Covid-19 infection and attributable mortality in UK Long Term Care Facilities: Cohort study using active surveillance and electronic records (March-June 2020)

Background: Rates of Covid-19 infection have declined in many countries, but outbreaks persist in residents of long-term care facilities (LTCFs) who are at high risk of severe outcomes. Epidemiological data from LTCFs are scarce. We used population-level active surveillance to estimate incidence of, and risk factors for Covid-19, and attributable mortality in elderly residents of LTCFs. Methods: Cohort study using individual-level electronic health records from 8,713 residents and daily counts of infection for 9,339 residents and 11,604 staff across 179 UK LTCFs. We modelled risk factors for infection and mortality using Cox proportional hazards and estimated attributable fractions. Findings: 2,075/9,339 residents developed Covid-19 symptoms (22.2% [95% confidence interval: 21.4%; 23.1%]), while 951 residents (10.2% [9.6%; 10.8%]) and 585 staff (5.0% [4.7%; 5.5%]) had laboratory confirmed infections. Confirmed infection incidence in residents and staff respectively was 152.6 [143.1; 162.6] and 62.3 [57.3; 67.5] per 100,000 person-days. 121/179 (67.6%) LTCFs had at least one Covid-19 infection or death. Lower staffing ratios and higher occupancy rates were independent risk factors for infection. 1,694 all-cause deaths occurred in 8,713 (19.4% [18.6%; 20.3%]) residents. 217 deaths occurred in 607 residents with confirmed infection (case-fatality rate: 35.7% [31.9%; 39.7%]). 567/1694 (33.5%) of all-cause deaths were attributable to Covid-19, 28.0% of which occurred in residents with laboratory-confirmed infection. The remainder of excess deaths occurred in asymptomatic or symptomatic residents in the context of limited testing for infection, suggesting substantial under-ascertainment. Interpretation: 1 in 5 residents had symptoms of infection during the pandemic, but many cases were not tested. Higher occupancy and lower staffing levels increase infection risk. Disease control measures should integrate active surveillance and testing with fundamental changes in staffing and care home occupancy to protect staff and residents from infection. Funding: Economic and Social Research Council [ES/V003887/1].

Background 45 Globally the number of Covid-19 cases continues to increase, but cases in Europe have 46 declined since April 2020, 1 following the introduction of lockdown measures. 2 Although the 47 incidence of infection in the general population in England is low (0.04%), 3 new infections 48 persist, with substantially higher rates of infection reported in both long-term care facilities 49 (LTCFs) and hospitals. 4 This raises the possibility that these settings represent a reservoir for 50 transmission of infection back to the community. 51 In the UK, there are an estimated 400,000 residents living in approximately 11,000 LTCFs for 52 the elderly. 5 Residents of LTCFs are particularly vulnerable to Covid-19 due to their advanced 53 age and high prevalence of comorbidity, 6 and their frequent exposure to infection through 54 close contact with staff members, other residents and contaminated surfaces in the care 55 facility. At the peak of the pandemic, the number of deaths in residents of LTCFs was three-56 fold higher than the equivalent period in in 2019. 7 Staff in LTCFs also have higher aged-57 standardised rates of Covid-19 related mortality compared to other occupations. 8 National 58 statistics suggests two-thirds of excess deaths recorded in residents of LTCFs in the last 6 59 months involved Covid-19, 7 but this is likely to be an underestimate because many residents 60 were not tested. Understanding the proportion of excess deaths that can be directly and 61 indirectly attributed to Covid-19 infection is important, to fully assess the impact of the 62 pandemic on LTCFs. 63 The development of public health strategies to protect the public, residents and staff from 64 Covid-19 requires knowledge of the burden of and risk factors for infection in residents and 65 staff in LTCFs, linked to outcomes. However, there is no syndromic surveillance for infection in 66 LTCFs in England, and widespread one-off testing for SARS-CoV-2 using reverse transcriptase 67 polymerase chain reaction (RT-PCR) was not established for staff and residents in LTCFs until 68 11 May 2020. 9 Prior to this, testing was only available for residents or staff who were admitted 69 to hospital, or as part of Public Health England's (PHE) outbreak investigations which permitted 70 a maximum of five tests per LTCF. Consequently national estimates of incidence and 71 prevalence will substantially underestimate the burden of infection in residents and staff in 72 LTCFs. 73 In the absence of cohort studies or active surveillance, outbreak investigations provide the 74 most reliable estimates of the burden of infection and case-fatality. 10,11 An estimated 44% of 75 English LTCFs have had at least one outbreak, with a living systematic review 12 reporting 76 substantial variation in cumulative incidence of infection (0%-72%) and case fatality (0-34%) 77 in residents of LTCFs. A major limitation of outbreak investigations is that follow-up is usually 78 less than 30 days, 12  Characteristics of each LTCF (number of beds, region, nursing versus residential care) were 131 collected from organisational data. We therefore extracted organisational data, individual-132 level data for 8713 residents and aggregate data for all staff and residents ( Figure 1

Risk factors for infection 139
Risk factors for infection included individual-level variables (age, sex, general or dementia care, 140 residential versus nursing care) and LTCF characteristics (number of beds, occupancy and bed 141 to staff ratio). Baseline LTCF occupancy was computed by averaging weekly occupancy in 142 January-March 2020, before the first Covid-19 case, in order to calculate a ratio of baseline 143 occupancy to the number of bedrooms. We also estimated the ratio of bed to staff as a 144 continuous variable. An outbreak in a LTCF was defined as at least one confirmed infection or 145 Covid-19 related death. 146

Statistical analysis 147
Infection in staff and residents in LTCF's 148 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 15, 2020. . https://doi.org/10.1101/2020.07.14.20152629 doi: medRxiv preprint Prevalence, incidence and cumulative incidence were calculated for residents and staff using 149 the aggregate daily tallies. These were the trusted source of information used for national 150 reporting of cases, and encompassed all residents and staff. 14 Infection incidence was also 151 estimated from Datix, but was subject to under-reporting (supplementary material 2). In order 152 to calculate infection and death incidence, we estimated the total number of residents in each 153 LTCF by extrapolating estimates of LTCF occupancy from the individual-level dataset (because 154 dates of entry and exit to/from the LTCF were not available for 855 beds, supplementary 155 material 1). Daily occupancy was inferred from the weekly report of bed occupancy using linear 156 interpolation. The total number of residents at risk of infection was unknown, so it was 157 approximated in a multiple decrements life table (supplementary material 1). The life table 158 allowed us to compute Kaplan-Meier product limit estimators of the cumulative incidence of 159 symptoms, confirmed infections, and Covid-19 related deaths by day. The rate ratio for LTCF 160 versus community infections was estimated by contrasting the cumulative incidence for 161 confirmed cases in England with estimates from a national household survey for the period 11 162 May-7 June 2020. 3,15 163

Mortality, attributable mortality and risk factors 164
Individual-level data were used to estimate rates of infection, all-cause mortality and case-165 fatality by age and gender in residents. Aggregate data were also used to estimate the crude 166 rate of Covid-19 related mortality. Cox proportional hazards models were used to test the 167 association between individual and organisational-level risk factors and confirmed infection. 168 In order to investigate the relationship between Covid-19 infection and excess mortality, we 169 assumed that residents in non-outbreak LTCFs had not been exposed to infection, and would 170 therefore not experience excess Covid-19 related mortality. We therefore compared all-cause was obtained by using the model to predict the counterfactual mortality, then computing the 179 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 15, 2020. . https://doi.org/10.1101/2020.07.14.20152629 doi: medRxiv preprint attributable fraction within study. 16 Ninety-five percent confidence intervals for proportions 180 and rates were computed from the exact Poisson and binomial limits. 181 Data were analysed in R3·5·0 using the epitool 17   is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 15, 2020. . https://doi.org/10.1101/2020.07.14.20152629 doi: medRxiv preprint In England, 179 infections were confirmed of a total 194,023 residents-days between 11 May 207 2020 and 7 June 2020. In comparison, the survey of English community households 3 found a 208 total of 35 confirmed cases out of 483,259 person-days during the same period. This implies 209 a confirmed infection rate ratio comparing LTCFs to the community of 12·7 [8·9; 18·3]. 210  CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 15, 2020. . https://doi.org/10.1101/2020.07.14.20152629 doi: medRxiv preprint CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

244
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Factors associated with confirmed infections in residents 245
Using  . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Factors associated with all-cause mortality 259
The time-dependent Cox proportional hazard models in Table 6  It is important to note these hazard ratio estimates do not give a comprehensive measure of 273 effect: hazards were not proportional across these categories (Figure 3). 274 275 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

280
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Attributable mortality 283
Model-based estimates of attributable mortality were derived from the individual-level data. 284 Overall, 567/1,694 (33%) deaths were attributed to Covid-19. In LTCFs with outbreaks only 28% 285 (159 residents) of the mortality attributable to COVID-19 occurred in people with confirmed 286 infection (Groups C and D), (Table 7). Exclusion of the early pandemic period in sensitivity 287 analysis increased attributable mortality to 560/1,343 (41·7%). Model-based estimates of 288 deaths based on individual-level were slightly higher (8%) than counts from the aggregate 289 data. 290 291 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 15, 2020. . https://doi.org/10.1101/2020.07.14.20152629 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 15, 2020. . https://doi.org/10.1101/2020.07.14.20152629 doi: medRxiv preprint Whereas two-thirds of LTCFs in our study reported at least one case of infection or death, just 315 44% of LTCFs have notified an outbreak to PHE. This suggests that nationally, local health 316 protection teams may be unaware of Covid-19 infections in up to 1 in 5 LTCFs. Integration of 317 data systems, so that test results can be accessed and acted upon by local public health teams 318 is fundamental to the pandemic response. 319 In common with a Canadian cohort study, 24 we found strong associations between infections 320 and LTCF occupancy. We also identified lower staff to resident ratios as a risk factor for 321 infection. These organisational factors, linked to chronic underfunding of the care sector, are 322 likely to facilitate the implementation of infection control procedures 25 such as isolating or 323 cohorting infected residents, staff training, and regular environmental deep cleaning. When 324 staff care for fewer residents they also have reduced likelihood of spreading infection between 325 residents. Higher staff to resident ratios may also decrease reliance on agency staff who may 326 spread infection between LTCFs, and indicate better resourced LTCFs. 327

Strengths and limitations 328
The unique surveillance system we established in partnership with FSHCG allowed us to track 329 infections throughout the entire pandemic period across a large number of LTCFs, and identify 330 symptomatic as well as confirmed and asymptomatic cases. To our knowledge, this is the most 331 complete reporting system for Covid-19 infections in LTCFs published to date. It is possible 332 that LTCFs that paid less attention to active surveillance to support control will have had higher 333 levels of uncontrolled outbreaks compared to those seen in this study. 334 A limitation is lack of access to information on comorbidity and ethnicity, both of which have 335 been shown to be important risk factors for adverse outcomes in Covid-19. 6 However, we were 336 able to identify individuals with dementia, and adjust for receipt of nursing care which will 337 partially capture comorbidity. We also lacked information on the overall rate of testing in each 338

Clinical, research and policy implications 340
In the UK the number of infected residents and staff has been underestimated, due to limited 341 availability of testing until late in the pandemic. High levels of asymptomatic infection will also 342 lead to under-ascertainment. It is important to note however, that mortality rates were also 343 increased in those recorded as asymptomatic infections suggesting that in an elderly cohort 344 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 15, 2020. . https://doi.org/10.1101/2020.07.14.20152629 doi: medRxiv preprint residents may have atypical presentations that do not conform to standard case definitions. 345 Although our findings support increased use of testing to improve case ascertainment, 346 frequent testing in residents of LTCFs may not always be desirable if the risk of infection is low, 347 because the testing procedure (nasopharyngeal swabs) is invasive and may distress vulnerable 348 residents. Since the incubation period and serial interval of COVID-19 is short, 26 the interval 349 between successive screens required to interrupt transmission may also need to be short. 350 Rapid early diagnosis of symptomatic cases in residents and staff and expansion of more 351 widespread testing after a case is identified may also be effective strategies to prevent 352 transmission. Such approaches depend on strengthened surveillance in LTCFs and would be 353 greatly facilitated by the availability of near patient testing platforms, which may be achievable 354 in larger LTCFs. 355 Our findings of excess deaths in those with no direct evidence of infection may be due to 356 under-ascertainment, direct effects of Covid-19 control measures on delivery of care, and/or 357 indirect effects due to additional disruption caused by the outbreak. Studies from other 358 healthcare settings 27,28 have highlighted the ways in which Covid-19 has impacted delivery of 359 care associated with excess mortality in individuals who are uninfected . Detailed analysis of 360 cause of death and reasons for hospital admission in residents of LTCFs will be important to 361 understand how the pandemic has affected the quality of care in LTCFs. Our analysis provides 362 a method that could be widely applied to estimate excess mortality, provided LTCF's with 363 outbreaks can be reliably identified. 364 Globally, there is an opportunity to mitigate the impact of future waves of infection on staff 365 and residents in LTCF's. Our findings suggest that countries can achieve this most effectively 366 by adopting a holistic approach, which integrates surveillance and focused testing for Covid-367 19 with increased investment to reduce LTCF occupancy and increase staffing. All authors interpreted the data and edited and revised the final manuscript. 374 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 15, 2020. . https://doi.org/10.1101/2020.07.14.20152629 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 15, 2020. . https://doi.org/10.1101/2020.07.14.20152629 doi: medRxiv preprint CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 15, 2020. . https://doi.org/10.1101/2020.07.14.20152629 doi: medRxiv preprint