Preterm birth is the leading cause of infant mortality in industrialised societies. Its incidence is greatly increased among the socially disadvantaged, but the reasons for this excess are unclear and have been relatively unexplored. We hypothesise two distinct sets of causal pathways and mechanisms that may explain social disparities in preterm birth. The first set involves chronic and acute psychosocial stressors, psychological distress caused by those stressors, increased secretion of placental corticotropin releasing hormone (CRH), changes in sexual behaviours or enhanced susceptibility to bacterial vaginosis and chorioamnionitis, cigarette smoking or cocaine use, and decidual vasculopathy. The second hypothesised pathway is a gene–environment interaction based on a highly prevalent mutation in the gene for methylenetetrahydrofolate reductase (MTHFR), combined with low folate intake from the diet and from prenatal vitamin supplements, consequent hyperhomocysteinemia, and decidual vasculopathy.
We propose to test these hypothesised pathways and mechanisms in a nested case-control study within a prospectively recruited and followed cohort of pregnant women with singleton pregnancies who deliver at one of four Montreal hospitals that serve an ethnically and socio-economically diverse population. Following recruitment during the late first or early second trimester, participating women are seen at 24–26 weeks, when a research nurse obtains a detailed medical and obstetric history; administers several scales to assess chronic and acute stressors and psychological function; obtains blood samples for CRH, red blood cell and plasma folate, homocysteine, and DNA for the MTHFR mutation; and performs a digital and speculum examination to measure cervical length and vaginal pH and to obtain swabs for bacterial vaginosis and fetal fibronectin. After delivery, each case (delivery at < 37 completed weeks following spontaneous onset of labour or prelabour rupture of membranes) and two controls are selected for placental pathological examination, hair analysis of cotinine, cocaine, and benzoylecgonine, and analysis of stored blood and vaginal specimens. Statistical analysis will be based on multiple logistic regression and structural equation modelling, with sequential construction of models of potential aetiological determinants and covariates to test the hypothesised causal pathways and mechanisms.
The research we propose should improve understanding of the factors and processes that mediate social disparities in preterm birth. This improved understanding should help not only in developing strategies to reduce the disparities but also in suggesting preventive interventions applicable across the entire
Among the most robust findings in perinatal epidemiological research are the large socio-economic disparities in such key pregnancy outcomes as perinatal and infant mortality, low birthweight, intrauterine growth restriction (IUGR), and preterm birth.1–4 The high rates of these adverse outcomes in developing countries provide a striking illustration,5 but even within developed countries, large differences in rates according to socio-economic status (SES) have been consistently reported.
Preterm birth (delivery at < 37 completed weeks) is a key outcome to consider in developed country settings because of its strong relationship to infant mortality, long-term morbidity, and high health-care costs. In a large, population-based study from the United States,2 gradients in risk of preterm birth by income, education, and occupation were observed in both black and white women. Large socio-economic disparities in risk of preterm birth have also been reported in recent population-based studies from the UK,6 the Czech Republic7 and Finland,8 while Sweden has reported smaller disparities according to maternal education and marital status.7
SES is a complex construct generally used to define social inequality, and usually measured by income and/or educational attainment. Education is the dimension of SES that most strongly and consistently predicts health. For example, maternal and paternal education were the strongest predictors of reproductive outcome in a US study.2 A low level of education limits a person’s access to jobs and other social resources, which in turn limits his/her capacity to integrate within society and thereby increases the risk of subsequent poverty.
Having more money or education probably has no direct effect on the duration of gestation. In other words, socio-economic disadvantage is unlikely to be a direct, independent determinant of preterm birth; instead, it probably leads to unhealthy behaviours, exposure to stress, and psychological reactions to stress that shorten gestation. Such exposures are the mediating variables that are the focus of our proposed research. The mediators most frequently mentioned in the literature are maternal anthropometry and nutrition (height, prepregnancy body mass index, gestational weight gain, and micronutrient status); cigarette smoking; genital tract infection and inflammation; cocaine and other drug use; physically demanding work during pregnancy; the quantity and quality of prenatal care; and stress, anxiety, depression, and other psychological factors.1,9 As we argue below, however, the evidence for some of these mediators is weak, while several other biologically plausible causal pathways have not been explored.
Determinants of preterm birth and their relationship to SES
Previous work by us1 and others8,10 has documented our incomplete understanding of the aetiological determinants of preterm delivery. From a public health perspective, and particularly concerning their role as potential mediators of the low SES–preterm relationship, these determinants must be understood in terms of their aetiological fractions, which depend both on their associated relative risks (RR) for preterm birth and on their prevalence (especially among the socially disadvantaged). The fact remains, however, that most cases of preterm birth occur without any known cause.1,10
A number of studies and reviews have focused on the potential role of genital tract colonisation, infection, and inflammation as causes of preterm birth.11–16 The most consistent associations and highest risks for preterm birth have been reported with bacterial vaginosis (BV), with most RRs reported between 1.5 and 2.5 but ranging as high as 6.9.13,14 The evidence is particularly strong for early preterm birth (< 32 weeks), when evidence of upper genital tract infection/inflammation (chorioamnionitis) is frequently present.15,16 Vaginal or cervical fetal fibronectin, which is strongly associated with early preterm delivery, may be a marker for such infection/inflammation.15–18 Randomised trials of antibiotic treatment of pregnant women with BV have yielded conflicting results; despite early promise,19 a recent multicentre National Institutes of Health trial found no evidence of reduction in risk of preterm birth.20 BV is more common among the socially disadvantaged,11,21 and if the BV–preterm birth relationship is indeed causal, it seems reasonable to speculate that genital tract infection could be an important mediating variable in explaining the high risk of preterm birth among the socially disadvantaged.
Multiple (twin and higher-order) births represent only about 2% of all births, but half of twins and almost all higher-order multiples are born before 37 weeks.22 Indirect evidence, however, suggests that the recent increase in multiple births attributable to infertility treatment has occurred to a greater extent among high-SES groups.23,24 Such a tendency would therefore tend to narrow socio-economic disparities rather than widen them.
Pregnancy-induced hypertension (PIH) and incompetent cervix are both well-recognised determinants and together account for a sizeable minority of preterm births.25 PIH is more frequent among the poor26 (although this association has not been adequately explained), whereas incompetent cervix is not known to vary by SES.
Cigarette smoking is commonly acknowledged to increase the risk of preterm birth, but the RR for preterm delivery associated with cigarette smoking is far lower than the corresponding RR for IUGR.1 Thus, even though cigarette smoking is more frequent and heavier among low SES women,27–29 it can account for only a small fraction of the much higher risk of preterm delivery among the socially disadvantaged.25 Cocaine use during pregnancy is a well-recognised determinant of preterm birth,10,25,30 and is strongly associated with low SES.31,32 But since it carries only a small aetiological fraction for preterm birth (owing to its low prevalence),1,25 cocaine use is unlikely to be a quantitatively important mediator of the low SES–preterm birth relationship. In poor, inner-city areas in the US, however, the prevalence of cocaine use during pregnancy may be extremely high,33 and in these settings, its mediating role may be substantial. Marijuana34,35 and narcotic36 use and high alcohol37–39 and caffeine (> 300 mg/day)40,41 consumption are also more common among the socially disadvantaged, but these factors have not been shown to have an impact on gestational duration.1,10,25,30,41–43
Although there is some controversy about whether women who are thin, short, or have low weight gains during pregnancy are at increased risk for preterm delivery, the available evidence strongly suggests that the effects of maternal anthropometric or nutritional factors, if they exist at all, are much smaller than their effects on IUGR.1,25,44,45 Moreover, in most developed countries, body mass index in women is inversely associated with SES (i.e. higher among the socially disadvantaged),46 so low prepregnancy weight-for-height could not explain the higher rates of preterm birth among low SES women.25
Although two recent reports suggest that low folic acid intake increases the risk for preterm delivery,47,48 there is little evidence that supplementation with specific micronutrients improves fetal growth or lowers the risk of IUGR or preterm birth. Based on a systematic review in the Cochrane Database of Systematic Reviews, reduced risks have been found with magnesium supplementation for both preterm birth (RR = 0.75 [0.58, 0.96]) and IUGR (RR = 0.70 [0.53, 0.90]), but the quality of the trials included in the review is poor.49 No significant effect on reducing preterm birth or IUGR has been found in a systematic review of trials of supplementation with iron50 or folate,51 even in combination.52 In the case of zinc, the most recent Cochrane review53 reports a significant effect of supplementation in reducing preterm birth (although not IUGR [RR = 0.90, 95% CI = 0.64,1.28]), but that encouraging result has not been confirmed in two recent trials from Peru54 and Bangladesh55 that are not included in the review. No effect of calcium supplementation on reducing preterm birth is evident (RR = 0.98 [95% CI = 0.85,1.14]) in the systematic review by Atallah et al.56
Several epidemiological studies57–59 and a recent meta-analysis60 have reported that prolonged standing or strenuous activity at work is associated with an increased risk of preterm birth. However, only a small fraction of employed women are exposed to these conditions at work,59 and the aetiological fraction associated with such work is therefore unlikely to be high. Physically demanding domestic work may make a larger, albeit unrecognised, contribution.61,62 For example, Misra et al.63 report increased risks of preterm birth associated with walking upstairs 10 or more times per day and with walking to do the grocery and other routine shopping.
The quantity and quality of prenatal care has been a popular subject for investigation by epidemiologists and clinical researchers for the last two decades. Observational studies have reported strong associations between late entry into prenatal care, or the gestational age-adjusted number of prenatal care visits, and the occurrence of preterm birth.63,64 Comparisons of participants and non-participants in special preterm birth prevention programmes have also yielded impressive results.65,66 By contrast, virtually all randomised trials of prenatal care, including those of ‘intensive’ prenatal care for low SES women at high risk for preterm delivery, have produced disappointingly negative results.67–69 Why this discrepancy between the encouraging results of observational studies and the disappointing findings of randomised trials? Late entry or infrequent use of prenatal care may be a marker for an unwanted pregnancy. Women who initiate prenatal care early and visit their obstetrician, family physician, or midwife on a regular basis may also differ from women who do not, in other fundamental ways that affect their risk for preterm birth – ways that are extremely difficult to measure (and therefore to control for) and that extend beyond conventional sociodemographic and psychological variables.
There seems little doubt that inadequate prenatal care is associated with socio-economic disadvantage.63,64,70 This appears to be true even in Canada, despite universal health care insurance.70 However, given the serious doubt about the effects of prenatal care on reducing the risk of preterm birth, it too seems an unlikely mediator of socio-economic disparities in this outcome.
The belief that stress and maternal psychological factors can affect the course and outcome of pregnancy can be traced to Biblical times,71 but it has been only recently that social scientists and epidemiologists have considered this area worthy of serious investigation. Associations have been reported between preterm birth and stressful life events, anxiety, depression, stressful work, physical abuse, and low level of social support.72–77 Recent reports suggest that the effects of psychosocial stressors may be mediated by cortisol-induced positive feedback-mediated increases in placental secretion of CRH,78–80 which can stimulate uterine contractility by potentiating the effects of oxytocin (perhaps via prostaglandins) or by fetal cortisol- or androgen-induced oestriol synthesis.80,81 However, as summarised by Hoffman and Hatch,82 several recent studies have not found associations between preterm birth and psychosocial factors (especially with stressful life events). Moreover, randomised trials of providing social support have yielded uniformly negative results,83 although the type of intimate support required to reduce perceptions and reactions to stress may not be possible to provide through experimental manipulation.82 In other words, effective support might require a pre-existing, ongoing relationship that cannot be ‘allocated’ in the context of a clinical trial.
The inconsistent results may be attributable to the many ways of operationalising stress. Most studies have relied on checklists of stressful life events, the number of which has not been robustly associated with preterm birth. More consistent associations have been reported between perceived stress and preterm birth, thus underlining the key role of stress appraisal.76,77 In fact, it has been suggested that ‘daily hassles’ and other chronic stressors might be better predictors of health problems than more dramatic, but less frequent, life events.82,84 These chronic stressors include environmental factors like poor and crowded housing conditions, unemployment and other severe financial problems, as well as interpersonal factors such as living without a partner and experiencing domestic violence. According to some recent evidence, it is cumulative, ‘lifecourse’ (i.e. long-term) exposure to adverse socio-economic environments that may have the largest impact on health.85,86 The effect of chronic stressors and perceived stress may be mediated by stress hormones such as CRH.78–80
Low SES pregnant women experience more stressful life events during their pregnancy.6,87,88 As we have shown, chronic stressors are embedded within and accrue from the environment of low SES women.89 Besides the chronic stress caused by financial insecurity, these stressors include poor and crowded housing conditions, living without a partner, unsatisfying marital relationships, domestic violence, and stressful working conditions.6,90,91
Although some low SES women may be emotionally ‘numbed’ (e.g. less reactive to stressful events) by their chronic difficulties,87 chronic stressors associated with low SES may lead to adverse intrapsychic processes that have an impact on perceived stress. Erosion of personal resilience is one of these potential processes. Personal resilience refers to a person’s level of self-esteem, optimism, and ‘mastery’ beliefs (i.e. beliefs about personal control).92 This cluster of personality traits has been associated with preterm birth and appears to be mediated by perceived stress.93 More generally, this cluster has been found to at least partially mediate the impact of SES on health.94,95 Living in a chronically stressful environment appears to erode personal resilience, which in turn may heighten perceived stress, anxiety, a sense of helplessness, a lack of optimism, and depression, thereby increasing the risk of preterm birth via the release of CRH, changes in sexual practices that lead to genital tract infection/inflammation, or greater use of cigarettes, cocaine, or other drugs of abuse.96 Although some evidence is available to support each of these links, no previous study has simultaneously examined each of these pathways.
Personal resilience captures an important cognitive-affective intrapsychic construct, but it is also important to consider a motivational construct implicated during pregnancy. Commitment represents the motivation to be pregnant, have a baby, and engage in behaviours that facilitate the achievement of that goal. It encompasses and incorporates both the intendedness and wantedness of a pregnancy.97 Unintended, unwanted pregnancy is far more common among low SES women98 and has been associated with physical violence.99,100 We101 and others102 have shown that low pregnancy commitment is also associated with cigarette smoking and use of alcohol and street drugs. Generally, commitment appears to be associated with health behaviours during pregnancy.103 Although hitherto unexplored, commitment seems likely to affect dietary habits and reduce the use of vitamin supplements during pregnancy, which may also lead to preterm birth.
Whereas personal resilience and pregnancy commitment are postulated to be mediators, social support may be a moderating (buffering) factor.104,105 Social support includes the size of one’s social network, the degree of social integration, and the availability and receipt of support when needed.
In reviewing the associations between social support and health, the most consistent results have been obtained from measures of network size and availability,106,107 perhaps because those who are distressed are more likely to elicit and receive social support.108 Because of these associations, efforts have been directed at understanding why the belief that ‘social support will be available should I need it’ appears so effective in buffering stress responses.109,110 Similar associations have been found with measures of adult attachment.111,112 Modelled after studies on mother–infant attachment, adult attachment assesses a person’s tendency to be anxious about relationships in general and to avoid closeness.113 Because of previous interactions with significant others, a person develops a generalised set of expectancies about others’ responsiveness to them that may influence his or her stress response. The reassurance that others will be there if needed helps the person to appraise the event as less stressful and to engage in more adaptive ways of coping.109,110
Especially significant is the presence and quality of an intimate relationship (i.e. relationship status and relationship quality). Previous research has found that these resources can act as a stress buffer, reducing the effects of chronic and severe stressors on intrapsychic factors mediating adverse health outcomes.104,105 We114 and others77 have shown that low SES women have a more restricted social support network, with social support less available during their pregnancy. Thus, low SES pregnant women are not only more exposed to both acute (life events) and chronic (difficult life conditions) stressors, but also have less social support to limit the impact of those stressors.
Several methodological problems limit the inferences that can be drawn from previous studies linking maternal psychosocial factors to preterm birth. First, as mentioned earlier and as emphasised by Hoffman and Hatch,82 chronic stressors have been relatively ignored in favour of stressful life events, anxiety, and depression. Second, several early studies ascertained exposure to psychosocial factors retrospectively (i.e. after delivery), with considerable potential for recall bias. In other words, women who delivered prematurely may have been more likely to report the occurrence of prior anxiety or stressful events. Third, many studies have assessed psychosocial factors individually or only a few at a time, thus ignoring the cumulative effect of the multiple adverse factors to which disadvantaged women are often exposed. Fourth, because of the large number of psychological factors that might be implicated and the even larger number of potential confounders, sample sizes have often been insufficient to detect moderate psychosocial effects. Fifth, the majority of published studies have been based on individual hospitals or small communities, raising questions about sample selection and thus the generalisability of the findings. Sixth, to our knowledge, none of the studies on maternal psychological factors have controlled for the effects of genital tract infection/inflammation.
In our view, however, the most important limitation of previous studies has been the absence of clear hypotheses linking the psychosocial factors in a clear causal hierarchy (pathway). Nor have the psychosocial factors been examined specifically with respect to their role in mediating the SES–preterm relationship or to the potential biological mechanisms (other than CRH) by which they might lead to preterm birth. What is needed therefore is not more studies that examine simple bivariate associations between individual social or psychological measures and preterm birth, but rather an integrated approach that tests hypotheses about specific causal pathways and biological mechanisms. We believe that our study fills this need by building on the previous experience of several members of our research team in studying psychological aspects of pregnancy and the chronic stressors experienced by poor pregnant women. This experience enables us to focus on the specific psychosocial factors that seem most likely to mediate the low SES–preterm birth relationship.
Apart from studies of risk of recurrence in individual women1,10,115–117 and intergenerational (mother-to–daughter) associations of risk,118,119 genetic aetiologies for preterm birth remain relatively unexplored.120 However, as genetic factors are not expected to vary by socio-economic status, their ability to explain social disparities in preterm birth must involve an interaction with one or more environmental factors. In addition, if genetic factors have been rarely studied as aetiological determinants of preterm delivery, then gene–environment interactions have been totally ignored.
Hyperhomocysteinaemia (HHC), which occurs in association with the rare condition hereditary homocystinuria, but more commonly results from a combination of vitamin B12 or folate deficiency and a cytosine-to-thymine mutation of nucleotide 677 (C677T) in the gene coding for MTHFR, has been associated with an increased risk of thromboembolism121,122 and of coronary heart disease.123,124 We have demonstrated that homozygotes for the MTHFR mutation are found in 10–12% of the population125,126 and that the mutant genotype results in hyperhomocysteinaemia when folate levels are low.126 We127 and others128 have also shown that an HHC-inducing interaction between MTHFR deficiency and low folate intake accounts for a substantial fraction of neural tube defects. Even more pertinent to this proposal are recent reports linking HHC to pre-eclampsia, abruptio placentae, antepartum stillbirth, and recurrent spontaneous abortion.129–134 To our knowledge, only one previous study has examined the relationship between thrombophilic mutations and preterm birth; significant associations were reported with both the Factor V Leiden and 20210A prothrombin mutations in the infant (not the mother), but neither MTFHR nor HHC was examined.135 A recent study reported that homocysteine levels during pregnancy, although slightly lower owing to plasma volume expansion, are responsive to folic acid supplementation.136
The potential importance of a vasculopathic pathway to preterm birth has recently been emphasised.137 Few studies of preterm birth have incorporated pathological examinations of the placenta, but two independent groups of investigators have reported that vascular lesions such as thrombosis, placental infarcts, and chronic vasculitis are present in one-third of women who deliver preterm compared with one-sixth of women who deliver at term;138,139 in one of these studies, these abnormalities were found in over two-thirds of deliveries between 22 and 28 weeks gestation.138 Thus, a gene–environment interaction leading to HHC and consequent placental vascular lesions that increase the risk for preterm birth is biologically plausible and merits further study. Of even greater importance for our research, low dietary folate intake is more prevalent among low SES women.140–142 Since maternal vitamin supplements are moderately expensive and are not covered by either Welfare or private insurance in Quebec, it is reasonable to assume (although undocumented) that low SES pregnant women in Quebec are less likely to use these supplements. Given recently reported associations between low folate intake and preterm birth,47,48 SES-based differences in folate intake might explain at least part of the low SES–preterm birth relation.
In summary, based on the available evidence, the strong association between low SES and preterm birth seems most likely to be mediated by chronic stressors, psychological distress, changes in health behaviours, and genital tract infection/inflammation. Although more indirect, several lines of evidence also indicate the biological plausibility of a different causal pathway: a link between a common genetic mutation and a low SES-induced reduction in folate intake.
Figure 1 shows the causal pathways that potentially mediate the strong association between low SES and preterm birth. The arrows in the figure link low SES, shown on the left, through the successive hypothesised mediating factors, to the study outcome (preterm birth) on the right. Arrows impinging on other arrows denote effect modification, i.e. a factor that modifies (increases or decreases) the effect of another factor.
We hypothesise two distinct sets of causal pathways comprising factors that mediate the increased risk of preterm birth among low SES women; arrows within these two pathways are labelled A and B and numbered consecutively in the figure. The numbers thus connote the sequence, but not the relative importance, of the hypothesised pathways. The first set of pathways (A) involves chronic and acute stressors (A1) and includes living without a partner, inadequate housing, insufficient funds for essential needs, marital maladjustment, domestic conflict and violence, stressful working conditions, and stressful life events. These chronic and acute stressors may stimulate placental CRH release directly (pathway A2–A10) or may operate via adverse psychological factors (depleted personal resilience, heightened perception of stress, or low commitment to the pregnancy) (A3), and either changes in sexual practices (A9–A12) or enhanced susceptibility (A11–A14) to bacterial vaginosis and chorioamnionitis (A15–A16); or increased cigarette smoking or cocaine use (A7 or A9) and decidual vasculopathy (A13–A17). Several of the pathways in set A may be abetted by low social support (A5 or A6).
The second set of pathways (B) reflects the combined effects of a genetic defect and low folate intake. We hypothesise that a common mutation (C677T) in the gene for MTHFR interacts (B3) with low folate intake (B1) to produce hyperhomocysteinae-mia (B2), which leads to decidual vasculopathy (B4) and consequent preterm delivery (B5).
Our study design combines features of a prospective cohort study and a nested case-control design. It also combines traditional interview methods with extensive biological measurements to explore the causal pathways and mechanisms that we hypothesise. The nested case-control design improves efficiency (costly laboratory analyses of biological markers are limited to all cases and two controls per case, rather than all non-cases), while providing statistical power nearly equivalent to analysis of the entire cohort. Only the psychosocial measures and other interview-derived variables (which require prospective data collection to avoid recall bias) are obtained for the entire cohort.
It is not feasible to recruit a population-based sample of pregnant women prior to delivery (no registry of pregnant women, or even those who begin prenatal care, is available). Research assistants therefore recruit women primarily when they present for routine ultrasound examination (16–20 weeks) at four hospitals affiliated with McGill University and l’Université de Montréal: the Royal Victoria Hospital, Jewish General Hospital, Centre Hospitalier de l’Université de Montréal, and Hôpital Maisonneuve-Rosemont. At the Royal Victoria Hospital, subjects are also occasionally recruited when they present for prenatal blood drawing (usually at 8–12 weeks), and at Hôpital Maisonneuve-Rosemont, subjects are additionally recruited in the hospital’s prenatal clinic (until 24 weeks). The study hospitals serve a wide socio-economic spectrum, including a large number of poor women and immigrant women, with good representation from both French- and English-speaking populations. The Canadian health-care system provides all women with free access to prenatal care irrespective of their socio-economic status. Notification of the study has been sent to all obstetricians and family physicians performing deliveries at the four study hospitals.
Women aged 18 years who are pregnant with a singleton fetus and who speak and understand either French or English are considered eligible for enrolment, provided they do not have a severe chronic illness (other than hypertension, asthma, or diabetes) requiring ongoing treatment, placenta praevia, a history of incompetent cervix diagnosed in a previous pregnancy, or a fetus affected by a major fetal anomaly already diagnosed in the current pregnancy. These conditions increase the risk of preterm birth but are not strongly associated with SES. Including them might therefore dilute the effects of the causal pathways under study.
Women who consent to participate in the study are requested to return to a special research clinic visit at 24–26 weeks of gestation, based on the date of the last normal menstrual period if confirmed within ± 7 days by the early ultrasound estimate (otherwise based on the early ultrasound estimate). This gestational age was chosen because it is early enough to precede the period of greatest risk for preterm birth but late enough to enhance the predictive value of the psychosocial measures and of several potentially important biological markers: bacterial vaginosis, fetal fibronectin, and CRH. The research assistant calls each participant 7–10 days before her scheduled visit to remind her of the appointment and to verify that neither placenta praevia nor a major fetal anomaly has been diagnosed since recruitment. The research nurse calls the participant the day before her appointment to introduce herself and to remind her of the time and exact location of the clinic visit.
At the clinic visit, which lasts approximately one and a half hours, a research nurse performs an interview that includes standard sociodemographic information, including birthplace of mother (and, as a measure of acculturation in immigrants, duration of residence in Canada); measures of socio-economic status, including education and occupation of the mother, her parents and partner, and family income; medical and obstetric history; cigarette, alcohol, and drug use prior to and since the beginning of the pregnancy; prepregnancy weight; current height and weight; use of prenatal care (gestational age at first visit and number of visits); sexual practices (number of previous-year sexual partners and frequency of sexual intercourse); and use of contraceptives and vaginal douching/hygiene practices (these last questions taken from the National Center for Health Statistics’ National Survey of Family Growth Cycle IV). The research nurse also administers the psychosocial measures discussed below.
Chronic and acute stressors
As recommended by Hoffman and Hatch,82 we focus on chronic stressful conditions, rather than (acute) life events. Crowding is assessed using the density ratio (number of persons per room). A subscale from the Daily Hassles84 is used to measure how often, and to what degree, the woman has lacked money for basic needs such as food, heating and electricity, since the beginning of pregnancy. This subscale has been translated into French and had adequate internal consistency (Cronbach’s α = 0.79) when administered to a sample of 144 pregnant women from the Montreal area.143
The Marital Strain Scale of Pearlin and Schooler144 is used to assess chronic stress with the romantic partner. The nine items represent the three items with the highest factor loadings for each of the three factors contributing to the measure of marital strain. The Abuse Assessment Screen is used to assess conjugal violence. This five-item instrument assesses the frequency, severity, perpetrator, and body sites of injury.145,146 In addition, the item on negative interaction from the Arizona Social Support Interview Schedule (ASSIS)147 (see below under ‘Social support’) is used to assess interpersonal conflicts.
Job-related stress and social support at work are assessed using an abbreviated version148 of the instrument developed by Karasek et al.149 This version contains 13 questions: four on job demand, five on job control, and four on social support at work. The abbreviated scales maintain internal consistencies of 0.60–0.88 (Cronbach’s α) vs. 0.61–0.81 for the entire instrument.149,150 Cronbach’s α of 0.69–0.84 have been reported for the French version of the entire instrument.151
Each woman is asked if she has received help from her partner or any other person to accomplish domestic tasks during her pregnancy and if there is any conflict over task sharing. The number of preschool children under her care is also ascertained as a crude measure of strain from domestic work.61 If she is employed outside the home, she is asked how many hours she works per week, how many hours per work day she spends standing and walking, and whether she carries heavy loads (> 20 lbs) regularly during working hours.57–59 She is also asked how many stairs she climbs per day, on average, at work, home, or elsewhere.
Acute stressors are assessed using the Prenatal Life Events Scale (PLES),152 which was recently derived from more general life-events scales that have been validated in French. The PLES also assesses the respondent’s ratings of the severity of events, which are included among the measures of perceived stress (see below).
Perceived stress, personal resilience, and commitment to the pregnancy
Because previous research linking stress with pregnancy outcomes has found that subjective appraisal of stressors is a key predictor of pregnancy outcomes,76,77 stress appraisal is assessed using multiple measures. First, severity ratings of life events are obtained using the PLES.152 Perceived stress is assessed using the short form of the Perceived Stress Scale.153 The short form has an internal consistency of 0.72 (Cronbach’s α) and test-retest reliability of 0.55 over 2 months. A four-item measure of pregnancy-related anxiety is also administered; this measure has been associated with preterm birth, after controlling for age, education, income, marital status, parity, and medical risk.154
Assessment of personal resilience includes measures of self-esteem, optimism/pessimism, and mastery beliefs of control. Self-esteem is measured with the Rosenberg Self-Esteem Scale; it is the current psychometric standard for adults and has excellent reported internal consistency (0.88) and test-retest reliability (0.82).155 A French translation has been validated and published.156 Optimism is assessed using a short form of the Life Orientation Test.157 The full scale has good internal consistency (Cronbach’s α = 0.76) and good test-retest reliability (0.79). The short form we are using includes all eight items from the original scale but deletes the four ‘filler’ items. Event-specific optimism is assessed by asking the woman about her expectations of birth complications, as well as five questions on her expectations about life once the baby is born (financial stress, sharing of household tasks with her partner, quality of the relationship with her partner, quality of the relationship with her new baby, and satisfaction with her social support).158 Mastery beliefs of control are assessed using four items from the Parental Beliefs Scale.159 The full scale has been used to predict health behaviours during pregnancy. More recently, the internality subscale has been reported to be significantly associated with preterm birth. Our items are from the internality subscale.
Depressed affect is assessed with the Center for Epidemiologic Studies, Depression Scale (CES-D).160 The CES-D is a 20-item instrument that has excellent internal consistency (0.90) and reasonable test-retest reliability (0.54) for a scale that is designed to be sensitive to adverse changes in the respondent’s environment. A French translation has been validated and published.161
Commitment to the pregnancy is measured using an 8-item scale, which we have found to have good internal consistency (0.91) and test-retest reliability (0.94).101 We also administer the adapted version of the Miller Intendedness Scale.97 This scale uses a decision-tree format in order to place participants on a 5-point scale: (0) using birth control all the time; (1) using birth control on and off; (2) not using birth control but neither intending to get pregnant nor intending to have a child if pregnant; (3) not using birth control and not intending to get pregnant but intending to have a child if pregnant; and (4) not using birth control and intending to get pregnant. Finally, we ask the participant about her desire to be pregnant, taken from the scale of Adams et al.162
The social support interview ASSIS147 is specific for the perinatal period and has been translated and validated in French.163 It measures the number of persons in the respondent’s social network who can be relied upon for five functions of support: instrumental, emotional, informative, normative, and companionship. In the English form, Cronbach’s α varies from 0.70 to 0.78 for these five functions,147 while in the French form these coefficients vary from 0.81 to 0.86.163 In order to assess support that is truly available to the respondent, questions about the availability of help when needed during the two preceding weeks are asked for each function of support. We have also included two questions (taken from the Santé Québec survey questionnaire on cardiovascular health) concerning the number of persons on whom the participant can count if she is in difficulty or wishes to confide.
Attachment is assessed with items from Brennan et al’s comprehensive review and distillation of the most commonly used measures of adult attachment.164 Their analysis of responses from 1076 respondents generated two subscales representing two theoretical dimensions of adult attachment: avoidance of closeness and anxiety about relationships. Earlier versions of attachment measures included measures of both fear of closeness and fear of dependency. Because both of these are closely related to avoidance of closeness, four items loading on the fear-of-closeness subscale were chosen (two fear-of-closeness and two fear-of-dependency items). Because anxiety has stood alone as a measure even in earlier versions, only three items with the highest factor loadings for this subscale were retained.
Study women are asked whether or not they are living with a partner. Those who reply in the affirmative are asked to complete two measures that assess the nature of the relationship with the partner, in addition to the two measures mentioned earlier bearing on the chronic stress associated with this relationship. Relationship satisfaction is assessed with two key items from the Quality of Marriage Index (QMI).165 The QMI has been found comparable to the Dyadic Adjustment Scale (DAS).166,167 Moreover, one of the items in the QMI (and retained for our study) is ‘Magical Question’ 31 from the DAS, a question that correlates closely with the other 31 items of the 32-item DAS at 0.90.168 The other item retained had the highest factor loading on the QMI. Wording of instructions and items were modified slightly so as to apply to women in cohabiting, rather than married, relationships, as Quebec has the highest rate of such relationships in North America.169
Following the interview and administration of the psychosocial measures, the research nurse performs a vaginal examination using a warmed and moistened (with water only) disposable speculum to measure the vaginal pH and to obtain swabs for analysis of fetal fibronectin and bacterial vaginosis. She first examines the cervix to ensure that it is not effaced or dilated > 2 cm, and to assess the presence and extent of erosions and discharge. The first swab (for fibronectin) is rolled in the posterior fornix, then placed in a special transport medium supplied by the manufacturer (Adeza, Sunnyvale, CA), and stored in the hospital biochemistry laboratory’s freezer at −20 °C along with the frozen blood specimens (see below). A second and third swab are rolled in the posterior fornix, and each is then smeared on to two glass slides. The slides are air-dried and are then fixed by dipping them in methanol for 10 s. All slides and swabs are then kept in the hospital biochemistry laboratory’s freezer at −20 °C and transported on dry ice with the fibronectin swab every week to Dr Libman’s laboratory, where they are stored at −70 °C for later analysis (see below). After removing the speculum, the research nurse estimates the cervical length using the index finger.
Next, the research nurse draws blood into two 7-mL EDTA-containing lavender-topped tubes (Becton-Dickinson) by venipuncture. The tubes are placed on ice and brought immediately to that hospital’s clinical biochemistry laboratory, where the plasma is separated by centrifugation (3500 r.p.m., 4 °C, 20 min). Five aliquots of 1.0 mL plasma are stored in colour-coded 2-mL cryovials. One blue-capped cryovial is kept for analysis of plasma homocysteine, folate, and vitamin B12; a green-capped tube is used for CRH; and three yellow-capped tubes are kept for back-up. The combined buffy coats from the two lavender-topped are placed in a sixth, red-capped cryovial. All six vials are labelled by the research nurse with the participant’s study identification number and placed in a Ziploc bag labelled with ‘Montreal Prematurity Study’ and stored in the laboratory’s −20 °C freezer. Every week, the frozen samples are transported on dry ice to a −80 °C freezer in Dr Genest’s laboratory for later analysis (see below).
Following the blood drawing, the nurse reminds the participant about the case and control selection and postpartum interview and hair sample procedure (see below). Finally, on completion of the clinic visit, each participant is given $50 CDN to reimburse her for transportation and child care expenses.
Identification of cases and controls
The case room (delivery ward) of each of the four study hospitals is monitored on a daily basis (including weekends) for deliveries of study subjects. The monitoring combines three approaches. First, our research nurse places a special green sticker on the prenatal chart of each study participant so that the delivery ward nurses know she is a participant. Second, we ask each participant to inform the delivery ward nurses that she is a study participant and to ask them to contact our study staff. Third, our study staff visits each case room daily on weekdays and telephones twice daily on weekends.
Each woman who delivers following spontaneous onset of labour or prelabour rupture of membranes before 37 completed weeks (based on the early ultrasound examination) is selected as a case for the nested case-control study. The two study women who deliver at the same hospital at 37 weeks immediately after each case subject are selected as controls.
For all cases and controls, data on stressful life events occurring since the study interview at 24–26 weeks and on interim pregnancy history are collected by interview and from the mother’s and infant’s hospital chart by a research assistant. This history includes information on substance use, sexual activity, employment, domestic violence, total gestational weight gain, the occurrence of pregnancy complications (such as pregnancy-induced hypertension, pre-eclampsia, gestational diabetes, urinary tract infection, antepartum haemorrhage, abruptio placentae, and prelabour rupture of membranes), treatment (if any) for preterm labour or preterm prelabour rupture of membranes, and the infant’s birthweight and gestational age.
For all cases and controls, 25–30 strands of maternal hair are cut near the scalp from the posterior vertex, using disposable gloves and scissors disinfected with alcohol, sealed in a labelled envelope, and sent to Dr Koren’s laboratory in Toronto for later analysis (see below). Placentas from all cases and controls are placed in a double plastic bag, refrigerated immediately following delivery, and marked with a special label by our research assistant. Each hospital’s pathology laboratory performs a gross examination and prepares paraffin blocks and slides for microscopic examination (see below).
All analyses are performed blindly, i.e. without identifying the source of each specimen as case vs. control.
Placentas of all cases and controls are weighed before and after trimming of the fetal membranes, umbilical cord, and blood clots. Gross examination includes assessment of overall form (spherical, bilobed, or circumvallate), appearance of membranes (clean, cloudy, or meconium-stained), retroplacental haematomas (none, < 10%, 10–30%, or > 30% of the maternal surface), pale or red infarction and thrombus (none, < 10%, 10–30%, or > 30% of the cut surface on multiple 1-cm slices), and inspection of the umbilical cord (central, eccentric, marginal, or velamentous). A minimum of six samples are cut from the fresh placenta: one from the membrane (membrane roll); a second and third from the umbilical cord 1 cm from the placental and distal ends, respectively; and three transmural sections of 3 mm thickness (one each near the insertion of the umbilical cord, near a placental margin, and midway in between). Extra samples are taken from any abnormal or suspicious sites. All samples are fixed in 10% buffered formalin overnight, then embedded in paraffin, and sent to Dr Chen’s laboratory every 1–2 weeks. There the blocks are cut in 3 μm thicknesses. Slides are then stained with haematoxylin and eosin for microscopic examination.
Microscopic examination is performed according to a modification of Benirschke’s method.170 Chorioamnionitis is graded in separate examinations of the extra-placental membrane roll and placental membranes according to the presence of definite inflammatory (leukocytic) infiltration: none, grade 1 (subchorionic space only), grade 2 (chorionic and subamniotic, with preservation of amniotic epithelium), or grade 3 (diffuse, with destruction of amniotic epithelium). Microscopic meconium staining is assessed as present or absent in the subamniotic macrophages. Placental villous structures are examined in four high-power fields (HPFs) for each of three regions (centre near cord insertion, near a margin, and half-way in between). Four histological features [(a) > 30% tertiary villi, (b) > 30% syncytiotrophoblastic knots, (c) hypercapillarisation (> 10 capillaries per tertiary villus), (d) nucleated red blood cells, (e) maternal decidual vasculopathy, and (f) fetal vasculopathy] are graded as present on 0, 1–5, or 6–12 HPFs. The same three regions are assessed for microscopic evidence of infarction and villitis (none, < 10%, 10–30%, or > 30% of the total cross-section examined). The umbilical cord is measured for length and diameter, and sections at both ends (one near the placenta, the other near the fetus) are assessed microscopically for the presence or absence of funisitis and (in each of the three vessels) vasculitis. A 10% random sample of all microscopic placental examinations will be (blindly) re-examined by Dr Chen for assessment of intra-rater reliability and sent to another experienced placental pathologist at McGill (Dr Indrojit Roy, Department of Pathology, St. Mary’s Hospital) for assessment of inter-rater reliability. Recent studies suggest that inter-rater reliability is high for chorioamnionitis but only modest for decidual vasculopathy.171,172
Vaginal slides for bacterial vaginosis of all cases and controls are analysed in Dr Libman’s laboratory. Two slides are used to prepare a standard Gram stain (Hucker modification). The third and fourth slides are stained using the Kopeloff modification of the Gram stain,173 which replaces the safranine counterstain with basic fuschin. This stain facilitates the distinction between Gardnerella and Bacteroides morphotypes. The stained slides are stored for later analysis (see below). Primary analysis uses the following variables: presence of bacterial vaginosis by the Nugent criteria (present, equivocal, or absent),174 bacterial vaginosis ‘score’ as a continuous variable using the Nugent criteria (0–10),174 and presence of BV by the Spiegel criteria (present or absent).175 In addition, the presence of ‘clue’ cells on Gram stain is noted and classified as absent, constituting less than 20% of epithelial cells, or constituting more than 20% of epithelial cells; polymorphonuclear cells are noted and classified as absent, 1–10 per high-power field, or > 10 per high-power field. The presence of yeasts, pseudohyphae, and trichomonads is noted. All slides are read by a single experienced laboratory technician. In order to assess inter- and intra-observer variability, 10% of slides will be randomly selected for blinded review both by a second technician and repeat reading by the original technician.176
Fetal fibronectin analyses are performed with a specialised immuno-serology technician using the enzyme immunoassay (EIA) kits supplied by the manufacturer (Adeza, Sunnyvale, CA) and read on a standard EIA microtiter plate reader (Sanofi Diagnostics Pasteur, Montreal, Canada).
Plasma folate, vitamin B12, and homocysteine are determined by IMx (Abbott Diagnostics, Abbott Park, IL). IMx measures homocysteine by fluorescence polarisation immunoassay, folate by ion capture technology, and B12 by microparticle enzyme immunoassay. CRH is determined using a radioimmunoassay in Dr John Challis’ laboratory at the University of Toronto.
Genomic DNA is isolated from the buffy coat using Quiagen columns and sent to Dr Rozen’s laboratory at the Montreal Children’s Hospital for analysis. We expect 50–200 mg DNA from the two 7-mL tubes. DNA is amplified by the polymerase chain reaction (PCR) using published primer sequences for the C677T MTHFR mutation.125 The PCR fragment for the MTHFR gene is 198 bp. Digestion of this fragment with HinfI, followed by polyacrylamide gel electrophoresis, yields the undigested 198 bp fragment for the normal allele and two fragments (175 and 23 bp) for the mutant allele.
Analysis of maternal hair for cotinine, cocaine, and benzoylecgonine is undertaken in Dr Koren’s laboratory at the Toronto Hospital for Sick Children using well-established procedures,177,178 many of which originated in Dr Koren’s laboratory. For women whose hair length permits, each hair is cut into three 3-cm segments closest to the scalp, which, in women who deliver at or near term, corresponds to each of the trimesters (assuming an average hair growth rate of 1 cm per month). Between 2 and 5 mg of hair from each segment is finely minced and weighed. The hair sample is placed in a 20-mL glass vial and washed with ethanol. The wash protocol consists of adding 2 mL of ethanol to the hair sample and incubating at 37 °C for 30 min. After 30 min, the ethanol is discarded and fresh ethanol is added. The procedure is repeated four times. After the washing, 1 mL of methanol is added to the washed hair, sonicated for 30 min, and incubated at 45 °C overnight. The next day, the methanol is pipetted off and the hair is rinsed with an additional 1 mL of methanol. After evaporating the methanol at 40 °C under a stream of nitrogen, the residue is reconstituted in phophate buffered saline and analysed for cocaine and benzoylecgonine by radiommunuoassay (RIA). Two commercially available RIA kits are used (Immunalysis Corporation, San Dimas, CA). One kit has an antibody specific to the cocaine metabolite benzoylecgonine; with the appropriate benzoylecgonine standards in drug-free hair, it is used for the quantification of benzoylecgonine in the unknown hair samples. The other kit has an antibody specific to both cocaine and benzoylecgonine and is used to quantitate the sum of cocaine and benzoylecgonine. The sensitivity of both kits is 0.1 ng/mg hair when 2 mg hair is used. Positive samples are confirmed by gas chromatography-mass spectroscopy.
After the cocaine analyses, the same hair sample is used for cotinine analysis. The hair is completely digested overnight using 1 mL of 0.6N NaOH. The following day, the solution is neutralised with 50–70 mL of concentrated HCell, and 100 mL aliquots of the neutral solution are used to measure cotinine using an RIA. The RIA materials are obtained from the Department of Biochemistry, Brandeis University, Waltham, MA. For quantification, cotinine standards in blank hair extracts are used. Results are expressed as ng of cotinine per mg of hair.
The primary study outcome is spontaneous preterm birth (delivery < 37 completed weeks), although subgroups of cases will also be selected according to degree (< 34 weeks and < 32 weeks) and type (spontaneous preterm labour vs. preterm prelabour rupture of membranes).179 The statistical analysis will parallel the general hierarchy of constructs that follows from our hypothesised causal pathways (see Fig. 1). From left to right in the figure, we posit SES as an exogenous construct (i.e. not explained by other constructs, that is, not receiving an arrow in the diagram), and several sets of endogenous constructs (i.e. explained by other constructs, that is, receiving at least one arrow in the diagram) successively mediating the causal links between SES and the final endogenous construct of preterm birth.
For reasons stated earlier,2,9 we will base our primary analysis of SES on maternal education ( 11 years) for low SES. Alternative criteria such as family income (above the Statistics Canada poverty line, 60–100% of poverty line, < 60% of poverty line) and maternal and paternal occupation (based on the Hollingshead classification) will be explored in secondary analyses.
Two types of analyses will be performed: multiple logistic regression and structural equation modelling. For both types of analyses, the dependent variable to be explained is preterm birth, and the independent variables are the potentially mediating variables (plus SES always included). Regression analyses will be performed to (1) identify the variables that mediate the increased risk of preterm birth among women of low SES, (2) quantify the strength of their mediation, (3) assess the contribution and statistical significance of relevant interactions, and (4) develop a multivariate model that ‘explains’ preterm birth by SES and its mediators. The main purpose of the regression analyses is sequential model building, based on the extent to which inclusion of each variable weakens or strengthens the association between SES and preterm birth. We have used this sequential model-building approach to advantage in two recent studies investigating the determinants of secular trends in preterm birth: one a Poisson regression-based analysis of Canadian vital statistics data,180 the other a logistic regression analysis of a hospital-based cohort study from Montreal.181
Although the order of entry of the dependent variables in the model and their relative contribution to the explanation of the outcome constitute important information about the general hierarchy of the dependent variables, they are only partial and indirect indications of the specific causal pathways (see Fig. 1) linking SES to preterm birth. Hence, a structural model expressing the nature and strength of each link between pairs of constructs integrated in a comprehensive and detailed model that can be statistically tested for fit would best meet our needs. In order to identify the ‘best’ model, a sequence of nested comprehensive models will be formulated, with each separately analysed by structural equation modelling.182–184 Each comprehensive structural model will be formulated on the basis of the conceptual framework complemented by the results of the logistic regression analyses. It is then assessed based on well-known statistical criteria182,183 pertaining to the overall statistical fit, tests of significance relative to each link, estimations of direct and indirect effects in the network of causal pathways, and the residuals. A comparison of competing models is possible within structural equation modelling, provided these models are nested. Hence, among similar (nested) models that are equally meaningful, the most parsimonious can be identified and is considered as ‘best’.
Sample size and statistical power
Our hypotheses are based on mediating variables and causal pathways, rather than simple associations between preterm birth and one or more risk factors. To our knowledge, no standard procedures or formulae are available to estimate sample sizes to test these types of hypotheses. Reported prevalences of psychosocial risk factors among low SES women range from 10 to 50%. We plan to recruit 5000 subjects, which should yield an approximate total of 200 cases, based on the spontaneous preterm birth rate of 4% observed among the 750 study women who have delivered thus far. The nested case-control study will be based on these 200 cases and 400 controls (two controls per case), which provides 80% power to detect odds ratios (OR) of 2.0 associated with prevalences of psychosocial risk factors among controls as low as 15%, even in the presence of moderate confounding (OR of 2.0 for confounding).185 Thus, considering mediating variables as ‘confounders’ of the SES–preterm relationship (for purposes of assessing statistical power), and based on an overall low SES prevalence of 30% and an OR as high as 2.0 for the low SES–preterm relationship, 200 cases and 400 controls should be adequate to detect an adjusted OR of 2.0 for associations between the psychosocial factors and preterm birth. The expected MTHFR mutation prevalence among controls is 10–12%; with about 25% of these (i.e. 3% of all controls) expected to have folate levels in the range previously reported to be associated with an increased risk of preterm birth,47 an OR of 3.0 for the MTHFR–folate interaction will be detectable with the proposed sample size. This sample size is also generally considered sufficient to satisfy the asymptotic requirements of structural equation modelling.183,186
To our knowledge, no previous study has permitted a thorough assessment of the mechanisms by which poverty and social disadvantage lead to increased risk of preterm birth. Our focus on chronic stressors and psychological distress with simultaneous evaluation of genital tract infection/inflammation, cigarette smoking, cocaine use, CRH, and comprehensive placental pathology will enable us, for the first time, to understand the relative importance of these factors as mediators of socio-economic disparities in preterm delivery. In addition, this is the first study of which we are aware that specifically examines a gene–environment interaction (pathway B in Fig. 1) as an aetiological determinant of preterm delivery. Because the available evidence suggests that low folate intake is more prevalent among low SES women,140–142 we believe that, as with neural tube defects, the effect of low folate may be limited to women with a genetic defect (a mutation in MTHFR).
Montreal is a particularly suitable site to investigate the causal pathways and mechanisms that we hypothesise. Its extremely high prevalence of poverty187 is (unlike the United States) unconfounded by race, because the vast majority of Montreal’s poor are white, Canadian-born, and French- or English-speaking. Moreover, free and universal access to prenatal and other health services is guaranteed by Canadian Medicare.
Regardless of whether our results confirm our two main hypothesised sets of causal pathways, we should succeed in placing potential mediators of social disparities in preterm birth in their proper context and thus provide a useful service to public health. No previous study has been sufficiently comprehensive and large to meet this goal. Understanding the causal pathways and mechanisms by which low SES leads to preterm delivery is important for two reasons. First, such an understanding should help in developing strategies (e.g. change in sexual behaviours, folate supplementation, or use of CRH antagonists) to reduce the social disparities. Second, it should provide important clues to the aetiology of preterm birth in general and hence suggest preventive interventions applicable across the entire socio-economic spectrum.
Drs Kramer and Platt are recipients of career awards from the Canadian Institutes of Health Research. The research reported here is supported in part by the March of Dimes Perinatal Epidemiological Research Institute grant #20-FY98-702.