in shared household | 92.9 | 36.4 | 9.4 | 98.6 | 0.293 |
Partner in separate household | 0 | 99.5 | 0 | 93.4 | < 0 |
78.6 | 14.1 | 6.1 | 90.3 | < 0 | |
0 | 82.3 | 0 | 92.1 | < 0 | |
Less than 1 month ago | 14.3 | 50.0 | 2.0 | 89.2 | < 0 |
Not less than 1 month ago | 85.7 | 50.0 | 10.8 | 98.0 | 0.357 |
Soreness of breasts | 35.7 | 82.8 | 12.8 | 94.8 | 0.185 |
Darkened areolas | 57.1 | 95.5 | 47.1 | 96.9 | 0.526 |
Increasing fatigue | 57.1 | 69.2 | 11.6 | 95.8 | 0.263 |
Nausea | 35.7 | 86.9 | 16.1 | 95.0 | 0.226 |
Vomiting | 14.3 | 95.5 | 18.2 | 94.0 | 0.098 |
PPV, positive predictive value; NPV, negative predictive value.
The misclassification in terms of the rate of false negatives can be reduced by asking a fourth question (Q4) (
Questions | Sensitivity (%) | Specificity (%) | PPV (%) | NPV (%) | Youden’s J |
---|---|---|---|---|---|
Single | 16.7 | 80.4 | 2.6 | 96.8 | < 0 |
Married | 0 | 83.1 | 0 | 96.2 | < 0 |
Partner in shared household | 83.3 | 36.5 | 4.0 | 98.6 | 0.198 |
Partner in separate household | 0 | 100 | NA | 96.9 | 0 |
83.3 | 13.8 | 3.0 | 96.3 | < 0 | |
0 | 83.6 | 0 | 96.3 | < 0 | |
Less than 1 month ago | 16.7 | 49.2 | 1.0 | 95.9 | < 0 |
Not less than 1 month ago | 83.3 | 50.8 | 5.1 | 99.0 | 0.341 |
Soreness of breasts | 33.3 | 83.1 | 5.9 | 97.5 | 0.164 |
Increasing fatigue | 50.0 | 68.8 | 4.8 | 97.7 | 0.188 |
Nausea | 50.0 | 87.3 | 11.1 | 98.2 | 0.373 |
Vomiting | 0 | 95.2 | 0 | 96.8 | < 0 |
PPV, positive predictive value; NPV, negative predictive value.
Additional analyses were conducted to examine whether alternative questions were more predictive of pregnancy in two distinct subgroups: 1) women who were currently breastfeeding and 2) women who had children. These did not provide any increased predictive accuracy. Subgroup analyses were also conducted to explore potential reasons for questionnaire responses (e.g. whether being younger, single or primiparous could explain a difference in responses). These analyses offered no additional insights.
Increasingly, girls and women of reproductive age will be included in large-scale deworming programs and it will be important to effectively rule out those who are in the early stages of pregnancy. While it is unlikely that all pregnant women will be able to be screened out unless a pregnancy test is administered, we have shown that by asking one or two questions, false negative rates can be minimized. Increases in positive predictive values (and decreases in the rate of false negatives) can only be achieved by adding one or more questions, which inevitably leads to misclassification in terms of an increase in the rate of false positives (i.e. more women would be screened out as being pregnant when they are truly not pregnant). As such, this trade-off must be balanced in light of the risks of inadvertently administering deworming treatment to pregnant women in their first trimester and the unrealized benefits stemming from eligible women not receiving the deworming treatment. To a certain extent, increases in the rate of false positives can be justified by decreases in the rate of false negatives given the relative importance of competing risks. Unfortunately, no objective standard currently exists to quantify this trade-off. One potential strategy to mitigate misclassification would be to administer pregnancy tests to those women who cannot be definitively classified as "pregnant" or "not pregnant" after one or two questions; however, pregnancy tests are not likely to be available in the context of large-scale deworming programs and the numbers of tests required might be quite high, making this approach infeasible.
Considerations other than the statistical approach used in this study should be taken into account during the implementation of a deworming program. Questions on sexual activity were not included in the present questionnaire because of acceptability issues. Cultural practices related to recognizing, and then disclosing, pregnancy status must also be considered [
The rationale for identifying women in early pregnancy in the implementation of deworming programs is the concern for adverse effects attributable to the deworming treatment in the first trimester of pregnancy. In large-scale deworming programs which include women of reproductive age, and where administering pregnancy tests to each woman would be prohibitively costly, this study has shown that inadvertent exposure to deworming treatment in early pregnancy can be considerably reduced by asking a parsimonious set of questions. The wording, number and order of questions may differ by cultural setting, in light of acceptability and feasibility concerns. These concerns can be appropriately captured in focus group discussions (ideally undertaken in all WRA subgroups) conducted prior to the launching of a WRA-targeted deworming program. This will not only enable adaptation of questions to context, including around norms linked to pregnancy disclosure, but it will also inform relevant and appropriate training for the front line health workers who will administer the questions and distribute the deworming treatment. Where possible, inadvertent exposure to deworming treatment may be further reduced, even to zero, where pregnancy tests can be administered to smaller numbers of women after answering an initial series of questions.
(XLSX)
Dr. Rubina Imtiaz, Director of Children Without Worms, provided valuable comments at the initial stages of this research. Ms. Isabela Fabri Karam assisted in the training of the field team. Global Health Programs, McGill University provided awards to Kariane St-Denis (McGill Global Health Scholar (Graduate)) and Isabela Fabri Karam (McGill Global Health Scholar (Undergraduate)).