Quantitative microbial risk assessment of drinking water is generally followed by sensitivity analysis for examining the relative importance of variables of the simulation model on the outcome. This study investigated the effect of the statistical methods applied to unboiled water consumption data on sensitivity analysis. The sensitivity analysis for concentration of Escherichia coli (E. coli) in treated water showed completely different results from the analysis for E. coli dose. This was due to the application of a Poisson model to the water consumption, which suggested that 27% of the people did not drink tap water. Our study then applied a different model—an exponential distribution—to the water consumption data. In addition, incidental water intake was assigned to non-consumers in the Poisson model. The results of sensitivity analyses for these cases were very different from the ones obtained from the first analysis. This study therefore demonstrated that the statistical methods used to analyze water consumption data have great impacts on sensitivity analysis, although they do not affect the yearly risk of infection. Specifically, statistical methods may devalue sensitivity analysis. To avoid this problem, it is preferable to apply a continuous model such as the exponential model, rather than a discrete one such as the Poisson model, to describe the variability in water consumption.

通常在饮用水的微生物风险定量评估之后进行敏感性分析，以检查模拟模型变量对结果的相对重要性。这项研究调查了统计方法应用于未煮沸的水消耗数据对敏感性分析的影响。对处理后的水中大肠杆菌（E. coli）浓度的敏感性分析表明，其结果与对大肠杆菌的分析完全不同。剂量。这是由于将Poisson模型应用于耗水量，这表明27％的人不喝自来水。然后，我们的研究对耗水量数据采用了不同的模型（指数分布）。此外，在Poisson模型中，偶然摄入的水被分配给了非消费者。这些情况下的敏感性分析结果与第一次分析的结果有很大不同。因此，这项研究表明，尽管不影响每年的感染风险，但用于分析耗水量数据的统计方法对敏感性分析有很大影响。具体而言，统计方法可能会使敏感性分析贬值。为避免此问题，最好应用连续模型，例如指数模型，