The Battle Creek Health Study: A retrospective cohort study of the health effects of volatile organic chemicals with individual estimate of exposure based on time-modeling

In Battle Creek, Michigan, groundwater was found contaminated with a number of volatile organic chemicals. The author conducted a study of the potential health effects of the water contamination. The target was to study all current and former residents of 85 households with contaminated private wells. The final study size was 251 exposed and 498 reference people. The design was that of a historic cohort study based on concepts of risk assessment for chemical toxicity, mainly a thorough exposure assessment, rigorous quality control of all study components, an analysis focusing on dose-response relationships, and an assessment of the impact of uncertainties in data and methods on the results. The core of the analysis was the introduction of the concept of time in exposure, disease development, and analytical models, while consistency in results was a central theme in the evaluation of the final study results. The exposure assessment consisted of modeling exposure values for individuals retrospectively, estimating when the exposure started, establishing dose equivalents for related chemicals, and using dose rather than "available" exposure. Multiple data sources were used to assess the final validated health status. The analysis was based on univariable and multivariable analysis. The latter made use of a newly developed modification of the proportional hazard model. In evaluating analytical results, emphasis was given to epidemiological criteria of causality rather than statistical tests. The study yielded no evidence of a toxic effect of exposure with a "no observed effect level" of accumulated maximum dose of 6.8 mg (all chemicals combined) per kg body weight per day for 99 months (median 3.7 mg/kg/day for 47 months). The study yielded hitherto unknown data on population-based incidence rates for hypertension, diabetes, cholelithiasis, peptic ulcer, epilepsy, neprolithiasis, and hypothyroidism. The first three disorders occurred frequently enough for multivariable analysis, which identified risk factors, some of which were previously unknown. Other disorders analyzed included birth defects, abnormal pregnancy outcomes, cancer, skin diseases, and allergies. Further, there was some evidence of deficiencies in the local health care system, which might explain some of the excess occurrence of disorders in the study population.