The results obtained by this study suggest that the plasma/blood Pb ratio could be viewed as a marker of susceptibility for lead toxicity. Our findings are consistent with studies relating blood lead and risk of spontaneous abortion [1
]; a high plasma/blood Pb ratio implies more circulating lead is free to cross the placenta at a given blood lead level. This is consistent with our finding of a correlation between plasma/blood Pb ratios and cord blood Pb concentration at birth. If inter-individual factors such as polymorphisms of the ALAD gene determine the plasma/blood lead ratio [5
], then some women, if exposed to lead during pregnancy, would be at increased risk of fetal lead exposure. The rationale for the latter idea is that if a pregnant woman is acutely exposed to lead from, for example, eating from a lead-glazed pot, her erithrocytic lead binding capacity will determine how much lead remains free in plasma and potentially reaches the fetus. This hypothesis is consistent with our observations: fetuses of women who tend to have a higher plasma/blood Pb ratio would be more susceptible to lead exposure and therefore these women would have a history of spontaneous abortions in previous pregnancies. In accordance with our hypothesis, a lack of a strong association between plasma Pb concentrations and history of previous occurrence of abortions was indeed observed in our sample as presented in table . The relatively large effect estimates of bone Pb are worthwhile mentioning, since it would be expected that women with a high lead burden are at elevated risk of spontaneous abortion. Furthermore, the positive patella Pb-ratio interaction term estimate suggests that the association between bone Pb and history of abortion is greater in those women with higher plasma/blood Pb ratios, which is in accordance to the hypothesis presented here. However, the small sample size of women with valid bone Pb measurements in our study prevented us from further exploring this hypothesis and reaching any definitive conclusion.
Our study has several limitations worth discussing. Given we are trying to find an association between a biomarker measured in the present time to events occurred years before, no causal relationship can be established. However, under the assumption that certain factors that determine the percentage of free circulating lead in humans remain constant between pregnancies, these findings would imply that women who have a lower red cell binding capacity for circulating lead may be at increased risk of suffering from spontaneous abortion. A second limitation relies on the fact that history of miscarriage was self-reported. This has the limitation that some women may suffer a spontaneous abortion without knowing that they were pregnant, while other women desirous of being pregnant may also report a spontaneous abortion without actually being pregnant. Nevertheless, this imprecision in the number of reported miscarriages is unlikely to be correlated to the plasma/blood Pb ratio and therefore our estimates are unlikely to be biased. This random measurement error of the outcome, however, may be reflected in imprecise effect estimates.
A different issue is that our sample constitutes a relatively small fraction of the cohorts which it arose from. If somehow the probability of being selected into the study sample was related to both plasma/blood Pb ratios and the number of miscarriages, selection bias could ensue. In order to test if being selected into the final sample depended on observed characteristics of the subjects, especially those related to the outcome and the exposure, we fitted the following empirical model by means of logistic regression:
ln(pi/1 - pi) = α0 + α1Ai + α2BPbi + α3Ci + X'iβ
Where pj is a the probability of participation in the study, Ai denotes the number of spontaneous abortions for the i-th subject; BPbi is the blood lead concentration of the subject, Ci is an indicator variable for the subject's cohort and X'i is a vector of individual characteristics such as age, number of years in school, number of pregnancies, weight and height (Table ). Using this approach, we found no evidence of selection bias; neither of the explored variables was associated to the probability of participating in the study except for the cohort indicator. The reason for the latter is the greater percentage of subjects in the first cohort who had a plasma Pb measurement and were therefore included in the study sample.
Logistic regression model of participation in the study. Mexico City, Mexico, 1997–2004
Another limitation was our inability to address the issue of lead and risk of spontaneous abortion in current pregnancies since women were recruited past the point at which spontaneous abortion usually occurs, and as a consequence the incidence of abortion in the study cohort was very small (1.3 percent), trying to estimate the effect of lead on the outcome of this particular pregnancy will be likely biased since women who miscarried before week 10 (including events possibly due to lead) had zero probability of being included in this sample. Abortions in past pregnancies could have occurred at any gestational age and thus our estimates would not be affected by this issue. However, there is a possibility that, if our hypothesis is true, women with very high plasma/blood Pb ratios could suffer from frequent miscarriages. For such women, the probability to reach 10 weeks of gestation and become eligible for this study would be low. If this were the case, it is likely that the true effect of the plasma/blood Pb ratio on the risk of miscarriage is underestimated by our study. This possibility is supported by the data as the mean of the plasma/blood Pb ratio in our study sample (0.22 percent) was lower than that found in a sample of comparable non-pregnant Mexico City residents (0.31 percent) using the same methodology for plasma and blood Pb measurement [12
Finally, we should note that social class could be viewed as an important potential confounder. Since both cohorts had the limitation of not registering a socioeconomic index, the models we presented adjust for schooling as a proxy of socioeconomic status. The fact that it did not seem to act as a confounder is consistent with the idea that plasma/blood Pb ratios are governed by biological individual characteristics rather than social factors. This is reflected on the finding that schooling was not correlated with the plasma/blood Pb ratio in our study sample (Spearman's ρ = -0.03, p = 0.67).