We used administrative data on hospital admissions from Secondary Uses Service, a national database of NHS-funded hospital activity in England. As care homes are not reliably recorded as admission source, care home residents were identified through linkage to the patient index from the National Health Applications and Infrastructure Services, as previously described [5, 7]. This database contains longitudinal records of patient registrations with general practices in England, including patient address and year and month of death. Data extracts are created on the first Sunday following the 13th day of the month. Patient addresses were matched to Unique Property Reference Numbers and subsequently cross-referenced to care home addresses and characteristics held by the Care Quality Commission, the regulator of social care services in England. Care home opening and closing dates were used to resolve address matches to multiple care homes, which can occur when recorded characteristics of care homes change. Small area-level socioeconomic deprivation (Index of Multiple Deprivation 2019, Office for National Statistics) was added using care home Lower Layer Super Output Areas (LSOA). All processing of addresses and linkage of patient information was carried out by the National Commissioning Data Repository. Data were anonymised in line with the Information Commissioner’s Office’s code of practice on anonymisation.

Individuals were included in the 2020 study cohort if they were recorded as living in a care home on 19 January 2020. These were compared to a cohort of individuals living in a care home on 20 January 2019. Residents of all ages and in all types of care homes were included, including specialist care homes. The cohorts were further split into subgroups based on care home type (residential or nursing) using information provided by the regulator that nursing care was being provided to some residents.

Age, sex, and month of death were taken from the patient index. The Charlson Comorbidity Index and previous dementia diagnoses for each resident were determined using primary and secondary diagnosis codes of all hospital admissions from up to three years prior to the respective study start date in January [21–23].

The follow-up period for the study was between 21 January and 30 June 2019 or between 20 January and 28 June 2020, respectively. Observations were censored at the date of death, or the end of the care home stay. The end of the care home stay was defined as the last day of the monthly data extract where an individual’s address matched to a care home. The date of death was defined as the last day of the month of death, or the last day of the data extract ending in the month of death, whichever was later.

Hospital admissions were not included if the administrative category was private patient, the admission method was a transfer or missing. Admissions were divided into emergency and elective admissions based on the method of admission. Admission rates resident per year or per 100 residents per year were calculated across all residents, controlling for the number of days spent in a care home. Primary admission reasons were categorised according to the International Classification of Diseases 10^th revision (ICD-10) chapter of the primary diagnosis code (Appendix Table S1).

Several acute conditions, as defined by ICD-10 code, were selected for further analysis: suspected or confirmed COVID-19 (U07.1, U07.2) [24], acute coronary syndromes (I20.0, I21.0-4, I21.9, I22.0-2, I22.8-9, I24.8-9) [25], and stroke (I61, I63, I64) [26]. A small number of COVID-19 admissions (n = 37, 0.59% of COVID-19 admissions) were coded as elective; these were not analysed separately. Cataract surgeries were defined using Office of Population Censuses Surveys Classification of Surgical Operations and Procedures (OPCS4) codes C71-C75 [27], based on all procedure codes for any given admission.

We also examined a subset of unplanned hospital admissions from care homes, which have been referred to as potentially avoidable emergency admissions: either because they are generally considered manageable, treatable or preventable outside a hospital setting or because they can be caused by poor care or neglect [28, 29]. Potentially avoidable emergency admissions have previously been used as primary outcome measure to evaluate the impact of several initiatives to improve health and care in care homes in England [5, 7]. This list of conditions, which was developed by the health and social care regulator, includes acute and chronic lower respiratory tract infections, pressure sores, diabetes, food and drink issues, food and liquid pneumonitis, fractures and sprains, intestinal infections, pneumonia, and urinary tract infections. ICD-10 code lists are shown in Appendix Table S2, including corrections made after consultation with the authors [29].

Although these admissions are not necessarily inappropriate and often cannot be avoided once the condition becomes acute, they may have been preventable at an earlier stage with additional support or better care coordination [5]. The degree to which potentially avoidable admissions could in fact have been avoided will also be influenced by residents’ other health conditions as well as organisational context. Nursing support available in a nursing home may enable staff to oversee treatments that might otherwise require an admission to hospital. However, enhanced primary and community care for care home residents during the COVID-19 outbreak could be expected to have an impact on this group of admissions. Due to the heterogeneity within this group of potentially avoidable admission reasons, we also considered them separately to examine changes in the number of admissions during the pandemic.

In Figure 1 showing trends in weekly admission rates, a locally estimated smoothing function was fitted for each group (using the geom_smooth function of the R package ggplot2 and the default loess smoothing function). To assess the similarity between years of baseline characteristics of study cohorts and of admitted patients, absolute standardised mean differences were calculated using the R package tableone. They are defined as the difference in means of a given characteristic as a percentage of the pooled standard deviation, with a cut-off value of 10% being widely adopted for negligible imbalance between groups [30]. Standardised mean differences allow comparisons of the magnitude of difference between groups, rather than statistical significance. They can be used across variables of different scales and have the advantage that they are not dependent on sample size. We used R (version 3.6.3) for data processing and analysis and SAS (version 7.12) for data cleaning and analysis of comorbidities [31].

We identified 252,432 residents living in a care home in January 2019 and 257,843 residents in January 2020 (45% in nursing homes in both years) who met the inclusion criteria. A data cleaning flow chart is shown in Appendix Figure S1. The demographic characteristics of individuals living in residential and nursing homes were broadly similar in January 2019 and 2020, but mortality during the follow-up period was higher in 2020 (Appendix Table S3 and Appendix Figure S2).

To examine trends in hospital admissions, we quantified weekly rates of COVID-19 admissions, other emergency admissions and elective admissions between 20 January 2020 and 28 June 2020 and compared these with the corresponding period in 2019 (Figure 1). For each type of admission, as well as for categories of primary admission reasons, we subsequently compared the number of admissions during the first three months of the COVID-19 outbreak in England, between 1 March and 31 May 2020, with the same period in 2019. This period was chosen as it corresponded to the first wave of the pandemic in care homes (COVID-19 admissions in Figure 1).

A locally estimated smoothing spline was fitted to the weekly admission rates for 2019 and 2020. The date of the UK-wide COVID-19 lockdown (23 March 2020) and the date on which NHS providers were asked to resume elective activity (29 April 2020) are shown as vertical dashed lines. Elective admissions include ordinary admissions and day cases. Emergency admissions (non-COVID-19) are defined as admissions that had a primary diagnosis code other than confirmed or suspected COVID-19. COVID-19 admissions are only shown for weeks where there were more than 10 admissions from both types of care home. w/c; week commencing.

To free hospital capacity for patients critically ill with COVID-19 and to reduce the risk of infection during the first wave of the pandemic, local health systems and the NHS in England rapidly reorganised care pathways. This led to concerns that care home residents may have faced barriers to accessing hospital care. During the first wave of the pandemic in care homes, between 1 March and 31 May 2020, there were 6,255 admissions for suspected or confirmed COVID-19 from residential care settings, which peaked during the first half of April. Admissions for COVID-19 were skewed towards larger care homes and care homes located in London and the West Midlands. London was the region with the highest levels of SARS-CoV-2 antibody prevalence in the general population in July 2020 and international evidence strongly suggests that there is a relationship between high community prevalence of COVID-19 and COVID-19-related mortality rate in care homes [32–35]. Compared to the same period in 2019, there were 13,191 fewer emergency admissions for reasons other than COVID-19, a decline of over a third. This was only partially due to a decrease in potentially avoidable conditions, as the proportion of emergency admissions that were potentially avoidable was broadly similar to the previous year. However, we observed substantial changes in the number of admissions for some potentially avoidable reasons, such as UTIs. The most substantial drop was seen for elective admissions, where there were 3,762 fewer admissions from residential care settings than in 2019, a decline of close to two thirds.

Our study has several strengths. It is based on a novel linkage methodology to identify care home residents in routinely collected hospital data, as care home residents are not consistently flagged in administrative hospital records. With a positive predictive value of over 99% compared to manual address matching, this method allowed us to only include admissions for patients that were highly likely to be genuine care home residents [36]. The study also captures all admissions to NHS hospitals in England. Therefore, it provides the first comprehensive overview over hospital inpatient activity of care home residents in England during the pandemic. Our study has some limitations. A recent validation study showed that, compared to manual address matching, the linkage methodology misses around 22% of care home residents [36]. However, due to the stratified sampling design used in the validation study, it does not allow us to assess whether the linkage quality varies by certain care home characteristics, such as size and rurality. As the linkage method relies on patient addresses held by general practices, we may not capture admissions for residents who moved into a care home around the time of the study start, who moved into a care home on a temporary basis or where the address had not been updated. There might also be variation in how timely patient addresses are updated. Due to these methodological constraints, this study was designed as a cross-sectional analysis of the population of residents who were living in a care home before the onset of the COVID-19 pandemic, with an adjustment for the time spend in a care home. A caveat to extrapolating these findings to the current population of care home residents is therefore that there may have been systemic changes in resident characteristics, due to the uneven impact of COVID-19 mortality across care homes, changes in care home occupancy levels or the influx of new residents [3, 37, 38]. While the study cohorts were comparable in observed characteristics, another limitation of observational studies is that unobserved differences between study cohorts or hospital records could have confounded the estimates. During the early stages of the outbreak, COVID-19 diagnoses might not have been accurately captured in hospital records, and for all admissions the severity of the condition would not be captured in the data. Following a COVID-19 incident in a care home, other residents in the same home may have been less likely to seek hospital treatment for other care needs due to isolation measures. This was not possible to be determined without access to linked data on care home outbreaks. The list of potentially avoidable reasons was developed by the regulator for an analysis focusing on trends in hospital care use by older people but was not specifically designed for care home residents [29]. While it is based on a list of commonly defined ambulatory care sensitive conditions and captures the most common preventable causes of harm in care homes [39], further validation of the individual conditions as a marker of potentially avoidable admissions in the care home population is needed.

The baseline hospital admission rates for our cohort of residents, as well as their relative proportions, are similar to a previous study of care home residents in England aged 65 or over, which reported 0.77 and 0.63 annual emergency admissions per residential and nursing home resident, respectively [7]. The small difference in observed rates might be explained by the different age cutoff and the fact that we used data from March, April and May, rather than data on admissions across a whole year. We also found that there were more COVID-19 admissions from nursing homes than from residential care homes. This is consistent with emerging evidence that care home size is an important risk factor for experiencing a COVID-19 outbreak and nursing homes on average have a greater number of beds [40]. In our study, 2.2% of residents experienced a hospital admission for COVID-19 between 1 March and 31 May 2020. A survey of 9,081 care homes in England between 26 May and 20 June 2020 reported that 11% of residents had tested positive for COVID-19 [41]. Excess mortality attributable to COVID-19 in care homes in England up to 7th August 2020 has been estimated to be equivalent to 6.5% of care home beds [3].

Our analysis shows that care home residents’ use of inpatient hospital care decreased during the first three months of the pandemic. The largest decreases were observed for elective care and some potentially avoidable admissions, such as urinary tract infections. We saw large variation within the group of conditions associated with potentially avoidable admissions and their changes during the COVID-19 outbreak, compared to the previous year. However, interpreting these differences is complex, as not all conditions within this group will be equally amenable to prevention and treatment in the community. Some acute emergency conditions, including acute coronary syndromes and stroke, also saw decreases admissions, indicating that fewer residents received appropriate urgent hospital care.

These changes are likely to be driven by several factors. Firstly, avoiding hospital admission has previously been seen as preferable by both clinicians and residents [42]. This sentiment may have grown stronger due to the risk of contracting COVID-19 in hospital [17]. Secondly, as previous evidence shows that some avoidable emergency admissions can be reduced through enhanced primary care in care homes [5], the additional support provided by primary and community care teams may have been successful in avoiding some of these during the COVID-19 outbreak [43]. Consistent with the observed variation in the changes of avoidable admission reasons seen in this study, some avoidable causes, such as UTIs, might be more amenable to intervention without an admission. Finally, visiting restrictions and enhanced infection control measures might have led to fewer admissions due to other communicable infections. However, our analysis does not allow us to differentiate whether admissions were prevented through out of hospital care, whether treatment in the care home was prioritised, or whether conditions remained undiagnosed. In addition, further research is needed to examine the outcomes for patients and residents during this period. The rapid and sustained changes in care pathways highlight the increased need care workers and families have for support from geriatricians and other health care staff in advance care planning with patients [44].

Elective admission rates for all conditions were lowest during April and rose through May and June but remained far below historical levels until the end of June. As non-urgent elective procedures were paused for all patients in March 2020, it is important to interpret these findings within the context of overall admission patterns. According to data published by the NHS, elective admissions across the healthcare system fell by 56% between March and May 2020, compared to the same period in the year before [45]. The findings of this study therefore suggest that care home residents were affected more strongly by the reduction in hospital admissions, with potentially concerning consequences for unmet elective care need. If unmet need is concentrated in high symptom burden conditions such as cataract then, while not prolonging life, lack of treatment will have a significant impact on quality of life [46].