In the surveillance system 795 outbreaks from 379 care homes were recorded. The attack rate of 37.1 outbreaks per 100 care homes per year is substantially higher than that observed in France (4.6 to 5.5 outbreaks per 100 facilities per year)  and Australia (16.8 outbreaks per 100 facilities per year) . The difference in reported outbreak rates may be due to different resident populations, different structural or organisational arrangements, or lower levels of circulating pathogens at the time of surveillance. This finding could, however, represent more complete ascertainment by community health staff in frequent contact with local care homes. Ascertainment could have also been improved by the use of an internet-based reporting system, the use of which has been shown to increase the level of reporting of hospital-based norovirus outbreaks .
We found that care homes that closed promptly had outbreaks of shorter duration. This supports similar findings in care homes in other European countries [9, 20] and is consistent with comparable work looking at norovirus outbreaks in hospital which also found that prompt closure led to a shorter duration of outbreaks [21, 22]. We based these findings on the date of closure and the onset date of the first case. The date on which the outbreak was identified was not collected in this surveillance system, and thus we used time to closure after the first case as a proxy for outbreak identification.
We also found that increased duration of closure was significantly associated with increased size of the home and increased attack rate; both findings are epidemiologically plausible and have been observed in other studies . The lack of significant association with CQC rating could be related to the scoring criteria used in this metric. Only a very minor part of the CQC rating covers topics such as infection prevention policies which have been shown to be important in preventing transmission in this setting . This lack of significant association with CQC rating could also reflect the timing of the rating information; CQC data was extracted in December 2016 and therefore the ratings of a care home included in this analysis may not correspond to the rating of the care home at the point when the outbreak occurred.
Duration of closure is an important outcome as it may have a direct impact on a care home in terms of delayed admissions, and a wider impact on hospitals that are prevented from discharging patients to the affected care home. Duration of closure could be influenced by many factors. These may be organizational issues within the care home such as the time required to complete cleaning prior to re-opening and occupancy levels. As we did not capture resident capacity and therefore occupation rate, we were not able to adjust for this in our analysis. Evidence has shown that infection control measures are most effective when implemented in care homes within three days . It is possible that good infection control measures slow down transmission but do not stop it, prolonging the outbreak and duration. However, information on the timing of infection control implementation, infection control policies, leadership or decontamination resources was not collected, so it was not possible to examine this. Some care homes may have taken longer to reopen as they did not have sufficient staff to undertake a deep clean promptly. This is less plausible, as the ratio of residents to staff was adjusted for in the analysis, and the significant relationship with other variables remained. In addition, it is possible that outbreaks with a high attack rate initially were more likely to be reported as they could be easier to recognize. It is plausible that the duration of these outbreaks may have been shorter due to the early onset of most cases, though it was not possible to test this, as onset dates for individual cases were not collected.
Over the four year surveillance period there were 11,568 cases, 75 hospitalisations and 29 deaths in this population; if this were extrapolated over the whole of England, this would represent a substantial burden of illness across the country. Although not directly comparable, this rate of hospitalization and mortality appears to be lower than that observed in similar settings ; the difference could be due to underreporting, a different population or different treatment practices. Unfortunately it was not possible to calculate incidence or morbidity measures per bed-day with the information collected in this system. Such information would be useful in order to model individual risks to residents.
We saw marked seasonality in the outbreaks, with more outbreaks occurring during the winter months (November to March). However, outbreaks were reported year round, highlighting the continuing need for good infection prevention and control practice. The winter increase we observed is in line with individual case data in hospitals  and the general population . This seasonality in care home outbreaks may reflect the increased levels of infection circulating in the community, which increases the risk that staff, visitors or admitted residents will introduce the infection into the home. Introductions of norovirus into care homes by people are far more frequent than through food .
The most commonly detected pathogen was norovirus, which is consistent with studies in similar settings in other countries [9, 12]. Other viral pathogens such as rotavirus, sapovirus and astrovirus were less commonly detected. These viruses are less frequently detected in the general UK population  but have previously been associated with gastroenteritis outbreaks in care homes [28,29,30]. One of the key limitations when interpreting these findings is the large proportion (85%) of outbreaks in which samples were taken but no result was recorded. This may have been due to samples not being sent to the laboratory, not being tested at the laboratory due to procedural issues, or not being tested for viruses. Another explanation is that the database was frequently not updated with positive results from laboratory testing due to these results being reported after the surveillance report was completed. Unfortunately it was not feasible with the information stored in the surveillance database to cross-reference these results with laboratories in the area.
The primary aim of the surveillance system was to capture outbreaks of viral gastroenteritis; outbreaks of bacterial aetiology or food poisoning should have been captured on a separate incident management system and therefore not be included in this system. However, due to the syndromic nature of the case and outbreak definitions used in this surveillance system, and the small proportion of outbreaks where a sample was taken and the result recorded, some such outbreaks may have been included in this system.
One of the strengths of this analysis is that the dataset covers a large population and a wide geographical area including urban, rural and urban/rural mixed areas. It also covers a 4 year period thereby avoiding periods of unusually high or low rates of illness. One of the main limitations of this work is the difficulty of formally ascertaining the completeness of these surveillance data, both over the study period and in the different geographical areas. It might have been that ascertainment improved in the winter, leading to the observed winter increases in recorded outbreaks. Due to the close collaboration between CMHPT and CIPCTs, the data completeness is perceived to be good. Without an external dataset to validate these findings it is difficult to formally assess the completeness of these data. Nevertheless, our findings from this surveillance system are broadly consistent with other studies. Another limitation of these data is that by the nature of the surveillance system, they only include cases which are part of outbreaks. Without collecting similar information on sporadic cases of gastroenteritis, it is impossible to estimate the full burden and cause of gastroenteritis in care homes.