Risk factors for outbreaks of COVID‐19 in care homes following hospital discharge: A national cohort analysis

Abstract Background The population of adult residential care homes has been shown to have high morbidity and mortality in relation to COVID‐19. Methods We examined 3115 hospital discharges to a national cohort of 1068 adult care homes and subsequent outbreaks of COVID‐19 occurring between 22 February and 27 June 2020. A Cox proportional hazards regression model was used to assess the impact of time‐dependent exposure to hospital discharge on incidence of the first known outbreak, over a window of 7‐21 days after discharge, and adjusted for care home characteristics, including size and type of provision. Results A total of 330 homes experienced an outbreak, and 544 homes received a discharge over the study period. Exposure to hospital discharge was not associated with a significant increase in the risk of a new outbreak (hazard ratio 1.15, 95% CI 0.89, 1.47, P = .29) after adjusting for care home characteristics. Care home size was the most significant predictor. Hazard ratios (95% CI) in comparison with homes of <10 residents were as follows: 3.40 (1.99, 5.80) for 10‐24 residents; 8.25 (4.93, 13.81) for 25‐49 residents; and 17.35 (9.65, 31.19) for 50+ residents. When stratified for care home size, the outbreak rates were similar for periods when homes were exposed to a hospital discharge, in comparison with periods when homes were unexposed. Conclusion Our analyses showed that large homes were at considerably greater risk of outbreaks throughout the epidemic, and after adjusting for care home size, a discharge from hospital was not associated with a significant increase in risk.


| BACKG ROU N D
Care homes are settings in which resident populations typically live in close proximity. Annually, they experience outbreaks of gastrointestinal and respiratory illnesses, including norovirus and influenza, with associated morbidity and mortality; 70% of acute respiratory infection outbreaks in the UK occurred in care homes in the winter of 2018/19. 1 Outbreak-associated infections may be introduced via human sources such as new admissions from home or hospital, via staff or via visitors.
Early evidence from the COVID-19 pandemic, later further corroborated, was that older people were more severely affected, with a case fatality proportion of 2.3% overall but 8% in those aged 70-79 and 14.8% in those aged over 80. 2 An assessment of international evidence from April estimated that in Italy and Spain, over half of reported deaths were in care home residents. 3 Preliminary studies from April in England found extensive spread amongst staff and residents in homes reporting incidents, and wide variation in symptom profiles. 4

A Scientific Pandemic Influenza
Group on Modelling (SPI-M) paper predicted that nearly all care homes would become affected if current conditions persisted and indicated a role for staff in introducing infections, particularly where staff worked across more than one home. 5 Recent studies indicate that Personal Protective Equipment (PPE) and number of staff employed 6,7 have an impact on the number of COVID-19 infections.
More recent data from the Care Quality Commission suggest that, to mid-June, 36% of care homes experienced an outbreak (defined as a single laboratory-confirmed case) 7 with a study of care homes across a large Scottish Health Board reporting a figure of 37%. 7 Early estimates of the impact of COVID-19 in the UK suggested that inpatient and critical care bed capacity could be overwhelmed. 8 Hospitals in the UK prepared rapidly for the increase in cases, including cancellation of elective procedures and expediting discharges to home or social care facilities. Testing for residents scheduled for discharge was not always available or done. 9 Media reports have implicated these discharges as the cause for many of the subsequent outbreaks in care homes [10][11][12][13] but we were unable to locate any studies either published or in preprint that linked data on discharges to outbreaks. Expert commentary on existing data identified care and non-care staff, visitors and resident discharges from hospital as possible vectors for the introduction of COVID-19 into care homes, particularly where testing is not available, 14 and the discharge back to their care home of untested SARS-CoV-2 positive individuals has been suggested as a risk factor for outbreaks in these settings. 15,16 Studies reporting evidence from testing all staff and residents in specific care homes have suggested high proportions of asymptomatic cases, particularly amongst older residents, than were initially assumed. 7,14,17 This suggests the risk of importing COVID-19 into care homes via hospital discharge of untested asymptomatic residents has been underestimated.
Several studies have provided further evidence of factors that may have increased the risk of outbreak in care homes. A large survey carried out by the Office for National Statistics suggested frequent use of agency staff or carers and staff working conditions, including provision of sick pay, influence the risk of an outbreak. 18 Two studies have used routine data to consider a range of risk factors, including resident need, evidenced by services provided (eg nursing care, dementia care), corporate ownership and pre-COVID-19 outbreak history. 7,19 Wales had its first case of COVID-19 confirmed on 28 February 2020, and also saw a subsequent rise in cases and outbreaks in care homes. Public Health Wales's (PHW) Communicable Disease Surveillance Centre has been undertaking surveillance for outbreaks in care settings since 2015.
We aimed to use our national surveillance framework to test whether the risk of a COVID-19 outbreak in the period following a discharge from hospital to a care home was increased compared with other periods, in order to better understand the sources of infection and prevent further incidents.

| Study population
The study population was all adults living in residential or nursing care homes in Wales, which has seven health boards and a population of 3 152 879. 20 We defined a care home as a premises registered  The testing policy for care home cases changed during the time period for this analysis. Initially, testing was offered for up to the three most recently symptomatic individuals in homes which had not already recorded a confirmed case. This was increased to up to 5 symptomatic individuals from 15 April, and to all symptomatic residents from 24 April. Due to likely under-ascertainment of cases in the earlier part of this period, we defined a COVID-19 outbreak as one resident testing positive for SARS-CoV-2 whilst resident (consistent with Burton et al 7 ), or within 14 days of being resident. All testing was performed by health boards in Wales and all samples processed by NHS laboratories. From 02 May, all hospital patients were required to have a negative COVID-19 test result before being allowed to be discharged back into a care home.

| Data linkage
Data were linked by care home, matching addresses on the CIW registration record with addresses recorded on individual hospital discharges from PEDW, and on test result records reported on Tarian.

| Exposure and outcome
The first notification of a case of COVID-19 in a care home was made to PHW on 15 March 2020 relating to a specimen collected on 14 March 2020; before this date, all homes were considered at risk.
Once homes had a case of COVID-19 confirmed by laboratory test result, they were excluded from further analysis. This was due to the considerable uncertainty in assigning subsequent cases to a chain of transmission within the home, or to external exposure.
Our outcome was the time (from 22 February) to the first laboratory-confirmed case of COVID-19 in each care home. We defined a baseline exposure period following a discharge from hospital as 7 to 21 days post-discharge. Thus, any first case appearing during this window was recorded as being associated with the discharge event. This window was chosen to approximately account for the potentially incubation and infectious period of an asymptomatic or pre-symptomatic (and thus untested) discharged resident and for subsequent incubation period of cases caused by onward transmission in the home. As described above, testing pathways for outbreak identification were only routinely available for symptomatic care home residents during the study period. We considered this baseline scenario the most likely to capture an outbreak if caused by a discharge event, but also considered a sensitivity analysis in which all 1-week, 2-week and 3-week windows between 0 and 31 days post-discharge event were analysed.

| Statistical analysis
We used a Cox proportional hazards regression model 21 to estimate the effect of discharge on the rate at which homes first became affected by COVID-19. Since we defined the (baseline) exposure period as 7-21 days post-discharge, we considered the factor "hospital discharge" as a time-dependent covariate in the model. Thus, any home could potentially move back and forth between the at-risk or not at-risk categories over time. Additional covariates investigated were obtained from CIW: size of home, services available (nursing, specialist care for dementia or learning disabilities) and region (health board). Hazard ratios were calculated for the unadjusted univariable models and for the mutually adjusted full model. In our sensitivity analysis, we considered the wide range of possible exposure windows (between 0 and 31 days), controlling for the false discovery rate using q values. 22 We also calculated outbreak event rates per 1000 days of exposure to hospital discharge compared with the event rate per 1000 days unexposed, and stratified these by care home size. This study period timeline is shown in Figure 1. The overview depicts how the 7-21 day risk period follows a discharge date. Each time there was a discharge to a home, a new risk period was added to the model. Depending on timing of discharges to a home, risk periods in that home could be consecutive (scenario A), not occur (scenario B) or be overlapping (scenario C). In the case of overlapping risk periods, these were considered cumulative in our model, extending the overall risk period. As such, our model accounted for the possibility of care homes having none, some or all risk periods overlapping. Our end point was time to first outbreak, hence, if an outbreak occurred in a home (scenarios B, C and D), it was censored at that point.

| Patient and public involvement
Due to the nature of this study (analysis of routine data) and the imperative to analyse data and report results rapidly to support public health responses to COVID-19, it was not possible to involve patients and the public in this study.

| RE SULTS
Of the 1068 care homes in the analysis, an outbreak was recorded  Table 1.

F I G U R E 1
Study period analysis timeline with risk period interaction scenarios. Scenarios: (A) Two non-overlapping exposure periods, no outbreak; (B) No exposure to hospital discharge, outbreak occurs; (C) Two overlapping periods of exposure, outbreak occurs later when not exposed; (D) Two non-overlapping periods of exposure, outbreak occurs during the second discharge period In the Cox regression, time-dependent exposure to hospital discharge in the univariable model, with no other factors, was associated with a significantly increased hazard ratio for the risk of an outbreak (2.47, 95% CI: 1.96, 3.11, see Table 2). Similarly, significant univariable effects of size, dementia care, service subtype (nursing care), learning disability provision and regional health board were detected. However, in the mutually adjusted model, there was no significant association for hospital discharge, service subtype, dementia care or learning disability provision. The adjusted hazard ratio for hospital discharge was slightly raised, at 1.15, but with a 95% CI from 0.89 to 1.47 (P = .29). The results indicate strong confounding in the raw data by care home size, which was by far the strongest independent predictor of outbreak risk. In comparison with the reference category of small care homes with 1-9 residents, the hazard ratio for homes with 10-24 residents was 3.40 (1.99, 5.80). For homes of 25-49, the hazard ratio was 8.25 (4.93, 13.81) and for the largest category of homes (50+) it was 17. 35 (9.65, 31.19). The effect of health board largely mirrored the regional size of the epidemic and therefore acted as a marker of prevalence. Proportional hazard assumptions were met The confounding effect of care home size on observed univariable effect of hospital size can clearly be seen by considering the outbreak event rate per 1000 days at risk from hospital discharge (within the window) and comparing it to the event rate when not exposed. Over all care homes, there was a recorded 6.67 outbreaks per 1000 days of exposure to hospital discharge, compared to 2.47 outbreaks per 1000 days not in the exposed window. However, after stratifying by home size there were no significant differences at any care home size category. For example, the largest (50+) care homes recorded 14.05 (95% CI 10.08, 18.22 per 1000 days when exposed to a hospital discharge, and a similar 11.69 (95% CI 8.53, 14.99) outbreaks per 1000 days when unexposed (see Table 3).
In our sensitivity analysis, considering a wide range of possible time-dependent exposure windows, no q values for hospital discharge reached significance at either the 5% or 10% level. The estimated overall proportion of true null hypotheses (π 0 ), was 1.0. The smallest q value was 0.14, associated with an observed hazard ratio of 1.43 at a window of 10 to 31 days, and implying a minimum false discovery rate (fdr) of 14% incurred if considered significant. Very similar results were obtained using the local false discovery rate, and the estimated π 0 remained 1.0 across all values of fdr tuning parameter λ. We note that when considering only hospital discharge and care home size in the model (omitting all other non-significant covariates) the results were almost identical. Finally, we considered the effect of the change in policy to mandate testing prior to discharge (02/05/20) by fitting the models with a factor for the two time periods. This factor was not found to be significant and did not significantly alter hazard ratios. It must be noted that although a large number of care homes and events were included in the analysis, the precision of our estimated hazard ratio for the effect of hospital discharge covers the confidence interval 0.9 to 1.5. Hence, an effect within this range cannot be ruled out, and in individual cases, the source of the introduction to the home could have been hospital discharge. Whilst it is possible that few infectious cases were discharged, or they were late in infection so not excreting, it is also possible that care home staff took specific action receiving discharged patients meaning these residents were successfully isolated in the homes. In addition, the potential increased risk of acquiring COVID-19 had they not been discharged to care homes should be considered. Remaining in hospital is not without risk, and there was a rationale for expediting discharges, given the expected influx of COVID-19 cases to hospitals.

| Limitations
Clearly, not all discharges would have had COVID-19, so the effect of our defined risk factor would be diluted by non-risk discharges.
However, the aim was to see an overall effect of the pattern and policy of discharges. It was not possible in this study to link data at an individual level and therefore to ascertain if the case in outbreaks was the resident who had been discharged from hospital within the period of interest. This will be the focus of further investigation. The TA B L E 1 Summary statistics for hospital discharges and positive SARS-CoV-2 tests in residents of care homes in Wales, 22 February to 27 June 2020

| CON CLUS I ON S AND RECOMMENDATIONS
The requirement for a hospital patients to provide a negative RT-PCR for SARS-CoV-2 remains in place at the time of writing (December 2020 However, overall, these discharges were not a significant factor in the spread of COVID-19 to residential care in Wales.

ACK N OWLED G EM ENT
The authors would like to acknowledge the support of Care Inspectorate Wales in providing information on dementia and specialist care within care homes in Wales.

CO N FLI C T O F I NTE R E S T S
None declared.

E TH I C S
Ethical oversight of the project was provided by PHW R&D Division.
As this work was carried out as part of the health protection response to a public health emergency in Wales, using routinely collected surveillance data, PHW R&D Division advised that NHS

D I SS EM I N ATI O N D ECL A R ATI O N
Public Health Wales is working closely with Welsh Government and other stakeholders including Care Inspectorate Wales in the response to COVID-19. The results of this study will be made available to all of these stakeholders through the appropriate channels.

DATA AVA I L A B I L I T Y S TAT E M E N T
Deidentified data available on request due to privacy/ethical restrictions. Code used to analyse data was written in R and is available to