European Journal of Obstetrics & Gynecology and Reproductive Biology
ReviewBiases in observational studies of the effect of prenatal treatment for congenital toxoplasmosis
Introduction
Toxoplasma infection is due to the parasite Toxoplasma gondii and is most often a benign disease. Two populations are at risk of severe disease: (1) immunocompromised individuals, such as HIV-infected patients and (2) fetuses or children with toxoplasmosis transmitted from their mother via the placenta. Congenital toxoplasmosis may result in severe manifestations, such as intrauterine death or stillbirth, retinochoroiditis or hydrocephalus [1], [2]. Prevention and clinical management of congenital toxoplasmosis varies widely among countries. Some countries, such as Austria and France, offer routine testing of pregnant women never infected by toxoplasmosis, and antibiotic treatment when seroconversion occurs during pregnancy. In other countries, screening of children is performed after birth (e.g. Denmark) or there is an explicit policy not to screen (e.g. United Kingdom). Explanations for the variation in prevention and clinical management policies include the lack of reliable information on treatment effectiveness [3], [4], [5], [6] and variation in disease prevalence. Congenital toxoplasmosis is rare on average, at less than one case per 1000 pregnancies, and the prevalence of susceptible women is highly variable among countries. The evidence about the efficacy of prenatal treatment on mother-to-child transmission of the parasite or on clinical manifestations in infected children is based only on observational studies [7], [8]. Moreover, published studies reported controversial results [1], [4], [5], [9], [10], [11], [12], [13], [14]. We hypothesized that most of the controversy in the interpretation of results could be explained by two problems: firstly, the series of events in mother-to-child transmission of toxoplasma infection is complex. Secondly, biases may be numerous in observational studies. In Fig. 1, we describe the clinical management of congenital toxoplasmosis as recommended in France with a monthly screening of pregnant women who have a first negative serology for toxoplasmosis. Women who seroconverted are most often treated by spiramycin (S). If mother to child transmission of the parasite is diagnosed (for example, by PCR analysis of amniotic fluid) then treatment is changed to pyrimethamine and a sulfonamide (PS). After birth, a potential bias is that infected children with clinical signs are more likely to be treated and followed up compared to asymptomatic children. Other methodological pitfalls include selection of cases, lack of standardisation of outcome measures, or confounding. Examples of such biases are shown in Fig. 1 and described latter in this paper. In addition to systematic errors, small sample sizes due to the rarity of the disease yield wide uncertainty around estimations of treatment effect.
This paper provides a review of the major epidemiological issues associated with observational studies on congenital toxoplasmosis, in the absence of randomised trials. We present several biases that might alter the comparability of treated and untreated groups in observational studies and therefore alter inferences regarding the effect of prenatal treatment. We propose different solutions that should be implemented to minimise or at least to quantify these biases.
This review aims to provide a way of interpreting observational data on this topic that may also be helpful for developing experimental or observational studies in this field in the future.
Section snippets
Principle
The choice of a reference group is important for the validity of the estimates of treatment effect as well as for the generalisability of the results. Basically, the unexposed group should be similar to the exposed group with respect to the important predictors of disease incidence [15]. Thus, a historical control group is not relevant because other factors associated with the disease incidence and management have changed with the exposure.
In congenital toxoplasmosis studies, cohorts have often
Principle
Another usual bias when evaluating a treatment with observational data is the indication bias (“confounding by indication”). This bias occurs when a drug is prescribed for reasons associated with the outcome of interest. For example, patients could be more likely treated when they are more likely to develop disease. The treatment effect is then underestimated.
Illustration
In the context of the transmission of T. gondii from mother-to-child, the effect of a regimen like spiramycin followed by
Principle
In a study of mother to child transmission of T. gondii, factors other than prenatal treatment may influence the risk of transmission. Gestational age at maternal seroconversion is one of those factors. Indeed, the risk of transmission is estimated to increase from 6% at 13 weeks of gestation to 72% at 36 weeks [17] (Fig. 2). Moreover, there is an association between treatment and age at seroconversion, as, for example, most of the untreated women seroconverted during the third trimester [5] (
Principle
The time when seroconversion occurs can never be known precisely, only the dates of last negative test and first positive test. Consequently, the observation of seroconversion is interval censored, that is one knows that seroconversion occurred between these two dates. A simple way to estimate the date of seroconversion is to impute the midpoint of the interval between the two tests. However, this simple imputation does not account for the uncertainty about the moment of seroconversion that
Principle
Errors in the measurement of outcome, primarily presence or absence of congenital toxoplasmosis or presence or absence of clinical signs in infected children, offer a potential for misclassification. When there is non-differential misclassification, meaning that infected can be classified as uninfected and visa versa regardless of the treatment status, the treatment effect is likely to be underestimated [15]. However, if detection of the outcome is insensitive (infected are classified
Principle
An attrition bias may occur when some patients are lost to follow-up. Thus, when the duration of follow-up differs according to the type of treatment given, the estimation of treatment effect could be biased.
Illustration
For example, in the study by Desmonts and Couvreur [1], 142 women among the 378 initially included (38%) were lost-to-follow-up. It is not known whether the women that were not included were infected or not, nor whether they were treated. Such a large amount of missing data might have
Conclusion
Although prenatal treatment of congenital toxoplasmosis has been used for almost 30 years in some countries, an increasing number of papers underline the uncertainty of its effectiveness particularly on mother-to-child transmission of the parasite [4], [5], [7], [8], [24]. Because of the methodological issues inherent in the evaluation of prenatal treatment of congenital toxoplasmosis using observational studies, and the variation in analytic methods and reporting of results, a systematic
Acknowledgment
A part of this work has been done in the context of the EUROTOXO project (http://eurotoxo.isped.u-bordeaux2.fr) funded by the EU in the 5th Framework Programme (Grant No. QLG4-CT-2002-30262).
References (31)
- et al.
What is known about the prevention of congenital toxoplasmosis?
Lancet
(1990) - et al.
Treatment of toxoplasmosis during pregnancy: a multicenter study of impact on fetal transmission and children's sequelae at age 1 year
Am J Obstet Gynecol
(1999) - et al.
Mother-to-child transmission of toxoplasmosis: risk estimates for clinical counselling
Lancet
(1999) - et al.
Prenatal diagnosis using polymerase chain reaction on amniotic fluid for congenital toxoplasmosis
Obstet Gynecol
(2001) - et al.
Eye manifestations of congenital toxoplasmosis
Am J Ophthalmol
(1996) - et al.
Feasibility of neonatal screening for toxoplasma infection in the absence of prenatal treatment
Lancet
(1999) - et al.
Congenital toxoplasmosis. A prospective study of 378 pregnancies
N Engl J Med
(1974) - et al.
Early and longitudinal evaluations of treated infants and children and untreated historical patients with congenital toxoplasmosis: the Chicago Collaborative Treatment Trial
Clin Infect Dis
(1994) - European Multicentre Study on Congenital Toxoplasmosis. Effect of timing and type of treatment on the risk of mother to...
- et al.
Prevention and treatment of congenital toxoplasmosis: organization protocol
Minerva Ginecol
(2004)