Microbiological Quality of Purified Water Assessment Using Two Different Trending Approaches: A Case Study

Statistical process control (SPC) is of prime importance for the evaluation of inspected characteristics of the monitored product or process. Microbiological quality of water that supplies healthcare facilities and plants is crucial to ensure the safety of its use in consumption and application. The current case aimed to investigate the microbiological stability and value of purified water produced from city water supply source that feeds water processing station through multiple purification stages. This study is part of monitoring project of water quality at different time periods from water treatment plant. Purified water (PW) produced from the processing of the water plant was assessed microbiologically using conventional culture technique on daily basis during working days in the facility. The results of the laboratory analysis were saved, interpreted statistically and trended using two different types of control charts. Statistical analysis and trending charts were interpreted using commercial statistical software programs. The analysis showed that microbiological results have met the acceptance criteria for bioburden limit with no fungi or waterborne pathogens were detected during the testing period of the third quarter period of the year (Q3). However, statistical interpretation demonstrated the non-Gaussian distribution of data. Moreover, the results did not follow even Poisson or negative binomial type of distribution. Interestingly, omitting aberrant results, square root and logarithmic transformations did not improve the normalization of data significantly. In such instances, Laney correction for overdispersion or underdispersion of data was considered. In parallel, Individual-Moving Range (I-MR) chart was constructed showing similarity with control limit parameters for the Laney-modified attribute Shewhart chart indicating that I-MR chart is robust with non-normal data with the advantage of greater sensitivity in detecting out-of-control alarm signals. In conclusion, both types of control charts may be considered for non-normal microbiological count data that resist the normalization process and does not follow Poisson or negative binomial distributions.