TERATOLOGY 46261-266 (1992)
Cardiovascular Birth Defects and Prenatal Exposure to Female Sex Hormones: A Reevaluation of Data Reanalysis From a Large Prospective Study ERNEST B. HOOK School of Public Health, University of California, Berkeley, Berkeley, California 94720 and DeDartrnent of Pediatrics, University - of , California. . San’Francisco, San Frarkisco, California 94143
ABSTRACT In data of the U.S. Collaborative Preinatal Study (CPS), the Drug Epidemiology Unit (DEU) reported a relative risk of about 2.3 between maternal female sex hormone exposure during months 1 to 4 of pregnancy and cardiovascular malformation in infants (Heinonen et al., ’77a N. Engl. J. Med., 296:67-70). Wiseman and Dodds-Smith (’84)reexamined the original CPS data and found the DEU had made errors in classification of exposure and disease of some cases. Also they challenged the classification of cases as “exposed” in those born to mothers who received the compounds outside the day 19 to 50 window of cardiovascular embryogenesis. Wiseman and Dodds-Smith stated that their reanalysis “clearly showed that there was [in the data used by the DEUI no statistically significant association between exposure in the critical organogenic period of pregnancy and cardiac malformation in offspring.” They did not undertake any statistical analysis, but their reanalysis resulted in a widespread nonacceptance of the association reported by the DEU. The study reported here reclassified the cases of the original DEU study in accord with the implications of the Wiseman and Dodds-Smith reanalysis of exposure and disease. After this reclassification, an effect magnitude measure of association, the relative risk rose from 2.33 to 2.48 and remained nominally significant statistically a t the .05 level. Thus, if anything, the quantitative consequences of the Wiseman and Dodds-Smith review of the data, when applied in an unbiased manner, result in an increase in the measure of effect. The increase is consistent with the theoretical epidemiological expectation that correction of random errors in a database and of other non-differential misclassification, will tend to raise the estimate of an underlying association in the population studied. While these results reestablish the reported association, they do not, of course, prove that the positive association represents causal induction of defects in conceptuses by female sex hormones. o 1992 Why-Liss, Inc. The relationship of ingestion of female tions with any drug (Heinonen et al., ’77b). sex hormones by the pregnant woman to The DEU used data collected by the Collabcongenital heart malformations in the off- orative Perinatal Study (CPS) of the Naspring remains an unresolved issue. Several tional Institutes of Health on over 50,000 studies have reported evidence for a positive births occurring at 12 medical centers in the asso-ciation; others reported no significant U.S.A. from 1958 through 1966 (Niswander association. The reasons for these differ- and Gordon, ’72). The DEU reported a relaences remain unexplained (for review, see tive risk of 2.3 for cardiovascular defects in Hook, ’86.) Members and affiliates of the offspring of pregnant women exposed to feDrug Epidemiology Unit (DEU) of Boston University undertook one such study (Heinonen et al., ’77a) as part of an examination of associations of all categories of malformaReceived January 14, 1991; accepted September 17, 1991 0 1992 WILEY-LISS, INC.
262
E.B. HOOK
TABLE 1 . Original data relating congenital cardiovascular defects and exposure to female sex hormones’
Exposed Unexposed Total
Affected 19 385 404
Unaffected 1,023 48,855 49,878
‘Modified from Heinonen et al. (‘77a). xz
=
Rate 1.82% 0.78%
Total 1,042 49,240 50,282
13.9; xf = 12.6;P
Relative risk 2.33
(90% confcence interval) (1.5 to 3.5)
(95% confidence interval) (1.4 to 3.7)
- ,0002.
male sex hormones in months 1 through 4 inclusive (Heinonen et al., ’77a), an association already implied, a t least for one category of heart defects from an earlier analysis of the same data source by Mitchell et al. (’71) and Mitchell and Berendes (’71). One virtue of the CPS study was that the mothers were enrolled and data collection carried out before conclusion of the pregnancy, thus avoiding many biases that arise from retrospective case-control studies in which enrollment of subjects is undertaken and information on exposure is gathered some time after exposure and delivery. The DEU report and data analyses were widely reported and cited. But subsequently, a reevaluation of the CPS files raised questions about the accuracy of the abstraction of the data and the validity of the conclusions reached by the DEU (Wiseman and DoddsSmith, ’84). Writing in this journal, the authors, one affiliated with a chemical company, maker of female sex hormones, and the other of an English legal firm, presumably associated with the manufacturer, concluded that “reexamination of the base data of the [DEU] does not support the reported association between the exposure to female sex hormones during pregnancy and the occurrence of cardiac malformations” (Wiseman and Dodds-Smith, ’84). In this note I examine the consequences and challenge the conclusions of that reanalysis. I suggest that the original claim of the association reported by the DEU is in fact not undermined but bolstered by the results of the reanalysis of Wiseman and Dodds-Smith, contrary to their arguments. I also consider further the problem of whether the apparent association represents a true causal relationship.
cluded not only twins but also infants of mothers with rubella and some other groups from the CPS data, which originally comprised about 56,000 births. Data on maternal drug exposure were reviewed and coded in the early 1970s as part of the large broad scale study of the association of all congenital anomalies with prepartum maternal drug exposure (Heinonen et al., ’77b). The results of the original analysis appear in Table 1. Heinonen et al. (’77a) only presented data pertinent to the crude relative risks of exposure to any female sex hormone but did present relative risks adjusted for some covariates for some subcategories of female sex hormones. As data are not now available to me on the covariates, I can reanalyze here only changes in the crude relative risks.
Background: The Wiseman and Dodds-Smith reanalysis Wiseman and Dodds-Smith went to the original data, inspected some of the original records, and disputed some of the original classifications of disease and exposure and the rationale for classification of some cases as exposed during the vulnerable period. They reexamined the records of all 19 individuals with alleged congenital cardiovascular disease whose mothers were allegedly exposed to female sex hormones in months 1 to 4 of gestation and a random sample of 100 of the 1,023 individuals whose mothers had the same alleged exposure but were purportedly unaffected. They classified purported exposure to female sex hormones as to whether, with regard to embryogenesis of the cardiovascular system, any had occurred at all, or if exposure had occurred whether it had been “too late,” i.e., after day 50, or “too early,” i.e., before day 19 meaMETHODS suring, presumably, from the estimated Background: The original study date of biological conception. The DEU study included 50,282 motherWiseman and Dodds-Smith also examchild singleton pairs from the U S . Collabo- ined the clinical findings in the same group rative Perinatal Study. This subset ex- of records and noted whether individuals
263
PRENATAL FEMALE SEX HORMONE EXPOSURE
TABLE 2 . Some consequences of the reanalysis by Wiseman and Dodds-Smith (‘84) of data of Heinonen et al. (‘77b) Exposed and affected Those cases with putative classification errors: Errors of assigned diagnosis Error of assigned exposure (no exposure in months 1-4) Exposure “too late” (after day 50) Exposure “too early” (before day 19) Subtotal Others4 Total:
Exposed and unaffected Sample ~113
21
o2 (0%)
0
2 5 2 11 8 19
12 (12%) 36 (36%) 13 (13%) 61 (61%) 39 (39%) 100 (100%)
123 368 133 624 399 1,023
‘False positive diagnosis in original. ‘False negative diagnosis in original (none observed). ‘Estimated from sample proportions and the total of 1023. q w o with Down syndrome in exposed-affectedand one in the sample of those exposed and unaffected.
classified as “affected” were in fact likely to be “unaffected” and vice versa. Table 2 summarizes the results of the reevaluation of the patients’ records. Because Wiseman and Dodds-Smith examined a random sample of 100 of the 1,023 exposed-unaffected cases, this table projects the maximum likelihood estimates of what the results in the entire group of 1,023 would be based upon the results of that sample. Reclassification of cases I made stepwise alterations in the results given in Table 1 based on the implications of the Wiseman and Dodds-Smith reanalysis. First, I reassigned the two cases with no heart defect discovered among those classified as “affected,” to the unaffected group, and I reassigned those classified erroneously as being exposed to female sex hormones to the unexposed group. With regard to the latter there were two such cases among the affected and exposed and 12%of the 1,023 exposed (123 cases). I also reassigned from the exposed to the unexposed categories those reported by Wiseman and Dodds-Smith as exposed “too late,” i.e., after day 50: 5 cases among the 19 affected-exposed and 36% of the 1,023 exposed-unaffected (368 cases). Similarly, I also reclassified to the unexposed category the two exposed-affected cases in which exposure was putatively too early and 13% of the 1,023 of the original “exposed and unaffected,” 133 cases, in whom exposure was also putatively “to early.” RESULTS
Table 3 presents the results of the reclassification of cases after the adjustments de-
scribed above. Of the original 19 classified as “affected-exposed,”there are now eight remaining. Of the original 1,023 classified as “unaffected-exposed,” there are now an estimated 409. The rate in the exposed has risen from 1.82 to 1.96%. There have also been proportionally smaller changes in the unexposed category and the background rate has risen from 0.78 to 0.79%.The relative risk of being affected given “exposure” has risen from 2.33 to 2.48. The nominal probability of this or a higher relative risk by chance alone (assuming that the estimates in Table 3 were actually observed) is still less than .05. DISCUSSI 0N
The trends in the results
All studies which involve review of records are likely to introduce some errors in classification because of typographical and/or recording errors if no others. Such errors, if random, or otherwise result from what epidemiologists term “nondifferential misclassiication,” will tend to introduce a bias towards the null. That is, any observed association on balance will be more likely than not to be present, despite, rather than because of, random errors or this type of misclassification. And the quantitative measure of any apparent effect observed will tend to be lower than the true effect, and correction of such errors will tend to result in an increase in the estimate of the effect magnitude (see for justification, e.g., Rothman, ’86, p. 88.) The results of the reclassification here illustrate that expectation, in that the relative risk rose from the
264
E.B. HOOK
TABLE 3. Results of the reclassification of Table 1 entries after correction of erroneous exposure information, erroneous diagnoses of cardiovascular defects, and “too late” and “too early” exposure among the exposed group, ver classification of Wiseman and Dodds-Smith’ (90% (95% Relative confidence confidence ~~~~~~~ Affected Unaffected Total Rate risk interval) interval) Exposed 8 401 409 1.96% 2.48 (1.2to 4.6) (1.1 to 4.9) Unexposed 394 49,479 49,873 0.79% Total 402 49,880 50,282 ~
x
1 2 --
7.0; xz
=
5.6; P
~~~
~~
~
~
~
~~~~
~
~
~~~
~
~
~
- .01(assuming estimated valucs were actually observed)
original estimate of 2.33 to 2.48 after correction of putative misassignments among the exposed group, consistent with a “real” association in the data. Of course, the exact choice of the “window” chosen to classify a mother as affected, day 19 to 50 in this case, will also affect the results. One might dispute the precise width of this for some cases, or question whether available knowledge on the date of conception makes a precise classification of dates of exposure possible in all cases. But as long as the window is chosen without bias (certainly without a post hoc knowledge of the results) and is at least an improvement upon the original exposure category scheme used, then the trend in the results after reclassification will tend to an increase in the relative risk, if indeed there is a “true” association in the data. The possible consequences of a bias in choice or application of the window are discussed further below. The reevaluation of data by Wiseman and Dodds-Smith was only of those in the exposed category. Had they been able to reevaluate also data on those in the nominally unexposed category, it is likely that correction of the detected errors of “non-differential misclassification” in this group would have raised the estimated relative risk even further in accord with the expectation noted above. The issue of Down syndrome cases Wiseman and Dodds-Smith also analyzed and reported on three other variables in the records they examined: the presence of malformations in offspring of previous pregnancies, the presence of vaginal bleeding in index pregnancies, and a diagnosis of Down syndrome in the index cases. Of these, only the data regarding Down syndrome led to suggestions that particular cases should be treated differently than they were in the previous study of Heinonen et al. (’77 a).
As noted, Wiseman and Dodds-Smith found two cases with Down syndrome among the exposed-affected. They suggested these cases should be excluded from the analysis not because they had Down syndrome but because of the timing of the exposure for these particular cases. Although the mothers of these cases were exposed within the 19 to 50 day window of cardiovascular embryogenesis chosen for their original analysis, Wiseman and DoddsSmith claimed that the actual malformations in these two particular cases could not have been causally related to exposure because it occurred after day 40. If one were to exclude these two cases from the analysis in accord with the suggestion of these authors (p. 365), then there would be indeed a major change in the results. The trend would no longer be of nominal statistical significance. The relative risk would drop from 2.48 in Table 3 t o about 1.87, but this would be an erroneous estimate. For their suggestion that these cases be excluded was an ad hoc decision made after their examination of the data and, moreover, introduces a bias into the results because they applied their altered timing criterion selectively only to those in the exposed-affectedgroup, but not to those in the exposed-unaffected group. If one is to exclude cases from the exposed-affected group on the grounds of an altered timing criterion, then one must also apply the same altered timing criterion to the exposed-unaffected group and exclude cases outside of the altered window. (An appropriate way to adjust for the differences of susceptibility of various malformations within the 19 to 50 day window would be to undertake two separate analyses: one analyzing severe malformations associated with earlier embryonic events using an earlier window, and a second considering milder cardiovascular malformations and a later
PRENATAL FEMALE SEX HORMONE EXPOSURE
window. Using this or more refined subdivisions, one could then attempt to undertake a stratified pooled analysis. But in each case the reclassification criteria would have to be applied both to affected and the unaffected groups.) Separate from the issue of timing, Wiseman and Dodds-Smith implied that there was another reason to exclude Down syndrome cases from the analysis. They stated that this syndrome is “a condition of chromosomal origin which cannot be affected by exposure to sex hormones (or other drugs) after the period of implantation’’ (p. 363). This claim, for which no reference was cited, has never been established in the medical literature and there are no grounds of which I am aware on which it can be supported. Indeed it appears highly implausible to believe, as they imply, that after implantation the Down syndrome conceptus is impervious to effects of abortifacients such as folic acid antagonists or of teratogens such as thalidomide.1
Further issues Wiseman and Dodds-Smith (‘84) stated: “We believe the data examined from the relevant groups clearly show that there was no statistically significant assocoation between exposure to female sex hormones in the critical organogenic period of pregnancy and cardiac malformations in the offspring” (emphasis added). But in fact no statistical reanalysis was offered to support this contention. Certainly, the authors found errors and inaccuracies in the database which, as they note, are likely to arise with all large scale studies, and grounds to reclassify some
‘There is actually a methodologically acceptable rationale, albiet one which Wiseman and Dodds-Smithdo not cite, for treating Down syndrome cases differently from the hulk of the population. Because this chromosomal condition is associated with a genetic predisposition to congenital heart malformations, perhaps 50% are so affected compared to about 1% of those with normal chromosomes, one could argue that they should be analyzed separately from the rest of the population. But down syndrome cases are only one of many such groups with such a predisposition. Cardiac malformations appear in high frequency in other chromosomal abnormalities such as 45,H and trisomies 13 and 18, in several monogenic conditions such as the Holt-Oram syndrome, and in multifactorial caused syndromes.An unbiased approach should separate out all such genetically predisposed cases if it is to distinguish any one such category. Were unbiased data available, it would indeed be of interest to consider the associations of female sex hormones separately in those with and without any predisposition, but in the absence of such information, any exclusion of Down syndrome cases, whether among the exposed-affected or all those in the population studied, represents simply an ad hoe and post hoc adjustment of results.
265
of those classified as “exposed.” But as illustrated here by the statistical reanalysis of the crude relative risk, their own results imply that rectification of non-differential misclassification, if anything, increases the estimate of the magnitude of the association, in accord with theoretical epidemiological expectation, and the association still remained nominally significant after the adjustment. While the existence of the association in the data base is not, I believe, challenged seriously by methodologically appropriate reevaluation, the question of whether there is an underlying causal relationship of maternal ingestion of female sex hormones to induction of congenital heart defects in the conceptus remains unresolved. There are several reasons for this. The first is a statistical one. The report by the DEU (Heinonen et al., ’77a) cannot be regarded as a confirmation of the hypothesis of a relationship between some congenital heart malformations and female sex hormone status because this was first proposed or a t least implied to my knowledge by Mitchell et al. (’71) after their review of the CPS data on cardiovascular birth defects. (They focused on mothers with diagnosed “estrogen deficiency” who had presumably received female sex hormones therapeutically.) It appears to be generally overlooked that the DEU report could not in any way have been regarded as an independent confirmation of a relationship because they used the same CPS data base (albeit analyzing a slightly different subset of the same study and considering all female sex hormone exposure not just estrogens). In the absence of prior hypotheses, the reported results of the studies of Mitchell et al. and the DEU can only be used to structure a hypothesis about an association, a hypothesis that must be evaluated with other data sets. The results of other studies on hormones and heart defects are conflicting (e.g., Nora et al., ’78; Ferencz et al., ’801, although with one exception (Katz et al., ’85) the results of all of those known to me are at least consistent with an overall relative risk of two, based upon the 95% confidence intervals of the values of this variable (see for review Hook, ’86). (The latter reference, incidentally, concluded erroneously that the results of the Wiseman and DoddsSmith reanalysis undermined the strength of the association in the DEU study, a sub-
266
E.B. HOOK
jective impression corrected by the quantitative analysis herein.) The second issue is whether, even ifthere were a generally accepted statistical association of say a two-fold effect, roughly what was seen in the DEU study, the association is a causal one. Undetected confounding variables may still account for the significant positive associations in this study, in the different analysis of the same data by Heinonen et al. ('77a), and in others that observed statistically significant positive associations. Also, as most studies of the association have been confined to livebirths, it is possible that maternal ingestion of female sex hormones is associated with the differential survival to livebirth of affected conceptuses that otherwise would be spontaneously aborted or stillborn. Indeed, if, as some have suggested, some female sex hormones are antiabortifacients (see MacDonald, '89 for references), then there could be valid positive associations of maternal exposure to female sex hormones both in livebirths and in embryonic and fetal deaths, but no overall association in all conceptuses and no causal relationship to induction of cardiac malformation (Hook and Regal, '91). Thus, while appropriate correction of exposure and diagnosis information in the CPS data based on the findings of Wiseman and Dodds-Smith results in a raised estimate -of the crude relative risk, a rise in accord with theoretical epidemiological correction, the results of the reanalysis presented here do not establish that there is a true causal connection between maternal female sex hormone ingestion and congenital cardiovascular malformation in the offspring, either in the CPS population or more generally.
LITERATURE CITED Ferencz, C., G.M. Matanowski, P.D. Wilson, J.D. Rubin, C.A. Neil, and R. Gutberlet (1980) Maternal hormone therapy and congenital heart disease. Teratology, 21: 225-239. Heinonen, O.P. (1976) Risk factors for congenital heart disease: A prospective study. In: Birth Defects: Risks and Consequences. S. Kelly, E.B. Hook, D.T. Janerich, and I.H. Porter, eds. Academic Press, New York, pp. 221-264. Heinonen, O.P., D. Slone, R.R. Monson, E.B. Hook, and S. Shapiro (1977a) Cardiovascular birth defects and antenatal exposure t o female sex hormones. N. Engl. J. Med., 296:67-70. Heinonen, O.P., D. Slone, and S. Shapiro (1977b) Birth Defects and Drugs in Pregnancy. Publishing Sciences Group Inc., Littleton (MA). Hook, E.B. (1986) Environmental factors and human birth defects: Interpretation of relative risks in clinical genetics. In: Perinatal Genetics: Diagnosis and Treatment. I.H. Porter, N.H. Hatcher, and A.M. Willey, eds. Academic Press, New York, pp. 201-218. Katz, Z., M. Lancet, G. Sharmik, G. Chimke, B.M. Mogilner, and M. Klingberg (1985) Teratogenicity of progestogens given during the first trimester of pregnancy. Obstet. Gynecol., 65:775-780. Mitchell, S.C., and H.W. Berendes (1971) Congenital heart disease in 56,109 births: Incidence and natural history. Circulation, 43:323-332. Mitchell, S.C., A.H. Sellmann, M.C. Westphal, and J. Park (1971) Etiologic correlates in a study of congenital heart disease in 56,109 births. Am. J . Cardiol., 28:653-657. MacDonald, R.R. (1989) Does treatment with progesterone prevent miscarriages? Br. J Obstet. Gynecol., 96: 257-264. Niswander, K., and M. Gordon (1972) The Women and their Pregnancies; The Collaborative Perinatal Study of the National Institute of Neurological Diseases and Stroke. W.B. Saunders, Philadelphia, pp. 2-30. Nora, J.J., A.H. Nora, J. Blu, J. Ingram, A. Fountain, M. Peterson, R.H. Lortscher, and W.J. Kimberling (1978) Exogenous progestogen and estrogen implicated in birth defects. JAMA, 240r837-843. Rothman, K.J. (1986) Modern Epidemiology. Little, Brown and Company, Boston, pp, 87-8. Wiseman, R.A., and I.C. Dodds-Smith (1984) Cardiovascular birth defects and antenatal exposure to female sex hormones: A reevaluation of some base data. Teratology, 30:359-370.
Cardiovascular Birth Defects and Prenatal Exposure to Female Sex Hormones: A Reevaluation of Data Reanalysis From a Large Prospective Study ERNEST B. HOOK School of Public Health, University of California, Berkeley, Berkeley, California 94720 and DeDartrnent of Pediatrics, University - of , California. . San’Francisco, San Frarkisco, California 94143
ABSTRACT In data of the U.S. Collaborative Preinatal Study (CPS), the Drug Epidemiology Unit (DEU) reported a relative risk of about 2.3 between maternal female sex hormone exposure during months 1 to 4 of pregnancy and cardiovascular malformation in infants (Heinonen et al., ’77a N. Engl. J. Med., 296:67-70). Wiseman and Dodds-Smith (’84)reexamined the original CPS data and found the DEU had made errors in classification of exposure and disease of some cases. Also they challenged the classification of cases as “exposed” in those born to mothers who received the compounds outside the day 19 to 50 window of cardiovascular embryogenesis. Wiseman and Dodds-Smith stated that their reanalysis “clearly showed that there was [in the data used by the DEUI no statistically significant association between exposure in the critical organogenic period of pregnancy and cardiac malformation in offspring.” They did not undertake any statistical analysis, but their reanalysis resulted in a widespread nonacceptance of the association reported by the DEU. The study reported here reclassified the cases of the original DEU study in accord with the implications of the Wiseman and Dodds-Smith reanalysis of exposure and disease. After this reclassification, an effect magnitude measure of association, the relative risk rose from 2.33 to 2.48 and remained nominally significant statistically a t the .05 level. Thus, if anything, the quantitative consequences of the Wiseman and Dodds-Smith review of the data, when applied in an unbiased manner, result in an increase in the measure of effect. The increase is consistent with the theoretical epidemiological expectation that correction of random errors in a database and of other non-differential misclassification, will tend to raise the estimate of an underlying association in the population studied. While these results reestablish the reported association, they do not, of course, prove that the positive association represents causal induction of defects in conceptuses by female sex hormones. o 1992 Why-Liss, Inc. The relationship of ingestion of female tions with any drug (Heinonen et al., ’77b). sex hormones by the pregnant woman to The DEU used data collected by the Collabcongenital heart malformations in the off- orative Perinatal Study (CPS) of the Naspring remains an unresolved issue. Several tional Institutes of Health on over 50,000 studies have reported evidence for a positive births occurring at 12 medical centers in the asso-ciation; others reported no significant U.S.A. from 1958 through 1966 (Niswander association. The reasons for these differ- and Gordon, ’72). The DEU reported a relaences remain unexplained (for review, see tive risk of 2.3 for cardiovascular defects in Hook, ’86.) Members and affiliates of the offspring of pregnant women exposed to feDrug Epidemiology Unit (DEU) of Boston University undertook one such study (Heinonen et al., ’77a) as part of an examination of associations of all categories of malformaReceived January 14, 1991; accepted September 17, 1991 0 1992 WILEY-LISS, INC.
262
E.B. HOOK
TABLE 1 . Original data relating congenital cardiovascular defects and exposure to female sex hormones’
Exposed Unexposed Total
Affected 19 385 404
Unaffected 1,023 48,855 49,878
‘Modified from Heinonen et al. (‘77a). xz
=
Rate 1.82% 0.78%
Total 1,042 49,240 50,282
13.9; xf = 12.6;P
Relative risk 2.33
(90% confcence interval) (1.5 to 3.5)
(95% confidence interval) (1.4 to 3.7)
- ,0002.
male sex hormones in months 1 through 4 inclusive (Heinonen et al., ’77a), an association already implied, a t least for one category of heart defects from an earlier analysis of the same data source by Mitchell et al. (’71) and Mitchell and Berendes (’71). One virtue of the CPS study was that the mothers were enrolled and data collection carried out before conclusion of the pregnancy, thus avoiding many biases that arise from retrospective case-control studies in which enrollment of subjects is undertaken and information on exposure is gathered some time after exposure and delivery. The DEU report and data analyses were widely reported and cited. But subsequently, a reevaluation of the CPS files raised questions about the accuracy of the abstraction of the data and the validity of the conclusions reached by the DEU (Wiseman and DoddsSmith, ’84). Writing in this journal, the authors, one affiliated with a chemical company, maker of female sex hormones, and the other of an English legal firm, presumably associated with the manufacturer, concluded that “reexamination of the base data of the [DEU] does not support the reported association between the exposure to female sex hormones during pregnancy and the occurrence of cardiac malformations” (Wiseman and Dodds-Smith, ’84). In this note I examine the consequences and challenge the conclusions of that reanalysis. I suggest that the original claim of the association reported by the DEU is in fact not undermined but bolstered by the results of the reanalysis of Wiseman and Dodds-Smith, contrary to their arguments. I also consider further the problem of whether the apparent association represents a true causal relationship.
cluded not only twins but also infants of mothers with rubella and some other groups from the CPS data, which originally comprised about 56,000 births. Data on maternal drug exposure were reviewed and coded in the early 1970s as part of the large broad scale study of the association of all congenital anomalies with prepartum maternal drug exposure (Heinonen et al., ’77b). The results of the original analysis appear in Table 1. Heinonen et al. (’77a) only presented data pertinent to the crude relative risks of exposure to any female sex hormone but did present relative risks adjusted for some covariates for some subcategories of female sex hormones. As data are not now available to me on the covariates, I can reanalyze here only changes in the crude relative risks.
Background: The Wiseman and Dodds-Smith reanalysis Wiseman and Dodds-Smith went to the original data, inspected some of the original records, and disputed some of the original classifications of disease and exposure and the rationale for classification of some cases as exposed during the vulnerable period. They reexamined the records of all 19 individuals with alleged congenital cardiovascular disease whose mothers were allegedly exposed to female sex hormones in months 1 to 4 of gestation and a random sample of 100 of the 1,023 individuals whose mothers had the same alleged exposure but were purportedly unaffected. They classified purported exposure to female sex hormones as to whether, with regard to embryogenesis of the cardiovascular system, any had occurred at all, or if exposure had occurred whether it had been “too late,” i.e., after day 50, or “too early,” i.e., before day 19 meaMETHODS suring, presumably, from the estimated Background: The original study date of biological conception. The DEU study included 50,282 motherWiseman and Dodds-Smith also examchild singleton pairs from the U S . Collabo- ined the clinical findings in the same group rative Perinatal Study. This subset ex- of records and noted whether individuals
263
PRENATAL FEMALE SEX HORMONE EXPOSURE
TABLE 2 . Some consequences of the reanalysis by Wiseman and Dodds-Smith (‘84) of data of Heinonen et al. (‘77b) Exposed and affected Those cases with putative classification errors: Errors of assigned diagnosis Error of assigned exposure (no exposure in months 1-4) Exposure “too late” (after day 50) Exposure “too early” (before day 19) Subtotal Others4 Total:
Exposed and unaffected Sample ~113
21
o2 (0%)
0
2 5 2 11 8 19
12 (12%) 36 (36%) 13 (13%) 61 (61%) 39 (39%) 100 (100%)
123 368 133 624 399 1,023
‘False positive diagnosis in original. ‘False negative diagnosis in original (none observed). ‘Estimated from sample proportions and the total of 1023. q w o with Down syndrome in exposed-affectedand one in the sample of those exposed and unaffected.
classified as “affected” were in fact likely to be “unaffected” and vice versa. Table 2 summarizes the results of the reevaluation of the patients’ records. Because Wiseman and Dodds-Smith examined a random sample of 100 of the 1,023 exposed-unaffected cases, this table projects the maximum likelihood estimates of what the results in the entire group of 1,023 would be based upon the results of that sample. Reclassification of cases I made stepwise alterations in the results given in Table 1 based on the implications of the Wiseman and Dodds-Smith reanalysis. First, I reassigned the two cases with no heart defect discovered among those classified as “affected,” to the unaffected group, and I reassigned those classified erroneously as being exposed to female sex hormones to the unexposed group. With regard to the latter there were two such cases among the affected and exposed and 12%of the 1,023 exposed (123 cases). I also reassigned from the exposed to the unexposed categories those reported by Wiseman and Dodds-Smith as exposed “too late,” i.e., after day 50: 5 cases among the 19 affected-exposed and 36% of the 1,023 exposed-unaffected (368 cases). Similarly, I also reclassified to the unexposed category the two exposed-affected cases in which exposure was putatively too early and 13% of the 1,023 of the original “exposed and unaffected,” 133 cases, in whom exposure was also putatively “to early.” RESULTS
Table 3 presents the results of the reclassification of cases after the adjustments de-
scribed above. Of the original 19 classified as “affected-exposed,”there are now eight remaining. Of the original 1,023 classified as “unaffected-exposed,” there are now an estimated 409. The rate in the exposed has risen from 1.82 to 1.96%. There have also been proportionally smaller changes in the unexposed category and the background rate has risen from 0.78 to 0.79%.The relative risk of being affected given “exposure” has risen from 2.33 to 2.48. The nominal probability of this or a higher relative risk by chance alone (assuming that the estimates in Table 3 were actually observed) is still less than .05. DISCUSSI 0N
The trends in the results
All studies which involve review of records are likely to introduce some errors in classification because of typographical and/or recording errors if no others. Such errors, if random, or otherwise result from what epidemiologists term “nondifferential misclassiication,” will tend to introduce a bias towards the null. That is, any observed association on balance will be more likely than not to be present, despite, rather than because of, random errors or this type of misclassification. And the quantitative measure of any apparent effect observed will tend to be lower than the true effect, and correction of such errors will tend to result in an increase in the estimate of the effect magnitude (see for justification, e.g., Rothman, ’86, p. 88.) The results of the reclassification here illustrate that expectation, in that the relative risk rose from the
264
E.B. HOOK
TABLE 3. Results of the reclassification of Table 1 entries after correction of erroneous exposure information, erroneous diagnoses of cardiovascular defects, and “too late” and “too early” exposure among the exposed group, ver classification of Wiseman and Dodds-Smith’ (90% (95% Relative confidence confidence ~~~~~~~ Affected Unaffected Total Rate risk interval) interval) Exposed 8 401 409 1.96% 2.48 (1.2to 4.6) (1.1 to 4.9) Unexposed 394 49,479 49,873 0.79% Total 402 49,880 50,282 ~
x
1 2 --
7.0; xz
=
5.6; P
~~~
~~
~
~
~
~~~~
~
~
~~~
~
~
~
- .01(assuming estimated valucs were actually observed)
original estimate of 2.33 to 2.48 after correction of putative misassignments among the exposed group, consistent with a “real” association in the data. Of course, the exact choice of the “window” chosen to classify a mother as affected, day 19 to 50 in this case, will also affect the results. One might dispute the precise width of this for some cases, or question whether available knowledge on the date of conception makes a precise classification of dates of exposure possible in all cases. But as long as the window is chosen without bias (certainly without a post hoc knowledge of the results) and is at least an improvement upon the original exposure category scheme used, then the trend in the results after reclassification will tend to an increase in the relative risk, if indeed there is a “true” association in the data. The possible consequences of a bias in choice or application of the window are discussed further below. The reevaluation of data by Wiseman and Dodds-Smith was only of those in the exposed category. Had they been able to reevaluate also data on those in the nominally unexposed category, it is likely that correction of the detected errors of “non-differential misclassification” in this group would have raised the estimated relative risk even further in accord with the expectation noted above. The issue of Down syndrome cases Wiseman and Dodds-Smith also analyzed and reported on three other variables in the records they examined: the presence of malformations in offspring of previous pregnancies, the presence of vaginal bleeding in index pregnancies, and a diagnosis of Down syndrome in the index cases. Of these, only the data regarding Down syndrome led to suggestions that particular cases should be treated differently than they were in the previous study of Heinonen et al. (’77 a).
As noted, Wiseman and Dodds-Smith found two cases with Down syndrome among the exposed-affected. They suggested these cases should be excluded from the analysis not because they had Down syndrome but because of the timing of the exposure for these particular cases. Although the mothers of these cases were exposed within the 19 to 50 day window of cardiovascular embryogenesis chosen for their original analysis, Wiseman and DoddsSmith claimed that the actual malformations in these two particular cases could not have been causally related to exposure because it occurred after day 40. If one were to exclude these two cases from the analysis in accord with the suggestion of these authors (p. 365), then there would be indeed a major change in the results. The trend would no longer be of nominal statistical significance. The relative risk would drop from 2.48 in Table 3 t o about 1.87, but this would be an erroneous estimate. For their suggestion that these cases be excluded was an ad hoc decision made after their examination of the data and, moreover, introduces a bias into the results because they applied their altered timing criterion selectively only to those in the exposed-affectedgroup, but not to those in the exposed-unaffected group. If one is to exclude cases from the exposed-affected group on the grounds of an altered timing criterion, then one must also apply the same altered timing criterion to the exposed-unaffected group and exclude cases outside of the altered window. (An appropriate way to adjust for the differences of susceptibility of various malformations within the 19 to 50 day window would be to undertake two separate analyses: one analyzing severe malformations associated with earlier embryonic events using an earlier window, and a second considering milder cardiovascular malformations and a later
PRENATAL FEMALE SEX HORMONE EXPOSURE
window. Using this or more refined subdivisions, one could then attempt to undertake a stratified pooled analysis. But in each case the reclassification criteria would have to be applied both to affected and the unaffected groups.) Separate from the issue of timing, Wiseman and Dodds-Smith implied that there was another reason to exclude Down syndrome cases from the analysis. They stated that this syndrome is “a condition of chromosomal origin which cannot be affected by exposure to sex hormones (or other drugs) after the period of implantation’’ (p. 363). This claim, for which no reference was cited, has never been established in the medical literature and there are no grounds of which I am aware on which it can be supported. Indeed it appears highly implausible to believe, as they imply, that after implantation the Down syndrome conceptus is impervious to effects of abortifacients such as folic acid antagonists or of teratogens such as thalidomide.1
Further issues Wiseman and Dodds-Smith (‘84) stated: “We believe the data examined from the relevant groups clearly show that there was no statistically significant assocoation between exposure to female sex hormones in the critical organogenic period of pregnancy and cardiac malformations in the offspring” (emphasis added). But in fact no statistical reanalysis was offered to support this contention. Certainly, the authors found errors and inaccuracies in the database which, as they note, are likely to arise with all large scale studies, and grounds to reclassify some
‘There is actually a methodologically acceptable rationale, albiet one which Wiseman and Dodds-Smithdo not cite, for treating Down syndrome cases differently from the hulk of the population. Because this chromosomal condition is associated with a genetic predisposition to congenital heart malformations, perhaps 50% are so affected compared to about 1% of those with normal chromosomes, one could argue that they should be analyzed separately from the rest of the population. But down syndrome cases are only one of many such groups with such a predisposition. Cardiac malformations appear in high frequency in other chromosomal abnormalities such as 45,H and trisomies 13 and 18, in several monogenic conditions such as the Holt-Oram syndrome, and in multifactorial caused syndromes.An unbiased approach should separate out all such genetically predisposed cases if it is to distinguish any one such category. Were unbiased data available, it would indeed be of interest to consider the associations of female sex hormones separately in those with and without any predisposition, but in the absence of such information, any exclusion of Down syndrome cases, whether among the exposed-affected or all those in the population studied, represents simply an ad hoe and post hoc adjustment of results.
265
of those classified as “exposed.” But as illustrated here by the statistical reanalysis of the crude relative risk, their own results imply that rectification of non-differential misclassification, if anything, increases the estimate of the magnitude of the association, in accord with theoretical epidemiological expectation, and the association still remained nominally significant after the adjustment. While the existence of the association in the data base is not, I believe, challenged seriously by methodologically appropriate reevaluation, the question of whether there is an underlying causal relationship of maternal ingestion of female sex hormones to induction of congenital heart defects in the conceptus remains unresolved. There are several reasons for this. The first is a statistical one. The report by the DEU (Heinonen et al., ’77a) cannot be regarded as a confirmation of the hypothesis of a relationship between some congenital heart malformations and female sex hormone status because this was first proposed or a t least implied to my knowledge by Mitchell et al. (’71) after their review of the CPS data on cardiovascular birth defects. (They focused on mothers with diagnosed “estrogen deficiency” who had presumably received female sex hormones therapeutically.) It appears to be generally overlooked that the DEU report could not in any way have been regarded as an independent confirmation of a relationship because they used the same CPS data base (albeit analyzing a slightly different subset of the same study and considering all female sex hormone exposure not just estrogens). In the absence of prior hypotheses, the reported results of the studies of Mitchell et al. and the DEU can only be used to structure a hypothesis about an association, a hypothesis that must be evaluated with other data sets. The results of other studies on hormones and heart defects are conflicting (e.g., Nora et al., ’78; Ferencz et al., ’801, although with one exception (Katz et al., ’85) the results of all of those known to me are at least consistent with an overall relative risk of two, based upon the 95% confidence intervals of the values of this variable (see for review Hook, ’86). (The latter reference, incidentally, concluded erroneously that the results of the Wiseman and DoddsSmith reanalysis undermined the strength of the association in the DEU study, a sub-
266
E.B. HOOK
jective impression corrected by the quantitative analysis herein.) The second issue is whether, even ifthere were a generally accepted statistical association of say a two-fold effect, roughly what was seen in the DEU study, the association is a causal one. Undetected confounding variables may still account for the significant positive associations in this study, in the different analysis of the same data by Heinonen et al. ('77a), and in others that observed statistically significant positive associations. Also, as most studies of the association have been confined to livebirths, it is possible that maternal ingestion of female sex hormones is associated with the differential survival to livebirth of affected conceptuses that otherwise would be spontaneously aborted or stillborn. Indeed, if, as some have suggested, some female sex hormones are antiabortifacients (see MacDonald, '89 for references), then there could be valid positive associations of maternal exposure to female sex hormones both in livebirths and in embryonic and fetal deaths, but no overall association in all conceptuses and no causal relationship to induction of cardiac malformation (Hook and Regal, '91). Thus, while appropriate correction of exposure and diagnosis information in the CPS data based on the findings of Wiseman and Dodds-Smith results in a raised estimate -of the crude relative risk, a rise in accord with theoretical epidemiological correction, the results of the reanalysis presented here do not establish that there is a true causal connection between maternal female sex hormone ingestion and congenital cardiovascular malformation in the offspring, either in the CPS population or more generally.
LITERATURE CITED Ferencz, C., G.M. Matanowski, P.D. Wilson, J.D. Rubin, C.A. Neil, and R. Gutberlet (1980) Maternal hormone therapy and congenital heart disease. Teratology, 21: 225-239. Heinonen, O.P. (1976) Risk factors for congenital heart disease: A prospective study. In: Birth Defects: Risks and Consequences. S. Kelly, E.B. Hook, D.T. Janerich, and I.H. Porter, eds. Academic Press, New York, pp. 221-264. Heinonen, O.P., D. Slone, R.R. Monson, E.B. Hook, and S. Shapiro (1977a) Cardiovascular birth defects and antenatal exposure t o female sex hormones. N. Engl. J. Med., 296:67-70. Heinonen, O.P., D. Slone, and S. Shapiro (1977b) Birth Defects and Drugs in Pregnancy. Publishing Sciences Group Inc., Littleton (MA). Hook, E.B. (1986) Environmental factors and human birth defects: Interpretation of relative risks in clinical genetics. In: Perinatal Genetics: Diagnosis and Treatment. I.H. Porter, N.H. Hatcher, and A.M. Willey, eds. Academic Press, New York, pp. 201-218. Katz, Z., M. Lancet, G. Sharmik, G. Chimke, B.M. Mogilner, and M. Klingberg (1985) Teratogenicity of progestogens given during the first trimester of pregnancy. Obstet. Gynecol., 65:775-780. Mitchell, S.C., and H.W. Berendes (1971) Congenital heart disease in 56,109 births: Incidence and natural history. Circulation, 43:323-332. Mitchell, S.C., A.H. Sellmann, M.C. Westphal, and J. Park (1971) Etiologic correlates in a study of congenital heart disease in 56,109 births. Am. J . Cardiol., 28:653-657. MacDonald, R.R. (1989) Does treatment with progesterone prevent miscarriages? Br. J Obstet. Gynecol., 96: 257-264. Niswander, K., and M. Gordon (1972) The Women and their Pregnancies; The Collaborative Perinatal Study of the National Institute of Neurological Diseases and Stroke. W.B. Saunders, Philadelphia, pp. 2-30. Nora, J.J., A.H. Nora, J. Blu, J. Ingram, A. Fountain, M. Peterson, R.H. Lortscher, and W.J. Kimberling (1978) Exogenous progestogen and estrogen implicated in birth defects. JAMA, 240r837-843. Rothman, K.J. (1986) Modern Epidemiology. Little, Brown and Company, Boston, pp, 87-8. Wiseman, R.A., and I.C. Dodds-Smith (1984) Cardiovascular birth defects and antenatal exposure to female sex hormones: A reevaluation of some base data. Teratology, 30:359-370.
