The size of the present study enabled detection of minimum odds ratios of 1.5 and 1.6 for exposure prevalence in controls of 30 and 20%, respectively, i.e. of the same order of magnitude as those of maternal alcohol drinking during pregnancy or parental smoking.

The cases were identified through the data collection system of the French National Registries of Childhood Cancer (NRCH, NRCST), making selection of cases at the identification step unlikely. Cases who had died or were receiving palliative care were not eligible, which might have lead to survival bias. Nevertheless, we were able to compare the prevalence of parental smoking of 19 cases that died after their mothers’ interviews with the prevalence of parental smoking of the survivors. The prevalence of parental smoking, either for maternal and paternal smoking, did not differ between the non-surviving and surviving children making a survival bias unlikely in the present study.

Eighteen percent of the eligible cases did not answer the questionnaire. However, the age and gender distributions were similar for the respondent and non-respondent cases.

The controls were randomly selected from the general population, based on the national telephone directory. Unlisted numbers were randomly generated in order to avoid the selection of controls with listed numbers.

There were no differences between the cases and controls with regard to gender or age (considering all the types of cancer covered by the ESCALE study), or between the controls and overall population, particularly with regard to birth order, number of children living in the household and socio-demographic characteristics, indicating that quota sampling was successful. The socioeconomic status and educational level of the control parents were very similar to those of the cases and the French population (Blondel et al., 1997; Blondel et al., 2006). Adjustments for those variables did not change the results. Additional adjusment for ethnic origin (countries of birth of the 4 grand-parents) of the index child did not change the results as well.

All the information was based on the child’s mother’s interview. The mothers could have misreported their smoking habits and those of the father. However, the number of cigarettes smoked by the control mothers was comparable to that of the French population (Blondel et al., 1997; Blondel et al., 2006), as was the number smoked by the control fathers (Guilbert P et al., 2005). The results are therefore unlikely to be explained by under-reporting of smoking by the controls. In addition, the case mothers may have under-reported their own smoking, leading to under-estimation of the association with maternal smoking. However, the results reported herein are consistent with 2 meta-analyses that found no association between CNS tumours and maternal smoking during pregnancy (Boffetta et al., 2000; Huncharek et al., 2002). One recent cohort study (Brooks et al., 2004) reported a significantly increased risk related to maternal smoking, but this association was found to be significant for low grade astrocytomas, which were not eligible in the present study, and not for malignant brain tumours. However, an association with maternal smoking during pregnancy cannot be completely ruled out, given the slightly increased OR with maternal smoking in the absence of paternal smoking observed herein.

Case mothers may have over-reported smoking by the father, which would explain the relationship with paternal smoking. However, the paternal smoking habits of the controls were very similar to those of the French population (Guilbert P et al., 2005). Moreover, several previous studies, that investigated paternal smoking by interviewing the father himself also found positive associations (McCredie et al., 1994; Ji et al., 1997; Fillipini et al., 2002; Cordier et al., 2004). Two meta-analyses estimated ORs of 1.22 [1.05–1.40] and 1.29 [1.07–1.53] for paternal smoking during pregnancy (Boffetta et al., 2000; Huncharek et al., 2001; respectively). Cordier et al. reported an increased risk of malignant astrocytoma among children whose father was exposed to PAH from smoking, before the conception (Cordier et al., 2004). In the study by McCredie et al., malignant brain tumours were associated with pre-conception paternal smoking and maternal exposure to side-stream smoke from the father during pregnancy (McCredie et al., 1994). Two biologically-plausible mechanisms have been proposed to explain those findings. A direct hypothesis, supported by both animal and human data (Baldwin and Preston-Martin, 2004), suggests an impact of pre-conception paternal germ-cell exposure. Spermatogenesis is a continuous process, with intense replication of DNA, making germ cells more vulnerable to mutagenic changes (Anderson et al., 2000). The second hypothesis suggests that paternal smoking may act through the mother’s passive exposure to side-stream smoke during pregnancy. Biochemical studies showed that some constituents of environmental tobacco smoke may cross the placenta and interact with fetal DNA (Tredaniel et al., 1994; Anderson et al., 2000; Rice, 2004). However, the epidemiological data on this question are contradictory. Among the studies that explicitly addressed the role of passive exposure to tobacco smoke, some evidenced an association between CNS tumor and maternal exposure to side-stream smoke (Preston-Martin et al., 1982; McCredie et al., 1994; Filippini et al., 2002) while as many others did not (Kuijten et al., 1990; Cordier et al., 1994; Hu et al., 2000). It is noteworthy that paternal smoking remains difficult to differentiate from maternal smoking. Similarly, there are strong correlations between pre-conception, gestational and post-natal maternal and paternal smoking that make it difficult to elucidate the actual exposure with certainty in the context of paternal pre-conception smoking. However, the literature has not reported any evidence of an association with passive smoking during childhood (Ji et al., 1997; Little, 1999; Filippini et al., 2002).

The present study did not generate any evidence of an association between alcohol consumption during pregnancy and malignant CNS tumours. Alcohol consumption is difficult to quantify by questionnaires, and an information bias may have occurred, even though its extent would have been reduced by the use of a detailed standardized questionnaire. Guilt feelings may have led the cases’ mothers to minimize their alcohol consumption during pregnancy and the resulting recall bias may have masked an association. However, the bias is more likely to affect the dose-response relationship than the ever/never relationship. The literature on maternal alcohol drinking during pregnancy and CNS tumours is very limited and shows no consistent association. However, two studies evidenced that beer drinking during pregnancy was associated with CNS tumours (Howe et al., 1989) or with primitive neuroectodermal tumours (PNET) (Bunin et al., 1993). Beer is a known source of N-nitroso compounds (NOC), which are suspected to play a role in childhood brain tumours. In the present study, beer consumption was slightly, but not significantly, related to CNS tumours and particularly to PNET and other gliomas.

The results for maternal coffee and tea consumption raise some questions. A recall bias is unlikely given that there is no particular public concern with those habits. Wilkins et al. reported no differences between cases and controls for mothers recalling their own diet in a case-control study on childhood brain tumours (Wilkins and Bunn, 1997). Furthermore, associations with maternal consumption of coffee and tea during pregnancy were limited to ependymomas, which makes bias less likely. Cordier et al. found an increased OR associated with coffee (1.9 [0.9–3.9]), but nor with tea (0.7 [0.3–1.4]) (Cordier et al., 1994).