Methodological considerations in the study of parental occupational exposures and congenital malformations in offspring

Print ISSN: 0355-3140 Electronic ISSN: 1795-990X Copyright (c) Scandinavian Journal of Work, Environment & Health Original article Scand J Work Environ Health 1988;14(6):344-355 doi:10.5271/sjweh.1908 Methodological considerations in the study of parental occupational exposures and congenital malformations in offspring. by Shaw GM, Gold EB Affiliation: California Birth Defects Monitoring Program, California Department of Health Services, Emeryville 94608.

In North America congenital malformations account for about 20 % of infant mortality (9). They are the seco nd leading cause of death among children aged one to four yea rs and the third leading cause among children aged 15 to 19 yea rs. Mos t co ngenita l malformations are believed to be due to the interac tio n of both en vironmental and genetic influences. Cur rently, approximatel y 20 % of human malformat ions can be ascr ibed to genetic transmission , 5 l1Jo to chromosome aberrations, and approximately IO l1Jo to all known environmental factors, leav ing approxim atel y 65-70 % of all malformations with no kn own or suspected cause (87) .
Several reviews of the literature have provided detailed accounts of the few suspected risk factors for malformations and have reviewed some of the issues specific to the epidemiology of congenital malformations (39,42,44). Our objective in this review was to address the many methodological co nsiderations inherent in studies of the potential relation between parental occupational exposure a nd congenital malformation s among offspring. The justification for studying a po ssible relation between a parent's expos ur e at work and subsequent development of a con genital malformation in his or her child is based on the notion that exposures to potential ha zards ma y affect parental germ cells before conception or may affect embr yon ic somatic cells after conception ; either action m ay induce cell death or dysfunction leading to malforma tio ns in the offspring (31). Th e importance of epide m iologic studies of rele vant oc cupational expo sures lies in th e potential for preventing co nge nita l mal fo rmation s if risks associated with such expo sures ar e ident ified and exposures are reduced or elim inated .

Biological considerations
The teratogenic effect of a specific occupational exposure is unlikely to result in a general increase in all types of malformations although it ma y increa se the risk of more than one malformation , particularly if the expos ure occurs during a specific time of gestation when more th an one organ system is developing. In table I severa l examples are given of inves tiga tio ns which in vol ved a ll malformations gro uped to gether rather than specif ic malformations as th e end po int of int er est (5,6,62). A generic problem fo r studies of m al fo rm ati on s is whether to group malformations . From a biological perspective , it is important not to group con genital malformations into a single ou tcome but rather to con sider them as heterogeneou s outcomes which ma y ha ve distinct etiologies. The choice of a specific mal formation or group o f mal fo rmation s for st udy sho u ld be ba sed upon embryologic in formation about wh ich organs and tissue s ar e developing at the tim e of exposure. In addition, decisions about whi ch Table 1. S e l ec t ed stud i e s o f pa re n tal o ccupati on al e xposu res and congenita l mal formati on s . (B D M P =birth d efe c t monitoring prog ram , CLP=c le ft lip a nd pa late, CNS=ce ntral nervou s syste m , GI =gastro in tes tinal, MACDP =Metropolitan Atlanta Congeni t a l Defect Pro gram , MS =musculoskeletal, NAAA =National Agricult u ra l A viat i on A ssoci a ti o n , OR =odds ratio, PV C =polyvinyl chlori d e , RR =relat ive ri s k , V DT =vid eo di s p lay t e rm inal)      (17,75). How ever, another generic problem for studi es of malformations is that, even for substa nces that have been described as teratogenic, controversy exists with respect to extrapolation from animal dat a to human experience. Such extrapol ation is hindered by differences in metabolism, pla cent al function, and patterns of development among ani mal species (68, 81). Thu s the critical tim-348 ing of exposure to pa rti cular teratogens may not be generalizable across species (81). Furthermore , Saxen (68) has noted th at , for some substances found to be teratogenic in hum an s, an effect has not been foun d in experimenta l studies.

Prevalence estimation
Th e prevalences of congenit al malformations can be influenced by variations in the source of info rmation from which malformat ions are ascertained, the length of time subsequent to birth during which malform atio ns are detected , an d the diagnostic definiti on of a specific malform ation. There is pot ential for biased study results jf such variations are differentially exerted in exposed and unexposed groups. While these issues are generic to most epidemiologic studies of malformations, they must be considered specifically for investigations of parental occupational exposures and malformations in offspring.
The ascertainment of children with malformations has been accomplished from a variety of information sources, including vital statistics records such as birth, death and fetal death certificates, interviews with parents, medical records , and both active and passive reporting or surveillance systems (table I). The use of different ascertainment sources may result in varying prevalence estimates for a given malformation in that greater care may be taken to ensure complete and accurate information from some sources of information than from others. For example, while birth certificates are frequently used to ascertain malformations (table I, references 60 and 66) because of the relative ease and low cost, this method is well-known to underestimate the prevalence of malformations (4, 10,25) when compared to, say, a surveillance system such as the California Birth Defects Monitoring Program (26) or the Metropolitan Atlanta Congenital Defects Program (19), which actively monitor for malformations up to one year of life. However, some of the underreporting associated with the use of birth certificates is merely due to failure to record malformations on the birth certificate (69). Reporting has also been found to vary by size of the hospital of birth (4). It has been suggested that data on malformations obtained from interviews with parents, particularly in prospective studies (a method of ascertainment for many of the studies in table I) be considered with caution (3).
An additional concern generic to studies of malformations is that the length of time into the postnatal period when a malformation is diagnosed varies considerably among studies (table I). Duration of postnatal follow-up needs to be considered because both the prevalence and types of congenital malformations diagnosed change with time subsequent to birth. For example , anencephaly is generally noted at time of delivery, but many of the cardiac malformations may not be diagnosed until the first six months of life.
Another concern pertains to how a particular malformation is defined. Reporting varies by the nature and severity of the malformation (4, 10, 25, 46). Although less of an issue with severe malformations because they are usually more easily diagnosed, if the diagnostic criteria differ by reporting source or between studies for a particular malformation, so will the prevalence estimates. As an example, if the case definition for a particular cardiac lesion in one study specifies that the confirmed diagnosis be based on an invasive diagnostic procedure, whereas in another study such stringent criteria are not part of the case definition, then different prevalence estimates may be observed . A further problem associated with the definition of congenital malformations is whether to include stillbirths (37). The prevalence of malformations among stillborn infants is higher than among infants born live (87). Thus including or excluding stillbirths will affect the magnitude of the reported prevalence.
In addition to the potential variability in prevalence estimates due to differences in ascertainment source, length of follow-up, and inconsistent case definitions, malformations may be differentially reported in certain study subgroups, such as certain ethnic, educational, or economic groups (38,53), that have differing access to medical care for the diagnosis of malformations. For example, for workers' families covered by a health plan there may be better ascertainment than for workers' families without such coverage. It has further been observed that the reporting of some congenital malformations is greater in rural than urban areas (54).

Biological considerations of occupational exposures
Specifying hypotheses with a firm biological rationale helps reduce both false positive and false negative results. Maternal or paternal exposures to putative agents preconceptionally or postconceptionally may be related to the birth of a malformed child (39,42). Crucial to this understanding is the specific timing of such exposures in the reproductive cycle, a factor critical for the assessment of an agent's teratogenic risk (17).
Preconceptional occupational exposures of either parent could theoretically contribute to the subsequent development of offspring with a malformation. Exposures in the workplace may cause chromosome damage or changes in molecular deoxyribonucleic acid among parental germ cells, men being somewhat more likely to be affected since there are numerous opportunities when agents can have an impact during cell divisions in sperm production (82). However, Pearn (63) noted that there is, to date, no experimental evidence to suggest that paternal exposure to agents can result in a teratogenic effect. Neither paternal nor maternal preconceptional occupational exposures have been extensively examined with regard to the risk of malformations in offspring (5,6,14,18,20,21,52,77,78,85). On the other hand, much more is known about how postconceptional occupational exposures may affect normal embryologic development (2, 7, 16, 22-24, 29, 30, 34-36,45, 47, 48, 50, 55, 59, 72).
The critical period of teratogenesis is thought to coincide the best with the period of embryonic organogenesis, from the third to the twelfth week of gestation (58, 86). Within this period, there is a critical period for each organ system during which dysmorphogenesis might occur. The exact time of vulnerability during the critical periods varies with the agent, as well as with dose and duration of exposure (58, 86). The application of the critical period concept may be epidemiologically useful in identifying potential occupational teratogens. If exposure to an agent occurs prior to the end of a given critical gestational period, it can be considered a possible teratogen. However, occupational exposure of a pregnant woman after closure of the fetal palate cannot, for example, be a cause of cleft palate.
Defining exposure time relative to embryologic development allows for a greater likelihood of identifying a particular agent or agents as possible teratogens, ie, random misclassification of exposure is less likely, and therefore the estimated effect is less likely to be underestimated. However, imprecision may exist in the estimation of fetal age, often because the actual date of conception is not known but is, rather, estimated according to the date of last menses derived from medical records or estimated by the recollection of the person in question. The latter estimation is susceptible to variations in a woman's self-awareness of the beginning of her pregnancy, eg, a mother may perceive her pregnancy to have begun when she felt pregnant, possibly several weeks subsequent to when her pregnancy actually began. Such imprecision may lead to variations in the estimation of critical periods (58) and the misclassification of exposure status. Unless the proportion of truly exposed among those classified as exposed is relatively large or the effect of the exposure on the particular studied outcome is large (80), the effect could go undetected. To minimize this type of misclassification, study instruments must be constructed to elicit more precise information about the time of conception.
An add itional issue of biological concern is that of the exposure-response relationship. In most studies of occupational exposures related to congenital malformations , it is difficult to assess the extent of exposure accurately. Thus surrogates of actual exposure are often used, such as duration of exposure . As a result, there is little information in the literature about exposure in humans, and it remains unclear if high exposures for short or intermittent periods or low exposures for longer periods are equivalent in risk even within the limits of the relevant embryologic timing of exposure. Other models of risk of adverse reproductive outcomes associated with exposure have been proposed (73).

Methodological considerations for occupational exposures
The validity of reported occupational exposures is crucial to risk estimation and depends on the following two related areas of concern: the source of the exposure information and the inclusion of validity checks on this information. As can be seen in table 1, the most frequent sources of information about occupational exposures in studies of congenital malformations are birth certificates, interviews with or questionnaires for 350 one or both parents, and, occasionally, company records. Each of these techniques has its advantages and disadvantages as a source of occupational exposure dat a.
Source of information. The major advantages of the birth certificate as a source are ease of access and reduced cost. However, the many disadvantages of this source tend to outweigh its advantages. Concern about using birth certificates to acquire occupational exposure data results from the following possibilities. The data may be incomplete since, in many area s, ascertainment of parental occupation is not mandatory for the birth certificate (82). In addition, the recorded parental occupat ion at the time of birth on the birth certificate may not be the parent's usual occupation or occupation at conception or at the relevant time during pregnancy. As an example , the mother's occupation often appears as housewife at the time of birth when in fact she may have worked in another capacity at the time of conception or during the early months of her pregnancy (11). Errors in classifying occupational titles into exposed or unexposed categories may also bias the effect measure since specific occupational exposures are not recorded on the birth certificate and occupational title must be used as a surrogate measure. Occupational titles used as a surrogate may be too crude for exposure estimation (35). Several occupationexposure linkage strategies have been proposed in which occupations or occupation-industry combinations are linked with specific exposures (13,33,57,76). Two indirect strategies (33, 57), however, have been shown to have poor agreement with direct reporting based on questionnaire-elicited data (51). Moreover, information on occupation may not be collected or coded in a standardized way, and thus it may be noncomparable and somewhat inappropriate for purposes of aggregating these data from various areas or hospitals . Finally, the information obtained from the birth certificate about potential confounders may be limited.
The primary advantages of occupational data obtained from interviews and questionnaires are that subjects can be queried directly, in a detailed and standardized way, about occupations and occupational exposures, including information on substances, timing, exposures, protective gear, and work habits. There are, however, a number of disadvantages with this approach. First , while respondents know about terms of employment and character of the job (64), they may not know with certainty about exposures they have had on the job, many of which change over the time of employment and thus introduce the possibility of misclassification. Second, one parent may be easier to find and more accessible and may thus become the respondent. Yet, it is often unclear if maternal or paternal occupational exposures, or both, are the most relevant, and the one parent may not provide valid and complete information regarding the other 's specific oceupa-tional exposures and is more likely to know a specific job title than specific workplace exposures (74).
Finally, this technique is probably the most susceptible to problems of information bias (70). The underlying reason for concern about this bias is that parents of children with malformations may recall exposures better than parents of nonaffected children, the result being a biased estimate of the effect measure away from the null. Thus the likelihood increases for incorrectly claiming that a difference in exposure exists when it really does not. However, a number of studies examining exposures during pregnancy and specifically testing for recall bias have been unable to demonstrate its presence (43,61,84,88). Several suggestions have appeared in the literature for minimizing recall bias. First, asking specific questions about exposures (eg, in the case of drugs, the conditions for which taken, and the names) tends to improve recall, whereas open-ended questions do not enhance information (56). With regard to occupational exposures, asking about specific exposures rather than trying to deduce exposure from job categories may yield more information (51).
The primary advantages of using company records are that they are not subject to problems of recall bias and tend to be more complete in recording occupational exposures. There are, however, some problems with this technique as well. Perhaps the two most looming potential problems are that of gaining access to company personnel records and that of the lack of recording changes in exposures over time. Since companies are not universally cooperative in sharing their personnel records for studies of potential adverse health effects of occupational exposures, lack of cooperation can compromise the completeness and validity of data obtained in this way. Using company records to determine occupational exposures is also a rather cumbersome technique to be used in a case-referent study, since it often involves obtaining cooperation from many different companies. In addition, exposures for any given job title or position may change over time, but exposure changes may not be routinely recorded in the personnel file. Thus company records might be of less value, in some instances, in obtaining valid exposure information than interviews with the workers themselves . Furthermore, information on potential confounders from company records may be insufficient.
Validating exposure information. Because no single technique for obtaining occupational exposure information is perfect, the need for building validity checks into studies of such exposures is critical. Experienced occupational hygienists can certainly contribute to the process of validating occupational exposure information . Validating not only positive reports of exposure, but negative ones as well, would minimize the chances of misclassification bias. Sometimes it may only be necessary to perform validity checks on a sample of the total population to determine that the selected method for obtaining data on occupational exposures is sufficiently valid and complete, while at other times it may be necessary to validate information on all subjects. There are, however, problems in attempting to validate information. First, sources for validation are not always available. Second, it is not always clear what information source should be considered the valid measure of exposure if two sources disagree. For example, medical records are often considered the valid source of information in epidemiologic studies , but some investigators have found that these records may be as incomplete or inaccurate as other sources of data (32). Finally, more recently the use of biological markers in epidemiologic research has been increasing. Although the recent study conducted by the Centers for Disease Control (12) involving serum levels of dioxin serves as an example of how useful biological measures may be for validating another source of exposure information, the validity and meaning of many biological markers will still need to be established before used in this regard (71).

Reference groups
Inclusion of a reference group which will provide an unbiased estimate of risk requires careful consideration of likely modes, mechanisms, and timing of exposures, ie, an estimate not biased due to differential error in the ascertainment of exposure status. In prospective studies of occupational exposures in general , and specifically those studying these exposures as related to the occurrence of congenital malformations, a frequent approach is to include as a reference group another group of workers in the same industry which is presumed not to have had the exposure under study. It is usually easier to obtain permission and cooperation for such a group because they have already been obtained for the exposed group. The ascertainment and the follow-up of subjects may also be easier. In addition, this method makes it more likely that the reference group will experience medical surveillance similar to that of the exposed group than a reference group selected from outside the industry.
Problems generic to occupational studies using a reference group from the same industry include (i) the possibility of the workers changing positions within the same industry, the result being changed exposure status over time, and (ii) the possibility of the reference or unexposed group being inadvertently exposed to the substance of interest. These types of problems result in a misclassification of exposure that biases risk estimates toward the null. One must determine whether the advantages of using this type of reference group outweigh the disadvantage of possibly providing false reassurance about a potential hazard.
The choice of an appropriate reference group in a case-referentstudy may also be problematic. The most frequently used reference groups are infants with malformations other than the one under study or normal infants (eg, siblings, neighbors, friends, a probability sample of all infants in the area under study, or random-digit-dialed referents). Using infants with other malformations helps to control recall bias because parents with affected infants may try harder to recall exposures which may have affected their offspring adversely. Another advantage of a reference group with other malformations is that cooperation may be better and the subjects may be easier to follow-up than normal infants, the result being a reduction in the possibility for response bias.
One drawback to consider in using a reference group comprised of infants with malformations is that, since the etiology of about 65 0J0 of all congenital malformations is unknown (87), some of the reference group malformations may be etiologically linked with the exposure under study. Because much of organogenesis occurs in the first trimester, it is likely that more than one organ system could be adversely affected by the same exposure during that 12-week period. One alternative is to use a reference group with a malformation whose etiology is known, eg, a chromosome malformation, although this may not be as good a control for recall bias if the etiology is well-known to the parents. One could also include a variety of diagnostic groups of malformations to minimize the impact of inadvertently including one that is associated with the exposure under study (40,70). However, because the etiology of most malformations is unknown, the likelihood of including more than one malformation that is associated with the exposure under study may still be great and may thus result in an underestimate of the effect measure. Another possibility is to use a non diseased reference group which would provide the opportunity to compare exposure frequency to that of a group that has not been influenced by disease but would enhance the possibility of recall bias.
An additional option would be to use two reference groups, one with other malformations and one without. This method would permit both recall bias and risk to be assessed (70). It, however, is more difficult and costly and could either complicate or elucidate (15,28) the interpretation of results if effect measures differ substantially for the different comparison groups or if the explanation of the discrepancy can be deduced.
One further issue regarding the selection of a reference group is which subjects should be excluded. Some questions that arise in considering exclusions are (i) whether referents should be excluded who have malformations which may be associated with the exposure under study, ie, how strong must the association be for the referent to be excluded; (ii) what should be done with referents for whom the association with exposure is speculative (perhaps they should be included as a separate group for analysis); and (iii) what referents should be excluded when the study is exploratory (ie, a number of exposures may be studied which may be associated with different malformations in the reference group).

Sample size and multiple tests
The issue of sufficient study power for examining specific malformations needs to be considered. Because specific malformations are rare, the practice of broadly grouping malformations is often undertaken and justified on the grounds of increasing sample size with concomitant increases in study power. However, one is hardpressed on biological grounds to support a practice that, even though it may result in the ability to detect a smaller effect measure as statistically significant, may well obscure interpretation of a potential specific exposure-malformation relation. However, grouping specific malformations may have biological validity if several malformations suspected to have a similar etiology or embryology are considered as a single group or if complex mixtures of exposures, eg, in toxic waste sites, are of interest or if exposure occurs when a number of organ systems are developing. A number of the prospective studies reviewed in table 1 considered congenital malformations as a single end point (1, 23,27,49).
Inherent in a prospective examination of congenital malformations in which the frequency of numerous malformations is examined between exposed and unexposed groups is not only the problem of small numbers and reduced power, but also the problem of interpreting findings from multiple testing. There are approximately 200 specific congenital malformations at the four-digit level of the International Classification of Diseases (ninth revision). If relative risks are calculated for each outcome, one would expect 10 to be statistically significant by chance alone at an alpha level of 0.05. Frequently, multiple exposures and multiple time periods are also under consideration, a process which further increases the number of statistical tests and potential false positives. Certainly, if several indicators of the same exposure are available, one can examine the consistency of the effect estimates across the various indicators. However, if more than one estimate of the same exposure is not available, one is then left with the difficult choice of which positive associations to pursue. Several approaches might be considered for reporting results when a large number of potential associations have been examined, including the reporting of all potential associations examined with the thought that each exposure-outcome relation is a separate investigation (8) or the reporting of only associations which are "significant" after an adjustment procedure such as the Bonferroni (79) procedure or the reporting of all potential associations examined but providing a ranking in order of priority for investigations using empirical Bayes techniques (83).
Potentially hazardous workplace exposures have raised concern regarding the adverse reproductive effects of such exposures to men and women. Epidemiologic studies of these relations are important for asses sing such risks so as to help in setting standards to reduce hazardous exposures and the risk of malformations. Addressing the methodologic concerns discussed in this review regarding exposure and outcome in future studies of occupational exposures and congenital malformations will likely result in greater specificity of findings and thus enhance the possibilities for prevention .