Dr Jun Zhang, Epidemiology Branch, National Institute of Child Health and Human Development, National Institutes of Health, Building 6100, Room 7B03, Bethesda, MD 20892, USA. E-mail: email@example.com
Immunology has been hypothesised to play a critical role in the development of pre-eclampsia. A number of epidemiological studies have shown that multiparous women who changed partner had an increased risk of pre-eclampsia in the following pregnancy compared with multiparous women with the same partner. However, partner change is often associated with a long birth interval. Two recent papers using data from the same birth registry reported that, after controlling for birth interval, partner change was associated with a reduced risk of pre-eclampsia. Based on a causal diagram, the author argues conceptually that birth interval is not a confounder but more likely to be a collider. Controlling for or stratifying birth interval in the association between partner change and risk of pre-eclampsia could be inappropriate and may have produced a spurious association.
Despite enormous efforts and years of research, the aetiology of pre-eclampsia remains elusive. Immunology has been hypothesised to play a critical role in its development, the most compelling evidence to support this belief being that it is a disease mostly of primiparous women. Its incidence in multiparae is only a quarter to one-fifth of that in primiparae.1 Pre-eclampsia in multiparous women is also less severe, and is often found to be associated with certain underlying conditions, such as chronic hypertension and thrombophilia.2 In the past 10 years, evidence has further suggested that repeated maternal exposure to paternal antigens before conception may enhance a maternal immune adaptation and tolerance to fetal antigens during pregnancy and, therefore, reduce the risk of pre-eclampsia.3 For instance, frequency of vaginal exposure to paternal sperm, measured by contraceptive methods and frequency of intercourse, has been inversely associated with the risk of pre-eclampsia in some studies4,5 but not all.6 Oral sex (intra-oral ejaculation) was reported to reduce the risk of pre-eclampsia,7 while donor insemination and oocyte donation substantially increased the risk, compared with women who used their own oocyte and/or their partner's sperm.3 Furthermore, some studies,8,9 but not all,10 showed that HIV-infected pregnant women had a significantly lower incidence of pre-eclampsia. More interestingly, among patients who received antiretroviral therapy, the incidence of pre-eclampsia rebounded.8 It even became a significant adverse event in these women.11,12
A number of epidemiological studies have shown that multiparous women who changed partner had a higher risk of pre-eclampsia in the following pregnancy than multiparous women with the same partner [relative risk (RR) ranging from 1.2 to 8.6].13 However, partner change is often associated with a longer birth interval. Here, birth interval refers to the time interval between two consecutive deliveries from two separate pregnancies. Two recent papers using the same birth registry data14,15 showed that in women with no history of pre-eclampsia, when birth interval was controlled in the analysis, the increased risk of pre-eclampsia associated with partner change not only disappeared but also reversed in direction, i.e. partner change was associated with a reduced risk of pre-eclampsia. Two studies15,16 further demonstrated that the incidence of pre-eclampsia was consistently lower across birth intervals (in years) among women with a new partner than among women with the same partner. These results directly challenge the notion that partner change increases the risk of pre-eclampsia,3 and the hypothesis of ‘primipaternity’ in particular.14 Some researchers remain unconvinced.17
Given the importance of birth interval in studies of consecutive pregnancies, the question of how to handle this variable in the analyses can be critical. The strategy for adjustment may well vary in different situations. The purpose of this paper is to examine, from an epidemiological point of view, whether the protective effect of partner change is likely to be a factor or an artifact. Given the complex and often unclear aetiology of pre-eclampsia, I will focus on women with no history of pre-eclampsia in previous pregnancies. Using a causal diagram, I will demonstrate conceptually why controlling for birth interval as a confounder in the association between partner change and risk of pre-eclampsia could be inappropriate and how it may have produced a spurious association.
Figure 1 is a simplified causal diagram. In the upper part of this model, smoking is a typical confounder because smokers were 2.5 times more likely to change partner and had a lower risk of pre-eclampsia.16 The common cause for this association might be the woman's attitude towards life and risks. Without controlling for smoking, women who changed partner may appear to have a lower risk of pre-eclampsia due to smoking. Thus, in assessing the relation between partner change and pre-eclampsia, smoking must be controlled. Unfortunately, information on smoking was not available in some of previous studies.14,15
In the lower part of the model, partner change and risk of pre-eclampsia is the question of interest. Recent epidemiological studies suggest that a longer birth interval is associated with up to a twofold increased risk of pre-eclampsia for women with a birth interval greater than 7 years.15,16 The most critical issue is whether partner change tends to increase birth interval or, conversely, couples who decide to have a longer birth interval have a higher accumulative probability of separation. In other words, does partner change prolong birth interval, or long birth interval lead to partner change? Both scenarios probably exist in real life. However, due to the nature of the question and study population selection, the former is likely to be the causal direction.
The question under study is whether partner change affects the risk of pre-eclampsia in the subsequent pregnancy. Temporal sequence dictates that partner change is the ‘exposure’ and subsequent pregnancy (i.e. birth interval) comes after partner change. In all previous studies, only women who had another birth were included.13 Women who did not have another birth (i.e. no information on birth interval), regardless of whether they changed partner or not, were excluded. Therefore, the causal direction is that partner change leads to long birth intervals in the current study.
There is well-established literature showing that some maternal underlying conditions, such as autoimmune disorders, insulin resistance and maternal metabolic syndrome, can cause women to be less fertile (i.e. take longer to conceive or have a higher risk of early pregnancy loss).18–20 These same factors may also increase the risk of pre-eclampsia. They are usually unmeasured and were unavailable in previous studies. These causal relations place birth interval into a category called ‘collider’ (a factor affected by two other factors, E→ C←D) rather than a confounder (a factor that affects two other factors, E←C→D) (see more detailed definitions by Greenland21). Stratifying/controlling for a collider may create a statistical artefact.21,22
Figure 2 illustrates that controlling for smoking as a confounder is appropriate. It blocks the pathway of the confounding effects caused by smoking. Because smoking reduces the risk of pre-eclampsia, controlling for smoking will increase the relative risk of the association between partner change and pre-eclampsia. However, controlling for birth interval may induce an association between partner change and maternal underlying conditions that did not previously exist. Then, maternal underlying conditions become a confounder for the relationship between partner change and pre-eclampsia risk. Therefore, the reported protective effect of partner change on the risk of pre-eclampsia14,15 may be a bias due to inappropriate control for birth interval.
In more practical terms, suppose there are two women, both of whom had two births and these two births were both 5 years apart (i.e. controlling for birth interval). Woman A did not change partner during the 5-year period while Woman B did. If maternal underlying conditions operate on both birth interval and risk of pre-eclampsia as indicated in Fig. 1, Woman A would be more likely to have those conditions that may have caused her to be less fertile than Woman B. When they became pregnant, Woman A would have a higher risk of pre-eclampsia than Woman B. Thus, it would appear that partner change has a protective effect on pre-eclampsia. But this association was biased by the maternal underlying conditions (a confounder). Therefore, this confounder must be controlled simultaneously with birth interval to avoid a potential bias.
Birth interval reflects a mixture of social, behavioural and physiological factors. If the birth interval in women with the same partner is contributed to more by the woman's fecundity, the birth interval in women with a different partner is probably more related to time to find another partner and become pregnant with the new partner. The different nature of ‘birth interval’ in women with and without partner change suggests that birth interval in previous studies may be measuring different underlying factors for different women. Thus, directly controlling for underlying conditions becomes more important to minimise differential residual confounding.
Questions may be raised because Li and Wi23 controlled for birth interval but not maternal underlying conditions in their cohort study and still found that, among women with no history of pre-eclampsia, partner change was associated with an increased risk of pre-eclampsia (adjusted RR = 1.3 [95% CI 1.1, 1.6]). It should be pointed out that the birth interval in their study was less than 3 years. The prevalence of underlying conditions is likely to be very low in women who have two births within 3 years. Basso et al.16 found a similar pattern, in which after they restricted the birth interval to within 3 years, non-smoking women with partner change but no history of pre-eclampsia had an increased risk of pre-eclampsia (adjusted RR = 1.40 [95% CI 0.73, 2.68]). The case–control study by Mostello et al.,24 using the Missouri maternally linked cohort from 1989 to 1997, controlled for birth interval as well as other factors such as previous pre-eclampsia, chronic hypertension, diabetes, renal disease and previous small-for-gestational-age newborn. The increased risk persisted (crude RR = 1.43 [95% CI 1.25, 1.67]; adjusted RR = 1.25 [95% CI 1.0, 1.43]).
The causal model in this study also presents a dilemma. Biological plausibility that a longer birth interval (as a surrogate for maternal underlying conditions) is associated with higher risk of pre-eclampsia would support the controlling for birth interval.16 Adjusting for birth interval will cut the indirect effect of partner change through birth interval to pre-eclampsia and provide a cleaner association between partner change and pre-eclampsia. The adjusted RR is likely to decrease towards null after the adjustment. On the other hand, because birth interval is a potential collider, it should not be controlled for without controlling for other underlying conditions. Here, controlling for birth interval would be likely to draw the RR below 1 (i.e. showing a protective effect). Unfortunately, in most cases information on the underlying conditions is either unavailable or poorly measured. Under such circumstances, one may have to weigh the pros and cons of controlling for birth interval in the analysis. It will be helpful to present results with and without controlling for such a factor and discuss the directions and posited magnitudes of the indirect effect and collider bias.
I acknowledge that my causal model is probably overly simplistic. Other factors, measured and unmeasured, may work through a complex network. For example, the reported protective effect of partner change on the risk of pre-eclampsia14 may be due to three factors: collider stratification, lack of adjustment for smoking, and cutting off the indirect pathway. In addition, a causal diagram may not be particularly helpful when discussing the magnitude of associations, effect modification, or even the direction of accumulated bias. The true effect of a collider bias may turn out to be negligible when additional variables are available and controlled. Finally, it is worth noting that I focused on women who did not have pre-eclampsia in their prior pregnancies, which account for over 95% of multiparous women.14 The main reason is that women who had pre-eclampsia during pregnancy also have a higher level of blood pressure while not pregnant and an elevated risk of chronic hypertension later in life.25 This suggests that some pre-eclampsia cases are due to maternal underlying vascular conditions. They may have a very high risk of recurrence irrespective of partner change.26
Two cohort studies suggest that paternal genes may contribute to maternal pre-eclampsia by passing pre-eclampsia susceptible gene(s) to the fetus.27,28 This observation seems to support the findings by Li and Wi23 and Basso et al.,16 in which women who had had pre-eclampsia in the previous pregnancy had a slightly but not significantly reduced risk of pre-eclampsia when the current pregnancy was with a new partner (OR = 0.7 [95% CI 0.4 , 1.2]). The authors proposed that partner change in these women may have removed the risk from the male partner and, consequently, reduced the risk of pre-eclampsia in the pregnancy with a new partner.23 While this hypothesis still needs to be confirmed, it cannot be applied to the majority of women, who did not have pre-eclampsia in previous pregnancies. Instead, two studies showed that among women who had a birth interval of less than 3 years and had no history of pre-eclampsia, their risk of pre-eclampsia increased with partner change.16,23
In summary, in the examination of the association between partner change and pre-eclampsia, controlling for or stratifying by birth interval may distort the association. This may in part explain why partner change appears to have a protective effect on the risk of pre-eclampsia after controlling for birth interval.