Bias or biology: evaluating the epidemiologic studies of L-tryptophan and the eosinophilia-myalgia syndrome.

When investigating the cause of a chronic disease, epidemiologists are forced to substitute observational, nonexperimental methods such as cohort or case-control studies for the scientifically preferred "gold standard," the randomized controlled trial. Because neither cohort nor case-control studies are done under experimental conditions, the results may not accurately reflect those that would be found in a randomized experiment. Before the results of epidemiologic research can be used for inference regarding a cause of disease, it is necessary to examine the design and conduct of such studies to ensure their results could not have been distorted by potential sources of bias. We evaluate the epidemiologic studies of the relationship between intake of L-tryptophan (LT) and the occurrence of eosinophilia-myalgia syndrome (EMS) from information provided in the published reports and underlying data and documentation of the studies obtained from the US Centers for Disease Control and state health departments. We reviewed separately the initial 2 studies that examined the link between LT and the risk for EMS and the subsequent studies that focused on the specific manufacturing process or chemical constitutents of LT that might have been responsible for EMS. The 2 initial studies compared cases of EMS with controls who were asymptomatic. The investigators concluded that ingestion of LT was associated with the occurrence of EMS. However, these studies contained methodologic problems, including diagnostic bias, publicity, recall and reporting bias, bias in the inclusion and exclusion of cases and controls, inequalities between cases and controls in the indications for the use of LT, and failure to ensure that the exposure to LT preceded the onset of symptoms of EMS. These potential biases make it difficult to use the data derived from these investigations to justify a causal inference. Subsequent studies were conducted under the assumption that there was a valid association between ingestion of LT and the occurrence of EMS. These studies focused on tracing back LT to the manufacturer. These studies also had a variety of methodologic shortcomings, including problems in the assembly of study subjects leading to the selected samples of cases and controls, the lack of information on the brand and lot of LT for many subjects, multiple brand use, differences in the timing of exposure between cases and controls, the difficulty of the process of tracing LT products to the appropriate manufacturer, inconsistent classification of symptoms depending on brand of LT used, and inconsistencies in the traceback procedures between cases and controls. In light of these analyses, it appears that the cause of EMS has not been definitively identified. The search for the cause of EMS should continue without the underlying assumption that LT or some contaminant is responsible.