Shoe manufacturing is a traditional industry in northern Portugal. The workforce of nearly 54 000 consists predominantly of young women. Some of the more than 100 process operations in shoe manufacturing are mechanised, but many are still manual. Shoe workers are exposed to adhesive solutions and finishing products that contain organic solvents. A survey conducted in Portuguese shoe manufacturing factories indicated that most of these workplaces have an inadequate air exhaust system at “gluing stations”, and almost half of the workers are exposed to organic solvents.1

Organic solvents are volatile liquids, and they or their metabolites are rapidly distributed by the circulating blood to various tissues, where they exert numerous acute and chronic toxic effects—for example, skin and mucous membrane irritation, toxic effects in the central nervous system and injurious action in the liver and kidneys.2 Epidemiological studies have linked female exposure to organic solvents to various adverse reproductive outcomes, including an increased risk of spontaneous abortions and congenital malformations.3^–9

Epidemiologic studies consistently show an association between exposure to solvents in general and reduced female fertility.10^–19 The risk is difficult to attribute to any individual solvents because the workers are commonly exposed to mixtures of solvents. However, there is evidence that toluene and ethylene glycol ethers may adversely affect female fertility.17 19 20

Studies nested within one industry have an advantage of more clearly defined solvent exposure or working conditions. In the present study we focus on the fertility of female shoe manufacturing workers who are exposed to the following organic compounds: n-hexane and hexane isomers, toluene, methyl ethyl ketone, acetone, ethyl acetate, dichloromethane and isopropyl alcohol. We use time to pregnancy (TTP) as a measure of fertility.21 TTP refers to the number of menstrual cycles or months that the women required to become pregnant. Validity studies suggest that women recall TTP adequately.22^–24

MATERIALS AND METHODS

TTP was studied among women working in five of the 55 shoe manufacturing units using solvents located in Gaia commune in northern Portugal (“exposed women”, n = 1655 workers). An unexposed comparison group consisted of women working in stores of food units (bakery, ice cream) and storehouses located near each shoe manufacturing unit (n = 753 workers). From each group, we randomly selected 250 women who fulfilled the following criteria: (1) has been married/cohabiting between 1986 and 1997 (1340 exposed women and 655 unexposed women); (2) worked day shifts (no exclusions); (3) were Caucasian (no exclusions); and (4) had ever been pregnant (no records available on exclusions). We selected the first TTP during the study period of 1 September 1986 to 31 December 1997 that started after the woman’s entry into shoe manufacturing or comparison work. This retrospective design with features of prospective design has been described25 and used16 previously. The study was performed in accordance with the Declaration of Helsinki of 1975 (revised in 1988) with all subjects having given their informed consent. The study was approved by the Ethics Commission of the Portuguese National Institute of Health.

Interview

Face-to-face interviews were conducted at the workplaces during October to December 1997 by doctors trained in occupational health and public health technicians. The mean duration of the interviews was about 50 min. The number of interviewed women and exclusions are shown in table 1. The participation rate was 92%; 36 women (20 exposed and 16 unexposed) were excluded due to various reasons, and only 16 women (10 exposed and 6 unexposed) were not able to report their TTP. The final study population consisted of 197 exposed women and 209 unexposed women.

Contraceptive failures were defined as continuing regularly or almost regularly used contraception until conception. Those who did not use any contraception at the time of conception were asked: “Did you get pregnant during the first menstrual cycle after starting sexual intercourse without birth control?” The following alternatives were given: (1) Yes; (2) I became pregnant during the second cycle; and (3) The pregnancy started later. Women who marked the third alternative were asked: “How long did it take for you to get pregnant? ____ (years) ____ months”. Ninety-one percent of both the exposed and unexposed women were certain of their waiting time. The women were asked about their last contraceptive method and lifestyle factors (smoking, consumption of coffee and alcohol) during TTP. We sought information on reproductive, medical and surgical histories of the women. The women were asked to mark the occupational exposures or work tasks of their husband/partner from a list of 17 items. For example, solvents, metals, welding, pesticides and heat were included. Age was defined for TTP starting time.

Exposure assessment

Assessment of exposure to organic solvents was based on air monitoring according to traditional industrial hygiene methodology; the measurements were conducted in 1998 and 1999. Air sampling was performed in the personal breathing zones of the exposed women, spanning roughly an 8-hour work shift. Personal samples and static samples, when four or more women worked together seated around the same table, were collected by drawing air through standard sized coconut shell charcoal tubes (SKC 226-01), with SKC constant low-flow personal pumps (SKC MODEL 22-3 and Gilian-GilAir multiflow) operating at a flow rate of 200 ml/min.

Measurements were taken during several days, and air sampling was carried out for representative working periods (minimum 90 min for each sample). During the field survey, 603 personal and 58 static air samples were collected and analysed according to the methods of the National Institute for Occupational Safety and Health (NIOSH) (methods 1300, 1400, 1500/1 and 1550).26

The industrial unit’s registries did not include any retrospective data on the workers’ exposure to organic solvents. However, the databases on purchasing management and production volumes, as well as the work process flow sheets indicated no essential changes in the use of solvents since 1986. Also, shoe materials and production volume have been similar. Managers and managing directors confirmed that no changes had been introduced to the buildings, ventilation systems and work practices during the study period. According to the doctors trained in occupational health, in the early 1990s, some training was organised on safety issues in three units that were not included in our study, and no information about “solvents and health” was provided.

Eight different organic compounds were present in the collected air samples: n-hexane and hexane isomers, toluene, methyl ethyl ketone, acetone, ethyl acetate, isopropyl alcohol and dichloromethane. n-hexane, hexane isomers and toluene were used in all the shoe factories studied. Methyl ethyl ketone and acetone (in 85% of the units), ethyl acetate (in 50%) and dichloromethane (in 33%) were also used frequently.

The occupational exposure of the women to organic solvents was evaluated according to the criteria of the American Conference of Industrial Hygienists (ACGIH)—threshold limit values (TLVs).27 The time-weighted average (TWA) of solvent vapours was calculated for the different organic compounds in each workplace. Due to the simultaneous exposure to mixtures of solvents, we used the general ACGIH mixture formula to calculate the exposure index (EI) for each exposed woman. EI is the sum of the ratios between the TWA of each compound found at the workplace and its TLV value. The EI TLV is 1.0. In our data, the EI ranged from 0.01 to 4.36 with a median value of 0.14. We divided the exposed women into two groups: a low exposure group, defined as an EI of 0.01–0.14, and a high exposure group, defined as an EI >0.14.

We estimated the duration of work both for exposed and non-exposed workers prior to the beginning of TTP by subtracting the year of starting employment from the year of starting TTP. We assessed the impact of the duration of exposure on fertility by separate analyses for those employed for a short time (<6 years) and for a long time (⩾6 years).

Statistical analyses

TTP data were analysed with discrete proportional hazards regression28 using the SAS PHREG procedure and EXACT handling of ties.29 TTP data were categorised into five groups: 1, 2–3, 4–6, 7–10 and ⩾11 months, in order to improve the model fit. We also estimated the length of TTP as menstrual cycles taking into account the menstrual cycle length. The findings did not change at all and we kept the outcome variable as months. The outcome parameter is a hazard ratio and will be called the “fecundability density ratio” (FDR). FDRs below unity reflect reduced fertility. The covariates associated with fertility in previous studies were considered in the analyses. Only those male variables are presented that showed at least a suggestive association with fertility. Variables not associated with fecundability were dropped from the full model if their exclusion did not change the association between solvent exposure and fecundability by at least 5%. Therefore, female smoking and female and male use of coffee were not included in the adjusted models. Pregnancy planning was strongly associated with improved fertility, and adjustment for it strengthened the association between solvent exposure and reduced fertility. However, the inclusion of this variable also violated the proportional hazard assumption (the association between pregnancy planning and fecundability varied across TTP distribution), and we therefore did not include pregnancy planning in the final model.

The association of solvent exposure and fecundability was examined across a series of sensitivity analyses. These included: (1) female age; (2) parity; (3) pregnancy outcome; (4) regularity of menstrual cycle; (5) duration of work; and (6) consideration of truncation bias (right truncation is a potential concern in our data: only short TTPs were possible at the end of the study period), planning bias (couples may vary in the consistency with which they use birth control), wantedness bias (possible differences in how couples respond to the question of birth control failure) and infertile worker effect (fertile women are absent from the workforce for some time after delivery). Possible interactions with solvent exposure were examined for female lifestyle factors.

RESULTS

The exposed and unexposed women differed with regard to many potential confounders (table 2). The exposed women were more likely (22%) than the unexposed women (10%) to have one or more previous children. On the other hand, exposed women also reported more often to have experienced difficulties in conceiving (19%) than unexposed women (11%). Long menstrual period (34% vs 12%), female consumption of ⩾7 alcoholic drinks a week (40% vs 16%) and use of coffee (73% vs 56%), as well as the husband’s smoking more than 15 cigarettes a day (42% vs 25%) and his consumption of ⩾7 alcoholic drinks a week (76% vs 57%) were more common among the exposed women than among the unexposed women. Pregnancy planners were highly fertile, as were, unexpectedly, men exposed to engine exhaust. Non-use of contraception before the studied TTP period and an irregular menstrual cycle were associated with reduced fertility.

The exposed women were less likely than the unexposed women to conceive in the first month of trying (21/35%). Also, exposed women were more likely to have waited for 7 months or longer to get pregnant (24/13%), but this difference no longer existed at 1 year or more of trying (about 6% in each group). In the age-adjusted model, female occupational exposure to solvents was associated with reduced fertility: FDR 0.64, 95% CI 0.50 to 0.83 for low; and FDR 0.76, CI 0.58 to 0.98 for high exposure.

In the adjusted model, female exposure to solvents was associated with reduced fertility (FDR 0.62, CI 0.48 to 0.80; and by level of exposure, FDR 0.56, CI 0.41 to 0.75 for low, and FDR 0.71, CI 0.53 to 0.96 for high exposure, respectively, table 3). An alternative exposure grouping yielded FDRs of 0.49, CI 0.31 to 0.79; 0.62, CI 0.46 to 0.80; and 0.82, CI 0.54 to 1.25 for EIs 0.01–0.09, 0.1–0.99 and ⩾1.0, respectively. The association between low solvent exposure and reduced fertility was consistent across the sensitivity analyses—for example, in the different age groups, according to pregnancy outcome, and in a subset excluding cycle-one pregnancies (table 3). The subset findings for high exposure also suggest an association, although the 95% CIs included unity in age group analyses. Splitting the data into three exposure categories did not change the tendency towards more elevated association for lower exposure. The risk of reduced fertility for high exposure was stronger when the duration of exposure was shorter, as compared with longer duration. The associations were slightly stronger among women with a regular menstrual cycle than in the whole data.

The length of maternity leave was 3 months in Portugal during the study period. However, women who have had many children may have left the workplace for years. Traditional analyses to guard against infertile worker effect (restriction to employed women or to primiparous women12) may not fully address this type of selection in our study. We obtained information on a very low annual turnover (1.2%) among all the workers (70% women) from the largest shoe factory unit in our study (126 exposed women), but there is no way to attribute this turnover to the number of children. To minimise the likelihood of this extensive infertile worker effect, we restricted our exposed group to workers from this unit. The adjusted FDRs were 0.53, CI 0.32 to 0.87, and 0.69, CI 0.44 to 1.10 for low and high exposure, respectively.

The association of solvent exposure with fecundability was modified by female smoking and drinking coffee (table 4). In the interaction model, solvent exposure was strongly associated with reduced fertility among those who did not use coffee. Also, female smoking and coffee drinking expectedly had a negative impact on female fertility among unexposed women. Highly exposed women who also drank coffee were highly fecund compared with unexposed women who did not drink coffee. Similarly, a high fecundability can be seen among exposed smokers, but the finding is based on small numbers. Also, all but four of the 47 smokers also drank coffee. Female use of alcohol did not modify the effect of solvent exposure.

DISCUSSION

Our finding of reduced fertility among women exposed to solvents is in accordance with several previous studies.10^–14 17^–19 However, two studies found no adverse effect of solvents.15 16 One needs to be cautious when comparing our findings with earlier studies. Solvents and exposure levels vary between studies. Interestingly, two studies with comparable exposure came up with similar findings.11 17 Exposure to toluene, a common solvent in the Portuguese shoe industry, was linked to reduced fertility in female workers in the printing industry.17 Also, a small group of shoe workers showed low fecundability in a Finnish study conducted among women biologically monitored for exposure to solvents.11 In that study, the most common exposures were toluene, hexane and acetone.

The strengths of our study are: (1) a large and homogenously exposed population as compared with many other studies; (2) the participation rate was high, and only a few women could not recall their TTP. Thus, we consider it unlikely that selective participation could have influenced the findings markedly. The studied TTPs occurred at most 11 years earlier, and recall of TTP is considered accurate at a group level even with a duration of recall up to 20 years.23 (3) Hygienic measurements conducted at the workplaces add to the reliability of exposure assessment. (4) Our data were not restricted to pregnancy planners (the association was stronger among pregnancy planners), rather, we excluded only contraceptive failures. Moreover, there were no difference in exposure in women experiencing contraceptive failure—a finding that suggests no essential role for planning bias.25 (5) We conducted several sensitivity analyses and found that the findings were most probably not related to wantedness bias (cycle-one exclusion29) or infertile worker effect (only employed women included, the consistent findings among primiparous subjects12 and focusing on low turnover). (6) The selected retrospective study design with features from prospective design (focus on first pregnancy after defined calendar time) reduces the potential for time-trend bias, and also renders the recall times comparable between the studied groups.

We considered several potential confounding factors. Solvent exposure was consistently related to reduced fertility across several sensitivity analyses, although the findings for high exposure were not statistically significant in all the subsets. Moreover, the restriction of the study to pregnancies ending in births, to primiparous women or to women with a regular or normal length menstrual cycle strengthened the findings. Thus our finding cannot be explained by the differences in these factors between exposed and unexposed women.

Some weaknesses in the study must be considered when interpreting the findings. Exposure to solvents was based on hygienic measurements at work departments in 1998 and 1999, and these were thus not taken during TTPs. Current hygienic measurements may not be adequate for the earliest years in our study. Therefore, the findings according to the level of exposure are of concern. However, based on information gathered from the workplaces, we assume that there were no essential changes over time in solvent exposure during the study period. Moreover, our findings remained virtually unchanged when we excluded the earliest years. On the other hand, the internal dose may differ regardless of a similar external dose, and the findings according to the level of exposure may therefore be diluted. All in all, our finding of reduced fertility among exposed women may not be explained by misclassification of exposure.

We found that the exposed and unexposed women differed as regards many potential confounders. However, we think that our findings are not related to these differences since adjustment for the known risk factors for reduced fertility slightly strengthened the association. Assessment of smoking was in part based on smoking history (years of starting and stopping smoking) and questions directly focused on TTP starting time. Women easily remember whether they have ever smoked regularly and whether they have stopped smoking. Still, the data on smoking and other potential confounders may be imprecise, and they leave room for residual confounding. However, we consider it unlikely that these misclassifications could have affected our findings to a great extent. Moreover, we lack information on frequency of intercourse. The exposed women may have been sexually less active than their unexposed counterparts. However, we see a stronger association in low exposure than in high exposure, and it is hard to understand why sexual activity would have been lower in the low exposure group. Also we think that residual confounding by factors unrecognised by us is a possible but unlikely explanation for the overall finding on reduced fertility among exposed women.

We see low fecundability among women exposed to solvents, but our findings do not show a traditional dose–response. In the following, we discuss the potential reasons behind this finding. It is now generally accepted that studies on TTP should have information on contraceptive failure and times not leading to pregnancy by exposure.30 Our study fulfils the first quality criterion, and we found that frequency of contraceptive failure was similar in exposed and unexposed women. However, we interviewed only ever pregnant women. In retrospective studies conditional on pregnancy, the lack of infertile couples may artificially mask the true effect of high exposure in particular.31 The most striking example of this bias is the increased fecundability with increasing female age.32

It has been hypothesised that a phenomenon called “dose–response fallacy” may occur in epidemiological studies on reproductive health when only recognised pregnancies are studied.33 Congenital malformations, spontaneous abortions, subclinical abortions, prolonged TTP and infertility fall on a continuum according to biological severity. If the risk of a biologically less severe health outcome falls due to the increasing risk of more severe health outcomes, along with increasing exposure, a non-traditional dose–response relationship may be observed. The findings for low exposure were stronger than those for high exposure, and we see only the minor effects of long-lasting high exposure. If solvent exposure strongly affects fecundability, and even causes infertility with higher or long-lasting exposure, or among susceptible or intrinsically sub-fecund women, then our findings are underestimates.

Another possibility for negative bias is that women who are highly sensitive to solvent exposure may have left the shoe industry before contributing to our data. However, data on low turnover are reassuring, and information collected during air monitoring campaigns indicates that exposed women have not been informed about occupational health issues and their training is concentrated on safety and prevention of accidents (Olga Mayan, personal communication, 2007). On the other hand, women’s own choices may also have biased the findings towards an increased risk by means of selective participation according to exposure level and fecundability. Nevertheless, because of the high participation in our study, we think that this potential discrepancy does not explain our main finding on solvent exposure and reduced fertility.

The studied shoe manufacturing workers were all exposed to mixtures of solvents, making it difficult to attribute the observed association to any specific solvents. Moreover, our findings do not show a traditional dose–response relationship; rather we see a slightly stronger effect for low exposure than for high exposure, and a more elevated association for exposure for a short time than for longer exposure times. Antagonistic effects have been observed in male rats exposed to mixtures of solvents. Simultaneous exposure to toluene and xylene was found to protect rats from n-hexane-induced testicular atrophy.34 However, to our knowledge there are no published experimental fertility studies on female solvent exposure indicating antagonistic effects. In general, an inverted U-shaped relationship curve has been proposed as a plausible outcome for endocrine-disrupting chemicals.35

We also sought possible interactions between solvent exposure and female lifestyle factors. In the interaction with coffee drinking, we found solvent exposure alone to be strongly associated with reduced fertility. Surprisingly, particularly high solvent exposure was related to increased fecundability among coffee-drinking women. Similarly, the effect of solvent exposure was modified by female smoking, and we see high fecundability among exposed smokers. However, this finding may well be attributed to coffee drinking because all but four smokers also drank coffee. We know of no epidemiological or experimental studies reporting similar antagonistic effects. However, a re-analysis of the Finnish data12 showed a similar, statistically non-significant interaction with female smoking but not with the use of coffee. Interestingly, smoking is known to increase the rate of caffeine metabolism, and the association between caffeine intake and reduced fertility may be restricted to non-smoking women.36 37 Whether a corresponding metabolic change is also true in the case of solvent exposure remains to be seen.

Significant associations might be found in subgroups which have nothing to do with causality due to multiple comparisons.38 However, we used subgroup analyses only to test the robustness of our main finding, and two out of three studied interactions were statistically significant. In addition to unknown biological mechanisms, the observed interactions are also compatible with the dose–response fallacy theory. In particular, our data may lack intrinsically subfecund exposed women who drink coffee or smoke. All in all, our findings indicate reduced fertility among solvent-exposed women. One can discriminate between the interaction and dose–response fallacy in a prospective study: a stronger association among solvent-exposed coffee consumers would favour dose–response fallacy whereas in the case of interaction a dilution of an effect would be expected.

Our findings provide further evidence that exposure to organic solvents is hazardous for female reproduction. The observed association may be related to any of the following solvents commonly used in shoe manufacturing: toluene, n-hexane and hexane isomers, methyl ethyl ketone, acetone, ethyl acetate and dichloromethane.