Associations of infection control measures and norovirus outbreak outcomes in healthcare settings: a systematic review and meta-analysis

Abstract

Background:

Although most norovirus outbreaks in high-income countries occur in healthcare facilities, information on associations between control measures and outbreak outcomes in these settings are lacking.

Methods:

We conducted a systematic review/meta-analysis of published papers to assess associations between norovirus outbreak control measures and outcomes in hospitals and long-term care facilities (LTCFs), globally. Using regression analyses stratified by setting (hospital/LTCF), we compared durations, attack rates and case counts for outbreaks in which control measures were reportedly implemented to those in which they were not.

Results:

We identified 102 papers describing 162 norovirus outbreaks. Control measures were reportedly implemented in 118 (73%) outbreaks and were associated with 0.6 (95% CI: 0.3–1.1) times smaller patient case counts and 0.7 (95% CI: 0.4, 1.0) times shorter durations in hospitals but 1.5 (95% CI: 1.1, 2.2), 1.5 (95% CI: 1.0–2.1) and 1.6 (95% CI: 1.0–2.6) times larger overall, resident and staff case counts, respectively, and 1.4 (95% CI: 1.0–2.0) times longer durations in LTCFs.

Conclusions:

Reported implementation of control measures was associated with smaller/shorter outbreaks in hospitals but larger/longer outbreaks in LTCFs. Control measures were likely implemented in response to larger/longer outbreaks in LTCFs, rather than causing them. Prospective observational or intervention studies are needed to determine effectiveness.

1. Introduction

Norovirus is the leading cause of outbreaks of acute gastroenteritis in the United States and other high-income countries, with more than 1,000 outbreaks and 40,000 associated illnesses reported each year in the U.S. alone [1]. The majority of outbreaks in high-income countries occur in healthcare facilities, including long-term care facilities (LTCFs) (52% of U.S. outbreaks) and hospitals (3% of U.S. outbreaks) [2,3]. Transmission in these settings is facilitated by high levels of contact, communal living, and immunocompromised populations. Patients in hospitals and residents in LTCFs are also at greater risk of more severe and fatal illness due to underlying medical conditions and/or older age [4]. Ideally, introduction of norovirus into healthcare facilities could be prevented, but the virus is common in communities, with an estimated 19–21 million norovirus illnesses occurring in the U.S. each year [5], and can easily be introduced into healthcare facilities through infected patients/residents, visitors and staff [1,6]. Therefore, effective infection control measures are needed to mitigate transmission in LTCFs and hospitals.

There is currently no licensed vaccine or specific antiviral therapy available to prevent or treat norovirus infection, so infection control measures are the mainstay for curtailing transmission [7]. In the U.S., guidelines on how to prevent and control norovirus outbreaks in healthcare settings are largely based on a 2011 literature review by the Centers of Disease Control and Prevention (CDC) and the Healthcare Infection Control Practices Advisory Committee (HICPAC), in which authors examined the evidence for norovirus control measure effectiveness in healthcare settings [8]. Guidelines on how to control norovirus outbreaks include, but are not limited to, the following measures: 1) enhanced hand hygiene (e.g., actively promoting adherence to hand hygiene, beyond routine practice), 2) enhanced environmental cleaning (e.g., increasing the frequency of cleaning and disinfection), 3) movement restrictions (e.g., patient cohorting, staff cohorting, limiting patient transfers and ward closures), and 4) exclusion of ill staff from work until a minimum of 48 hours after resolution of symptoms (i.e., staff exclusions) [9]. However, published evidence for the effectiveness of these control measures in mitigating norovirus transmission is lacking. While handwashing is well known to reduce the risk of diarrheal illness among individuals [10–15], and environmental cleaning has been shown to reduce the risk of norovirus transmission [14], the effectiveness of enhanced hand hygiene and environmental cleaning measures in controlling norovirus outbreaks in healthcare facilities has not been established [8]. Other recommended measures, such as movement restrictions and staff exclusions, may intuitively limit exposures between infectious and susceptible individuals, but have also not been proven effective in controlling norovirus outbreaks in healthcare facilities [8]. Moreover, in a systematic review by Harris et al. (2009), authors found no evidence that implementing any infection control measures affected norovirus outbreak duration or attack rates in enclosed and semi-enclosed settings [16]. They concluded that the body of published literature at that time did not provide an evidence-base for the value of norovirus outbreak control measures.

Because the CDC/HICPAC prevention and control guidelines are updated as new information becomes available [8,17], investigation into the associations between norovirus outbreak control measures and outbreak outcomes is warranted. To this end, we performed a systematic review of published healthcare facility norovirus outbreaks globally and a meta-analysis to assess the associations between the implementation of any control measures and specific control measures and outbreak outcomes: duration, attack rate and size.

2. Methods

2.1. Systematic Review

The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement was used as the guideline for conducting this review [18]. Literature search strategies were developed using medical subject headings (MeSH) and text words related to norovirus outbreaks in healthcare settings. PubMed/MEDLINE, Embase (Elsevier), and Scopus (Elsevier) were searched for papers describing healthcare-associated norovirus outbreaks globally that were published from August 1, 2008 (the day after the last date included in the systematic review by Harris et al. [16]) to July 31, 2019. The results were also limited to those written in the English language. For the purposes of this review, we defined healthcare facilities as hospitals and LTCFs (nursing homes, skilled nursing facilities, and assisted living facilities) [19,20]. Norovirus outbreaks were defined as two or more cases within a facility either suspected or laboratory-confirmed to be caused by norovirus infection [21]. To reduce reporting bias, the following gray literature sources were also searched (on September 13, 2019) for unpublished outbreak reports: the CDC Stacks, the World Health Organization (WHO) Institutional Repository for Information Sharing (IRIS), and the National Technical Reports Library (NTRL). The search result records were imported into EndNote X9 for data management and deduplication. Seven hundred fifty-four (754) records were imported into the web-based application Covidence for screening (search details are available in Supplemental Table 1).

Papers were eligible for inclusion if they contained any of the following information on one or more norovirus outbreaks occurring in a hospital or LTCF: attack rates (or information on numbers at-risk and affected so that attack rates could be calculated), duration (or start and end dates so that duration could be calculated), and/or final sizes (i.e., the number of cases in an outbreak). Because outbreak attack rates and final case counts were often reported separately for staff and patients/residents, and/or as an overall measure (i.e., for both patients/residents and staff combined), papers were eligible for inclusion if they contained information on any of these measures. Outbreaks in which the mode of transmission was reported as foodborne or waterborne in origin were included only if there was also secondary spread.

All identified studies underwent a title and abstract screen by two independent reviewers based on the inclusion criteria described above. During full-text review, the reason for excluding papers was recorded and any discrepancies that arose were resolved by a third reviewer. Sixty-three (63) papers were identified for data extraction. The number of articles screened during this process can be found in the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flowchart (Figure 1).

Figure 1.

Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flowchart of review process.

The following data were extracted by one reviewer from all included studies: outbreak duration; attack rates for patients/residents, staff, and overall (and the number of cases and individuals at-risk from which attack rates could be calculated); case counts for patients/residents, staff, and overall; whether or not any infection control measures were implemented, per reporting in the paper; information on specific control measures implemented, per reporting in the paper; country where the outbreak took place; and outbreak setting (i.e., hospital or LTCF). Lastly, the dataset from our current review was combined with that from the previous review by Harris et al. (2009) [16]. Therefore, papers published on or prior to July 31, 2019 were also included in this review and meta-analysis. Because the previous review included information on norovirus outbreaks from all enclosed and semi-enclosed settings, previous review data were restricted to healthcare facility outbreaks and combined with the current review data. Any questions we had about the previous review data were clarified through direct communication with the first author.

2.2. Meta-analysis

To examine the associations between infection control measures and norovirus outbreak outcomes, we compared outbreak duration, attack rates and case counts for outbreaks in the combined dataset in which infection control measures were implemented to outbreaks in which they were not implemented, per reporting in the papers. The main exposure of interest was implementation of any control measures and this dichotomous predictor variable was defined as follows: if papers reported that control measures were implemented at any point during the outbreak, outbreaks were classified as having had control measures implemented; otherwise outbreaks were classified as having not had control measures implemented. Because information on timing of control measure implementation was missing for the majority (72%) of outbreaks in which control measures were reportedly implemented, it was not included in the analyses.

For this meta-analysis, there were seven outcomes of interest: outbreak duration, outbreak attack rates among patients/residents, staff, and overall, and case counts among patients/residents, staff, and overall (Table 1). Outbreak outcomes were calculated from the raw data whenever possible. When calculating attack rates, all patients/residents and staff in the entire facility (i.e., not only those in affected wings/units/floors) were considered at-risk. If calculated outcomes differed from those reported in the paper, we used calculated, rather than reported, outcome values for our analyses.

Table 1.

Outbreak outcomes and their definitions

Outbreak outcome	Definition
Duration	Difference in days between first and last illness onset dates for a given outbreak, including the first illness onset date (considered outbreak day 1)
Attack ratesa,b	Number of symptomatic cases divided by total number of individuals at-risk
Overall	Patients/residents and staff; does not include facility visitors
Patients/residents	Hospital patients or LTCF residents only
Staff	Hospital or LTCF staff only
Case countsa,b	Total number of symptomatic cases reported for a given outbreak
Overall	Patients/residents and staff; does not include facility visitors
Patients/residents	Hospital patients or LTCF residents only
Staff	Hospital or LTCF staff only

^aExcludes individuals who tested positive for norovirus but did not exhibit symptoms

^bIf reported for patients/residents only (and not staff), or staff only (and not patients/residents), unreported outcomes were coded as missing (and not 0)

To visually examine heterogeneity in outcomes, we constructed forest plots, in which individual outbreak outcomes, stratified by the reported implementation of control measures, were shown. Variances and confidence intervals were calculated by assuming attack rates for each outbreak followed binomial distributions and case counts and outbreak durations for each outbreak followed Poisson distributions. We were unable to calculate variances and confidence intervals for outbreak attack rates if the number of cases or individuals at-risk were not reported in the paper. Furthermore, we calculated Q-statistics to quantitatively assess the heterogeneity of each outcome. Because some outbreaks had 0 reported cases among staff, the variance for staff attack rate and case count were also equal to 0 for these outbreaks. We added 0.01 to the variance of these variables for Q-statistic calculations. Additionally, to assess the quality of evidence, we characterized outbreak reports based on the following measures of quality: 1) a case definition was provided, 2) an outbreak definition was provided, and 3) the day on which control measures were implemented was reported.

Meta-regression mixed effects models were used to assess the associations between the implementation of any and specific control measures, per reporting in the paper, and all outcome variables (duration, attack rates and case counts). We used the following regression models: 1) log-linear for duration (the natural log of duration was taken so that it met assumptions of normality), 2) linear for attack rates, and 3) negative binomial generalized linear for case counts, which were over-dispersed. When presenting results, we exponentiated the regression coefficients from the log-linear and negative binomial models so that results are on the multiplicative scale. Using a directed acyclic graph (DAG) to inform the modeling approach, outbreak setting and country were determined a priori to be potentially confounding variables for all associations (Supplemental Figure 1). Setting was controlled for through stratification and a random intercept for country was included in all models to account for within- and across- country variability. Because we intended to make inferences beyond the particular countries included in this review, we chose a random, rather than fixed effect for country. However, in a separate analysis, we included country in the model as a fixed effect and assessed the interaction between country and control measures and found no evidence of an interaction, which may be due to insufficient power. When examining associations between control measures and outbreak outcomes, we weighted all outbreaks equally, rather than using inverse variance weighting, as inverse variance weighting would inherently assign greater weight to shorter outbreaks in smaller facilities (i.e., smaller numbers at-risk). We categorized specific control measures into four categories, which were determined a priori: 1) enhanced hand hygiene, 2) enhanced environmental cleaning, 3) movement restrictions, and 4) staff exclusions. We assumed the residual heterogeneity, e_ij, and random slopes, α_0i, b_0i, and ε_0i, were independent and identically distributed (iid) with mean zero and their respective variances. The models used for these regression analyses are below:

$$\ln \left({Duration}_{ij}\right)=\left(A_0+{\alpha }_{0i}\right)+A_1*{ControlMeasure}_{ij}+e_{ij}$$

$$AttackRat{e}_{ij}=\left({\beta }_0+b_{0i}\right)+{\beta }_1*ControlMeasur{e}_{ij}+e_{ij}$$

$$\begin{gathered} \ln \left(CaseCount{s}_{ij}\right)=\left(E_0+{\epsilon }_{0i}\right)+E_1*ControlMeasur{e}_{ij}+e_{ij} \\ j=individualoutbreak \\ i=countryofoutbreak \end{gathered}$$

$$AttackRateandCaseCounts=overall,patients\text{/}residentsonly,orstaffonly$$

$$ControlMeasure=any,enhancedhandhygiene,enhancedenvironmentalcleaning,movementrestrictions,orstaffexclusions$$

We assessed the sensitivity of results using the following restricted datasets: 1) outbreaks with 10 or more cases, 2) outbreaks from the current review only (i.e., excluding outbreaks from the previous review [16]), and 3) full paper outbreak reports (from the current review only). The first sensitivity analysis was used to address the issue of reverse causation. Reverse causation occurs when the exposure-outcome process is reversed, and associations are seen because the outcome causes the exposure, rather than the exposure causing the outcome [22]. In this analysis, reverse causation would mean that control measures were implemented in response to larger and longer outbreaks, rather than control measures affecting outbreak outcomes. The second sensitivity analysis was used to address discrepancies in the previous and current review data that may have arisen from the systematic reviews and data extractions being performed by different research groups at different times. The third sensitivity analysis was used to address bias from exposure misclassification, in which outbreaks that implemented control measures were potentially misclassified as not having implemented control measures because control measures were not reported in the paper. Full papers (i.e., excluding abstracts), and particularly full paper outbreak reports, defined here as any paper in which the primary purpose was to describe one or more outbreaks (i.e., excluding research papers and surveillance reports), are probably more likely to include full outbreak information, and therefore less subject to information bias. To examine this further, we compared the percent of outbreaks with reported control measures and outbreak outcome information for outbreaks described in full paper outbreak reports to those not described in full paper outbreak reports. Information on whether papers were full paper outbreak reports was only available in the current review data.

3. Results

3.1. Systematic Review

Sixty-three (63) papers from the current review were combined with 39 papers from the previous review, for a total of 102 papers included in the analyses (Figure 1). Twenty-two (22) of these papers included information on two or more outbreaks, resulting in a total of 162 outbreaks: 107 (66%) from the current review and 55 (34%) from the previous review. There were approximately equal numbers of hospital outbreaks (78, 48%) and LTCF outbreaks (80, 49%) in the dataset. Four (4) outbreaks (3%) took place in a combined hospital and LTCF setting and were included in the restricted datasets for both hospital and LTCF outbreaks. Of the 84 outbreaks that took place in a LTCF or LTCF and hospital, 70 (83%) included information on the type of LTCF, the majority of which were nursing homes (56, 80%). Other LTCFs included adult group care (5, 7%), psychiatric care (5, 7%), assisted living (3, 4%), and a rehabilitation center (1, 1%). Lastly, the majority (88%) of outbreaks occurred in high-income countries, with the rest occurring in upper middle-income countries (Supplemental Table 2).

Of the 162 reported outbreaks, at least one control measure was implemented in 118 (73%), per reporting in the paper. The implementation of any control measures, per reporting in the paper, was more common in hospital outbreaks (82%) than LTCF outbreaks (65%). Furthermore, the most common specific control measure implemented, per reporting in the paper, in all settings was enhanced hand hygiene (67%), followed by movement restrictions (63%), enhanced environmental cleaning (59%), and staff exclusions (43%) (Table 2). The day on which control measures were implemented was only reported for 28% of outbreaks in which control measures were reportedly implemented, and therefore was not included in regression analyses. For those outbreaks that did include information on the day control measures were first implemented, the median day of implementation was day 5 (IQR: 3–10 days) of an outbreak. Lastly, full paper outbreak reports were more likely to include information on any control measures and all specific control measures, with the exception of staff exclusions, compared to non-full paper outbreak reports (Supplemental Table 3).

Table 2.

Control measures reported to be implemented by setting

Control measure	No. outbreaks (%) with reported control measure / setting
	Hospital (n = 78)	LTCF (n = 80)	All settingsa (n = 162)
Any control measures	64 (82)	52 (65)	118 (73)
Enhanced hand hygiene	56 (72)	50 (63)	108 (67)
Enhanced environmental cleaning	53 (68)	41 (51)	96 (59)
Movement restrictions	53 (68)	47 (59)	102 (63)
Staff exclusions	31 (40)	36 (45)	69 (43)

^aFour outbreaks occurred in both hospital and LTCF settings and were included only under “All settings”.

3.2. Meta-analysis

We examined associations between reported control measures and each outbreak outcome separately. Of the 162 outbreaks in the dataset, the majority (64%) of outbreaks were missing at least one of the seven outcome variables: duration, attack rates (overall, among patients, and among staff), and case counts (overall, among patients, and among staff). As expected, outcome variables were highly correlated. Attack rates and case counts were correlated by definition, as case counts are used to calculate attack rates. Outbreak duration was positively associated with case counts (R² = 0.4, 0.3, and 0.3 for overall, patient and staff case counts, respectively), but not with attack rates. Lastly, patient attack rate was positively associated with staff attack rate, as was patient case count with staff case count (R² = 0.6 each).

We did not find evidence that outbreak duration and overall and patient case counts were heterogeneous, but we did find evidence for heterogeneity among attack rates (overall, patients, and staff) and staff case counts (Supplemental Figure 2). We therefore stratified the dataset by outbreak setting (hospitals and LTCFs) for all regression analyses. While some outcomes were still heterogeneous even in the stratified datasets, the heterogeneity was reduced.

In our quality of evidence assessment, we found that only 33 (28%) outbreaks in which control measures were reportedly implemented had information on when control measures were implemented. The majority (113, 70%) of outbreaks included a case definition, however outbreaks in which control measures were reportedly implemented were more likely to include a case definition compared to those in which they were not (76% vs. 55% of outbreaks, respectively). Finally, few outbreaks (44, 27%) included an outbreak definition, and this information was again more likely to be reported for outbreaks in which control measures were reportedly implemented compared to those in which they were not (31% vs. 18% of outbreaks, respectively).

Among hospital outbreaks, we found that patient case counts were 0.6 (95% CI: 0.3, 1.1) times smaller for outbreaks in which control measures were implemented compared to those in which they were not, per reporting in the paper. Conversely, among LTCF outbreaks, we found that overall, resident and staff case counts were 1.5 (95% CI: 1.1, 2.2), 1.5 [95% CI: 1.0, 2.1), and 1.6 (95% CI: 1.0, 2.6) times larger, respectively, for outbreaks in which control measures were implemented compared to those in which they were not, per reporting in the paper. Lastly, among hospital outbreaks, we found that outbreaks in which control measures were implemented were 0.7 (95% CI: 0.4, 1.0) times shorter than those in which they were not implemented, per reporting in the paper. Conversely, among LTCF outbreaks, we found that outbreaks in which control measures were implemented were 1.4 (95% CI: 1.0, 2.0) times longer than those in which they were not implemented, per reporting in the paper. While results also suggest that control measures may be associated with smaller staff attack rates in hospital outbreaks, confidence intervals for associations between control measures and all attack rates were extremely wide and therefore associations could not be determined (Table 3).

Table 3.

Comparison of outbreaks by reported implementation of any infection control measures

Outcome/Settinga	No. outbreaks (%) reporting any control measures	Median (IQR) Any control measures reported		Associationb (95% CI) with reported implementation of any control measures
		Yes	No
Hospitals
Duration (days)	58 (84)	15 (12, 21.3)	26 (23, 54)	0.7 (0.4, 1.0)c
Attack Rates (%)
Overall	25 (83)	21.5 (11.5, 29.0)	22.5 (21.1, 44.3)	−7.4 (−23.9, 9.8)d
Patient	51 (88)	25.8 (15, 41.7)	22.5 (15.4, 30.8)	3.8 (−13.1, 21.1)d
Staff	31 (82)	16.9 (3.3, 30)	31 (24.7, 50)	−14.8 (−27.8, −1.8)d
Case Counts
Overall	53 (85)	28 (11, 60)	38 (20, 122)	0.8 (0.4, 1.6)c
Patient	61 (85)	14 (9, 32)	44 (7.5, 81)	0.6 (0.3, 1.1)c
Staff	55 (86)	16 (3, 32)	28 (15, 60)	0.8 (0.3, 2.0)c
LTCFs
Duration (days)	42 (64)	22 (11.3, 29.8)	14.5 (9, 24.5)	1.4 (1.0, 2.0)c
Attack Rates (%)
Overall	20 (49)	42.7 (19.2, 48.5)	30.1 (19.3, 45.7)	6.4 (−3.9, 17.0)d
Resident	45 (67)	41.3 (22, 54.3)	34.0 (21.4, 50.3)	4.5 (−5.9, 14.5)d
Staff	26 (54)	23.7 (11.6, 38.3)	23.2 (10.8, 43.0)	−2.7 (−11.8, 6.8)d
Case Counts
Overall	36 (55)	68.5 (31, 107.3)	53 (27.3, 76.3)	1.5 (1.1, 2.2)c
Resident	45 (62)	41 (21, 74)	30 (16, 52.3)	1.5 (1.0, 2.1)c
Staff	39 (58)	25 (7.5, 50)	16 (10.5, 28.3)	1.6 (1.0, 2.6)c

^aFour outbreaks took place in both a hospital and LTCF and were included in both hospital and LTCF settings.

^bLog-normal mixed linear regression was used for duration, mixed linear regression for attack rates, and mixed negative binomial regression for case counts. All regression models included the any control measure and a random intercept for country of outbreak as independent variables.

^cOn the multiplicative scale; exponentiated regression coefficients are shown

^dOn the additive scale

When examining specific control measures in hospital outbreaks, we found that enhanced hand hygiene measures and enhanced environmental cleaning were associated with 0.6 (95% CI: 0.4, 0.8) and 0.7 (95% CI: 0.5, 1.0) times shorter outbreak durations, respectively. Furthermore, we found the following associations between specific control measures and final case counts: 1) enhanced hand hygiene measures were associated with 0.5 (95% CI: 0.3, 0.9) times smaller patient case counts, 2) movement restrictions were associated with 1.7 (95% CI: 0.9, 3.0) and 1.7 (95% CI: 0.9, 3.5) times larger overall and staff case counts, respectively, and 3) staff exclusions were associated with 1.4 (95% CI: 0.8, 2.3) and 1.5 (95% CI: 0.8, 2.9) times larger overall and staff case counts, respectively. Lastly, while results suggest that enhanced hand hygiene measures, enhanced environmental cleaning and movement restrictions may also be associated with smaller staff attack rates, results were too imprecise to make conclusions about associations between any specific control measures and attack rates (Table 4).

Table 4.

Associations between implementation of specific infection control measures and outcome variables by setting

	Associationb (95% CI) with reported implementation of specific control measure
Outcome/Settinga	Enhanced hand Hygiene	Enhanced environmental cleaning	Movement restrictions	Staff Exclusions
Hospitals
Duration (days)	0.6 (0.4, 0.8)	0.7 (0.5, 1.0)	1.0 (0.7, 1.4)	0.9 (0.6, 1.2)c
Attack Rates (%)
Overall	−0.2 (−16.7, 15.8)	−0.2 (−16.7, 15.8)	−10.0 (−26.4, 7.6)	−9.2 (−22.1, 4.8)d
Patient	−6.2 (−19.6, 6.8)	−1.9 (−15.1, 10.9)	6.0 (−7.4, 18.3)	9.1 (−1.3, 19.3)d
Staff	−16.7 (−28.9, −4.7)	−13.8 (−24.9, −2.9)	−19.0 (−32.4, −5.4)	−8.7 (−19.2, 2.1)d
Case Counts
Overall	0.7 (0.4, 1.3)	0.8 (0.5, 1.5)	1.7 (0.9, 3.0)	1.4 (0.8, 2.3)c
Patient	0.5 (0.3, 0.9)	0.9 (0.5, 1.5)	0.9 (0.5, 1.6)	1.0 (0.6, 1.7)c
Staff	0.7 (0.3, 1.5)	0.7 (0.3, 1.4)	1.7 (0.9, 3.5)	1.5 (0.8, 2.9)c
LTCFs
Duration (days)	1.4 (1.0, 2.0)	1.2 (0.8, 1.7)	1.6 (1.1, 2.2)	2.1 (1.5, 2.8)c
Attack Rates (%)
Overall	7.7 (−2.4, 18.3)	−2.8 (−17.0, 10.8)	1.6 (−10.2, 12.9)	1.5 (−10.5, 12.3)d
Resident	4.9 (−5.4, 14.6)	−8.6 (−18.4, 1.2)	−1.6 (−11.1, 7.8)	−3.4 (−12.7, 6.0)d
Staff	−0.9 (−10.1, 8.4)	−10.2 (−19.4, −0.7)	−3.5 (−12.4, 6.2)	−0.4 (−9.8, 9.3)d
Case Counts
Overall	1.5 (1.1, 2.2)	1.2 (0.8, 1.8)	1.6 (1.1, 2.3)	1.7 (1.2, 2.4)c
Resident	1.5 (1.0, 2.1)	1.2 (0.8, 1.7)	1.5 (1.1, 2.2)	1.4 (1.0, 2.0)c
Staff	1.7 (1.1, 2.7)	1.5 (1.0, 2.5)	1.6 (1.0, 2.5)	1.9 (1.2, 3.0)c

^aFour outbreaks took place in both a hospital and LTCF and were included in both hospital and LTCF settings.

^bLog-normal mixed linear regression was used for duration, mixed linear regression for attack rates, and mixed negative binomial regression for case counts. All regression models included the specific control measure and a random intercept for country of outbreak as independent variables.

^cOn the multiplicative scale; exponentiated regression coefficients are shown

^dOn the additive scale

When examining specific control measures in LTCF outbreaks, we similarly found that results suggest enhanced environmental cleaning may be associated with smaller staff attack rates, but that results were too imprecise to make conclusions about the associations between specific control measures and attack rates. However, we did find the following associations between specific control measures and outbreak duration and case counts: 1) enhanced hand hygiene measures were associated with 1.4 (95% CI: 1.0, 2.0) times longer durations and 1.5 (95% CI: 1.1, 2.2), 1.5 (95% CI: 1.0, 2.1) and 1.7 (95% CI: 1.1, 2.7) times larger overall, resident and staff case counts, respectively, 2) enhanced environmental cleaning was associated with 1.5 (95% CI: 1.0, 2.5) times larger staff case counts, 3) movement restrictions were associated with 1.6 (95% CI: 1.1, 2.2) times longer durations and 1.6 (95% CI: 1.1, 2.3), 1.5 (95% CI: 1.1, 2.2) and 1.6 (95% CI: 1.0, 2.5) times larger overall, resident and staff case counts, respectively, and 4) staff exclusions were associated with 2.1 (95% CI: 1.5, 2.8) times longer durations and 1.7 (95% CI: 1.2, 2.4), 1.4 (95% CI: 1.0, 2.0) and 1.9 (95% CI: 1.2, 3.0) times larger overall, resident and staff case counts, respectively (Table 4).

In sensitivity analyses, results for the associations between outbreak outcomes and the reported implementation of any control measures were robust, with one exception when the data were restricted to full paper outbreak reports. In this sensitivity analysis for hospital outbreaks, the association between any control measures and duration disappeared and the association between any control measures and smaller patient case counts persisted but the confidence intervals became too wide to make conclusions. In sensitivity analyses examining associations between outbreak outcomes and the reported implementation of specific control measures, results were generally robust, with a few exceptions for hospital outbreaks. First, when data were restricted to full paper outbreak reports, the associations between enhanced hand hygiene measures and duration and patient case counts became inconclusive due to large confidence intervals. Second, the association between enhanced environmental cleaning and duration disappeared when data were restricted to new data only, and reversed when data were restricted to full paper outbreak reports. Lastly, the association between movement restrictions and overall and staff case counts largely disappeared when data were restricted to outbreaks with 10 or more cases and to full paper outbreak reports (Supplemental Figures 4–8).

4. Discussion

Our aim was to search the scientific literature to identify norovirus outbreaks in healthcare facilities globally and examine associations between control measures and outbreak duration, attack rate and size. From these reports, in hospital outbreaks, we found that patient case counts were smaller and durations were shorter when control measures were implemented compared to when they were not, per reporting in the paper. Conversely, in LTCF outbreaks, case counts (overall, residents, and staff) were larger and durations were longer when control measures were implemented compared to when they were not, per reporting in the paper. Both findings were robust in sensitivity analyses.

The direction of the association between control measures and outbreak outcomes was as expected in hospitals, but was opposite of expected in LTCFs. We hypothesize that outbreak control measures in LTCFs are more likely to be implemented for larger and longer outbreaks than smaller and shorter outbreaks. While this may also be true in hospitals, we believe it is to a lesser extent, as control measures were not associated with larger or longer outbreaks in this setting. It may be that LTCFs have more limited resources and personnel compared to hospitals, and therefore control measures may only be implemented once LTCF outbreaks reach a certain size or duration (i.e., a threshold of outbreak severity). In other words, control measures may be implemented later in LTCF outbreaks compared to hospital outbreaks. However, while there is evidence that LTCFs in the U.S. and other high-income countries are underfunded and understaffed [125–129], studies comparing funding and staffing levels in LTCFs to those in hospitals are lacking, as are studies comparing the implementation of norovirus outbreak control measures in these two settings, so we can only speculate. In addition, there are several other important differences between hospitals and LTCFs. First, hospitals typically provide acute care to patients requiring immediate yet brief medical treatment, whereas LTCFs typically provide long-term care, including both medical and personal, to people who are unable to live independently. Therefore, the average length of stay for hospital patients is much shorter than that for LTCF residents (4.6 vs. 485 days, respectively) [130,131]. Second, LTCFs and hospitals have different physical designs, with LTCFs often emulating a residential, non-institutional environment that includes common areas where residents can socialize [132]. Lastly, hospitals are typically larger than LTCFs, with an average of 160 beds per hospital compared to 110 beds per nursing home in the U.S. [131,133]. While these are all important differences, we do not believe they explain the opposite association between control measures and outbreak outcomes in LTCFs. We accounted for setting (hospital vs. LTCF) through stratification, so while these are not confounding factors, they, in part, may underlie different patterns observed. However, without more detailed outbreak data, including case counts by day (i.e., outbreak curves) and timing of control measure implementation, we are unable to examine reasons for the opposite association in LTCFs further.

In general, we also found that specific control measures were associated with smaller outbreak size and shorter duration in hospitals but larger outbreak size and longer duration in LTCFs. Among hospital outbreaks, enhanced hand hygiene measures were associated with smaller outbreak size and shorter durations, findings that were generally robust in sensitivity analyses. Among LTCF outbreaks, enhanced hand hygiene measures, enhanced environmental cleaning, movement restrictions and staff exclusions were associated with larger outbreak size and longer durations, findings that were also robust in sensitivity analyses. A notable exception, however, was the association between movement restrictions and larger outbreak size and/or longer duration in both hospitals and LTCFs. We hypothesize that hospitals are more likely than LTCFs to implement any control measures in response to smaller outbreaks, but that movement restrictions, such as closure of affected clinical areas (i.e., ward closures), which are generally more extreme and costlier to implement [134], may only be implemented for larger outbreaks. While studies examining the threshold (e.g., certain outbreak size) at which ward closures are typically implemented are lacking, the current CDC/HICPAC prevention and control guidelines mentions that the threshold for ward closure should vary and depend on facility risk assessments [8]. Thresholds for implementing any other control measures are not mentioned, suggesting that ward closures, unlike other control measures, may typically be implemented only once outbreaks reach a certain threshold. In sensitivity analyses, the association between movement restrictions and larger outbreak size in hospitals largely disappeared, supporting this hypothesis. However, as noted above, without more detailed outbreak data, we are unable to examine this further.

We note a number of limitations in our study. First, in our quality of evidence assessment, we found that a substantial number of outbreaks had missing information for variables we examined as indicators of quality, and that outbreaks in which control measures were not reportedly implemented were more likely to be missing this information. Similarly, some outbreaks were likely misclassified as not having had control measures implemented. However, expected bias from this misclassification would be toward the null, and control measures were found to be associated with smaller outbreak size in hospitals despite this potential bias. Second, analyses were subject to bias from reverse causation. In sensitivity analyses, we found associations between control measures and larger and longer outbreaks in LTCFs persisted, leading us to believe that we were unsuccessful in completely removing this bias. Third, there was insufficient information on timing of control measure implementation, so we could not examine this further. Control measures implemented earlier in an outbreak, before transmission is well established, are likely more effective in mitigating transmission compared to control measures implemented later. Similarly, we only had information on whether control measures were reportedly implemented, and not on the quality of or adherence to control measures, and were therefore unable to include this information in our analyses. Fourth, healthcare facilities, and LTCFs in particular, can be highly heterogeneous, both within and between countries. However, due to insufficient sample size, we were unable to examine control measure effectiveness in more specific settings (e.g., nursing homes, assisted living facilities, etc.). Fifth, attack rate calculations can vary substantially depending on the definition used for individuals at-risk (e.g., the entire facility, affected wards/units only, etc.). While we calculated attack rates whenever possible (using a consistent definition of individuals at-risk), some attack rates could not be calculated due to insufficient information, and the definition used for individuals at-risk was not always available. Lastly, while we searched PubMed/MEDLINE, Embase (Elsevier), Scopus (Elsevier) and the gray literature, some data may be missing from this review if eligible papers were published exclusively on other databases.

Owing to the limitations described above, we were unable to examine the causal effect of control measure implementation on norovirus outbreak outcomes in healthcare settings. Instead, our study provides a summary of these associations. More research is needed to determine the effectiveness, as well as cost-effectiveness, of outbreak control measures in these settings. For example, interventions such as closure of affected wards/units are very costly, so their effectiveness should be considered in light of impacts on the provisions of health services. While randomized control trials, in which healthcare facilities are randomized to implement specific control measures upon detection of norovirus cases, could lead to unbiased estimates of control measure effectiveness, these trials are not ethical nor feasible. However, hundreds of norovirus outbreaks are reported annually in the US alone. If a focused effort was made to record the timing and characteristics of control measures implemented during these outbreaks, more could be learned about control measure efficacy. Future studies should focus on prospectively collecting detailed information on outbreak control measures (e.g., specific control measures implemented, when they were implemented, and adherence to control measure protocols) from a representative sample of healthcare facility norovirus outbreaks and examining differences in outbreak outcomes. In particular, timing of control measure implementation is likely an important predictor of outbreak outcomes and should be considered in future studies.

5. Conclusions

By reviewing the relevant literature, we found that hospital outbreaks in which control measures were implemented were smaller in size and shorter in duration compared to hospital outbreaks in which control measures were not implemented, per reporting. Conversely, we found that LTCF outbreaks in which control measures were implemented were larger in size and longer in duration compared to LTCF outbreaks in which control measures were not implemented, per reporting. Control measures in LTCFs may be more likely to be implemented in response to larger and longer outbreaks, therefore explaining the reversed association. Longitudinal observational or intervention studies are needed to determine any causal associations and the effectiveness of norovirus outbreak control measures in healthcare settings.