Terence Hines - Pseudoscience and the Paranormal

Before discussing the research on the possible effects of PCBs on infants, it is appropriate to discuss the relationship between PCB exposure and cancer. While it is the case that giving massive doses of PCBs to animals can cause cancer, these doses are vastly greater than those to which any human would ever be exposed (Kimbrough 1993). Kimbrough et al. (1999) addressed the hypothesis that occupational levels of exposure to PCBs resulted in a greater incidence of cancer and other health problems. This study examined the causes of death of 1,157 industrial workers who had been exposed to high levels of PCBs as a consequence of their work. This sample is the largest ever studied in this regard. There was no increase in the rate of cancer among these workers.

The possible effects of PCBs on the children of women who ate fish with PCBs have been extensively studied. The results show a very minimal risk, if any. The earliest study to examine this issue was that of Fein et al. (1984), who found that, compared to infants of mothers who had not eaten PCB-contaminated fish, the infants of mothers who had showed a smaller birth weight, a smaller head circumference, and had lower “neuromuscular maturity.” These differences, while statistically significant, were very small. For example, the actual difference in head circumference was only 1.6 percent. Using the same group of infants, Jacobson et al. (1984) reported that the exposed children showed “worrisome” results on three out of seven measures of neonatal behavior at three days of age. The authors noted that their results “must be interpreted with caution” because such behavioral differences “are frequently transitory [and] their long-term developmental implications are uncertain” (p. 530). In a 1996 study Jacobson and Jacobson reported that greater prenatal exposure to PCBs was associated with significantly lower scores on five out of eleven measures of intelligence. Even on the five measures where the differences were statistically significant, they were tiny. The largest association had a correlation with an absolute value of .18. This means that only a little over 3 percent of differences among the children on this measure were due to PCB exposure.

Contrary to the findings noted above, Rogan et al. (1987) did not find any effect of exposure to PCBs or another chemical pollutant (DDE) on infant weight, head circumference, or how often the children suffered from various childhood illnesses. A later study (Gladden et al. 1988) examined various behavioral and intelligence variables in PCB and DDE exposed infants. Out of sixteen comparisons made, only two showed any effects of PCB exposure. In both cases, exposure was related to lower psychomotor development scores. As would be expected based on the quote from Jacobson et al. (1984) above that such findings are “transitory,” Gladden and Rogan (1991) found that, when tested at ages three to five, exposed children showed no effects of exposure. The small effects seen earlier had vanished.

Jacobson et al. (1985) reported an effect of PCB exposure in utero on a measure of visual recognition memory in infant. Gladen et al. (1988), however found no effect of PCB exposure on several measures of infant cognitive function.

Lonky et al. (1996) examined the relationship between prenatal PCB exposure and several measures of infant behavior. Infants were tested one and two days after birth. It is not at all clear from the published report just how many statistical comparisons were done, but in their Table 8 (p. 208), twenty-four are reported. Of these, only five were significant, the definition of significance being adjusted to take into account the number of tests performed. Nonetheless, if there really was an effect of PCBs, one would expect it to be more pervasive and show up in more than five out of twenty-four results. Making the interpretation of these results even more problematic is the fact that Lonky et al. did not analyze the actual scores on the tests from day one and day two; rather, they took the difference between day one and day two scores and analyzed these difference scores. This is a serious problem. For one thing, using difference scores throws away a great deal of actual data. For another, difference scores are much less reliable than using the original scores. For example, Consider baby A and baby B. Both are given a test of behavior on which the lowest possible score is 0 and the highest is 10. Baby A scores .5 on day one and 2.3 on day two. Baby B scores 7.7 on day one and 9.5 on day two. Both babies have a difference score of 1.8, but obviously there is a huge difference between the two babies on whatever aspect of behavior the test measures. Using only difference scores obliterates this information.

Stewart et al. (2000) reported a slightly different analysis of what appears to be basically the same data reported by Lonky et al (1996). While Lonky et al. found PCB exposure to be associated with poorer performance on three out of seven measures of infant behavior, Stewart et al. found it associated with poorer performance on only two of seven measures. Publishing repeated analyses of the same data results in a spurious impression of replication.

Using older infants (between six and twelve months of age) Darvill et al. (2000) reported that PCB exposure impairs performance on a test of infant intelligence. But, as usual, the effects were inconsistent and very small, such that exposure to PCBs accounted for only about 2 percent of differences on test scores.

The Gladen et al. (1988) study was mentioned above, but I did not describe one of their most interesting results. This was not an effect of PCB exposure, but an effect of exposure to DDE—and the effect was not a deleterious one. At six months of age, children exposed prenatally to DDE scored significantly higher than nonexposed children on a test of intelligence. Now it seems to me highly unlikely that prenatal exposure to DDE really makes kids smarter. Rather, this finding is yet another example of the fact that if one does a large number of statistical comparisons, just by chance alone, some are going to turn out to be “significant”; much as if one calculates enough risk ratios, some will look frighteningly high just by chance. Similarly, the scattered findings of small deleterious effects of PCBs are most likely due to chance. The observed effects are small, as noted above, and inconsistent. This is just what would be expected if PCBs have no real effect, but hundreds of statistical tests are performed looking for such effects. To be consistent, those who believe that PCBs are really dangerous should recommend that pregnant women take small doses of DDE during pregnancy to increase the intelligence of their children!

In a more recent study, Schantz et al. (2001) have claimed that exposure to PCBs via eating fish causes memory problems in older adults. When this study was reported on by the media in the summer of 2001, it was usually simply stated that the study proved that PCBs caused such impairments in adults—but an actual reading of the study shows something quite different and far less alarming. In this study, the effect of exposure to PCBs and DDE was examined using twenty-four different tests of memory and cognitive function. This resulted in fourty-eight different statistical tests, of which only three showed an effect of PCB exposure and two an effect of DDE exposure. Of these latter two, in one case the effect was deleterious. But the other was in the opposite direction—greater exposure to DDEs was associated with better performance. At least three other tests were also carried out that showed no effect of exposure to either chemical. This is another example of the shotgun approach to this sort of problem: Take a whole bunch of tests, give them to a whole bunch of subjects, and then highlight the few significant effects that were obtained and downplay the much greater number of tests that showed no effect. If PCBs and/or DDEs really had a detrimental effect on memory, one would expect that more than just a small fraction of the tests would have picked it up.