This study has just been published in the Journal of Hopsital Infection, showing that the introduction of hydrogen peroxide vapour (HPV) for the terminal disinfection of rooms vacated by patients with CDI was assocaited with a significant reduction in the rate of CDI, from 1.0 to 0.4 cases per 1000 patient days.
The study was performed over 4 years in a US hospital (in Tupelo, for the record – the birthplace of Elvis no less), and I think it’s the ‘cleanest’ data demonstrating that improving terminal disinfecion of hospital rooms using HPV reduces the spread of C. difficile. There were no other interventions, although worth noting that antibiotic usage was not measured or reported. The magnitude of the reduction (60%) is worth discussing. Although it has been established that admission to a room where the previous room occupant has had CDI is a risk factor for CDI, I don’t think that blocking this room-to-room transmission of C. difficile is the whole story. I suspect that HPV resulted in a reduction of the environmental load of C. difficile throughout the hospital and so reduced the ‘acquistion’ of C. difficile spores on the hands of HCWs. Also, the study team somehow managed to get to every single room following a C. difficile discharge with HPV; not a single one was missed over 2 years – in contrast to other studies where out of hours discharges and, less commonly, the need to admit another patient sometimes resulted in reverting to the usual terminal disinfeciton protocol (give it a quick wipe down and get the next patient admitted stat…I jest, of course).
One important aspect of the study is the ‘breakpoint’ time series analysis that was used to examine the rate of CDI. In a traditional time series analysis you say “we put in an intervention in January 2011 – let’s see whether there was a significant change in rate trend before and after the intervention”. In a breakpoint model, you say to the model “you tell me whether / when a significant change in rate trend occurred, and let’s see whether this matches the time of intervention”. Clearly, there’s a lot less room for bias in the breakpoint version of the model. In this case, there were 2 breakpoints in the rate of CDI see Figure). The second breakpoint is easier to explain – this was exactly when the hospital started using HPV for terminal disinfection following cases of CDI. The first breakpoint is more tricky – and doesn’t match a specific infection control intervention, but it is probably explained by seasonal variation.
Figure: The monthly rate of CDI in the 24 months prior to the implementation of HPV (2010 and 2011, study months 1-24) and the first 24 months of HPV usage (2012 and 2013, study months 25-48). The breakpoint model identified two significant changes in rate indicated by vertical lines at months 19 and 25; dotted vertical lines are 95% confidence intervals around the breakpoints.
This study didn’t start out as a breakpoint model: the original submission included a less sophisticated statistical test of the change in CDI rate. In response to a positive, challenging review from the JHI (for which I am grateful), I asked some of my epi team (Sid and Elie) to learn how to run a breakpoint model, which they did in double-quick time (if anybody is interested in finding out more, let me know). There are a couple of other examles of similar analyses published here and here – and I really think breakpoint time series analyses are the way forward for analysing trends in infection/colonisation rate data.
If the introduction of HPV for terminal room disinfection really does reduce the rate of CDI by 60% in a low prevalence setting, then we really should be looking very carefully at introducing it – or at least drastically improving the level of disinfection following the discharge of patients with CDI.