Nondifferential Exposure Misclassification Research Articles

Streptococcal infection, Tourette syndrome, and OCD

A causal relationship of common streptococcal infections and childhood neuropsychiatric disorders has been postulated. To test the hypothesis of an increased rate of streptococcal infections preceding the onset of neuropsychiatric disorders. Case-control study of a large primary care database comparing the rate of possible streptococcal infections in patients aged 2-25 years with obsessive-compulsive disorder (OCD), Tourette syndrome (TS), and tics with that in controls matched for age, gender, and practice (20 per case). We also examined the influence of sociodemographic factors. There was no overall increased risk of prior possible streptococcal infection in patients with a diagnosis of OCD, TS, or tics. Subgroup analysis showed that patients with OCD had a slightly higher risk than controls of having had possible streptococcal infections without prescription of antibiotics in the 2 years prior to the onset of OCD (odds ratio 2.59, 95% confidence interval 1.18, 5.69; p = 0.02). Cases with TS or tics were not more likely to come from more affluent or urban areas, but more cases lived in areas with a greater proportion of white population (p value for trend = 0.05). The present study does not support a strong relationship between streptococcal infections and neuropsychiatric syndromes such as obsessive-compulsive disorder and Tourette syndrome. However, it is possible that a weak association (or a stronger association in a small susceptible subpopulation) was not detected due to nondifferential misclassification of exposure and limited statistical power. The data are consistent with previous reports of greater rates of diagnosis of Tourette syndrome or tics in white populations.

Case-Crossover Designs Compared With Dynamic Follow-up Designs

In the late 1980s, when we started developing the case crossover design, confusion about the relation between case-control studies and cohort or follow-up studies was still common. Some epidemiologists were still teaching students that a case-control study is like a backwards cohort study the "trohoc" fallacy.' The turning point was the realization that most case control studies are (or ought to be) the same as dynamic population follow-up studies, except for a key difference: controls are a small sample rather than a complete census of the study base-the population-time that produced the cases.2 This realization played a role in our formulation of the case-crossover design, as we puzzled over how to identify triggers of myocardial infarction (MI) onset by retrospective interviews with patients.3 Here the underlying dynamic fol low-up study was like a crossover experiment, except that the time at which subjects crossed back and forth between expo sure and nonexposure was not controlled by an experimenter but influenced by natural events or subjects' personal decisions. The impetus for using cases as their own controls was our finding that all other control groups are too vulnerable to selection or information bias.4 By interviewing patients about the day of their heart attack (the case day) as well as the day before (the matching control day), we could rule out con founding by constant characteristics of patients. Also the subjects' interpretations of our questions about exposures would be the same for both days. Initially, we saw little reason why case-crossover de signs might be useful in absence of those biases, such as in the context of database studies of the health effects of air pollution. Such studies typically capture all deaths or hospi talizations in a population, minimizing selection bias. Air pollution data are comprehensive and collected independently from the health outcome data, so differential information bias is negligible. However, it rapidly became apparent that case-cross over studies do offer an additional research strategy for air pollution researchers, especially when data are collected from individual subjects. Even with no additional data besides what are already available in health and air pollution data bases, a case-crossover analysis can complement time-series analyses. By redefining the time scale so "time zero" is the onset of the outcome, a case-crossover analysis enables the investigator to directly conceptualize the causal relation, and may thereby reveal associations as well as problems that were initially overlooked in time-series analyses. For example, long-term, seasonal and weekly patterns of traffic density, air pollution and timing of disease onset make it necessary to control for time trends in exposure. In a case-crossover analysis, this can be achieved by design, ie, by selecting control intervals on the same day of week and from the calendar month as each case interval, and ignoring exposure data from other times. This controls confounding by time trends to the extent that the dimensions of the strata capture the temporal variation. In a time-series analysis, the same can be achieved by modeling, ie, by keeping the exposure data from all other times in the follow-up period and adding terms to a multivariate model to control for such time trends. If the model specification happens to be right, then using exposure data from other times preserves statistical information in the analysis and improves statistical precision. However, model ing can add misinformation if the modeling assumptions happen to be wrongeg, uncontrolled confounding can re sult from incorrectly specifying the shape of the longand short-term time trends. Some people understand and trust analyses more when confounding is controlled by design. Simple crude analyses from case-crossover studies-espe cially raw counts of events in case and control periods-are easy to understand, despite their involving control for many confounders by self-matching. Does a case-crossover analysis actually uncover new information, or does it just help investigators to perceive it? First, we must understand the nature of this question by contrasting a case-crossover study with a traditional case control study. The latter asks "Why them?" (Why did these people become cases whereas those people did not?). A Submitted 6 November 2007; accepted 8 November 2007. From the *Pharmaceutical Services Division, British Columbia Ministry of Health, Victoria, BC, Canada; tEpidemiology Department, Harvard School of Public Health, Boston, MA; tSchool of Health Information Science, University of Victoria, Victoria, BC; and ?Cardiovascular Epidemiology Research Unit, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, MA. Correspondence: Malcolm Maclure, University of Victoria, 1774 Armstrong Street, Victoria BC V8R 5S6 Canada. E-mail: malcolmmaclure@shaw.ca. Copyright C) 2008 by Lippincott Williams & Wilkins ISSN: 1044-3983/08/1902-0176 DOI: 10.1097/EDE.Ob013e318162afb9

Brief Report

A well-known heuristic in epidemiology is that non-differential exposure or disease misclassification biases the expected values of an estimator toward the null value. This heuristic works correctly only when additional conditions are met, such as independence of classification errors. We present examples to show that, even when the additional conditions are met, if the misclassification is only approximately non-differential, then bias is not guaranteed to be toward the null. In light of such examples, we advise that evaluation of misclassification should not be based on the assumption of exact non-differentiality unless the latter can be deduced logically from the facts of the situation.

Occupational Noise Exposure and Risk of Acoustic Neuroma

A small number of prior epidemiologic studies of occupational noise exposure based on self-report have suggested an association with acoustic neuroma. The goal of the present study was to further examine the association between noise exposure and acoustic neuroma by using an objective measure of exposure in the form of a job exposure matrix. A total of 793 acoustic neuroma cases aged 21-84 years were identified between 1987 and 1999 from the Swedish Cancer Registry. The 101,756 controls randomly selected from the study base were frequency matched to cases on age, sex, and calendar year of diagnosis. Occupational information, available for 599 of the cases and 73,432 of the controls, was obtained from censuses and was linked to a job exposure matrix based on actual noise measurements. All risk estimates were close to unity, regardless of noise exposure level or parameter. The overall odds ratio for exposure to > or = 85 dB of noise was 0.89 (95% confidence interval: 0.64, 1.23). Contrary to previous study results, the present findings did not demonstrate an increased acoustic neuroma risk related to occupational noise exposure even after allowing for a long latency period. The effect of nondifferential misclassification of exposure must be considered a potential cause of the negative findings.

Bias magnification in ecologic studies: a methodological investigation

BackgroundAs ecologic studies are often inexpensive to conduct, consideration of the magnitude and direction of ecologic biases may be useful in both study design and sensitivity analysis of results. This paper examines three types of ecologic bias: confounding by group, effect measure modification by group, and non-differential exposure misclassification.MethodsBias of the risk difference on the individual and ecologic levels are compared using two-by-two tables, simple equations, and risk diagrams. Risk diagrams provide a convenient way to simultaneously display information from both levels.ResultsConfounding by group and effect measure modification by group act in the same direction on the individual and group levels, but have larger impact on the latter. The reduction in exposure variance caused by aggregation magnifies the individual level bias due to ignoring groups. For some studies, the magnification factor can be calculated from the ecologic data alone. Small magnification factors indicate little bias beyond that occurring at the individual level. Aggregation is also responsible for the different impacts of non-differential exposure misclassification on individual and ecologic studies.ConclusionThe analytical tools developed here are useful in analyzing ecologic bias. The concept of bias magnification may be helpful in designing ecologic studies and performing sensitivity analysis of their results.

Methodological issues regarding confounding and exposure misclassification in epidemiological studies of occupational exposures

Confounding and exposure misclassification are issues that concern epidemiologists because of their potential to bias results of studies and complicate interpretations. In occupational epidemiology both are routinely raised to argue that an observed result is either a false positive or a false negative finding. Although it is important to consider the potential for limitations of epidemiologic investigations, judgment regarding their importance should be based on their actual likelihood of occurrence. This paper is based on our experience in epidemiologic analyses and a brief review of the literature regarding confounding and exposure misclassification. Examples of substantial confounding are rare in occupational epidemiology. In fact, even for studies of occupational exposures and lung cancer, tobacco-adjusted relative risks rarely differ appreciably from the unadjusted estimates. This is surprising because it seems the perfect situation for confounding to occur. Yet, despite the lack of evidence that confounding is a common problem, nearly every epidemiologic paper includes a lengthy discussion on uncontrolled or residual confounding. On the other hand, exposure misclassification probably occurs in all studies. The only question is, how much? The direction and magnitude of nondifferential exposure misclassification (the type most likely to occur in cohort studies) on estimates of relative risks can be largely predicted given knowledge on the degree of misclassification, that is, relatively small amounts of misclassification can bias relative risks substantially towards the null. The literature, however, is full of discussions implying that misclassification of exposure is an explanation for a positive finding. These comments are not to suggest that all potential limitations for epidemiologic studies should not be considered and evaluated. We do believe, however, that the likelihood of occurrence and the direction and magnitude of the effect should be more carefully and realistically considered when making judgments about study design or data interpretation.

Residential mobility patterns and exposure misclassification in epidemiologic studies of birth defects

Many studies of environmental exposures and birth defects use mothers' addresses at delivery as a proxy for the exposure. The validity of these studies is questionable because birth defects generally occur within 8 weeks of conception and the mother's address at delivery may differ from her address early in pregnancy. In order to assess the extent of this bias, we examined the pattern of maternal residential mobility over the span of 3 months prior to conception through delivery, and associated maternal socio-demographic characteristics. We linked Texas subjects from a national case-control study of birth defects with their corresponding records from the Texas Birth Defects Registry and the Texas live birth certificates. Logistic regression analyses were conducted to assess maternal socio-demographic factors related to mobility during pregnancy. Overall, 33% of case and 31% of control mothers changed residence between conception and delivery. The pattern of mobility was similar for both case and control mothers for each pregnancy period. Multivariate analyses indicated that for case mothers, older age (OR=0.39, 95% CI=0.21-0.70), higher household income (OR=0.35, 95% CI=0.18-0.68), Hispanic ethnicity (OR=0.64, 95% CI=0.44-0.92), and higher parity (OR=0.59, 95% CI=0.38-0.94) were indicators of lower mobility during pregnancy. For control mothers, the same pattern of association was present, however, only older age was significantly associated with low rates of mobility. Studies of birth defects using maternal address at delivery as a proxy for maternal environmental exposures during pregnancy may be subject to considerable nondifferential exposure misclassification due to maternal mobility during pregnancy.

Air pollution and childhood cancer: A review of the epidemiological literature

The authors evaluated support in the literature for the hypothesis that ambient air pollution causes childhood cancer. The PubMed database was searched for original articles, which were reviewed for evidence of a relation with the main types of childhood cancer, using criteria including sample size, magnitude and precision of relative risk estimates, presence of a dose-response pattern and potential for bias. The hypothesis has been studied almost entirely with respect to traffic-related air pollution. Since derivation of the hypothesis from 2 case-control studies in Denver, USA, two further case-control studies have provided new positive evidence and 4 case-control and 7 ecological studies mainly negative evidence. The 4 case-control studies providing positive evidence were relatively small and tended to have more methodological limitations than those showing no association. Publication bias is possible. The weight of the epidemiological evidence indicates no increased risk for childhood cancer associated with exposure to traffic-related residential air pollution. Nevertheless, the limited number of studies, the methodological limitations of both positive and negative studies and the absence of consistency in the results obviate a firm conclusion of no effect. In particular, nondifferential misclassification of exposure might have masked true, weak associations.

Curious phenomena in Bayesian adjustment for exposure misclassification

Many epidemiologic investigations involve some discussion of exposure misclassification, but rarely is there an attempt to adjust for misclassification formally in the statistical analysis. Rather, investigators tend to rely on intuition to comment qualitatively on how misclassification might impact their findings. We point out several ways in which intuition might fail, in the context of unmatched case-control analysis with non-differential exposure misclassification. Particularly, we focus on how intuition can conflict with the results of a Bayesian analysis that accounts for the various uncertainties at hand. First, the Bayesian adjustment for misclassification can weaken the evidence about the direction of an exposure-disease association. Second, admitting uncertainty about the misclassification parameters can lead to narrower interval estimates concerning the association. We focus on the simple setting of unmatched case-control analysis with binary exposure and without adjustment for confounders, though much of our discussion should be relevant more generally.

Proper interpretation of non-differential misclassification effects: expectations vs observations

Many investigators write as if non-differential exposure misclassification inevitably leads to a reduction in the strength of an estimated exposure-disease association. Unfortunately, non-differentiality alone is insufficient to guarantee bias towards the null. Furthermore, because bias refers to the average estimate across study repetitions rather than the result of a single study, bias towards the null is insufficient to guarantee that an observed estimate will be an underestimate. Thus, as noted before, exposure misclassification can spuriously increase the observed strength of an association even when the misclassification process is non-differential and the bias it produced is towards the null. We present additional results on this topic, including a simulation study of how often an observed relative risk is an overestimate of the true relative risk when the bias is towards the null. The frequency of overestimation depends on many factors: the value of the true relative risk, exposure prevalence, baseline (unexposed) risk, misclassification rates, and other factors that influence bias and random error. Non-differentiality of exposure misclassification does not justify claims that the observed estimate must be an underestimate; further conditions must hold to get bias towards the null, and even when they do hold the observed estimate may by chance be an overestimate.

The Need for a Systematic Approach to Complex Pathways in Molecular Epidemiology

Thinking about biochemical pathways has become an increasingly important part of molecular epidemiology. The field is rapidly moving from evaluation of single candidate genes, one at a time, to consideration of entire pathways comprising perhaps dozens of genes and their environmental substrates,

The effects of joint misclassification of exposure and disease on the attributable risk

While there is extensive methodological literature analysing the effects of misclassification on the relative risk under various misclassification scenarios, for the attributable risk only the effects of non-differential misclassification either of exposure or disease, and the effects of non-differential independent misclassification of exposure and disease have been discussed for the 2 x 2-situation. The paper investigates the effects of non-differential correlated misclassification of exposure and disease on the attributable risk taking possible correlations of both types of misclassification into account. Furthermore, a comparison with the corresponding effects on the relative risk is drawn. We propose a matrix-based approach to describe the underlying structure of non-differential misclassification. The bias arising from non-differential misclassification in the attributable risk and relative risk is evaluated in four examples assuming under- or overreporting of exposure and disease. In each of the four examples we found scenarios where pronounced differences in degree and, more importantly, in direction of bias occurred. Our results clearly demonstrate the danger lying in the stereotype transfer of findings regarding misclassification effects on the relative risk to other epidemiologic risk measures and underline the necessity of specific analyses of the effects of misclassification on the attributable risk.

Studies with low response proportions may be less biased than studies with high response proportions.

The association between the response proportion in epidemiologic studies and nonresponse bias is complicated, because exposure prevalences and misclassification errors may vary by recruitment wave. In this paper, the authors illustrate the effect of varying degrees of wave-specific nondifferential exposure measurement error in a dichotomous exposure on the relative risk in a hypothetical cohort study of 5,000 participants, by recruitment wave. The field phase of the hypothetical cohort study consisted of five consecutive recruitment waves. The authors assigned response proportions to each wave (wave 1: 30%; wave 2: 10%; wave 3: 10%; wave 4: 30%; wave 5: 20%) and studied three different wave-specific patterns of true exposure prevalence: 1) the true exposure prevalence remains the same in all waves; 2) the true exposure prevalence increases by wave; and 3) the true exposure prevalence decreases by wave. The authors assumed three corresponding patterns of nondifferential misclassification error in exposure status across waves. If the nondifferential exposure misclassification increases by wave, the cumulative relative risk estimate is increasingly biased towards the null. This bias is intensified if the true exposure prevalence increases by wave. Studies with low response proportions may be less biased than studies with high response proportions if the nondifferential misclassification error in a dichotomous exposure increases by recruitment wave.

Historical pesticide exposure in California using pesticide use reports and land-use surveys: an assessment of misclassification error and bias.

We used California's Pesticide Use Report (PUR) and land-use survey data to conduct a simulation study evaluating the potential consequences of misclassifying residential exposure from proximity to agricultural pesticide application in health effect studies. We developed a geographic model linking the PUR with crop location data from land-use surveys to assess the impact of exposure misclassification from simpler exposure models based solely on PUR or land-use data. We simulated the random selection of population controls recruited into a hypothetical case-control study within an agricultural region. Using residential parcel data, we derived annual exposure prevalences, sensitivity, and specificity for five pesticides and relied on the PUR plus land-use model as the "gold standard." Based on these estimates, we calculated the attenuation of prespecified true odds ratios (ORs), assuming nondifferential exposure misclassification. True ORs were severely attenuated a) when residential exposure status was based on a larger geographic area yielding higher sensitivity but low specificity for exposure, in contrast to relying on a smaller area and increasing specificity; b) for less frequently applied pesticides; and c) with increasing mobility of residents among the study population. Considerable effect estimate attenuation also occurred when we used residential distance to crops as a proxy for pesticide exposure. Finally, exposure classifications based on annual instead of seasonal summaries of PUR resulted in highly attenuated ORs, especially during seasons when applications of specific pesticides were unlikely to occur. These results underscore the importance of increasing the spatiotemporal resolution of pesticide exposure models to minimize misclassification.

Exposure to asbestos and lung and pleural cancer mortality among pulp and paper industry workers.

We studied the mortality from lung and pleural cancers in a cohort of 62,937 male workers employed for at least 1 year in the pulp and paper industry in 13 countries during 1945 to 1996. Mill departments were classified according to probability and level of exposure to asbestos on the basis of available dust measurements and mill-specific information on exposure circumstances. Thirty-six percent of workers were classified as ever exposed to asbestos. Standardized mortality ratios of lung cancer were 0.99 (95% confidence interval [CI], 0.90 to 1.08) among unexposed and 1.00 (95% CI, 0.90 to 1.11) among ever exposed workers. The number of pleural cancer deaths among unexposed workers was 10; that among exposed workers was 14, most of which occurred among maintenance workers. In internal analyses, a trend in mortality from either neoplasm was suggested for estimated cumulative exposure to asbestos, weighted for the individual probability of exposure within the department and for duration of exposure (relative risk for lung cancer for 0.78+ f/cc-years, as compared with < or = 0.01 f/cc-years: 1.44; 95% CI, 0.85 to 2.45; corresponding relative risk for pleural cancer: 2.43; 95% CI, 0.43 to 13.63). Despite a possible nondifferential misclassification of exposure and outcome, this study suggests that the carcinogenic effect of asbestos can be detected among workers employed in industries such as the pulp and paper industry, in which it is not considered to be a major hazard.

Commentary: does the spectre of ecologic bias haunt epidemiology?

Beginning students of epidemiology learn that non-differential exposure misclassification biases results (on average) toward the null. Ecologic studies are one of the most interesting exceptions. As shown by Brenner et al. in 1992, ecologic studies are biased away from the null when exposure is non-differentially misclassified with the same sensitivity and specificity in each group. 1 In this issue, Bjork and Stromberg 2 extend this result in an interesting and disconcerting direction: ecologic bias in an important type of occupational study. In an individual-level study, investigators measure outcome, exposure and covariates for each subject. In contrast, ecologic studies measure these variables at the group level. Mixed designs are also possible. For example, one might measure exposure at the group level but outcome and covariates at the individual level. Such partially ecologic studies are actually rather common. In occupational and environmental epidemiology measurement of exposure for each individual is often infeasible. Instead, exposure is assigned based on group membership, for example job title or pollution as measured at central monitors. These types of partially ecologic studies have been designated semi-individual by some authors. 3,4

Implications of Nondifferential Misclassification on Estimates of Attributable Risk

Only the effects of isolated nondifferential misclassification of exposure or disease on the estimates of attributable risk have been discussed in the literature. The aim of this paper is to broaden the spectrum of scenarios for which implications of misclassification are available. For this purpose, a matrix-based approach allowing a comprehensive, unified analysis of various structures of misclassification is introduced. The relative bias or--in the situation of covariate misclassification--the relative adjustment are presented for the different misclassification scenarios. Under nondifferential misclassification of exposure or disease, the attributable risk is biased towards the null with the only exception of perfect sensitivity of exposure classification or perfect specificity of disease classification both leading to an unbiased attributable risk. From these two marginal effects, the consequences of simultaneous nondifferential independent misclassification of exposure and disease on the attributable risk are derived in the matrix-based approach. Misclassification of a dichotomous covariate leads to partial adjustment. To a large extent, the results for the attributable risk are in accordance with the well-known results for the relative risk. The algebraic differences between the two risk measures, however, make it necessary to repeat the methodological considerations for the attributable risk.

Pesticides and Fetal Death due to Congenital Abnormalities

The authors reply: Acquavella and Burns raise two issues regarding exposure assessment in studies of residential proximity to agricultural pesticide applications and birth outcomes: the magnitude of drift from pesticide applications to residential areas, and whether a biologically-active dose level would harm vegetation and insect populations. Several studies have shown that, when compared to the general population, individuals living near agricultural crops have an increased exposure to agricultural pesticides due to drift at the time of application. 1–3 These studies have measured detectable levels of pesticides in house dust from homes located within mile of agricultural crops. However, detection decreased as the distance between the homes and crops increased. In our study, we assigned proximity to exposure by using the “TRS” (township, range, and section) from the USGS (US Geological Survey) to define a fixed unit of one square mile. 4 If a woman’s residence was near the edge of the TRS in which she lived, and an application occurred near her home but in the adjacent TRS, then our narrow definition of exposure would classify her as “unexposed.” Therefore, we used the broad definition, which included her TRS of residence and the eight surrounding TRSs, in order to increase the sensitivity of the exposure categorization. By using this definition, we could have misclassified some individuals outside the range of drift as exposed. However, given that nondifferential exposure misclassification tends to bias estimates toward the null, it is unlikely that the observed increase in risk was due to exposure misclassification. At the time of our study, we were limited to a resolution of 1 square mile. Recently, geographic-based exposure metrics have been developed that can be linked to the California pesticide use database in order to more precisely estimate the proximity of agricultural pesticide applications to residences. 5 These new methods could serve as valuable exposure assessment tools in future studies. The authors also suggest that the validity of our exposure model is dependent upon having an environmentally relevant dose (ie, the insect population and vegetation between the application and residence would be noticeably damaged). Given that the agricultural pesticides are applied to crops in order to control for insects or weeds, we can assume that this criterion would be met for mothers living in very close proximity to agricultural crops. However, this criterion assumes that the biologically and environmentally relevant doses are equal. Several studies have shown that the risk of adverse birth outcomes (birth defects in particular) is a factor of both dose and fetal age at the time of exposure. 6–8 Thus, we cannot rule out the possibility that the environmentally safe dose is greater than the biologically relevant dose for a fetus, particularly when the exposure occurs during a vulnerable period of fetal development. We agree with Acquavella and Burns that studies of residential proximity to agricultural pesticide applications would benefit from an improved measure of exposure. Nevertheless, our study represented a significant improvement in the exposure assessment over the majority of studies evaluating reproductive health effects of pesticides. Geographic proximity was not subject to recall bias, and gestational age at the time of exposure was known. As discussed above, methods that incorporate geographic mapping techniques in conjunction with pesticide application data have been developed that could improve the specificity of exposure classification. Finally, biomonitoring studies that characterize pesticide exposure due to drift, controlling for variability in wind and weather conditions, would lead to better understanding of environmental exposure to agricultural pesticides. Erin M. Bell Irva Hertz-Picciotto James J. Beaumont

Locational uncertainty in georeferencing public health datasets

The assignment of locational attributes to a study subject in epidemiologic analyses is commonly referred to as georeferencing. When georeferencing study subjects to a point location using their residential street address, most researchers rely on the street centerline data model. This study assessed the potential locational bias introduced using street centerline data. It also evaluated georeferencing effects on a location-dependent, exposure assessment process. For comparison purposes, subjects were georeferenced to the center of their residential parcel of land using digitized parcel maps. A total of 10,026 study subjects residing in Jefferson County, Alabama were georeferenced using both street centerline and residential parcel methods. The mean nondirectional, linear distance between points georeferenced using both methods was 246 ft with a range of 11 to 13,260 ft. Correlation coefficients comparing differences in exposure estimates were generated for all 10,026 subjects. Coefficients increased as the geographic areas of analysis around study subjects increased, indicating the influence of nondifferential exposure misclassification.

Birth defects and paternal occupational exposure. Hypotheses tested in a record linkage based dataset

MAIN QUESTION: To test previously established hypotheses on associations of birth defects with paternal occupation on the basis of a Norwegian registry material. The study comprised all births in Norway 1970 -1993 for which linkage with population censuses 1970, -80 and -90 on parents' job title could be obtained--about 1 million births (75% all births). The reference population was offspring of the group that did not belong to the actual occupation. Vehicle mechanics had an association with hypospadias--OR 5.19 (CI 1.31-14.24), painters had a non-significant association with spina bifida--OR 2.03 (CI 0.99-3.75) and printers with club foot--OR 1.61 (CI 0.89-2.90). Associations observed previously in off-spring of fathers in large occupational groups such as teachers, drivers, electricity related occupations, sales related occupations and agricultural workers were not confirmed in this dataset. The study gave further evidence of cause effect relationships in the confirmed positive associations, though without any clarification of possible mechanisms involved. Possible false negative findings might be caused by low statistical power due to small occupational groups or non-differential misclassification of exposure.