Dear Editor, This letter is in response to the article "An ecological study of the association between air pollution and hepatocellular carcinoma incidence in Texas" by Cicalese et al. [1], where the authors used nonparametric generalized additive logistic regression and gamma regression models to examine the association between county-level concentrations of air pollutants and hepatocellular carcinoma (HCC) incidence rates, concluding that vinyl chloride is a significant contributor to the incidence of HCC in Texas. We believe that these findings are unsubstantiated and misleading given the severe methodological limitations of the paper that we describe below.As an ecological study, this study is limited by measuring both exposure and outcome at the population level rather than the individual level. Ecological studies are useful for generating hypotheses but are severely limited in their ability to establish causality given their many methodologic problems, including the lack of adequate data, uncontrolled confounding, specification bias, aggregation bias, multicollinearity, and temporal ambiguity [2]. In addition, several potential confounders were not considered by the authors. Therefore, the findings may be a statistical artifact of uncontrolled confounding. For example, gender, race, alcohol use, type II diabetes, and smoking are all important risk factors for HCC that were not included in the model analyses. While the authors did include an obesity term, the reported findings could have been influenced by heterogeneity and the distribution of uncontrolled risk factors at the county level.The authors' attempt to lag exposure by using National Air Toxics Assessment ( However, these time points (range 1-17 years) do not allow for sufficient induction and latency periods for HCC. Specifically, a study examining vinyl chloride exposure and HCC risk reported that the median latency for HCC deaths was 48 years (range 31.5-66.6) [3]. Due to this, the exposure time points evaluated in this analysis are likely biologically irrelevant and inappropriate, as any impact on disease would not yet have been clinically diagnosed or detected.By crudely assigning exposure values at the county level, the NATA data is a spatially diluted average value that is not an adequate representation of exposures at more refined, local levels, such as residences. Specifically, the "EPA suggests that the results of (NATA) be used cautiously… (and) (i)n many cases more localized assessments, including monitoring and modeling, may be needed to better characterize local-level risk" [4]. Regarding the 1996 NATA data, the "EPA strongly cautions that these modeling results should not be used to draw conclusions about local exposure concentrations or risk" [5].