Methods of a study of terminal cleaning of patient rooms

To the Editor—It is encouraging to see that people have reviewed our article “Increased Time Spent on Terminal Cleaning of Patient Rooms May Not Improve Disinfection of High-Touch Surfaces.”¹ However, a related Letter to the Editor raises concerns that some may be misinterpreting both the thrust of our paper and our study methodology.²

In our pragmatic report, we aimed to promote better cleaning by presenting research results that suggest that more than adequate time spent on terminal cleaning may not result in additionally lower bioburden on high-touch surfaces. We hope this information will cause practitioners to focus on other important factors such as proper training for environmental services staff (EVS), proper use of appropriate chemicals, and targeting high-touch surfaces that pose the greatest risk for transmission of pathogens to patients. We reiterated that adequate cleaning time is crucial, and we certainly do not advocate taking any shortcuts in the terminal cleaning process. Yet, as in many things, it is the quality of the process not the quantity that counts.

As to methodology, EVS were well-trained and experienced, and they voluntarily collaborated on the project. They were instructed to follow the manufacturer’s guidelines for application and contact time. We did not monitor EVS during room cleaning to avoid the Hawthorne effect and to obtain data on unmonitored cleaning.

The 5 high-touch surfaces chosen were the highest-touch surfaces according to published papers at the time of the study.³ We omitted details on the culture process and instead referenced a prior paper.⁴

Our analysis plan followed best practices for analyzing count data: use a generalized linear model with appropriate choice of family and link function, and avoid log transforming the data.⁵ We used Bayesian models and reported uncertainty in our estimates, rather than rely on a p-value. Recent articles highlight the pitfalls of statistical significance, which can be particularly problematic in small observational studies without preregistration.⁶ Major journals are now requiring some form of uncertainty interval rather than P values.⁷ We also chose to include model estimates on the actual outcome scale. This makes interpretation easy for those familiar with the outcome (ABC counts from press plates) but not familiar with statistical terminology like incident rate ratios. Our goal was to apply the best methods of analysis and interpretation.

Finally, we provided a full financial support disclosure statement in our article. The salary support for this study was provided by the authors’ employers.