Modelling characteristics to predict Legionella contamination risk – Surveillance of drinking water plumbing systems and identification of risk areas

•We assessed DWPS characteristics which were pertinent for early risk predictions of Legionella contamination.•Incidences of Legionella have a strong spatial and temporal (short- and long-term) variability.•Single sampling point-specific parameters cannot predict the occurrence of contamination.•For single outlets a good estimate of the risk of Legionella contamination can be calculated via a logistic regression model. For the surveillance of drinking water plumbing systems (DWPS) and the identification of risk factors, there is a need for an early estimation of the risk of Legionella contamination within a building, using efficient and assessable parameters to estimate hazards and to prioritize risks. The precision, accuracy and effectiveness of ways of estimating the risk of higher Legionella numbers (temperature, stagnation, pipe materials, etc.) have only rarely been empirically assessed in practice, although there is a broad consensus about the impact of these risk factors. We collected n=807 drinking water samples from 9 buildings which had had Legionella spp. occurrences of >100CFU/100mL within the last 12 months, and tested for Legionella spp., L. pneumophila, HPC 20°C and 36°C (culture-based). Each building was sampled for 6 months under standard operating conditions in the DWPS. We discovered high variability (up to 4 log10 steps) in the presence of Legionella spp. (CFU/100mL) within all buildings over a half year period as well as over the course of a day. Occurrences were significantly correlated with temperature, pipe length measures, and stagnation. Logistic regression modelling revealed three parameters (temperature after flushing until no significant changes in temperatures can be obtained, stagnation (low withdrawal, qualitatively assessed), pipe length proportion) to be the best predictors of Legionella contamination (>100CFU/100mL) at single outlets (precision=66.7%; accuracy=72.1%; F0.5 score=0.59).