Parental exposure, adverse pregnancy and offspring effects perspectives in developmental epidemiolgy

Developmental health is governed by biology and the societal culture that shapes family planning. This context should not be ignored in epidemiologic studies that address the effect of parental exposure from environmental and occupational agents on prenatal development, growth and survival, on adverse birth effects, or on postnatal events. It is important to have a thorough basic knowledge of developmental health in the population under study and to consider this basis in study design and performance. One way of accomplishing this task is to combine strengths of population-based cohort studies and nested case-control studies. Sibship-based cohorts and case-control studies that make use of biomarkers may provide particular advantages. Future research on the impact of parental exposure should be more strongly based on biological knowledge about the genetic, immune, and endocrine regulation of prenatal growth and development.

the conceiving spermatozoon or oocyte or their precursors or the stem cells from which they originated, so exposure could cover a period from the parents' own fetal life until conception of their offspring. Maternal exposure can act during pregnancy or postnatally during lactation. Manifestations of adverse outcomes could occur at any point in the life-span of the offspring; during pregnancy, delivery, or the postnatal period including adolescence and adulthood: death (spontaneous abortion, stillbirth, postnatal death), structural abnormalities (birth defects), and alterations in growth or functional competence of organ or organ systems (birthweight, nervous system development, cancer). Chronic disease in adult life could be a result of parental exposure, a concept based on the "programming hypothesis" of Barker and his colleagues (4).
The objective of this paper is not to provide a review, but rather to point out research needs and future directions of design and performance in epidemiology. Reproductive health is governed by biology, and the societal culture that shapes family planning. The impact of parental exposure cannot be understood without a basic knowledge of these contexts. Furthermore, specific validity concerns are of crucial importance. Therefore, I will first outline some aspects of family planning, biology, and validity matters.

Family planning aspects of developmental health
Declining infant mortality and societal changes have altered childbearing habits profoundly throughout this century (5). The current reproductive pattern in developed countries is still influenced by cultural and social conditions. One recent and substantial change is the postponing of women's reproductive career. In Norway, the proportion of primiparous women 230 years of age increased from about 5% to 15% between 1970 and 1990 (6). Combined with new developments in assisted fertilization techniques, this has resulted in a considerable increase in multiple pregnancies in several countries (7,8).
Family planning is also influenced by past reproductive history. Bjerkedal & Erickson (9) demonstrated that a subsequent pregnancy was more likely to follow among mothers who have experienced death of a previous infant than among mothers of surviving infants. Mothers with 2 children of the same gender were also more likely to have another pregnancy than mothers with 2 opposite-gendered children (9). A subsequent pregnancy is more likely for a woman who has experienced spontaneous abortion (lo), whereas birth defects may have the opposite effect (1 1).

Biological aspects of developmental health
Pregnancy involves biological processes that are timed and regulated in complex fashions. These processes involve development of parental germ cells, interaction between conceptus and mother to ensure the implantation and establishment of the placenta, biological processes within the mother and fetus that result in appropriate growth and development, and successful parturition. Development is dependent on genetic regulation and the intricate regulation and maternal redirection of immune and endocrine balance during pregnancy.

Heterogeneity
Epidemiologic studies have documented strong variability between women for most developmental outcomes. This propensity is manifested as differing probabilities of repeating pregnancy outcomes and, on the basis of past reproductive history, the identification of subgroups with heterogeneity in underlying risk. Such heterogeneity has been demonstrated for spontaneous abortion (10,(12)(13)(14)(15), perinatal mortality (16)(17)(18), gestational age (19)(20)(21), and birthweight and fetal growth (22,23). It is not surprising that these manifestations of fetal growth and survival are interrelated (20,22,(24)(25)(26). Recurrence has also been shown for birth defects (27). Repeating patterns of fetal growth and survival also extend across generations. Intergenerational birth record linkages demonstrate that birthweight is correlated between mother and child (28)(29)(30). Furthermore, offspring survival is dependent on maternal birthweight (30).
The biological basis for heterogeneity in fetal growth and survival is not fully understood. Seemingly, it does not represent a dichotomy where some mothers are doomed and others are protected (31,32). Although the basis for heterogeneity could be exogenous, it is probably explained by stable endogenous components involving genetic, immune, and endocrine expression and regulation. A biomarker of heterogeneity has still not been found, but, if it is, it could have profound implications in the study of environmental exposure effects, given the exposure effects resulting from concerted action with the biological factors in question (33).

Germ cells and genes
Environmental mutagens can act on parental germ cells or transplacentally and cause embryonic loss, birth defects or transgeneration cancer (34)(35)(36). Although this has been a traditional conceptual frame, the supportive evidence from humans is scarce. This is particularly the case for male-mediated developmental effects (37,38), in spite of the male being more prone than the female for mutagenic germ cell alterations that are compatible with successful conception (39). Furthermore, male-mediated postconception effects are easily accomplished in animal assays (40).
Genomic imprinting, parent-specific epigenetic marking of genes that results in monoallelic expression, has more recently been focused on as a potential mechanism for paternal influence on fetal growth and development (41). Clinical and experimental data suggest that the paternal genome is essential for placental development, whereas the maternal genome is essential for fetal growth regulation (42). Imprinting could play a role in the placental development and regulation of growth factors (43--46), but investigations of environmental agent influences are still lacking.
Gene-environment interaction is a rapidly growing branch of cancer research. The rationale for interaction is the hypothesis that allelic subtypes of genes that play a role in genetic or metabolic regulation influence the individual's susceptibility towards environmental agents. More recently, this interest has been extended to developmental effects as well (47). The discovery of periconceptional folic acid deficiency as a preventable cause of neural tube defects (48,49), and the investigations that followed on the role of allelic subtypes in folic acid metabolism (50,51) boosted further birth defect studies. One example is pursuit of the hypothesis that folic acid deficiency and maternal smoking during pregnancy cause facial clefts and that the risk from vitamin deficiency or fetal tobacco exposure is influenced by genotype (52)(53)(54)(55)(56)(57)(58)(59)(60)(61). The questions have still not been answered, but the problems have accelerated the development of both epidemiologic designs and genetic susceptibility markers (62,63).

Endocrine and immune factors
Normal intrauterine development and growth are dependent on endocrine and immune regulation and adaptation. The processes involve both mother and offspring, and the placenta plays a crucial role (64). Prenatal exposure to endocrine-disrupting xenobiotics are hypothesized to interfere with normal regulation and to cause genital birth defects and hormone-dependent cancer in adult life (65,66). The maternal immune system is redirected in PI-egnancy and shifts balance from a predominant Thl-driven cell-mediated immunity toward Th2-driven humoral immunity (67). Dysregulation of this equilibrium may constitute a threat to fetal survival and growth (67)(68)(69). Antiphospholipid antibodies are the only well-substantiated humoral causes of pregnancy loss or growth retardation (69)(70). Immunogenetic manifestations (allelic HLA subtypes) have been linked with habitual abortion (71). Considering the vast knowledge of immune effects from environmental factors, it is strange that virtually no study has addressed immune effects from environmental agents as a mechanism of developmental effects.

Validity issues in developmental epidemiology
I will emphasize some validity aspects and refer to textbooks (72) for more comprehensive reviews.

Selective loss
The appropriate denominator in a pregnancy outcome study is all conceptions in a study base. Such a denominator is generally not achieved, as most studies observe outcome after pregnancy as clinically ascertained, or they use births or live births. Selective loss of the outcome under study before observation begins results in an underestimation of the true incidence, and it may bias exposure-outcome associations. This is the basis for the "dose-response fallacy" in reproductive epidemiology (73), and it forces the investigation of prevalence rather than incidence. It is a notable problem in studies of chromosome aberrations, birth defects, and spontaneous abortion and could produce severe bias under realistic conditions (74)(75)(76). The same problem could occur if selective termination of prenatally diagnosed birth defects is not accounted for (77). Ignoring induced abortion in a spontaneous abortion study will also lead to a denominator problem, and it could bias the results (78).

Exposure
The quality problem of exposure information is shared with other fields of epidemiology. Timing of exposure is of particular importance. The choice of time windows of exposure will always have implications as to mechanistic inferences (79), and relevant time windows are often narrow in developmental epidemiology.

Study outcomes
There are no standard diagnostic criteria for several birth defects, and diagnosis may come long after the birth.
Validation using double registration systems have demonstrated that even major birth defects are not completely ascertained (80, 8 1). Gestational age based on the last menstrual period is also error prone (82), and it could produce bias in studies in which gestational age itself or intrauterine growth retardation or small-for-gestationalage were study outcomes.
Birthweight and gestational age are continuous variables that are often dichotomized as low birthweight (<2500 grams), preterm birth (<37 completed weeks), and small-for-gestational-age (below the 10th percentile of birthweight for gestational age). One obvious problem with the dichotomies is the possibility to lose important information about what is going on in the left tail of the distribution curve. Besides, considering the tendency of sibship repetition of preterm birth (20), low birthweight (20,22), small-for-gestational-age (23), and birthweight repetition over generations (30), this classification strategy may reflect adversity in a wrong way. Sibship and family histories indicate that a given birthweight may be normal for some newborns and adverse for others. This possibility is confirmed when perinatal mortality (as an indicator of adversity) is examined in relation to birthweight and sibship birthweight. In Norway the overall mortality for second births was 12.8 per 1000 in 1967-1984 (22); for second-born children weighing 2500 grams, perinatal mortality was only about 11 per 1000 if the older sibling had a birthweight of 2000-2499 grams, but 85 per 1000 if the older sibling weighed 4000-4499 grams at birth. This classification problem also operates on the population level, and it may yield distorted results when populations with differing base-line birthweight distributions are compared (83)(84)(85). There are analytic ways of overcoming the problems both on the population (83) and the family (22,23) level, but these are seldom used in studies addressing environmental exposures.
Proper grouping of outcomes in etiologic homogeneous categories is an ascertained problem. Preterm birth (86)(87)(88) and neural tube defects (89)(90) are etiologically heterogeneous, and subgroups should be considered. On the other hand, it has also been suggested that differing birth defects could have the same etiology (91, 92), so combining outcomes may be an option.

Covariates
Developmental health is strongly influenced by family planning and past reproductive history. In studies that address effects of parental exposure, these determinants should be regarded as covariates being considered in design and analysis. This is not an easy matter. Intuitively, it is not clear whether such covariates should be regarded as potential confounders or effect modifiers or whether they should simply be disregarded. A particular problem is the potential impact of selective fertility: the selection of women with adverse reproductive histories in 'ork Environ Health 1999, "0125, no 6, special issue high-order pregnancies, created by a desire for a certain family size (9,10,93), in combination with the heterogeneity of risk (10,93).
Socioeconomic status and social drift are obvious determinants for pregnancy outcome (94) and could create comparison problems. However, the nature of the influence of these factors is still poorly understood (95). Another problem is choosing a proper reference population for occupationally exposed women. Women who have an adverse reproductive history and who are at a high base-line risk of adverse outcomes tend to stay in work and have higher opportunities of exposure than women who stay home tending children. Thus reverse causality is one component of this "infertile worker effect" (96).
Covariate information quality has received little attention. Women can have a deficient recall of past reproductive history (97) or family history as a genetic indicator (98). A comparison of information of reproductive history from 2 sources indicates that covariate information quality may have impact on the relation between environmental exposure and pregnancy outcome (15).

Strategies of design and analysis
The investigator has several options in choosing the unit of analysis in a study of pregnancy outcome (eg, the couple, the pregnancy, and the fetus or infant). Studies including several pregnancies per couple may violate the assumption of independence. Analytic tools to sort out this problem exist (32). Analysis disregarding dependence seems not to matter much under usual assumptions (32), but analytic decisions should not be taken without careful assessment of the data (33).
Handling past reproductive history in design and analysis in studies addressing parental exposure and pregnancy outcome has been debated (31,(99)(100)(101). Suggestions to restrict studies to first pregnancies have been forwarded, as well as suggestions to disregard reproductive history in analysis. Studying only first pregnancies would be a way to escape the problem of selective fertility that could be intermingled with the exposure under study. However, this restriction has severe disadvantages, the lack of possibility to investigate interaction between exposure and reproductive history (102) to name one. Besides, quantification suggests that selective fertility may have a limited total impact and account for less than 2% of total perinatal mortality in a modern Western society (93). The exposure under study could cause adverse outcomes in previous pregnancies, and adjustment could produce rather than resolve bias. Handling reproductive history as a covariate must be based on mechanistic assumptions that should be thoroughly considered and made clear (100,101).

An example: maternal smoking in pregnancy and sudden infant death syndrome
Several factors, including maternal susceptibility, prenatal factors (growth retardation, short gestational age, maternal smoking), and postnatal factors (nonsupine sleeping position, parental smoking) are recognized or suspected causes of sudden infant death syndrome (SIDS). The impact of maternal smoking in pregnancy is naturally difficult to estimate because exposure is highly correlated with postnatal exposure; besides, prenatal smoking is a recognized cause of growth retardation (103,104). In a sibship-based Swedish birth registry cohort (105) and an inter-Nordic population-based case-control study (106), it has been possible to sort out some questions albeit some are still unanswered. Maternal smoking in pregnancy is a strong determinant of SIDS, along with other prenatal factors associated with prenatal growth retardation, factors seemingly not shared by SIDS siblings. Postnatal parental smoking seems to play a modest role if any. These results have been possible due to designs that allow the mother's reproductive history to be assessed and have smoking data of sufficient quality to disentangle prenatal and postnatal exposure.

The future
In the early 1980s, optimism prevailed: epidemiology should become a powerful tool to uncover environmental and occupational causes of reproductive failure (107,108). Expectations have not been fulfilled. Pesticides can serve as an example: numerous childhood cancer (log), fetal death (1 10) and birth defect (1 11) studies have been performed, and pesticides have been targeted as a priority research issue (1). The reward has not been substantial. In a recent editorial (112), reasons for failing advancement and future options were discussed, including the provoking question of whether epidemiologists should wait. These concerns deserve consideration at the turn of the century.

Research topics
Hopefully, there will be a shift away from risk factor epidemiology toward studies with hypotheses that are founded in biological knowledge, studies designed to investigate interference from environmental exposures on immune and endocrine regulation, and placental function, during pregnancy (65,113). We have some knowledge on the role of vitamins in embryonic and fetal development, and it is warranted to investigate potential disruptions of these mechanisms by environmental exposure (1 14, 115). We need more studies to investigate geneenvironment interaction (54,55,(58)(59)(60)(61)(62)116), and malemediated effects in proper designs (1 17).
Studies on fetal origins of adult disease (4,118,119) should be extended to acquire knowledge on the role of parental exposure. Today, most studies are restricted to diet, and correlations and associations between fetal antecedents (birthweight, etc) and adverse conditions later in life (1 19). Designs should allow separate assessment of prenatal and postnatal factors (120,121). As yet, parental exposure studies have concentrated on disorders in infancy (105,106) and childhood cancer (109). Cancer in adult life, autoimmune disorders, allergy, and mental development could be challenging study topics. There are animal models to investigate the effects of environmental agents on germ cell development in fetal life and fertility consequences in adult life (122,123). Similar hypotheses should be investigated for humans.
Joffe (34) has emphasized the virtue of studying several outcomes in concert. The problem of competing reproductive outcomes and the dose-response fallacy (73) could turn to strength if different outcomes that correspond to a single biological process were studied in the same cohort, providing mechanistic clues (124)(125)(126)(127)(128). Studying fertility and pregnancy outcome (124)(125)(126) could be particularly rewarding. Mechanistic clues could also be achieved by studying different outcomes in subsequent pregnancies (129). This concept could also be extended to whole families and consider alleged developmental and nondevelopmental outcomes in combination (130).

Designs
During the last several years, we have witnessed progress in design and learned much about study performance, and not the least about caveats and limitations.
National, population-based cohorts provide a valuable basis for parental exposure studies. The Nordic birth registries have been operative for decades, and they provide information with which to understand the sociocultural and biological context of developmental health at the population level (131)(132)(133). The establishment of sibship-based cohorts (102, 105) is a particular asset because the consideration of past reproductive history and selective fertility is made possible. Sibship-based cohorts have also been established in non-Nordic populations (134,135). The challenge in using the registers in parental exposure studies is to obtain exposure data from other sources (102). One option is to perform case-control studies based on populations in the registers (106).
Another challenging development is biomarkers and molecular epidemiology (136,137). Biomarkers in epidemiology certainly may have their problems (138,139). Increased cost and feasibility problems leave case-control studies as the most realistic options, but strategies have been developed to perform studies on several levels (140,141). Markers of susceptibility could prove to be the most valuable. The development of genetic susceptibility markers and designs to study gene-environment interaction have been exploding during recent years (142)(143)(144)(145). In parental exposure studies, several traditional designs in genetic epidemiology have not been satisfactory due to the failure to obtain estimates of exposure-outcome association (143,144). However, the development of gene-environment designs has led to merging with other fields of epidemiology. In particular, the "case-parent triad" case-control design seems promising since it allows separation of maternal and offspring gene impact and the assessment of imprinting (146)(147). These designs have potential validity problems (148)(149)(150), and actual experience with them is limited.

Do we need a new paradigm?
The prevailing paradigm in modern epidemiology is built on natural science. The concept of biological causality is governed by universal laws in the manner of theoretical physics. Research strategies are led by reductionism in the sense that we try to understand separate components of a process and use this information as building blocks to gain knowledge about complex systems. Susses & Susser (151) criticized these strategies and proposed the new paradigm of ecoepidemiology. Their suggested metaphor for ecoepidemiology is Chinese boxes, to integrate epidemiologic research on the molecular and the societal level (15 1). The proposed paradigm shift seems attractive for developmental epidemiology because of the impact of sociocultural conditions and biology. Parental exposure research that disregards this social and biological context could end up without perspective, to be ". . . oriented to explaining and quantifying the bobbing of corks on the surface waters, while largely disregarding the stronger undercurrents that determine where, on average, the cluster of corks end up along the shoreline of risk" [p 6341 (152). What should be done? Susser & Susses's proposal (15 1) has been extensively discussed, but the designs and analytic consequences of ecoepidemiology are yet not settled (153)(154). Meanwhile, we could do well to gain knowledge from different levels, in cohorts and population-based case-control studies, and escalate cooperation with molecular biologists, geneticists, and demographers.