Understanding the distribution of positive Legionella samples in healthcare-premise water systems: Using statistical analysis to determine a distribution for Legionella and to support sample size recommendations.

Abstract

To significantly fit a statistical distribution to the proportion of positive Legionella samples in a series of water samples from multiple facility-premise water systems. Statistical fit test. A hospital and associated long-term care facility (LTCF) in New York State, as well as temporal and culture data from a deidentified hospital site supplied by one of the vendor laboratories. Culture samples (n = 1,393) were segmented into 139 test cycles with roughly 10 samples in each. The proportion of positive samples was standardized to 25 total samples per test to give a distribution of discrete values. These values were analyzed for fit with the following discrete distributions: Poisson, negative binomial, geometric, and zero-inflated Poisson. The zero-inflated Poisson distribution fitted to the copper-silver ionization (CSI)-treated and untreated test cycles indicates that 88% of the expected positive proportions should occur by the 30% cutoff (rounded up to 8 positive samples among 25 total samples), similar to the 93% expectation for just CSI-treated test cycles. The other treatment in these data (chlorine dioxide) was not effective in treating Legionella in the sampled buildings, and if there is an underlying distribution to these specific test cycles, it is not the zero-inflated Poisson distribution. In a well-maintained or well-treated premise water distribution system, ~30% or lower proportion of positive Legionella samples should occur. Anything above that cutoff is either very unlikely or not expected at all and indicates a problem in the water system.