METHODS

The design was a cross sectional survey, but with retrospective assessment of exposures and outcomes over respondents’ reproductive histories, which approximates a retrospective cohort study, albeit based on current workers only. Data collection took place in four European countries. We approached companies whose employees are exposed to lead, and non-lead using companies as a control group, with a view to participation in the research; 28 companies agree to take part (for details, see table 1).

Data collection took place in 1997–98 (except for Finland: 1995–96). Men were eligible for the study if they had ever been married (or living as married). The Asclepios Project questionnaire¹⁸ was developed for the purpose of the study. The method of data collection varied according to local conditions, and included telephone and face-to-face interviews, postal questionnaires, and consultations carried out by the factory medical service. We attempted to identify non-eligible subjects among the non-respondents, using a one page questionnaire.

TTP was measured as the reply to the question, “How long did it take your wife/partner to get pregnant?”, preceded by filter questions on contraceptive behaviour. The Asclepios questionnaire focuses on the most recent birth (youngest child) for detailed questioning, as it is the most likely to have occurred during employment in the selected industry. The questionnaire was originally written in English and translated as necessary. Only pregnancies that resulted in births were included, as in the case of miscarriage or other pregnancy loss recall may be less good for year of occurrence and covariates as well as for TTP.²⁰ TTP values were eligible for inclusion if conception was reported to be not as a result of contraceptive failure, and if the target time (see below for definition) occurred after 1975 as insufficient information is available on lead exposure before then. Since TTP is a property of the couple, confounding factors were included for both male and female partners.

The exposed group consisted of those who were working in a lead exposed industry at the target time. There were two non-exposed populations: (a) external control group: men recruited in industries without lead exposure; (b) internal control group: men recruited through companies working with lead, but for whom the timing indicated that lead exposure could not have influenced their fertility when their youngest child was conceived.

Blood lead measurements were available for all the exposed groups as far back as the late 1970s, and the laboratories were subject to quality control programmes that would have minimised any spatial or temporal variation. Results are presented for three exposure models: (1) short term, reflecting recent exposure; (2) duration of exposure irrespective of its level; and (3) cumulative, a composite measure that integrates level and duration of exposure. For (1) and (3), exposure assessment was based on blood lead measurements when available, and imputed values when the measurements were missing but there was sufficient other information for an occupational hygienist to reliably allocate an exposure category (n = 78). This imputation was based on information on the place where the man worked, together with information on the blood lead levels prevailing there at each time.

In each case, the timing of exposure has to be considered relative to the period of unprotected intercourse leading to that particular conception. The duration of spermatogenesis and spermiogenesis in the human is approximately three months. We assumed that the probability of an effect of lead exposure was equally distributed throughout this period, and the “target time” was therefore defined as six weeks before the beginning of unprotected intercourse (fig 1).

For the short term measure, the mean of up to four blood lead values was used: the closest value to the target time, if within 6 months; the next value after that, if within 12 months; and the two values before the closest value, if within 24 months. The use of up to four values was designed to produce a more stable estimate than would be possible with only one; the asymmetry of timing is designed to reflect the time during which lead might exert an effect. These short term mean values were analysed in terms of four exposure categories: <20 μg/dl, 20–29 μg/dl, 30–39 μg/dl, and 40+ μg/dl.

Duration was measured from the month/year of starting work in the industry to the target time, ignoring any gaps (that is, taking them as being “exposed”). The cumulative measure was taken as the sum of the mean value in each year multiplied by the duration of exposure in that year, that is:

where y_i = the length of exposure in year i (up to 1, =12 months). The reason for taking means within a fixed time period (years in this instance) is to remove any possible effect of the density of measurements, as blood lead values may be determined once yearly, twice yearly, quarterly, or more often, with more frequent measurement typically being triggered by high values.

Statistical analysis

A Cox proportional hazards model with discrete ties was used.²⁷ This generates a hazard ratio; higher values (>1) indicate higher fertility. The results are presented separately with the external and internal control groups as baseline, both as unadjusted analyses and as multivariate analyses, adjusting for the effects of the following covariates: country, mother’s age, mother’s smoking, and whether it was the mother’s first pregnancy. Father’s age and smoking were considered for inclusion, but these variables were not associated with TTP, and were therefore not included in the models. The use of oral contraception was not included in the formal model, as the contraceptive pill tends to reduce fertility specifically in the early months after discontinuation of use, violating the assumption of proportional hazards. Therefore, as alternate analyses, separate models were estimated for men whose wives had or had not taken the pill immediately prior to commencing unprotected intercourse. In addition, the models were repeated after exclusion of all men with imputed blood lead data.

A number of precautions were taken to ensure unbiased analyses. First, the analysis was censored at 14 months (14 rather than 12 months because of digit preference), because values become sparse as TTP increases, making estimation of the upper tail of the TTP distribution unreliable. This also avoids problems due to the possible effects of fertility treatment.¹⁸ In addition, recall of the duration of relatively long periods of infertility is less accurate.²⁰ Alternate analyses with censoring at 10 or 7 months were also carried out, to check whether the findings were sensitive to this.

Secondly, TTP values of 0 and 1 were grouped together, as it may be difficult for respondents to make a clear distinction when conception occurs very quickly; analyses were also repeated without this grouping of values. The following additional checks were carried out, to ensure that there were no unrecognised problems: (a) the proportion of pregnancies categorised as contraceptive failures was tabulated against exposure category, in case the discrimination of contraceptive failures from very quick conceptions happens to vary with lead category; (b) for the same reason, the basic analysis was repeated after deletion of all cases with a TTP of 0 or 1, as recommended by Weinberg and colleagues.²⁸

Finally, it was necessary to avoid truncation bias,²⁹ due to the imposition of a calendar time cut off date (when data were collected), after which pregnancies are no longer included. This is a selection effect, affecting who is included in the analysis, and is completely separate from the issue of censoring. For example, a man who was interviewed in May 1997 and who had commenced unprotected intercourse with his wife in May 1996 would have a pregnancy only if the couple conceived within 12 months; the less fertile couples are thus excluded from the population at risk. This bias has the effect of artificially overestimating fertility in relation to births in the most recent category. As lead exposure has decreased during the relevant period, blood lead is highly correlated with calendar time, and truncation bias would be expected to create an artefactual tendency towards higher fertility in those with lower blood lead values. To ensure that the findings were not distorted by this bias, an alternate analysis was carried out excluding all those TTP values that related to a target time later than 1994. This cut off date was chosen because there are few instances of TTP values beyond 2–3 years, so that the bias becomes small at the population level. There was no equivalent bias at the beginning of the period studied, as inclusion was defined by the target time, not time of conception.

RESULTS

Table 1 lists the types of product covered by the study. Indications of the range of company sizes and duration of employment are also given. Mean values of blood lead were approximately 30 μg/dl in each of the participating countries. Although values up to 70 μg/dl were observed, such high values were rare, and levels of blood lead above 50 μg/dl were uncommon.

Response rates varied considerably: country specific rates varied from little over 50% to 85% or more. The attempt to distinguish non-responders from those known to be non-eligible was not very successful. Where doubt existed, we have assumed non-responders to have been eligible, thus inflating the denominators. The response rates shown in table 1 may therefore be misleadingly low.

Out of 2293 questionnaires, 1899 men reported that they had fathered a child; 303 of these were reported to have followed a contraceptive failure, and of the remaining 1596, the TTP value was unknown (n = 139), or the target time was missing (n = 2) or occurred prior to 1976 (n = 351), leaving 1104 values for analysis (fig 2). Of these, 236 were from the external control group and 230 from the internal control group.

Table 2 tabulates respondents’ characteristics in the different exposure groups (using the short term measure), including the two control groups. Men in the exposed groups tended to be older at the target time than those in the control groups, especially the internal control group, and this was true also of their wives. Some heterogeneity was noted between the exposure categories in the proportion of first pregnancies, smoking by either partner, and prior use of oral contraception, but no systematic pattern could be identified. Genitourinary disorders and treatment for infertility were infrequent, and did not vary systematically with exposure group.

Table 3 shows unadjusted TTP distributions by exposure category using the short term measure. The external control group and those with missing blood lead values were found to have a distribution that suggests considerably lower fertility than the other lead exposed workers including the internal control group. This is reinforced by a higher proportion of phases of infertility not leading to a pregnancy in the two former categories (8% and 9%, respectively), compared with 3–5% in the lead exposed groups and internal control group. However, these same two categories also reported a high proportion of contraceptive failures, so that part of their apparent difference from the rest of the population may be behavioural rather than biological.

The same results are given in the top half of table 4, in the form of a Cox proportional hazards model with discrete ties, adjusted for country (fertility was higher in Italy than in northern Europe). The second half of table 4 shows the same findings, adjusted also for maternal age, smoking, and first pregnancy. These confirm the apparently lower fertility of the external control group, which remains after adjustment. On the basis of the internal control group, no consistent or statistically significant decrease in fertility was observed in the exposed groups. The suspicion of a weak trend of decreasing fertility with increasing blood lead level within the exposed groups (table 3) disappeared when adjustment was made for country. The findings were closely similar when alternative measures of exposure were used, either the highest of the four values as previously defined, or the closest single value.

Tables 5 and 6, in the same format as table 4, present the findings when the exposure measures used were respectively duration and cumulative lead exposure. Again, the external exposure group differed from all the lead workers. Within the latter group, men in the third category had lower fertility than those in the other three categories, but no clear pattern of association was seen between lead exposure and fertility.

The findings were unchanged when each of the alternate and sensitivity analyses was carried out. In addition, repeating the analysis after exclusion of all worksites with a response rate less than 75% gave virtually identical results.

DISCUSSION

This study found no evidence that low fertility is associated with the levels of occupational lead exposure that currently exist. We need to consider whether these negative findings could be due to misclassification of either exposure or outcome, to insufficient numbers or to some other type of bias.

Retrospective epidemiological studies of environmental or occupational exposures are frequently handicapped by inadequate exposure measurements. In this instance, we were able to make use of the blood lead values that have been a statutory requirement for all worksites with exposure to lead in the European Community since the late 1970s. Misclassification of exposure is therefore less likely to have been a problem in this study than in most others of the same design.

In the case of TTP, validation studies have shown that misclassification occurs,^20–23 but that it is non-differential. Nevertheless, recall bias could have occurred in this population, and could possibly have influenced the results. However, it is difficult to explain why reporting of TTP should be distorted in any particular direction by the degree of lead exposure of the respondents. In addition, the expected associations of TTP with maternal age, smoking habit, and recent use of the contraceptive pill were found, which increases the confidence in our data.

The sample size was greater than 100 in most of the exposure groups. While a larger number would have been desirable, it seems unlikely that this would have altered the findings, but this cannot be ruled out.

It could be that response bias has influenced the results. A low response rate in many of the worksites, especially in England, was partly due to restrictions placed on the fieldwork; for example, men having to participate during their free time, and partly to the prevailing attitude that the researchers were looking for “volunteers”. An additional factor in the Belgian external control group was study fatigue, as several other studies and enquiries had immediately preceded this one. The motivation whether or not to take part in a workplace survey on fertility is likely to be related to fertility experience and/or to exposure status (although such problems are more to be expected in a study of semen quality for example, in which the men have the opportunity to be tested).³⁰ If men are more likely to have participated if they (a) have experienced fertility related problems, or (b) are aware of a possibly relevant occupational exposure, this would have the effect of biasing the results towards the null: those in the non-exposed groups would only be motivated to take part for reason (a), and would therefore be a selectively less fertile group for reasons unconnected with toxicological effects. The apparently low fertility of the external control group is compatible with this suggestion. On the other hand, the analysis that excluded worksites with low response rates (<75%) gave the same results as the full analysis, suggesting that this interpretation is less likely to be important.

Values of TTP were available only for pregnancies that ended in births; although TTP values tend to be longer for miscarriages,³¹ their exclusion does not bias the results.¹² In addition, only one pregnancy per couple was studied; this corresponds with the statistical method used, which assumes independence of events. Now that frailty models are available for survival analysis, it would be possible to increase the statistical power by including multiple pregnancies per couple, although this would require a longer and more complex questionnaire to obtain all the necessary information, including on covariates.

The results based on the external control group are more reliable, as comparison with the internal control group may be biased. For example, men who had been fertile before, but not after starting lead exposed work, would be represented in the internal control group despite unsuccessfully continuing unprotected intercourse after starting their new job.³²

On the other hand, because lead exposure declined over time, longer values of TTP will have been associated with an earlier period during which exposures were higher. The resulting bias would have led to an overestimation of the association of lead exposure and biological fertility.³³

Associations were observed between TTP and variables other than lead exposure, for example international differences, which have been observed previously.³⁴ An apparent difference was also found between men recruited through the lead industry and those recruited as the external control group, unrelated to exposure status. A differential healthy worker effect is unlikely to be the explanation, as (a) reduced fertility would not be directly associated with work status, and (b) in this age group, the prevalence of chronic diseases that could affect both fertility and the ability to work would be expected to be rather low. This unexplained observation underlines the current lack of understanding of the determinants of heterogeneity in human biological fertility.

Other studies of TTP in men exposed to lead at similar exposure levels have had mixed results. Apostoli et al found decreased fertility among men with blood lead concentrations of at least 40 μg/dl, but this was statistically significant only in a subgroup analysis restricted to subjects with just one child.³ Fertility was not at all reduced in men with blood lead concentrations in the range 30–40 μg/dl. Sallmén et al found lower fertility among all lead workers compared with non-exposed controls; within the lead exposed group, men with blood lead concentrations above 30 μg/dl had relatively low fertility, a finding that was almost statistically significant at the 5% level in the whole population, and highly significant when the analysis was confined to pregnancies that ended in a birth.¹² The reasons for the discrepancies between the findings in the different studies are unclear at this stage.

In conclusion, no detectable effects on male fertility were found at the levels of lead exposure current in the types of western European worksite represented by this study population. In the parallel study previously mentioned, which also formed part of the Asclepios Project, we examined semen quality in relation to current blood lead level. We found that sperm density was slightly reduced above a threshold that approximates to the highest exposure category in the present study.¹⁷ It would not be surprising that such a subtle change is not associated with an alteration in TTP.³⁵