On 2020-09-01 09:40:12, user Roland Salmon wrote:
This is a thorough piece of field epidemiology, although like much field epidemiology today, the data substantially comes from existing information sources. As a former director of the Communicable Disease Surveillance Centre Wales (CDSC), I am pleased that Public Health Wales, via CDSC staff, past and present, produces work of this quality.
The study demonstrates, persuasively, that much of the problem with infection in care homes, resulted from the care home's size, rather than from receiving infected patients, discharged from hospital. Nevertheless, I do not think that it should be stated ("Research in context"p.3) that "Our analysis found no effect of hospital discharges on care home outbreaks once care home size had been adjusted for" (my underline). In fact, as the discussion section makes clearer (p11), the observed hazard ratio is 1.15 and the effect could be as high as 1,47 (Table 2), although the result is not statistically significant at the 5% level. (It would be interesting inter alia to know the actual probability of this, the most probable estimate of hazard of 1.15.) Table 3, looking at the risk of outbreaks, by care home capacity, further, implies that the effect of discharges might be particularly marked in the smaller homes (<10 beds) where I calculate that the crude relative risk of an outbreak in the post hospital discharge risk period is 3.2. compared with around 1.2 for larger homes. Anyway, an intervention that reduced the risk of outbreaks, in this vulnerable population, by some 15% would be considered by most people as well worth having.
It's thus important to reflect whether the failure to demonstrate an effect of this size merely reflects a lack of statistical power, some of which could be due to misclassification of the outcome. The study authors recommend, in "Conclusions and recommendations" (p12), that, "further analyses should investigate the risk where discharges were confirmed or probable cases of Covid-19, and also consider additional evidence on likely chains of transmission that may become available from sources such as.....viral genetic sequence data". This is an important supplementary piece of work. In addition, the risk from hospital discharges, unlike that from home size, does not extend over the whole period of the study. I note that 16 outbreaks that occurred before certain homes received any discharges are included in the dataset so homes, therefore, enter the study before they are at risk of any infection introduced by receiving patients discharged from hospital. Secondly, homes remain in the study after 2nd May, when universal testing of hospital patients for SARS CoV2, prior to discharge to care homes, is introduced. Thus, from, a few days after this until the 27th June, the study's end date, effectively, risk from hospital discharges is eliminated whereas the risk from home-size remains. The authors consider this and report that they fitted their model, with a factor for the two time periods (before and after 2nd May). They tell us that, "this factor was found not to be significant, and did not significantly alter the hazard ratios". Whilst I understand that any alteration of the hazard ratios was not significant at alpha =5%, I would like to actually see the change in the observed hazard ratios. It might be expected that the hazard of receiving hospital discharges was higher in the period up to 2nd May, than in the period from 2nd May to the study's end.
I was curious as to why Cox's Proportional Hazard was the test used. I don't altogether see that the risk of outbreaks following introduction, by hospital discharge is particularly time dependent, given how readily and for how long SARS CoV 2 can spread in institutional settings. Thus, I don't really see why that risk factor could not be expressed as a categorical variable (outbreak, no-outbreak) which would allow a much simpler analytical approach. I, frankly, also, don't understand the detail of the sensitivity analyses, presented, for choosing different at-risk time periods which, I feel, for a general readership, certainly, merits being explained more fully.
Finally, I think that the discussion section could be more robust. If home size is the issue, then shouldn't the authors be saying that larger homes need to consider having dedicated areas, facilities and staff for smaller subsets of their residents. Maybe larger homes should have more stringent planning requirements. I also think that rather more should be made of the contribution of hospital discharge (notwithstanding it's failure to achieve conventional 5% levels of statistical significance) than the rather anodyne paragraph at the foot of page 11 which bears all the hallmarks of the dead hand of the corporate public relations department.
Nonetheless, overall, this is an accomplished piece of epidemiology with important practical implications.
Dr Roland Salmon