You are here:
A MODEL FOR ESTIMATING THE INCIDENCE OF SWIMMING-RELATED GASROINTESTINAL ILLNESS AS A FUNCTION OF WATER QUALITY INDICATORS
Citation:
Wymer, L J. AND A P. Dufour. A MODEL FOR ESTIMATING THE INCIDENCE OF SWIMMING-RELATED GASROINTESTINAL ILLNESS AS A FUNCTION OF WATER QUALITY INDICATORS. Presented at 4th International Conference on Environmetrics and Chemometrics, Las Vegas, NV, September 18-20, 2000.
Impact/Purpose:
Develop new bathing beach monitoring protocols and new approaches for communicating risks associated with swimming and other recreational water activities.
Description:
Several studies have demonstrated association between gastrointestinal illness (GI) in swimmers and sewage pollution as measured by the density of indicator organisms, such as e. coli and enterococci, in recreational waters. These studies generally classify illnesses into two categories according to the subjectivity of the reported symptoms and utilize separate analyses on the incidence of total illness and the incidence of objective symptoms of gastroenteritis. In addition, non-swimmer illness rates are available from these studies as an indicator of the background illness rates, but are not always utilized in the model. Ordinal logistic regression using response levels corresponding to the severity of illness is shown herein to be a potentially useful technique for modeling such data when background rates are included. Data from two prospective epidemiological studies conducted by the EPA and evidencing relationships between the incidence of swimming-associated GI and enterococcus or e. coli density in marine and fresh water, respectively, are used as examples. Initially, analysis of these data consisted of linear regression of log10 enterococcus density on the difference in illness rates between swimmers and non-swimmers. Subsequent published analysis of the marine study utilized logistic regression, but did not take background illness rates into account. both analyses produced separate models for rates of "highly credible" and total GI symptoms. The present analysis indicates that including the background rates improves the fit and that a proportional odds assumption for the dose-response is justified for these data.