Re-evaluation of precipitation rates and autism prevalence in the western US

Background: The diagnosis of autism spectrum disorder has shown a steady dramatic increase over the past three decades. Environmental causes or triggers have been proposed to account for part of the increase in diagnosis, but published epidemiologic studies of risk factors are few and contradictory. An ecologic association between precipitation and autism was recently reported for Washington, Oregon, and California. We hypothesize that the apparent association may be a spurious result of administratively censored low autism counts. Inadequate handling of censoring can cause bias and inflate type I error rates. Methods: Censoring definitions and appropriate statistical methods are reviewed. Data for the prevalence of autism is obtained from state agencies. Demographic information is obtained from the US Census. Precipitation data is obtained from the National Climatic Data Center. Ecologic analyses of county based prevalence of autism are performed using left censored Poisson models. Covariates include relative precipitation, total population, per capita income, ethnicity and urbanicity. A separate analysis of the same covariates is performed with relative precipitation replaced by total annual precipitation. Results: Results will be presented as covariate coefficients and relative risk estimates with 95% confidence intervals for Poisson models for autism prevalence. Conclusion: Studies of risk factors for autism often rely on administratively censored data sets provided by state agencies; censored data require special statistical methods. The importance of this study is to test the significance of precipitation as a trigger for development of autism.