Nursing home crowding and its association with outbreak-associated respiratory infection in Ontario, Canada before the COVID-19 pandemic (2014–19): a retrospective cohort study

Background Studies conducted during the COVID-19 pandemic have shown that crowding in nursing homes is associated with high incidence of SARS-CoV-2 infections, but this effect has not been shown for other respiratory pathogens. We aimed to measure the association between crowding in nursing homes and outbreak-associated respiratory infection incidence and related mortality before the COVID-19 pandemic. Methods We conducted a retrospective cohort study of nursing homes in Ontario, Canada. We identified, characterised, and selected nursing homes through the Ontario Ministry of Long-Term Care datasets. Nursing homes that were not funded by the Ontario Ministry of Long-Term Care and homes that closed before January, 2020 were excluded. Outcomes consisting of respiratory infection outbreaks were obtained from the Integrated Public Health Information System of Ontario. The crowding index equalled the mean number of residents per bedroom and bathroom. The primary outcomes were the incidence of outbreak-associated infections and mortality per 100 nursing home residents per year. We examined the incidence of infections and deaths as a function of the crowding index by use of negative binomial regression with adjustment for three home characteristics (ie, ownership, number of beds, and region) and nine mean resident characteristics (ie, age, female sex, dementia, diabetes, chronic heart failure, renal failure, cancer, chronic obstructive pulmonary disease, and activities of daily living score). Findings Between Sept 1, 2014, and Aug 31, 2019, 5107 respiratory infection outbreaks in 588 nursing homes were recorded, of which 4921 (96·4%), involving 64 829 cases of respiratory infection and 1969 deaths, were included in this analysis. Nursing homes with a high crowding index had higher incidences of respiratory infection (26·4% vs 13·8%; adjusted rate ratio per one resident per room increase in crowding 1·89 [95% CI 1·64–2·17]) and mortality (0·8% vs 0·4%; 2·34 [1·88–2·92]) than did homes with a low crowding index. Interpretation Respiratory infection and mortality rates were higher in nursing homes with high crowding index than in homes with low crowding index, and the association was consistent across various respiratory pathogens. Decreasing crowding is an important safety target beyond the COVID-19 pandemic to help to promote resident wellbeing and decrease the transmission of prevalent respiratory pathogens. Funding None.


Introduction
The COVID-19 pandemic has disproportionately affected residents of nursing homes in various countries, such as Canada and the UK. 1,2 Even before the COVID-19 pandemic, nursing homes were frequently and severely affected by outbreaks of respiratory infections, particularly influenza. 3,4 The congregate nature of nursing homes means that residents are often exposed to infection and are more likely than the general population to have severe negative outcomes of infection, due to their advanced age, comorbidities, and high degree of frailty. 5 Several design features of nursing homes could aggravate or mitigate the potential for transmission in a home; one such important feature is the degree of crowding. Crowding has previously been defined in terms of occupants per square foot, 6 proportion of occupants in single-bed rooms, ratio of occupancy to design capacity, 7 and mean number of occupants per room. 8 For nursing homes, one simple measure of the degree of crowding is the mean number of residents per occupied bedroom and bathroom in a home. 9 Increased crowding is positively associated with SARS-CoV-2 infection and mortality rates in nursing homes. 9,10 Substantial evidence suggests that crowding is associated with transmission of various infections, particularly tuberculosis, in private dwellings and in institutional settings. 6,8 Evidence on whether crowding is also a risk factor for non-SARS-CoV-2 acute respiratory infections could influence the extension of crowding restrictions beyond the pandemic period and the design standards for the construction of new nursing homes. 11 With this study, we aimed to examine the association between crowding and incidence and mortality of outbreak-associated respiratory infections in nursing homes in Ontario before the COVID-19 pandemic. We hypothesised that highly crowded nursing homes would have a greater incidence of outbreak-associated respiratory infections and deaths than would nursing homes with less crowding.

Study design and participants
We conducted a retrospective cohort study of nursing homes in Ontario. All nursing homes administered by the Ontario Ministry of Long-Term Care were included, except those homes that closed before January, 2020. The data included in this study were home-level respiratory infection outbreak data, nursing home characteristics, and aggregate resident characteristics, routinely collected and used in accordance with Ontario's Personal Health Information Protection Act. Data were anonymised before being shared with the project team. Individual consent was not required for the secondary use of nonidentifiable information (Canadian Tri-Council Policy Statement version 2: 5.5B). This study received ethics approval from Public Health Ontario's Research Ethics Board.

Procedures
Data on the distribution of bed types in nursing homes were obtained from the Ontario Ministry of Long-Term Care, Inspections Branch, extracted on Nov 17, 2020. The exposure of interest was the nursing home crowding index, which we defined as the mean number of residents per bedroom and bathroom in each home (ie, residents / [0·5 × bedrooms + 0·5 × bathrooms]). 9 Only bathrooms intended exclusively for resident use and located in sleeping quarters were included in the calculation. A home composed exclusively of single-bed rooms with private bathrooms would have a crowding index of 1, whereas a home composed exclusively of fourbed rooms with one bathroom per room would have a crowding index of 4. If half of residents resided in singlebed rooms with private bathrooms, and the other half resided in four-bed rooms each with its own shared bathroom, then the crowding index would be 2·5. Because we did not have direct measurements of crowding in nursing homes in Ontario, we obtained information from the Ontario Ministry of Long-Term Care on the distribution of new beds and type A beds (ie, meeting the 1999 design standard) 12 versus type B, C, or D beds (not meeting the 1999 design standard), and bed class according to the 1999 design standards: private beds (ie, in one-bed rooms with a private bathroom; crowding weight=1); semi-private beds (ie, in one-bed rooms with a shared bathroom; crowding weight=1·5); and basic beds (ie, in two-bed rooms with one bathroom; crowding weight=2). For older homes not meeting the 1999 design standard, private rooms are one-bed rooms that can have shared bathrooms (crowding weight=1·5) and semiprivate rooms are two-bed rooms (crowding weight=2). The number of residents per basic-bed room in older homes varied substantially (ie, one room could contain up to five beds). We requested survey data from 2022 from the Ontario Ministry of Long-Term Care to establish the mean number of residents per basic-bed room in older homes. The obtained figures of 2·04 residents per room for municipal homes, 2·50 residents per room for non-profit homes, and 3·20 residents per room in for-

Research in context
Evidence before this study Viral respiratory infections are an important cause of morbidity and mortality among older people, particularly residents of nursing homes. Crowded living conditions are associated with SARS-CoV-2 incidence, specifically among residents of nursing homes. We searched PubMed and Google Scholar for original research articles published in any language from inception to Aug 15, 2022, examining the association between crowding and respiratory infection incidence or mortality among nursing home residents, using the following terms: [("crowding" OR "overcrowding") AND ("respiratory infection" OR "COVID" OR "SARS-CoV-2" OR "influenza" OR "respiratory syncytial virus" OR "metapneumovirus" OR "parainfluenza" OR "rhinovirus" OR "coronavirus") AND ("long-term care" OR "care home" OR "nursing home")].

Added value of this study
To our knowledge, this is the first study measuring the association between crowded living conditions in nursing homes and increased incidence of respiratory infection and mortality for non-SARS-CoV-2 viral respiratory pathogens, including influenza A, influenza B, non-SARS-CoV-2 coronavirus, respiratory syncytial virus, human metapneumovirus, human parainfluenza, and rhinovirus or enterovirus.

Implications of all the available evidence
Taken together, evidence suggests that residents of crowded nursing homes with a high proportion of shared bedrooms and bathrooms are at a higher risk of a wide range of viral respiratory infections and associated mortality, including but not limited to SARS-CoV-2, than are nursing homes with a low proportion of shared bedrooms and bathrooms. Crowding in nursing homes is an important public health and patient safety target and should be monitored and reduced.
profit homes were used as weights applied to the proportion of basic beds in old homes.
Data for respiratory infection outbreaks were retrieved from the integrated Public Health Information System of Ontario held by Public Health Ontario and were extracted by MW on June 18, 2021. Under the Ontario Health Protection and Promotion Act, respiratory infection outbreaks in nursing homes must be reported provincially. Respiratory infection outbreaks are declared when two cases of acute respiratory infection (ie, new or worsening cough or shortness of breath) with an epidemiological link (eg, on the same unit or floor) have onset within 48 h in a nursing home, including one laboratory-confirmed case, or three cases regardless of laboratory confirmation. Up to four early specimens per outbreak are tested by real-time RT-PCR for a panel of respiratory pathogens at the Public Health Ontario laboratory; most outbreaks have at least two laboratoryconfirmed infections. 13 Data for the distribution of nursing home characteristics were obtained from the Ministry of Long-Term Care, Inspections Branch, extracted on Nov 17, 2020. We included data for ownership (ie, private for-profit entity, private non-profit, or owned by a municipality), number of beds, and health region (ie, east, central-east, Toronto, central-west, south-west, and north) of nursing homes.
Aggregate characteristics of residents in nursing homes were obtained from the Resident Assessment Instrument Minimum Data, extracted for January, 2020, on Nov 17, 2020. 14 The included variables were mean resident age, proportion of women, prevalence of each of six comorbidities (ie, dementia, diabetes, chronic heart failure, renal failure, cancer, and chronic obstructive pulmonary disease), and mean activities of daily living score (in which 0 indicates independence and 6 indicates total dependence).

Outcomes
Our primary outcomes were the incidence rate of outbreak-associated respiratory infections per 100 nursing home residents per year, recorded as part of a respiratory infection outbreak in a nursing home, and the incidence rate of outbreak-associated deaths per 100 nursing home residents per year, including only deaths as a result of the infection (as decided by the outbreak investigator or the most responsible physician). Secondary outcomes included outbreak frequency per year and outbreak size as a proportion of the number of residents in the home. We also examined incidence of infections and deaths for pathogens, classified into ten categories: coronavirus (strains OC43, 229E, NL63, and HKU1); influenza A; influenza B; human metapneumovirus; human parainfluenza virus; respiratory syncytial virus; rhinovirus or enterovirus; other (adenovirus); more than one agent; and unidentified. For descriptive purposes, we also examined case fatality (ie, the proportion of infections resulting in death).

Statistical analysis
We measured descriptive statistics (ie, median, IQR, and proportions) for homes with a high crowding index (ie, ≥2) and low crowding index (ie, <2) for each nursing home characteristic.
We used negative binomial regression with an offset consisting of the logarithm of the number of beds in the home to model the incidence rate of outbreak-associated respiratory infections and deaths occurring within a home and outbreak size, and we used negative binomial regression without an offset to model outbreak frequency. Negative binomial models were used for all outcomes because preliminary analyses suggested overdispersion relative to the Poisson distribution. All unadjusted models included the continuous crowding index as the only covariate. All adjusted models included the continuous crowding index, in addition to home characteristics (ie, ownership, number of beds, and region) and aggregate resident characteristics (ie, age, female sex, dementia, diabetes, chronic heart failure, renal failure, cancer, chronic obstructive pulmonary disease, and activities of daily living score). Number of beds in nursing homes was included as a 3-df-restricted cubic spline with knots at the 33rd and 66th percentiles. The continuous crowding index effect reflected estimated associations per one resident per room increase in the crowding index.
Statistical analysis was conducted using R version 4.1.0; negative binomial models were fit with the gam and glmer.nb functions in the mgcv and lme4 packages, and random-effects meta-analysis was done with the metafor package. Post-hoc model validation steps included a comparison of the negative binomial modelling approach with quasipoisson; a comparison of the R mgcv and R MASS packages for fitting negative binomial regression; a comparison of binary (ie, <100 vs ≥100), quintile, linear, and spline-based adjustments for number of beds in nursing homes; and a comparison of binary, linear, and spline-based modelling of the crowding effect.
We ran separate models for incidence of infections and deaths for each of ten pathogen groups. For each outcome, we recombined pathogen-specific models using random-effects meta-analysis and reported the combined meta-analytic estimate and the degree of heterogeneity between the pathogen specific estimates, using the Higgins (I²) statistic. 15 Additionally, we conducted a sensitivity analysis restricted to the eight single identified pathogens (ie, excluding outbreaks due to unidentified pathogens and outbreaks due to more than one confirmed pathogen, which could have been subject to outcome misclassification).
We used marginal standardisation to simulate the annual incidence of respiratory infections and deaths for the same population of nursing homes over the same time period, but wherein the crowding index was reduced among the more crowded homes. 16 We used the fitted models for each outcome to generate predictions for the estimated incidence of infections and deaths in scenarios where the population was identical but the crowding index of nursing homes was reduced. We produced estimates for two scenarios, where all rooms with more than two beds were replaced with two-bed rooms, and where rooms with two or more beds were replaced with one-bed rooms (more than 95% of nursing home residents in Ontario are housed in one-bed, twobed, or four-bed rooms); 95% CIs were based on the 2·5th and 97·5th percentiles of 10 000 parametric bootstrap samples. 17

Role of the funding source
There was no funding source for this study.

Results
Between Sept 1, 2014, and Aug 31, 2019, 5107 respiratory infection outbreaks across 588 nursing homes in Ontario were recorded. 186 outbreaks were excluded because they corresponded to homes that were not included in the study, since they either closed or were not administered by the province, and 4921 (96·4%) were included in the study. In these outbreaks, 64 829 cases of acute respiratory infection across 376 586 resident-years of follow-up (17·2 cases per 100 resident-years) and 1969 deaths occurred (table 1), for a case fatality of 3·04% (1969 of 64 829).
Influenza A was the most common pathogen identified among outbreak-associated cases and had the highest incidence rate, followed by rhinovirus or enterovirus and influenza B (table 2). Among single identified pathogens, influenza A caused the most outbreak-associated deaths, followed by influenza B and respiratory syncytial virus. 251 (42·7%) of the 588 nursing homes included in this study were classified as crowded (ie, crowding index ≥2; Data are n (%) or median (IQR). For resident characteristics and outcomes, we first measured the facility means (for continuous variables) or percentages (for binary outcomes) for all 588 facilities, and then reported the median and IQR of these facility-level variables. *Score of 0 (independence) to 6 (total dependence).  often private for-profit homes and tended to have fewer beds than homes with a low crowding index. Compared with homes with a low crowding index, homes with a high crowding index tended to have younger residents, fewer women, and fewer residents with clinically diagnosed dementia, but more residents with diabetes. Race and ethnicity were not recorded in the data sources used for this study.
In previous work, 9 we originally used weights of 4 for all basic beds in old homes; our new calibrated approach was strongly correlated with the original approach (Pearson's test r=0·96) but yielded lower overall estimates of crowding (median calibrated crowding index=1·7 [IQR 1·4-2·5]; median original crowding index=1·9 [1·4-2·9]) and of the overall number of beds in shared rooms with three or more beds (8402 beds vs 15 812 beds) than our original approach.
Compared with nursing homes with a low crowding index, homes with a high crowding index had higher numbers of outbreak-associated infections and deaths per 100 residents per year and a higher mean outbreak size; the frequency of outbreaks was similar between the two groups (table 1). Adjusted models ( figure 1, table 3) suggested that nursing homes with higher crowding index had a higher incidence of outbreak-associated infections and deaths, outbreak frequency, and outbreak size. Results of the post-hoc model validation analyses are shown in the appendix (p 2).
We ran simulations to examine the potential effects of an intervention to change the number of people residing in shared rooms. A cap of two beds per room would have reduced infections over the 5-year period by an estimated See Online for appendix

Figure 1: Associations between nursing home crowding index and incidence of outbreak-associated respiratory infections and deaths Associations between crowding index and infections (A) and deaths (B).
To visualise trends in data for nursing homes, homes were combined into 20 equal-sized groups of 29-30 homes according to crowding index (black points). The red line shows the negative binomial model-based association with the crowding index, the shaded area showing the 95% CI. Each one-point increase in the crowding index was associated with a 1·92-times increase in incidence of infection (95% CI 1·74-2·11) and a 1·96-times increase in respiratory infection-associated mortality (1·68-2·28). Data are risk ratio (95% CI). Models for infection incidence, death incidence, and outbreak frequency were based on negative binomial regression, and outbreak size was based on logistic regression. Unadjusted analyses were univariate, and adjusted analyses included covariates for ownership type (ie, municipal, private for-profit, private non-profit), size of home, mean age of residents, proportion of women, proportion of residents with each of six comorbidities (ie, dementia, diabetes, chronic heart failure, renal failure, cancer, and chronic obstructive pulmonary disease), mean activities of daily living score, and health region of the nursing home (six regions). Coefficient estimates for all variables included in the above models are available in the appendix (p 2).

Discussion
Over a 5-year period, 64 829 cases and 1969 deaths related to respiratory infection outbreaks in 588 nursing homes in Ontario were recorded. Crowded nursing homes had a greater incidence of outbreak-related acute respiratory infections and associated mortality than did less crowded nursing homes. Crowding was positively associated with increased incidence for all ten of the respiratory virus subgroups examined, including (but not limited to) influenza A, influenza B, respiratory syncytial virus, and rhinovirus or enterovirus. Simulations estimated that deaths associated with respiratory infection outbreaks would be reduced by more than 50% if nursing home occupancy was reduced to one person per room. Systematic data on the presence of shared rooms (socalled ward rooms) in nursing homes are scarce. Although national statistics agencies in several countries collect measures of crowding for residents of private dwellings (eg, the Canadian National Occupancy Standard), the same measures are usually not collected or reported for residents of nursing homes. In the absence of systematic data from the USA, a survey from 40 randomly selected nursing homes across five US states (CA, FL, MN, NJ, and NY) indicated that 71% (1408 of 1988) of residents in nursing homes lived in shared rooms, and 28% (549 of 1988) of residents shared a bathroom between four or more residents; notably, all homes with private rooms were non-profit facilities. 18 Similarly, no systematic, country-wide data for crowding in nursing homes are available in Canada. In this study of Canadian nursing homes, crowded homes were substantially more likely to be private for-profit facilities, as has been reported previously. 19 Regulations, reimbursement schemes, and design standards for nursing homes have allowed crowding to persist, particularly in older homes. In the USA, Medicaid provides no additional funds for private rooms, inadvertently disincentivising construction of single-bed rooms. In the state of NY, USA, design standards indicate that only a minimum of 10% of rooms need to be singlebed. 20 In Ontario, Canada's largest province, the design standards of 1999 and 2015 interdict construction of rooms for three or more residents, but set no limits on the proportion of rooms with two beds, and four-bed rooms were allowed to persist through a legacy clause. 12,21,22 On June 10, 2020, between the first and second COVID-19 waves in Ontario, a temporary cap of two people per room was put into place for new admissions; the occupancy cap remains in place as of Feb 7, 2023. 11 Consultations for new design standards for nursing homes in development in Canada indicate a strong preference for single-occupancy rooms. 23 Evidence suggests that crowding is associated with increased SARS-CoV-2 incidence across a range of residential settings, including nursing homes, prisons, and households. 7,9,24,25 Our results support these findings, suggesting that crowding is not only a risk factor for SARS-CoV-2 virus but is also associated with increased incidence of other acute respiratory infections. We hypothesise that crowded sleeping quarters have increased transmission across a range of mechanisms, including aerosols, droplets, and direct and indirect contact. Regardless of the specific mechanisms of transmission, a reduction in crowding can be expected to reduce transmission rates, because nursing home residents spend a mean of 15 h per day in bed (11 h at night and 30% of the remaining hours of the day). 26 Furthermore, crowding impedes the ability to quarantine and self-isolate, which can only be partially mitigated by infection control measures. 27 A systematic review on the burden of respiratory infection in nursing homes from before the COVID-19 pandemic indicated that, for nursing home settings, "little useful guidance for decision-making to decrease respiratory infection burden" was available. 28 This absence of guidance was apparent during the COVID-19 pandemic, which had devastating effects on residents of nursing homes in many countries. Our study identified an important and modifiable risk factor for non-COVID-19 respiratory infections and deaths in nursing Crowding index risk ratio for pathogen-specific outbreak-associated infections (A) and deaths (B). Points represent the risk ratios (per one resident per room increase in crowding) and line ranges show the corresponding 95% CIs. For each pathogen, the adjustment models included number of beds in the nursing home, ownership type, and mean resident age. The overall estimate is based on random-effects meta-analysis. *Coronavirus strains OC43, 229E, NL63, and HKU1 were tested. †Human parainfluenza strains 1-4 were tested. homes, with little heterogeneity across specific respiratory infections, similar to associations identified between crowding and SARS-CoV-2. 9 Although it increases costs, reducing crowding in nursing homes brings additional benefits, such as respecting the preference of most older adults of being housed in single-bed rooms 29 and potentially reducing night-time disturbances (although sharing a room might not be a primary driver of poor sleep among residents of nursing homes). 30 The primary cost associated with decreasing crowding is financial, with the estimated construction costs of single-bed rooms being 44% higher per bed compared with two-bed rooms and 82% higher per bed compared with four-bed rooms. 29 Furthermore, nursing staff in hospitals indicate that one perceived benefit of shared rooms is decreased walking distances during nursing shifts. 31 A limitation of our study is that our analysis was based on standardised outbreak surveillance procedures that might have missed some outbreaks and cases, leading to errors in the total outbreak size, and that causal pathogens might have misattributed because only the first four outbreak cases were tested in accordance with provincial guidance. In comparison, our analyses based on deaths might be less subject to misclassification errors and showed stronger associations with less heterogeneity. Data for resident and nursing home characteristics, including crowding, were extracted in November, 2020, and respiratory infection outbreaks in nursing homes included in this analysis occurred between 2014 and 2019, suggesting that the estimated resident and home characteristics, including the occupancy rate, might have been inaccurate if the resident population changed substantially between the study and data extraction. But, at a minimum, the occupancy rate of nursing homes in Ontario has been extremely stable through time; the median occupancy rate of 98·2% that we observed is consistent with that found in a 2012 Auditor General report, and the Canadian Government strongly incentivises homes to keep occupancy rates greater than 97%. 32 Furthermore, the finding that 96·4% of total outbreaks identified were linked to an included nursing home suggests that few nursing homes closed during this time, since we excluded homes that closed before January, 2020. An additional limitation of this work is that nursing homes with a higher crowding index are likely to be different in terms of other building characteristics (ie, be older, have smaller rooms, be more crowded in common areas of the facility, and have lower ventilation rates), and might have differed in terms of other unmeasured characteristics, such as staffing, so that it might be difficult to attribute all differences to the crowding index. Finally, the estimated effects of reductions in crowding should be interpreted with caution, as this study was not preregistered or randomised and did not directly evaluate an intervention that reduced crowding.
The results of this analysis can inform decisions on the design and construction of nursing homes and on the use of multi-bed rooms in the future. Decreasing crowding in nursing homes is an important initiative to improve resident quality of life and patient safety beyond the COVID-19 pandemic.

Contributors
PL and KAB conceived and designed the study and drafted the manuscript. All authors were involved in the acquisition, analysis, or interpretation of the data and critical revision of the manuscript for important intellectual content. Statistical analyses were conducted by KAB. All authors contributed to, revised, and approved the final version of the manuscript. KAB and MW accessed and verified all the underlying data in the study. All authors had full access to all the data in the study and had final responsibility for the decision to submit for publication.

Declaration of interests
AMG reports consulting fees from Sienna Senior Living; honoraria from AstraZeneca, Merck, Biogen, and Moderna; and advisory board participation for Pfizer, GlaxoSmithKline, Moderna, Medicago, Janssen, AstraZeneca, Novavax, and Sanofi. All other authors declare no competing interests.

Data sharing
Patient and nursing home-level data contain personal health information and cannot be shared publicly due to provincial data protection and confidentiality requirements.