HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

Correspondence to:

ARTICLES: B. A. Racette, S. D. Tabbal, D. Jennings, L. Good, J. S. Perlmutter, and B. Evanoff Prevalence of parkinsonism and relationship to exposure in a large sample of Alabama welders Neurology 2005; 64: 230-235 [Abstract] [Full text] [PDF]

Correspondence: Submit a response to this article

Correspondence published:

Prevalence of parkinsonism and relationship to exposure in a large sample of Alabama welders

Louis Anthony Cox, Jr.   (29 June 2005)

Reply to Cox

Brad A. Racette, Joel S. Perlmutter, and Bradley A. Evanoff   (29 June 2005)

Prevalence of parkinsonism and relationship to exposure in a large sample of Alabama welders 29 June 2005
Louis Anthony Cox, Jr., University of Colorado Health Sciences Center and Cox Associates, Inc. 503 Franklin Street, Denver, Colorado, 80218 Send Correspondence to journal: Re: Prevalence of parkinsonism and relationship to exposure in a large sample of Alabama welders tony{at}cox-associates.com Louis Anthony Cox, Jr.	A recent paper by Racette et al [1] reported that “The estimated prevalence of parkinsonism was higher within a sample of male Alabama welders vs the general population of male residents of Copiah County, MS”. The authors mention potential reviewer bias and a proxy for age stratification as caveats. These and other threats to valid causal inference, such as multiple testing biases from selection of occupational codes counted as “welders”, can potentially create significant statistical associations between welding and Parkinson disease (PD), even if welding does not cause increased PD risk. To test whether a study provides evidence that a statistical association is causal, it is common to refute plausible rival (non-causal) hypotheses [2], especially confounding and biases. For example, if older people are both more likely to have PD and to be welders, as shown in the following causal graph [3]: PD <-- AGE --> Exposure (1), this provides an alternative to the causal hypothesis that Exposure (welding) increases risk of PD: AGE --> Exposure --> PD. (2) Modern conditional independence tests [3,4] can help determine objectively which hypothesis is more consistent with available data by examining whether Exposure provides information about PD risk after conditioning on Age (as in (1) but not (2)). Racette et al note that “When age was included in the regression equation [for PD], only age remained associated with definite PD. Age was modestly correlated with exposure. "…with adjustment for age, there was no significant association between mean exposure hours and diagnosis.” This suggests that PD risk is conditionally independent of Exposure given Age – consistent with (1) but not (2). By this criterion, the Alabama data set provides evidence of an association between Exposure and Risk, but not of a causal relation between them. Although Racette et al “speculate that welding exposure either increases the prevalence of PD at all ages or may shift the distribution of PD to a younger age,” the study data apparently show no evidence of a causal impact. The current state of causal analysis now allows conditional independence tests to be routinely applied to help interpret epidemiological associations by separating those that might reflect genuine causation from those fully explained by incompletely controlled confounders [3,4]. Such tests can assist policy makers and risk managers, who must rely upon cause-and-effect links rather than statistical associations to achieve public health benefits. References 1. Racette BA, Tabbal SD, Jennings D, Good L, Perlmutter JS, Evanoff B. Prevalence of parkinsonism and relationship to exposure in a large sample of Alabama welders. Neurology. 2005 Jan 25;64:230-5. 2. Maclure M.Taxonomic axes of epidemiologic study designs: a refutationist perspective. J Clin Epidemiol. 1991;44:1045-53. 3. Shipley, B. (2000). Cause and Correlation in Biology. A User’s Guide to Path Analysis, Structural Equations and Causal Inference. Cambridge University Press. http://callisto.si.usherb.ca:8080/bshipley/my%20book.htm 4. Greenland S, Brumback B. An overview of relations among causal modeling methods. Int J Epidemiol. 2002 Oct;31:1030-7.
Reply to Cox 29 June 2005
Brad A. Racette, Washington University School of Medicine 660 South Euclid Avenue, Box 8111, St. Louis, MO 63130, Joel S. Perlmutter, and Bradley A. Evanoff Send Correspondence to journal: Re: Reply to Cox racetteb{at}neuro.wustl.edu Brad A. Racette, et al.	We thank Dr. Cox for his interest in our manuscript. In his comments, he points out useful published references for statistical methods used in the analysis of causal relationships in epidemiology. However, we disagree with his suggestion that our study data show “no evidence” of a causal relation between welding and PD. Dr. Cox focuses his comments on the secondary analysis of exposure and disease within the cohort of welders, all of whom are exposed. He does not address the main results of the study that showed a large and statistically meaningful difference in the age- adjusted prevalence of PD between a cohort of welders and the general population. This comparison of a highly-exposed population to a low- exposed or non-exposed general population supports our speculation “that welding exposure either increases the prevalence of PD at all ages or may shift the distribution of PD to a younger age.” [1] In this case, the absence of a monotonic relationship between exposure and disease among the highest exposed group does not refute the causal hypothesis that high lifetime exposure to welding is related to the development of PD. There are many reasons, including threshold effects, why exposure-response relationships may be in an entire study population versus a subset of the population that is restricted by disease or exposure status. As emphasized in one of the references cited by Dr. Cox, “it is essential to consider the distribution of exposures and confounders in the combined study population of all treatment (or exposure) groups that are under comparison, not in some specific target group of policy interest.…For example, a study of vinyl chloride effects may have as its target only workers actually exposed; nonetheless, to evaluate confounding one needs to include the unexposed group (as well as exposed group) in the population being modelled.” [2] Dr. Cox neglects this critical point when he implies that our study observations are “fully explained by incompletely controlled confounders.” References 1. Racette BA, Tabbal SD, Jennings D, Good L, Perlmutter JS, Evanoff B. Prevalence of parkinsonism and relationship to exposure in a large sample of Alabama welders. Neurology 2005; 64(2):230-235. 2. Greenland S, Brumback B. An overview of relations among causal modelling methods. Int J Epidemiol 2002; 31(5):1030-1037.