Objectives: Approximately 1,100 communities in the United States have combined sewer and stormwater systems whose capacity may be exceeded during moderate or heavy rainfall. Outflows may occur that can deposit water with varying concentrations of the components of sewage onto public areas, where contact with residents or workers is possible, potentially resulting in a range of adverse health effects. This study proposes and applies three analytic methods to evaluate the impact of such outflows on public health.
Methods: The work attendance records of 449 U.S Postal Service letter carriers in Sacramento, CA, were reviewed to determine the frequency of sick leave use (as a surrogate measure of illness) in relationship to rainfall and potential exposure to sewage-contaminated outflows in two distinct groups of letter carriers from October 1, 1992, through April 30, 1993. Rainfall was a surrogate measure of outflows from the combined sewer and stormwater system because no information existed about the extent of exposure of these (or any other) workers. One group of letter carriers delivered mail exclusively within an area served by the combined sewer and stormwater collection system; the second group delivered mail exclusively outside this area where sewage and stormwater collection are separated. The first approach to the assessment of the data was a description of the temporal relationship between rainfall patterns and absentee rates in the two groups of workers. The second approach used logistic regression modelling with varying lags between rainfall and sick leave usage. The third approach used Poisson regression analysis of the entire study period to examine the differential impact of rainfall on the two groups.
Results: The descriptive analyses detected no relationship between rainfall and sick leave use. The logistic regression analyses detected evidence (as measured by the interaction coefficient in logistic models) of an increased use of sick leave by the letter carriers within the Combined System area at lag periods of one, four, and five days after rainfall. These estimates were not, however, statistically significant (p > 0.05). The Poisson regression analysis showed no evidence of a differential impact of rainfall on the two groups (incidence rate ratio = 1.19 in both groups for periods of rain versus no rain).
Conclusions: These three methods can be used to investigate the public health impact of combined system outflows. Ideally, however, these approaches would be applied to prospectively collected surveillance data that would rely upon direct measurements of exposure and illness, rather than surrogate variables.