Beyond LOEL's, p values, and vote counting: methods for looking at the shapes and strengths of associations

• Published in: Neurotoxicology (Park Forest South) Vol. 14; no. 2-3; p. 237
• Main Author: Schwartz, J
• Format: Journal Article
• Language: English
• Published: Netherlands 1993
• Subjects: Animals; Child, Preschool; Data Interpretation, Statistical; Dose-Response Relationship, Drug; Humans; Infant; Intelligence; Lead Poisoning - psychology; Randomized Controlled Trials as Topic; Risk Factors
• ISSN: 0161-813X

Determining the shape of dose-response relationships, including the possible existence of thresholds, is a central concern in the epidemiology of neurotoxins. Assessing the strengths of associations within a paper, and for the literature as a whole, are also critical concerns. Modern statistical and epidemiologic methods have considerable advantages for answering these questions. This is illustrated by using non-parametric smoothing to look for the existence of thresholds or non-linearities in the neurotoxic effects of lead, the use of multiple analyses and robust methods to test the stability of the association between dentine lead and children's IQ in the study of Needleman et al., 1979, and the examination of the stability of the association between full scale IQ and blood lead across studies. Non-parametric smoothing shows the dose response relationship between McCarthy score and blood lead reported by Bellinger et al., (1991) has no threshold down to blood lead levels of 1 microgram/dl. The association between blood lead and hearing threshold reported by Schwartz and Otto (1991) also has no threshold down to the lowest concentrations seen in that study. The association between dentine lead and IQ in the Needleman et al. (1979) data holds in the full data set with no exclusions and with control for age, is insensitive to exclusions of subjects or inclusions of other covariates, and to the use of robust estimation techniques. Nonparametric smoothing shows the association is supported across the entire range of lead exposure and is not driven by a few outliers. Examination of effect size estimates from the IQ-blood lead studies shows moderately good concordance. This suggests a causal relationship. A meta-analysis of those studies indicates that an increase in blood lead from 10 micrograms/dl to 20 micrograms/dl is associated with a reduction of 2.5 points in full scale IQ. The effect was highly significant.