Principal findings
Using provider-level administrative data on all care homes in England, we estimated that there were over 29,500 excess deaths of care home residents during the first 23 weeks of COVID-19, equivalent to 6.5% of all care home beds. Almost 65% of the excess deaths were reported to be directly attributable (confirmed/suspected) to COVID-19. Our analysis shows that almost all excess deaths were recorded in the quarter of care homes which reported COVID-19 fatalities. This highlights that (i) non-COVID-19-attributed excess deaths were likely to be directly due to COVID-19 and/or (ii) any indirect negative effects of COVID-19 and enacted policies on mortality were predominantly constrained to those homes experiencing an outbreak. Non-COVID-19-attributed deaths being reported mainly during the early stages of the pandemic, when CQC recording of COVID-19 death was missing (before 10th April), guidance focused on a narrower set of symptoms and there was a shortage of testing, providing support for the former hypothesis.
Excess deaths were mainly concentrated amongst large and branded homes that provide services to older people and people with dementia. Adjusted care home level analysis confirmed these findings.
Strengths and limitations
To our knowledge, this is the first independent analysis that uses national administrative records from all care homes in England to estimate the impact of COVID-19. We find comparable total deaths to official estimates [5], adding stratifications of excess deaths by key care home characteristics and multivariable analysis to add a more nuanced understanding of these deaths. Local authority fixed effects were used to account for time-invariant measured and unmeasured determinants and confounders that differ across the local authority.
Our study also has limitations. Firstly, due to a high incidence of zeros at the individual care home level, it is not reliable to calculate the number of excess deaths per care home. Instead, we aggregated excess deaths to the local authority level and stratify by univariate care home characteristics in turn. To incorporate multivariable analysis with care home characteristics, we instead estimate odds of COVID-19 care home deaths as a proxy for odds of excess deaths. The univariate analysis suggests this should be a good proxy since almost all excess deaths occur in a care home with at least one recorded COVID-19 death.
We can observe the counts of COVID-19-attributed fatalities across care homes but not whether non-fatal COVID-19 cases occurred. This case data is not available, though serological and whole-genome sequencing studies give insights into this [24]. The attribution of COVID-19-related deaths is based on statements from providers to the CQC starting from 10 April 2020 and not always testing-confirmed or reflected in the death certificate. COVID-19-attributable deaths that occurred before 10 April would have been miscoded. The reported lower rates of testing could lead to some relevant deaths not having COVID-19 listed as a contributory factor, leading to apparently higher non-COVID-19 excess deaths [5, 10, 20].
No data was available on occupancy rates at the care home level. We instead used maximum bed capacity as reported in March 2020, assuming full occupancy. In the UK, occupancy rates were estimated to be on average 90% in nursing homes and 91% in residential homes [13]. It is very likely that occupancy rates declined during the COVID-19 period. However, assuming an arbitrary lower occupancy would increase excess mortality rates only proportionally, unless further breakdowns by time and care home types became available.
Measures of staffing and working conditions, and individual care home shortages of equipment would have been relevant for this analysis but there is no national care home-level data available. We also lacked data on residents’ case mix and their socio-demographic status. Our analysis is instead based on providers’ characteristics as reported to CQC. However, arguably, providers, rather than individual patients, are the targets of policy intervention and therefore these are the most relevant to include.
This study used administrative data and so sample sizes were not under our control. Despite the relatively large sample size available for examining the associations between the odds of one care home death and care home characteristics (> 15,000), it is possible that our analysis may not be powered to detect statistically significant associations with some characteristics, such as for-profit status.
We did not account for exposure and incidence of COVID-19 in the local area where each care home is located, or local policy responses to the pandemic, which changed over time. Wider community testing was negligible in the early parts of our analysis period [25] and likely differed by local authority capacity which would bias results if included. Furthermore, staff, healthcare professionals and any other individuals entering the care homes are not necessarily from or have only interacted with the immediate local areas [26], so fully capturing this would require location data for multiple individuals over time. Good-quality data on local policy responses was also unavailable.
Finally, as the number of deaths in the absence of the outbreak cannot be observed but only predicted, there is the potential that market dynamics and prediction errors could have influenced excess deaths estimates. However, we estimated small prediction errors in the pre-COVID-19 period relative to the size of excess deaths in the COVID period. Excess death estimates were also robust to different modelling approaches.
Study in context
By comparing observed deaths against averages over a historical 5-year period, the ONS estimated 25,876 excess deaths in English care homes up to 8 August [5]. Our estimates exceed this slightly. In addition to differences in methods, this is likely due to our data including deaths of care home residents occurring outside of a care home setting (e.g. in hospital).
Consistent with previous studies, we find that excess deaths occur overwhelmingly in the minority of care homes that experience COVID-19 fatalities [12]. This might suggest higher proportions of COVID-19-related excess deaths than reported [27] and that some deaths are potentially avoidable if initial care home outbreaks had been prevented. Although national lockdowns have the potential to displace care for care home residents with health conditions other than COVID-19, coupled with evidence of increases in mood and behavioural problems [28], our finding of no excess deaths in care homes without care home fatalities suggests that these issues may not impact mortality in the short run. Whether this type of excess mortality emerges in the longer-term in a subject for future research.
However, our results suggest that other care home characteristics, relating to the type of residents, staffing, ownership and size, are also important.
Care homes providing services to older people/with dementia suffered most deaths. This is unsurprising given the increased risk of contracting SARS-CoV-2 (difficulties complying with physical distancing, masking and hand hygiene) and increased risk of morbidity/mortality (comorbid illnesses), frailty and age. However, for care homes serving this group, there were smaller odds of COVID-19-related deaths in nursing compared with residential care homes. This might suggest a protective effect of the presence of staff with nursing backgrounds and infection, prevention and control (IPC) training, as found in other settings [9].
Overall, though, nursing homes had the most excess deaths and odds of COVID-19-confirmed/suspected deaths. This is likely due to these homes containing residents at high risk of contracting and dying of SARS-CoV-2, increased frailty and higher prevalence of co-morbidities, and therefore a greater likelihood of being in contact with other healthcare settings and practitioners [29].
In line with the existing literature, we found that large care homes are more likely to experience negative outcomes [10,11,12]. A likely contributor is that larger homes have a higher footfall altogether, of staff, healthcare workers, residents flowing in and out of hospitals, and visitors in non-pandemic times. This increases their chances of exposure to an infected individual, particularly in the absence of rigorous testing. Furthermore, it might be easier to ensure patient-centred management protocols in small care homes where policies around staff and patients contacts are set for smaller scales [30].
We find no significant differences between for-profit and non-for-profit providers, although for-profit providers experienced the most excess deaths because they account for the majority of the market. A Canadian study showed for-profit status was not associated with the odds of an outbreak, although it was associated with the extent of an outbreak (number of cases and deaths) [8]. However, we find that branded care homes had greater odds of COVID-19-confirmed/suspected deaths and rates of excess deaths. Branded homes could have policies around staff and patient movement across facilities that could potentially aid the spread of infection [7], particularly in the earliest parts of the pandemic before policy caught up and/or in the face of staff absence.