Main

Hospital transmission had a central role in the spread of Middle East respiratory syndrome coronavirus (MERS-CoV) and severe acute respiratory syndrome coronavirus (SARS-CoV) in human populations1,2, and multiple reports have indicated that SARS-CoV-2 is capable of spreading efficiently in healthcare settings3,4,5,6,7,8,9,10,11 and is associated with poor outcomes12,13. However, attempts to fully document the extent of hospital transmission using systematically collected national data or to take a data-driven approach to quantifying the drivers of such transmission are lacking. Addressing these knowledge gaps is important: hospital transmission directly affects patients likely to have multiple factors associated with poor outcomes; it puts healthcare workers (HCWs) at risk and compromises their ability to provide safe patient care; it disrupts service delivery; and it can have a major role in disseminating infection to vulnerable groups in the community. Moreover, because non-pharmaceutical interventions in the community do not affect rates of transmission from infected patients and HCWs in hospitals, hospital transmission can have notable effects on epidemic dynamics during lockdown periods. Understanding such transmission has implications for both continuing epidemics and threats from new variants even in highly vaccinated populations. We use data from 145 English National Health Service (NHS) acute hospital trusts (organizational units containing one or more acute care hospitals), excluding only those caring exclusively for children. These trusts contained 356 hospitals, had a combined bed capacity of about 100,000 (over 98% of the total NHS general and acute care bed capacity in England in 2020) and employed 859,000 full-time equivalent HCWs, 2.5% of the working-age population of England. From 20 March 2020, all such trusts completed a daily situation report that included essential information on the prevalence and incidence of SARS-CoV-2 infection, the number of patients admitted with SARS-CoV-2 infection and staff absences due to SARS-CoV-2. From 5 June 2020, a classification of the likely source of SARS-CoV-2 infection on the basis of European Centre for Disease Prevention and Control (ECDC) criteria was also required14. This was determined by the interval between hospital admission and date of onset of PCR-confirmed infection in hospitalized patients: community-onset infections were defined as those with an interval of 2 d or less; an interval of 3–7 d led to a classification of indeterminate healthcare-associated; intervals of 8–14 d were classified as probable healthcare-associated; and intervals of 15 d or more were classified as definite healthcare-associated. As few patients have hospital stays exceeding 7 d and many nosocomially infected patients will be discharged before testing positive, such definitions necessarily capture only a proportion of hospital-acquired infections15.

We make use of these data, linked with other national datasets, to infer the number of hospital-acquired infections in England between June 2020 and February 2021, the pathways of nosocomial transmission and factors potentially modulating such transmission, including hospital characteristics, vaccination coverage and prevalence of relevant variants. Using a model coupling hospital and community dynamics, we then explore the consequences of such nosocomial transmission for the effectiveness of community lockdown measures in averting infections.

Between 10 June 2020 and 17 February 2021, a total of 16,950 and 19,355 SARS-CoV-2 infections in hospital inpatients met the criteria for definite and probable healthcare-associated infections, respectively, corresponding to a median (interquartile range) of 1.7 (1.1, 2.5) detected infections per 1,000 occupied bed days. To estimate the total number of hospital-acquired infections, we multiply the recorded number of definite healthcare-associated infections by the reciprocal of the proportion of hospital-acquired infections that we expect to meet these ‘definite healthcare-associated’ criteria. Using the empirical length-of-stay distribution, the estimated incubation period distribution and the profile of PCR test sensitivity as a function of time since infection16 (Fig. 1a–c), we estimate that a policy of PCR testing of symptomatic patients would detect 26% (90% credible interval (90% CrI) 21%, 30%) of hospital-acquired infections, with 12% (10%, 14%) of all hospital-acquired infections meeting criteria for definite healthcare-associated infection (Fig. 1d–f). Adding asymptomatic PCR testing on days of stay 3 and 6 (as recommended by national screening guidance in England at the time) increases the proportion detected to 33% (26%, 38%) but does not substantively alter the proportion classified as definite healthcare-associated. Augmenting symptomatic PCR tests with testing for all patients at 7 d intervals (a policy adopted by some hospitals in England) increases the proportion of hospital-acquired infections detected to 44% (39%, 47%), and the proportion classified as definite healthcare-associated to 17% (16%, 18%). These low probabilities for detection and classification as definite healthcare-associated are a consequence of the typically short lengths of patient stay and low PCR sensitivities early in the course of infection (Fig. 1b,c).

Fig. 1: Quantifying the probability of observing hospital-acquired infections and estimating the total number of such infections.
figure 1

a–c, Model inputs are shown in the top row and include the incubation period distribution32 (a), the PCR sensitivity profile16 (b) and the length-of-stay distribution (c) for patients who were not admitted with COVID-19 between June 2020 and February 2021 (solid lines in a and b show expected values and shaded regions show 95% CrIs). In c, the minimum lengths of stays needed to be classified as a probable or definite healthcare-associated infection are shown by dashed and solid vertical lines. d–f, Estimates of the probabilities that patients with hospital-acquired SARS-CoV-2 infections have a PCR positive test while in hospital under different screening policies (d), and estimates of the probabilities that they both screen positive and meet the post-14 d onset criteria to be considered a ‘definite’ healthcare-associated infection (e) or the post-7 d criteria to be classified as a probable or definite healthcare-associated infection (f) are shown in the middle row. Panels d–f are on the basis of 1,000 Monte Carlo samples, with violin plots showing median values (points), interquartile ranges (rectangles) and densities. The Public Health England screening recommendations are highlighted in green and the policy of screening all patients at 7 d intervals after admission is highlighted in blue (note that in contrast to this policy, weekly and 2× and 3× weekly policies screen on fixed days of the week). g, The estimated total number of hospital-acquired infections across adult acute NHS trusts in England linked to observed weekly number of detected post-14 d onset infections, assuming the screening policies highlighted in the middle row on the basis of recorded ‘definite healthcare-associated infections’; week numbers are counted as 1 plus the number of complete 7 d periods since 1 January 2020. Green and blue shaded regions indicate 90% CrIs and white lines are posterior means.

Full size image

Combining these estimates with the number of reported definite healthcare-associated infections, we infer the number of hospital-acquired infections under two sets of assumptions. First, we assume patient testing followed national guidance at the time, which specified testing of symptomatic patients (without retesting) and included asymptomatic testing on two occasions in the first week but none after day 7 postadmission. This provides a plausible lower bound for the chance of identifying hospital-acquired infections and thus an upper bound for the estimated numbers of such infections. Second, we assume testing for all patients at 7 d intervals postadmission in addition to symptomatic testing of patients (the maximal testing policy known to be used in practice). This provides a plausible upper bound for the chance of identifying hospital-acquired infections and thus a lower bound on the estimated numbers of such infections. Using definite healthcare-associated infections only, this yields as an upper bound a mean (90% CrI) estimate for the number of hospital-acquired infections of 143,000 (123,000, 167,000) and a lower bound of 99,000 (95,000, 104,000). During this period there were 9.2 million hospital admissions from 5.0 million individual patients, so we estimate that between 1% and 2% of admissions developed a hospital-acquired SARS-CoV-2 infection. Similar estimates are obtained when using more granular length-of-stay data and in other sensitivity analyses, whereas repeating the analysis using probable and definite healthcare-associated infections yields estimates that are 20–30% higher (Supplementary Information section 2.1).

There is considerable variation in cumulative rates of hospital-associated infection between trusts, with the highest rates seen in the North West NHS region, and the lowest in the South West and London regions (Extended Data Fig. 1). There is a strong positive correlation between rates of definite and probable hospital-associated infections (r = 0.76), and weak positive correlation between definite hospital-associated infection and HCW infection (r = 0.31), but only a very weak correlation between definite hospital-associated infection and community-acquired infection (r = 0.16). Three hospital characteristics are weakly correlated with cumulative rates of definite hospital-associated infection: bed occupancy (r = 0.25), availability of single-bedded rooms (r = −0.39) and heated volume per bed, a measure of the volume of heated areas of trust buildings divided by the number of beds (r = −0.34).

To quantify drivers of transmission to patients and HCWs we link these data to national datasets (Fig. 2e–l), capturing information on hospital characteristics potentially affecting transmission, alongside regional variation in HCW vaccination and prevalence of the Alpha variant. As no direct measurements of hospital ventilation are available, we use hospital building heated volume per bed as a proxy. This analysis is restricted to 96 of the 145 trusts for which complete data are available and uses negative binomial auto-regression models for which the dependent variable is either the weekly number of patients with healthcare-associated infections or the imputed weekly number of HCWs with confirmed SARS-CoV-2 infection. Independent variables are selected on the basis of biological plausibility. Mechanistic considerations inform the parameterization of the dispersion terms and the inclusion of additive effects for exposures to community-acquired patient infections, hospital-acquired patient infections and infected HCWs (Fig. 2, top row), combined with multiplicative effects of trust characteristics (Fig. 2, middle row), HCW vaccine coverage and Alpha variant prevalence (Fig. 2, bottom row). Posterior predictive distributions from fitted models are shown in Fig. 3 and Extended Data Figs. 2 and 3.

Fig. 2: Summary of data used in the analysis.
figure 2

a–d, Data from situation reports related to SARS-CoV-2 infection in England showing variation between trusts. Each circle corresponds to one NHS trust scaled by the number of available beds. Shading indicates cumulative totals to the end of the period considered (17 February 2021). Geographic locations are approximate. Cumulative number of hospital-associated SARS-CoV-2 infections in patients per 100 hospital beds with first positive sample more than 14 d after admission (a); hospital-associated infections in patients with first positive sample more than 7 d after admission (b); imputed cumulative number of cases in HCWs (c) with grey shading indicating missing data; infections in hospitalized patients with community-onset (d). e–h, Trust-level data characteristics from the third quarter of 2020: bed occupancy (e); age of acute hospital buildings in the trust expressed as a weighted average of the percentage of hospital buildings constructed in 1964 or earlier, for which weights are the hospital gross internal floor areas (f); number of single room beds per trust (including isolation rooms) as a percentage of the number of general and acute beds available in the last quarter of 2020 (g); heated volume per bed (h). i–l, A snapshot of regional HCW immunization data at two time points showing the proportion of HCWs who had received at least one vaccine dose at least 3 weeks earlier at week 52 (i) and week 59 (j), and regional data on the proportion of PCR-confirmed infections due to the Alpha variant at week 42 (k) and week 52 (l) (in both cases, Voronoi tessellations centred on the location of the largest hospital in each trust are shown).

Full size image
Fig. 3: Factors associated with healthcare-associated SARS-CoV-2 in patients, HCWs and predictive distributions.
figure 3

a,b, Additive effects associated with categories of host infections and multiplicative effects of vaccine coverage in HCWs, Alpha prevalence and trust characteristics (posterior means and 50% and 90% CrIs are shown and results are on the basis of 1,661 hospital trust weeks of data from 96 different hospital trusts). For multiplicative effects, values below 1 indicate an association with reduced infection rates. Note that in the model for infections in patients (a), HCW vaccine coverage acts by modulating transmission associated with infected HCWs, whereas in the model for infections in HCWs (b) it has a global effect, modulating the overall rate of infection. c,d, Associated posterior predictive distributions for the number of detected infections by week in the 20 largest trusts are shown for infections in patients (c) and in infections in HCWs (d). Bold solid lines correspond to observed values, shaded regions correspond to 50% and 90% CrIs and the central lines within the shaded regions are median values from the posterior sample. e–g, For all trusts, classifications of detected infections by week (e) and contributions to predicted hospital-acquired infections in patients (f) and HCWs (g) from the three categories of infected hosts predicted by the full negative binomial regression models accounting for HCW vaccination and Alpha variant effects are shown. When the dependent variable is healthcare-associated SARS-CoV-2 infection in patients, these results use the ECDC definitions of definite and probable healthcare-associated infection (see Supplementary Information section 2 and the Supplementary Results for models using other definitions).

Full size image

Among the additive terms, the strongest predictor of new healthcare-associated infections is the number of patients in the same trust with healthcare-associated infections the previous week (Fig. 3); thus, one patient with a newly identified healthcare-associated infection the previous week is associated with a further 1.07 (95% CrI 0.93, 1.19) hospital-acquired infections in patients the following week (setting variables representing hospital characteristics to their mean values, and in the absence of the Alpha variant or vaccine effects). Additive effects associated with patient exposures to infected HCWs and patients admitted with SARS-CoV-2 are smaller, although the larger number of such exposures increases their contribution to patient infections (Fig. 3f).

Considering multiplicative effects associated with trust characteristics, increased availability of single rooms is associated with reduced incidence of healthcare-associated infections in patients with an incidence rate ratio (IRR) for a 1 s.d. increase in single room availability (corresponding to a 15% increase in the percentage of beds as single rooms) of 0.91 (0.87, 0.97), whereas heated volume per bed is associated with a similar reduction (IRR 0.90 (0.84, 0.97)) for a 1 s.d. increase, corresponding to an increase per bed of 207 m3, and older hospital buildings were also associated with reduced hospital transmission, although in this case 95% CrIs include the null value of 1.00 (IRR 0.96 (0.92, 1.00)) (Fig. 3). These effects were not seen for infections in HCWs. HCW vaccination was associated with substantial reduction in transmission to patients linked to exposures to infected HCWs, and large reductions in the overall rate of infection in HCWs. Increased Alpha variant prevalence was associated with large increases in the rates of infection in both patients and HCWs.

Negative controls can help to assess the likelihood that associations between exposures and outcomes in observational studies result from relationships that are not directly causal (Extended Data Fig. 4)17. We use as a negative control outcome the number of patients admitted meeting ECDC definitions for community-acquired SARS-CoV-2 infection. Assuming most hospital admissions with SARS-CoV-2 result from community transmission, this outcome would not be expected to have a strong association with hospital-based exposures. If associations between hospital characteristics (exposures) and this control outcome are similar to those for hospital-acquired infections, it would indicate that confounding is a plausible explanation for observed associations with hospital-acquired infections (for example, owing to differences in hospital characteristics not accounted for in the model). Note, however, that as some SARS-CoV-2 admissions from the community will result from the readmission of patients infected in hospital, some link is expected. In all models considered with this control outcome, there is no strong association with the number of healthcare-associated infections or with the single room provision, strengthening the evidence that these both play a causal role in the incidence of hospital-acquired infections (Supplementary Tables 15–17). However, both heated volume per bed and HCW vaccination coverage show similar negative associations with the control outcome, as reported for healthcare-associated infection outcomes, indicating the need for caution when considering whether these reported associations might reflect direct causal effects.

To help to interpret estimated regression coefficients we perform a series of simulation studies, generating synthetic transmission datasets from a multitype branching process model, applying an observation model to obtain partially observed infection data and replicating the above analysis (Supplementary Information section 2.3). This analysis indicates that when the outcome is patient hospital-acquired infections, regression coefficients typically underestimate the expected number of secondary cases per case when only a proportion of hospital-acquired infections are observed, although represent good approximations as the proportion approaches 1 (Extended Data Fig. 5).

We use estimates from these analyses and the wider literature on hospital-acquired SARS-CoV-2 transmission to inform a dynamic model coupling hospital and community dynamics (Methods and Supplementary Information section 1.2). We consider three scenarios: high hospital transmission, corresponding to self-sustaining within-hospital transmission; and intermediate and low hospital transmission, in which all hospital transmission rates were reduced by 25% and 50%, respectively, compared with the high hospital transmission scenario (Fig. 4). Community transmission rates were identical in all scenarios.

Fig. 4: Dynamics of community and hospital infections.
figure 4

a–i, Results from simulation runs showing infected patients under high (a), intermediate (b) and low (c) rates of hospital transmission, in which rates of hospital transmission in intermediate and low scenarios are, respectively, 25% and 50% lower than the high hospital transmission scenario without altering parameters related to community transmission. Corresponding numbers of infected HCWs (d–f) and total numbers infected in the community (g–i) under the same scenarios are also shown. Assumed population sizes for community, hospital inpatients and HCWs are 500,000, 1,000 and 4,000, respectively. Solid vertical lines correspond to initiation of ‘lockdown’ measures, which are assumed to reduce person-to-person transmission rates in the community by 80% for the first lockdown and 70% for the second. The two broken vertical lines correspond to progressive release of lockdown measures, here assumed to result in transmission rates in the community that are reduced by 70% (after 100 d) and 40% (after a further 50 d) compared with the pre-intervention rate. j–u, The same three hospital transmission scenarios are used when considering threshold-driven lockdown measures (j–u), when lockdown measures are initiated and released on the basis of per capita infection rates in the community being above or below prespecified thresholds. In these scenarios, when lockdown is in place person-to-person transmission rates in the community are assumed to be reduced by 90% compared with pre-intervention levels, whereas release of lockdown is followed by community transmission rates that are 50% of those before the first lockdown. Outputs shown are number of lockdowns under high (j), intermediate (k) and low (l) hospital transmission scenarios, the total proportion of the population infected (m–o), total days in lockdown (p–r), and infections averted per day in lockdown (s–u) under the same three scenarios.

Full size image

The level of hospital transmission has little overall impact on an unmitigated epidemic or an epidemic controlled by a single lockdown, modelled here as a policy that substantially reduces community transmission (Extended Data Fig. 6). However, when community transmission is controlled through punctuated lockdowns, the extent of hospital transmission can have a profound impact on overall epidemic dynamics. If lockdowns are put in place for a fixed time period and then released in a stepwise manner (Fig. 4a–i), the total infected population in the community decreases from 27% in the high hospital transmission scenario to 12% and 7% in the intermediate and low transmission scenarios (Fig. 4g–i), with corresponding decreases in the percentages of HCWs infected from 91% to 52% and 21% (Fig. 4d–f). Conversely, if instigation and release of lockdowns is driven by threshold infection rates in the community (Fig. 4j–u), the total number infected does not depend strongly on levels of hospital transmission (Fig. 4m–o), but the time spent in lockdown is reduced (Fig. 4p–r) and the efficiency with which lockdown averts infections compared with an unmitigated epidemic (Fig. 4s–u) is enhanced by reducing hospital transmission. These effects can be substantial despite the fact that, at any one time, the number of patients and HCWs is less than 2% of the total population.

Discussion

Between 1% and 2% of hospital admissions are likely to have acquired SARS-CoV-2 infection while in hospital during the ‘second wave’ in England, with only a minority of these infections correctly classified as ‘healthcare-associated’ based purely on the time elapsed between admission and positive test. Investigation of the time series of hospital-acquired infections with a regression model indicated that patients who themselves acquired SARS-CoV-2 infection in hospital were the main drivers of transmission to patients, whereas transmission from both HCWs and nosocomially infected patients were of similar importance for transmission to HCWs (Fig. 3f,g). HCW vaccination was associated with large reductions in infection rates and there was evidence that aspects of hospital building design could modulate such transmission; in particular, a higher proportion of beds in single rooms was associated with decreased transmission risk, as was increased hospital building heated volume per bed, consistent with predictions from theoretical models for the spread of airborne infections in enclosed spaces18.

Although lack of genomic data means we cannot conclusively demonstrate transmission, our findings accord with focused local investigations with densely sampled viral genome sequences. Such studies indicate that many hospital-onset infections not meeting ECDC definitions for healthcare-associated infection are hospital-acquired and highlight the importance of superspreading5,19. Such superspreading is implicit in our negative binomial models, which attribute 80% of detected patient–patient transmission events from nosocomially infected patients to about 20% of infected patients (Extended Data Fig. 7). Also aligned with our findings are conclusions from local studies that hospital-acquired infection in patients was primarily due to transmission from nosocomially infected patients, whereas sources for HCW infections came from patients and HCWs in approximately equal proportions9,19,20.

National infection prevention and control (IPC) guidance in England at the start of June 2020 emphasized respiratory and hand hygiene, use of face masks for patients and HCWs, cohorting of patients and staff, environmental decontamination, ventilation and staff social distancing. Screening of all patients for SARS-CoV-2 during the first 7 d of their hospital stay was recommended throughout the period, but some trusts went beyond these requirements by performing weekly testing. Records of such measures were not kept at a national level and lack of centrally collected data on trust-specific IPC measures means that effective interventions may have gone unrecognized and may potentially confound observed associations. Simulation studies, however, indicate that high-frequency asymptomatic screening and rapid isolation of patients with suspected SARS-CoV-2 can substantially reduce SARS-CoV-2 transmission in healthcare settings21,22, and highlight the importance of contact tracing23. Further limitations include the lack of PCR sensitivity estimates specific to the Alpha variant or conditioned on symptoms, and lack of consideration of vaccination in the patient population for which we lacked data. Although vaccine rollout to the over 70s and clinically extremely vulnerable began on 18 January 2021 in England, residents in care homes for older adults and their carers and those aged 80 and over were first eligible for vaccination on 8 December 2020; we estimate that 18% of those aged 80 and over and no more than 10% of those aged 70–79 may have had some degree of vaccine protection by the last week of the study (Supplementary Information section 2.4). We did not consider outpatients in this work as they are typically cared for in separate outpatient clinic settings distinct from the wards of acute hospitals.

The factors that make it hard to prevent SARS-CoV-2 transmission are relevant for hospitals everywhere. Although some well-resourced hospitals avoided large-scale nosocomial transmission in early 2020 (refs. 24,25,26), even in high-income settings the extent of such transmission showed considerable variation between hospitals8. Seroprevalence data before vaccination in HCWs also indicate a high degree of heterogeneity between hospitals even in the same countries and are consistent with high levels of nosocomial transmission in many settings (Extended Data Fig. 8). Hospitals in resource-limited settings face particular challenges due to poorly funded IPC activities, lack of capacity to carry out routine testing, lack of isolation facilities and high levels of patient crowding, but attempts to systematically quantify the extent of such transmission outside high-income countries are currently lacking.

Our findings have implications for control policies. First, they highlight the importance of early identification and prompt initiation of control measures for patients with new hospital-acquired infections and for other patients they may have infected. Second, they reinforce the need for measures that reduce transmission from patients with asymptomatic infection in non-COVID-19 hospital areas, including improved ventilation, use of face coverings by patients and staff, increased distancing between beds, minimizing patient movements within and between wards and promotion of hand hygiene27,28. Third, our findings support efforts to prioritize HCWs for COVID-19 vaccination both due to direct protection to HCWs and due to indirect protection offered to patients. Fourth, the findings highlight the need to prioritize research into effective methods of reducing hospital transmission of airborne pathogens for which evidence is currently lacking29, including ward design and air filtration systems30. Although our analysis focuses on nosocomial transmission early in the pandemic and before widespread vaccine coverage, the emergence of the highly contagious Omicron variants of SARS-CoV-2 has presented further infection control challenges, with high rates of hospital-onset infection reported despite high vaccine coverage, universal masking, admission testing and symptom-based screening; anecdotal reports indicate that heightened control measures may be needed to suppress nosocomial spread31.

Finally, our findings show that hospital transmission can have a substantial impact on epidemic dynamics in the wider community. In particular, the role of hospital transmission in seeding COVID-19 into care homes and other vulnerable groups in the community must be further investigated in light of the finding that much of the hospital transmission is likely to be unobserved.

Methods

Quantifying the number of hospital-acquired infections

Inferential approach

We estimate the total number of hospital-acquired infections in trust i (combining observed and unobserved infections), zi, by applying Bayes’ formula:

$$P({z}_{i}|\,{y}_{i},{{\rm{\pi }}{}^{{\prime} }}_{i})=P(\,{y}_{i}|{z}_{i},{{\rm{\pi }}{}^{{\prime} }}_{i})P({z}_{i})/P(\,{y}_{i}|{{\rm{\pi }}{}^{{\prime} }}_{i})$$

where \({{\rm{\pi }}{}^{{\prime} }}_{i}\), represents the probability that an infection acquired by a patient in trust i is both detected by a PCR test and meets the definition of a hospital-acquired infection (which requires the first positive sample to be taken 15 or more days after the day the patient is admitted to the trust and before patient discharge), assumed independent of zi. Here, \({P({y}_{i}|z}_{i},{{\rm{\pi }}{}^{{\prime} }}_{i})\) represents the binomial likelihood of observing yi identified hospital-acquired infections, \(P({z}_{i})\) is the prior distribution for the total number of infections, which we take to be uniform (bounded by 0 and 20,000), and we calculate \(P(\,{y}_{i}|{{\rm{\pi }}{}^{{\prime} }}_{i})\) using the law of total probability \(P(\,{y}_{i}|{{\rm{\pi }}{}^{{\prime} }}_{i})={\sum }_{l}P(\,{y}_{i}|{{\rm{\pi }}{}^{{\prime} }}_{i},{z}_{i}=l)P({z}_{i}=l)\).

Effect of testing policy

The probability that a new hospital-acquired infection in trust i is detected is given by \({{\rm{\pi }}}_{i}={\sum }_{m,{d}}{\gamma }_{{imd}}{P}_{{imd}}\), where \({P}_{{imd}}\) is the probability that a patient admitted to trust i with length of stay m and infected on day of stay d (where d ≤ m) has a positive PCR test while in hospital and \({\gamma }_{{imd}}\) is the probability that, given a new hospital-acquired infection in trust i occurs, it occurs in a patient with length of stay m on day of stay d. Similarly, the probability that a new hospital-acquired infection is both detected and meets the definition of a hospital-acquired infection is

$${{\rm{\pi }}{}^{{\prime} }}_{i}={\sum }_{m,d}{\gamma }_{imd}{P{}^{{\prime} }}_{imd}$$

where \({P{}^{{\prime} }}_{imd}\) is the probability that an infection in a patient admitted to trust i with length of stay m infected on day of stay d is both detected and meets the definition of a hospital-acquired infection.

Consider an infection that a patient acquires d days after the day the patient is admitted to the hospital. The testing policy in place in the trust during the patient’s stay, the day of infection and the incubation period distribution together determine the probability that a patient is tested on day k after the patient is infected (for k = 0, 1, 2, 3 …). We assume the test has a specificity of 1. Let ϕk represent the sensitivity of a PCR test taken k days after the date of infection, and let \({\tau }_{{ik}}\) represent the probability that such a test is performed k days after the infection event, assumed to be independent for each value of k of whether a test is performed on any other day. Then, \({P}_{{imd}}=1-{\prod }_{k=d\ldots m}({1-{\tau }_{i\left(k-d\right)}\phi }_{k-d})\).

The corresponding probability, \({P{}^{{\prime} }}_{imd}\), is zero for m < 15 (because in that case the definition of hospital-acquired infection is not met); otherwise, it is given by the probability that there is no positive test before day 15 and at least one positive test after. For d ≥ 15 this probability is identical to \({P}_{{imd}}\); otherwise, it is given by

$${P{}^{{\prime} }}_{imd}={\prod }_{k=d...14}(1-{\tau }_{i(k-d)}{\phi }_{k-d})(1-{\prod }_{k=\mathrm{15...}m}(1-{\tau }_{i(k-d)}{\phi }_{k-d})).$$

If \({\lambda }_{{im}}\) represents the probability that a patient at risk of nosocomial infection with SARS-CoV-2 admitted to trust i has a length of stay of m days, then, on a given day, the expected proportion of patients who both have a length of stay of m days and are currently on day of stay d is given by \({\psi }_{{imd}}=[\frac{{\lambda }_{{im}}m}{{\sum }_{n}{\lambda }_{{in}}n}]\frac{1}{m}I(m\ge d)\), where \(I\left(m\ge d\right)\) is the indicator function, \(\left[\frac{{\lambda }_{{im}}m}{{\sum }_{n}{\lambda }_{{in}}n}\right]\) is the probability that on a randomly chosen day a randomly chosen patient has a length of stay m and \(\frac{1}{m}\) is the probability that this randomly chosen day is day d of stay. Analysis of individual-level patient data indicates that although daily risk of infection changes over calendar time, it does not vary appreciably with day of stay d for typical lengths of stays9, and we therefore approximate \({\gamma }_{{imd}}\) by \({\psi }_{{imd}}\) which we estimate on the basis of the reported lengths of stays of completed episodes of patients admitted to each trust over the time period considered. This will represent a reasonable approximation provided that the infection hazard is small and roughly constant over a patient’s hospital stay.

Testing policies considered

We consider several different testing policies, which determine the probability values that the test is performed on day k after infection in trust i \({(\tau }_{{ik}})\), as exact data on what policies were available in each trust are unavailable.

The minimal testing policy, which involves the fewest tests, requires only that patients displaying symptoms of COVID-19 are tested, and we assume all such patients are tested on a single occasion, the date of symptom onset. When this policy is in place, the time of testing of patients with hospital-acquired infections, in relation to the time of infection, is determined by the incubation period and such a test is assumed to be performed if and only if the patient develops symptoms on or before the day of discharge. A second testing policy extends this by assuming that in the event of a negative screening result from a patient with symptoms, daily testing will continue to be performed until patient discharge, the first positive test or three consecutive negative tests (whichever occurs first). We consider further testing policies which combine symptomatic testing (without retesting if negative) with routine asymptomatic testing. In these policies all patients who have not already tested positive are screened at predetermined intervals using the same PCR test. We consider weekly, twice weekly, three times weekly and daily testing of all in-patients as well as a policy of testing twice in the first week of stay (in accordance with national guidance in England).

Accounting for uncertainty in test sensitivity, incubation period distribution and the proportion of infections that are symptomatic

For a given length-of-stay distribution, incubation period distribution, PCR sensitivity profile and probability that infection is symptomatic, the calculations outlined above to determine the probability that an infection is detected or both detected and classified as a hospital-acquired infection are deterministic, and require no simulation. We account for uncertainty in these quantities through a Monte Carlo sampling scheme, at each iteration sampling new values for PCR sensitivities, the incubation period distribution and the proportion of infections that are symptomatic. For PCR sensitivities, we directly sample from the posterior distribution reported by Hellewell et al.16. For the incubation period we assume a lognormal distribution, and sample the parameters from normal distributions with means (s.d.) of 1.621 (0.064) and 0.418 (0.069) as estimated by Lauer et al.32. Estimates of the proportions of infections that are symptomatic are taken from Mizumoto et al.33 and this quantity is sampled from a normal distribution with mean (s.d.) of 0.82 (0.012). Length-of-stay distributions are directly obtained from the Secondary Uses Service for NHS acute trusts, excluding: (1) patients who were admitted with PCR-confirmed COVID-19; (2) patients who had samples taken in the first 7 d of their hospital stay that were PCR positive for SARS-CoV-2; and (3) patients with a length of stay of less than 1 d. In the primary analysis we use aggregate length-of-stay data for all trusts taken from the 12 month period from 1 March 2020. We also present results from two sensitivity analyses: in the first we use trust-specific \({\lambda }_{{im}}\) values; in the second we allow for the possibility that length-of-stay distributions change over time and use period-specific empirical length-of-stay distributions from the periods: June to August 2020; September to November 2020; and December 2020 to February 2021.

Quantifying drivers of nosocomial transmission

We used generalized linear mixed models to quantify factors associated with nosocomial transmission. In these models the dependent variable was either the observed number of healthcare-associated infections in trust i and week j among patients, \({y}_{{ij}}\), or the imputed number of infections in HCWs, \({y}_{ij}^{{\prime} }\). When the dependent variable was healthcare-associated infections in patients, we used ECDC criteria, repeating the analysis using three different classifications of healthcare-associated infection: (1) definite; (2) definite and probable; (3) definite, probable and indeterminate. Three classes of independent variables were considered: (1) known exposures to others in the same trust infected with SARS-CoV-2 to account for within-trust temporal dependencies, with separate terms corresponding to exposures in the previous week to patients with community-onset SARS-CoV-2 infections \(({z}_{i(j-1)})\), patients with hospital-acquired SARS-CoV-2 \((\,{y}_{i(j-1)})\) and HCWs with SARS-CoV-2 \((\,{y{}^{{\prime} }}_{i(j-1)})\); (2) characteristics of the trusts that were considered, a priori, to be plausibly linked to hospital transmission: bed occupancy, provision of single rooms, age of hospital buildings, heated hospital building air volume per bed and size (number of acute care beds); (3) regional data including vaccine coverage among HCWs and the proportion of isolates represented by the Alpha variant. Models were formulated to reflect presumed mechanisms generating the data, and we used negative binomial models with identity link functions, allowing the number of exposures to different categories of SARS-CoV-2 infections to contribute additively to the predicted number of weekly detected infections, while allowing for multiplicative effects of the other terms. In models for which the dependent variable represented hospital-acquired infections in patients, the HCW vaccination effect was assumed to act only through a multiplicative term affecting transmission related to exposures to HCWs. By contrast, when the dependent variable represented infections in HCWs, vaccine exposure was allowed to have a multiplicative effect on the overall expected number of infections. Formally, we define the full model for infections in patients in trust i and week j (which we refer to as model P1.1.1) as:

$${y}_{ij}\sim {\rm{n}}{\rm{e}}{\rm{g}}\,{\rm{b}}{\rm{i}}{\rm{n}}({\mu }_{ij},{\varphi }_{ij}),$$

where \({\mu }_{{ij}}\) represents the mean and the variance is given by \({\mu }_{{ij}}+{\mu }_{{ij}}^{2}/{\varphi }_{{ij}}\).

In the full model \({\mu }_{ij}=({a}_{i}+b{y}_{i(j-1)}+{c}_{ij}{y}_{i(\,j-1)}^{{\prime} }+d{z}_{i(j-1)}){m}_{ij}{n}_{ij}\)

$$\begin{array}{l}{m}_{ij}\,=\exp (q\times {\rm{s}}{\rm{i}}{\rm{n}}{\rm{g}}{\rm{l}}{\rm{e}}\,{{\rm{r}}{\rm{o}}{\rm{o}}{\rm{m}}{\rm{s}}}_{i}+r\times {\rm{t}}{\rm{r}}{\rm{u}}{\rm{s}}{\rm{t}}\,{{\rm{s}}{\rm{i}}{\rm{z}}{\rm{e}}}_{i}+s\times {{\rm{o}}{\rm{c}}{\rm{c}}{\rm{u}}{\rm{p}}{\rm{a}}{\rm{n}}{\rm{c}}{\rm{y}}}_{i(j-1)}\\ \,\,\,\,\,\,\,+t\times {\rm{t}}{\rm{r}}{\rm{u}}{\rm{s}}{\rm{t}}\,{{\rm{a}}{\rm{g}}{\rm{e}}}_{ij}+u\times {\rm{t}}{\rm{r}}{\rm{u}}{\rm{s}}{\rm{t}}\,{\rm{v}}{\rm{o}}{\rm{l}}{\rm{u}}{\rm{m}}{\rm{e}}\,{\rm{p}}{\rm{e}}{\rm{r}}\,{{\rm{b}}{\rm{e}}{\rm{d}}}_{ij})\\ \,\,{n}_{ij}=\exp (w\times {\rm{p}}{\rm{r}}{\rm{o}}{\rm{p}}{\rm{o}}{\rm{r}}{\rm{t}}{\rm{i}}{\rm{o}}{\rm{n}}\,{\rm{A}}{\rm{l}}{\rm{p}}{\rm{h}}{\rm{a}}\,{{\rm{v}}{\rm{a}}{\rm{r}}{\rm{i}}{\rm{a}}{\rm{n}}{\rm{t}}}_{ij})\\ \,\,\,{c}_{ij}=c\times \exp (v\times {\rm{H}}{\rm{C}}{\rm{W}}\,{{\rm{v}}{\rm{a}}{\rm{x}}}_{i(j-1)})\\ \,\,{\varphi }_{ij}={\varphi }_{0}+{k}_{i}\,{y}_{i(j-1).}\\ \,\,\,{a}_{i}\sim N({a}_{0},{\sigma }_{a}^{2})\\ \,\,\,\,{k}_{i}\sim N({k}_{0},{\sigma }_{k}^{2}).\end{array}$$

The expression for the dispersion parameter of the negative binomial distribution, \({\varphi }_{{ij}}\), reflects the fact that the sum of n independent negative binomially distributed random variables with mean μ and dispersion parameter φ will itself have a negative binomial distribution with mean nμ and dispersion parameter nφ. Thus, in the idealized case that each of n nosocomially infected patients in 1 week has a fully observed negative binomially distributed offspring distribution the next week with mean μ and dispersion parameter φ, then the total number of nosocomial infections observed would have a negative binomial distribution with parameters nμ and nφ. The \({a}_{i}\) represents a trust-level random effect term to account for within-trust dependency. We also considered two nested models, P1.1.0 and P1.0.0, obtained by setting the terms q, r, s, t and u to zero in both cases (that is, removing the trust-level terms) and by additionally setting the terms v and w to zero in the latter case (that is, removing regional vaccine- and variant-related terms). As a further sensitivity analysis, we also considered a model that allowed for time-varying changes in the number of hospital-acquired infections not accounted for by the covariates, by setting

$${\mu }_{ij}=(1+s(\,j))({a}_{i}+b{y}_{i(j-1)}+{c}_{ij}\,{{y}^{{\prime} }}_{i(j-1)}+d{z}_{i(j-1)}){m}_{ij}{n}_{ij}$$

where \(s(\,j)\) is a degree 3 spline with 6 equally spaced knots. We refer to this model as P1.1.1.tv. Similar models were used when the dependent variable was HCW infections, except that the HCW vaccine effect was included in the multiplicative term \({m}_{{ij}}\) instead of operating only through the \({c}_{{ij}}\) term.

We used normal(0,1) prior distributions by default for model parameters, except for variance terms \({\sigma }_{a}^{2}\) and \({\sigma }_{k}^{2}\) for which we used half-Cauchy(0,1) prior distributions, and φ for which a half-normal(0,1) prior distribution was specified for the transformed parameter \(1/\,\sqrt{{\varphi }_{0}}\). All analyses were performed in Stan34 using the rstan package v.2.21.1 in R (ref. 35), running each model with four chains using 1,000 iterations for warm-up and 5,000 iterations for sampling.

In the main analysis, we used weekly aggregated data, counting week numbers as 1 plus the number of complete 7 d periods since 1 January 2020. We included only acute hospital trusts in this analysis, and excluded trusts that predominantly admitted children.

Imputation method for weekly number of infections in HCWs

Situation reports included fields allowing quantification of nosocomial transmission and number of HCWs isolated due to COVID-19 from 5 June 2020, but analysis here is restricted to data from week 42 (beginning 14 October 2020) to week 55 (beginning 13 January 2021), reflecting the date range for which all fields used in the analysis were consistently reported. Because situation reports did not explicitly include data on the number of infections in HCWs, only the number of HCWs absent due to COVID-19 on each day, we imputed the weekly number of infections among HCWs at each trust. We did this by first subtracting from the number of reported HCW COVID-19 absences in each trust on each day the reported number of such absences due to contact tracing and isolation policies (reflecting likely COVID-19 exposures in the community) to give \({a}_{t}\), the number of HCWs absent on day t due to COVID-19 infection potentially arising from occupational exposure. Then, assuming that each HCW with COVID-19 was isolated for 10 d and assuming that durations of these absences were initially uniformly distributed (starting from week 36), the number imputed to have entered isolation on day t, \({x}_{t}\), was taken as \({x}_{t}={a}_{t+1}+{x}_{t-10}-{a}_{t}\). For each trust we performed these calculations ten times, sampling the initial duration of staff absences from a multinomial distribution assigning equal probabilities to durations of 1 … 10 d, and then took the average (rounded to the nearest integer) of these samples. In some trusts it was evident that some days with missing HCW isolation data had been coded as zeroes. When such zeroes fell between daily counts in excess of ten we treated them as missing data and replaced them with the last number carried forward. Any negative numbers for daily imputed HCW infections resulting from the above procedure were replaced with zeroes.

Although data on healthcare-associated infections in patients were recorded consistently by all trusts throughout the inclusion period, in some trusts data on HCW absences due to COVID-19 were missing or had been recorded inconsistently throughout the inclusion period. Excluding such trusts and those with missing data for independent variables left 96 of the original 145 trusts included in the analysis.

Negative control outcomes

We used as a negative outcome control the number of patients admitted with community-acquired SARS-CoV-2 infection as the outcome variable. We performed three analyses in which we adopted this negative control as our dependent variable, corresponding to models P1.1.1, P1.1.0 and P1.0.0 as defined above.

Hospital–community interaction model

We modelled hospital–community interaction using ordinary differential equations for an expanded susceptible/exposed/infectious/removed model (Extended Data Fig. 9). This model included separate compartments for people in the community (SC, E1C, E2C, I1C, I2C, I′C, RC), patients in hospital (SH, E1H, E2H, I1H, I2H, I′H, RH) and HCWs (SHCW, E1HCW, E2HCW, I1HCW, I2HCW, I′HCW, RHCW), in which the two exposed compartments (E1 and E2) and the two infectious compartments (I1 and I2) for each subpopulation correspond to assumptions of an Erlang-distributed latent and infectious period with shape parameter 2, whereas the I′ compartments represent people with severe disease potentially requiring hospitalization. The model allowed for patient–patient, HCW–HCW, HCW–patient and community–HCW transmission, as well as movements of people between the community and hospital. In the interest of simplicity, we neglect hospitalization of HCWs who account for about 1% of the total population.

We used the model to explore the impact of hospital transmission on overall epidemic dynamics with the aim of providing qualitative insights. We considered outcomes from high, intermediate and low hospital transmission scenarios in which the primary epidemic control measure was restricting rates of contact in the community (‘lockdowns’). This community control measure was assumed to have no direct impact on contact rates within hospitals as hospital infection control measures were in force throughout the study period irrespective of efforts aiming to limit community transmission. Full model details are provided in Supplementary Information section 1.2 and Supplementary Tables 1 and 2.

Ethics approval

The study did not involve the collection of new patient data, or use any personal identifiable information, but used a combination of anonymized national aggregate data sources including C19SR01–COVID-19 Daily NHS Provider SitRep, and regionally aggregated vaccine coverage data from the SARS-CoV-2 immunity and reinfection evaluation (SIREN)  study for which the study protocol was approved by the Berkshire Research Ethics Committee on 22 May 2020 with the vaccine amendment approved on 23 December 2020.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.