The findings of this study indicate that parental occupation on the birth certificate may have some utility in the surveillance of occupational hazards in the workplace. Restricting comparisons to mothers who were gainfully employed according to the birth certificate and by interview, maternal occupation on these records exhibited excellent specificity and agreement (as measured by the kappa statistic) for most SOC occupational groups in which sufficient numbers of workers were available for analysis. We noted less agreement of the birth certificate with interview information regarding paternal occupation and maternal and paternal place of work than for maternal occupation. The limitations in using the birth certificate for occupational surveillance may differ for maternal and paternal occupations. In this study, mothers were frequently misclassified as homemakers or otherwise unemployed while the fathers' work information was often missing altogether. Furthermore, the problems of misclassification and missing values varied across occupational and industrial groups.
Because some parents change occupations between conception and delivery, the designation of parental occupation on the birth certificate as "usual occupation" limits the utility of using these records in the surveillance of reproductive hazards having effects early in pregnancy. Among mothers and fathers in this study population who held more than one position during the index pregnancy, the job held at delivery usually agreed between the birth certificate and interview.
Comparison of Findings with Other Studies
As in other studies, we found overreporting of "homemaker" on the birth certificate [6, 7]. This misclassification was particularly problematic among women who, according to the interview, held jobs requiring less formal education and/or technical skill such as sales and food service occupations with 44% and 65% respectively misclassified as homemakers or students on the birth certificate. In their study of New York State birth certificates, Marshall et al.  found that women in service occupations were the most likely to be misclassified as not employed (33%) while women in managerial/professional occupations were the least likely to be misclassified regarding employment status (6%).
The findings in this study were similar to those in New York  and California  with respect to the accuracy of maternal occupation on the birth certificate during the first trimester. Shaw et al.  noted that 71% of the maternal occupations on the birth certificate were similar to those obtained from an interview among case-mothers of births with severe cardiac disease and control-mothers who were residents of Santa Clara County. In the New York State study, 72% of maternal occupations and 77% of maternal industries or place of work on the birth certificate agreed with information from a mailed questionnaire. Both studies found negligible differences in agreement between cases and controls.
In the Santa Clara County study, the birth certificate was also examined for accuracy of paternal occupation for the period of three months prior to pregnancy and was noted to be comparable to interview information for approximately 80% of the records. In contrast, the birth certificate agreed for paternal occupation in only 63% of the records in the present study (with the missing cases included) in which we also included the first trimester as part of the exposure window. Part of this discrepancy might be explained by the higher proportion of unknown occupations (16%) in this study compared with the Santa Clara county study (6%). Using interview information, we examined types of occupations represented in the unknown category for birth certificates and noted a disproportionate number of occupations in food service, cleaning, construction, and the military according to the interview.
Based on finding similar misclassification of occupation on the birth certificate by case control status in previous studies, investigators [7, 8] concluded that misclassification was nondifferential, and use of the birth certificate information for case-control studies would lead to risk effect estimates closer to the null (odds ratio of 1.0). In the present study, we also noted similar proportions of misclassification for maternal and paternal occupations by case-control status. Although the numbers of specific birth defects were limited in the present study, we examined the relation between several maternal and paternal occupational groups and oral clefts. Risk effect estimates obtained for selected birth defects in relation to maternal occupations as reported on the birth certificate supported the prediction of weaker associations than those obtained from a more accurate source such as by questionnaire. Using all other paternal occupations as the referent group, we noted a positive association between paternal occupations in installation, maintenance, and repair and oral clefts; this association was stronger in the birth certificate group (odds ratio [OR] 2.7, 95% confidence interval [CI] 1.3, 5.6) than by interview (OR 2.0, 95% CI 0.96, 4.4) Although misclassification of paternal occupation appeared nondifferential in this study, the proportion of specific occupational groups that were unknown varied by case-control status. For example, control-fathers were more likely than case-fathers to be missing occupation on the birth certificate if they were employed in food service, cleaning or military occupations. Since the referent group for risk estimates included fathers who worked in occupations other than the occupational group of interest, these occupational groups that were more likely to be missing on the birth certificate would be underrepresented in the referent group for the controls. This bias could result in risk estimates farther away from the null even though the misclassification of paternal occupation by case control status was equivalent. Without the interview data, this potential bias would have been missed even if the percentage of unknown paternal occupations were compared by case-control status; in this study, the percentages of missing occupations were essentially the same between the two groups (15.7% versus 15.8%). We also looked at the effect of using fathers in professional or management occupations as a referent group; in the present study, these groups had few missing data on the birth certificate and showed substantial to almost perfect agreement with the interview. With these fathers as the referent group, paternal occupations in installation, maintenance, and repair were more strongly associated with oral clefts by interview (OR 3.4, 95% CI 1.2, 9.3) than by using the birth certificate information (OR 2.6, 95% CI 0.95, 7.0).
Furthermore, the proportion of missing paternal occupations on the birth certificate varied by reported maternal folic acid use during the periconceptional period. This differential in missing information by folic acid use could introduce bias in studies of paternal occupation and neural tube defects, oral clefts, and other defects for which insufficient folic acid is a risk factor. These findings underscore the need for caution when interpreting associations of parental occupation and birth defects based on birth certificate information. Choice of referent group might lead to overestimates as well as underestimates of associations if missing occupations vary by case-control status.
Previous studies did not specifically examine the proportion of missing values on the birth certificate by parental occupation/industry. In the present study, paternal professional or managerial occupations that require more formal education were less likely to have missing information on the birth certificate than occupations requiring less education such as construction or food service occupations. Very few birth certificates were missing information regarding maternal occupation or industrial sector, but, as already discussed, mothers were frequently misclassified as homemakers.
Limitations of the Study
This study had several limitations including use of the maternal interview as the gold standard to evaluate the quality of paternal occupational information on the birth certificate. In telephone interviews, mothers provided information about fathers' occupations and places of work. Schnitzer et al.  found mothers' reports of fathers' occupations to be subject to error with an exact agreement of 59% between mother's and father's reports of father's jobs in a metropolitan Atlanta population. However, mothers in the Atlanta study were interviewed 2 to 15 years after the index birth  in contrast to the NBDPS mothers who are interviewed 6 weeks to 2 years after the index birth. Therefore, it is likely that the mothers' reports of fathers' occupations in the NBDPS are more accurate than those of the 1968 – 1980 Atlanta birth cohort mothers from which Schnitzer et al. derived their study data, though some misclassification will still remain. Length of time from pregnancy to interview and multiple jobs during the exposure period of interest may increase errors in maternal recall of paternal occupations.
We also did not evaluate the accuracy of parental occupation in the fetal death records because this information was not available in the computerized fetal death record files in Texas. Although this exclusion amounted to only 20 cases in this study, it would be important to include fetal death records for occupational disease surveillance for several types of birth defects such as anencephaly in which fetal deaths represent a high proportion of cases.