The primary purpose of an environmental monitoring program is to provide oversight for microbiological cleanliness of manufacturing operation and document the state of control of the facility. Key to the success of the program is the establishment of alert and action limits. In practice, several statistical methods including normal, Poisson, and negative binomial modeling have been routinely used to set these limits. However, data collected from clean rooms or controlled locations often display excess of zeros and overdispersion, caused by sampling population heterogeneity. Such data render it inappropriate to use the traditional methods to set alert and action levels. In this paper, a method based on a zero-inflated negative binomial model is proposed for the above instances. The method provides an enhanced alternative for trending environmental data of classified rooms, and it is demonstrated to show a clear improvement in terms of model fitting and parameter estimation.
Lay abstract: The primary purpose of an environmental monitoring program is to provide oversight for microbiological cleanliness of manufacturing operation and document the state of control of the facility. Key to the success of the program is the establishment of alert and action limits. In practice, several statistical methods including normal, Poisson, and negative binomial modeling have been routinely used to set these limits. However, data collected from clean rooms or controlled locations often display excess of zeros and overdispersion, caused by sampling population heterogeneity. Such data render it inappropriate to use the traditional methods to set alert and action levels. In this paper, a method based on a zero-inflated negative binomial model is proposed for the above instances. The method provides an enhanced alternative for trending environmental data of classified rooms, and it is demonstrated to show a clear improvement in terms of model fitting and parameter estimation.