distribution of HH compliance among HCWs in a population affects the ensuing transmission risk. A single HCW with low HH compliance could play a significant role in pathogen transmission, especially if such a low‐performing HCW provides care to many patients consecutively. This effect was demonstrated in two agent‐based models by Temime et al. and by Hornbeck et al [44, 45]. Exemplarily, the proportion of low‐ performing staff in our study were predominantly doctors who typically deliver care to many different patients. This ‘weakest link’ mechanism challenges the usefulness of pooled means of HH compliance infectious risk in each care unit. Our real‐world data support the idea that similar pooled mean HH compliance rates between observed settings can be the result of quite different distributions of high‐ and low‐performing HCWs. This study has limitations. First, not all observed HCWs could be included in the analysis due to the required number of observations. This could have led to an overrepresentation of permanent staff. Both the pooled baseline and intervention compliance were, however, comparable between the group of study HCWs and the excluded HCWs. Second, the definition of Improving HCWs as those with ≥20% HH improvement precluded the inclusion of HCWs with >80% compliance at baseline in this category. To circumvent this problem, we evaluated the possibility of using the change in non‐ compliance, rather than compliance to distinguish HCWs. This resulted in slightly higher correlations and similar effect estimates in the univariable regression model. We therefore decided to remain with the traditional definition of HH compliance. Third, the chosen cut‐off value of 20% to distinguish HCWs into the three HH compliance change categories was somewhat arbitrary. However, a multivariable model with a cut‐off value of 10% provided similar results (data not shown). Fourth, some of the observed between‐ HCW variability could be explained by chance due to a limited number of HH observation sessions per HCW and a limited number of opportunities per session. However, our study is the largest of its kind to date and demonstrates the feasibility and benefit of this approach. It might take an advanced automatic HH monitoring system to collect a larger number of opportunities per identified HCW. Finally, like other HH observation studies, observer and observation biases cannot be entirely excluded, especially a desirability bias by the observers also being involved in the promotion of the intervention, and observation bias, also known as Hawthorne effect [46‐48]. However, given the long study duration of 30 months and the focus on improvement dynamics, it is likely that neither biases influenced the results to a degree that would invalidate our findings.
RkJQdWJsaXNoZXIy MTk4NDMw