Quantifying the roles of vomiting, diarrhea, and residents vs. staff in norovirus transmission in U.S. nursing home outbreaks

The role of individual case characteristics, such as symptoms or demographics, in norovirus transmissibility is poorly understood. Six nursing home norovirus outbreaks occurring in South Carolina, U.S. from 2014 to 2016 were examined. We aimed to quantify the contribution of symptoms and other case characteristics in norovirus transmission using the reproduction number (REi) as an estimate of individual case infectivity and to examine how transmission changes over the course of an outbreak. Individual estimates of REi were calculated using a maximum likelihood procedure to infer the average number of secondary cases generated by each case. The associations between case characteristics and REi were estimated using a multivariate mixed linear model. Outbreaks began with one to three index case(s) with large estimated REi’s (range: 1.48 to 8.70) relative to subsequent cases. Of the 209 cases, 155 (75%) vomited, 164 (79%) had diarrhea, and 158 (76%) were nursing home residents (vs. staff). Cases who vomited infected 2.74 (95% CI: 1.90, 3.94) more individuals than non-vomiters, cases with diarrhea infected 1.62 (95% CI: 1.09, 2.41) more individuals than cases without diarrhea, and resident-cases infected 1.69 (95% CI: 1.18, 2.42) more individuals than staff-cases. Index cases tended to be residents (vs. staff) who vomited and infected considerably more secondary cases compared to non-index cases. Results suggest that individuals, particularly residents, who vomit are more infectious and tend to drive norovirus transmission in U.S. nursing home norovirus outbreaks. While diarrhea also plays a role in norovirus transmission, it is to a lesser degree than vomiting in these settings. Results lend support for prevention and control measures that focus on cases who vomit, particularly if those cases are residents. Author summary The majority of all norovirus outbreaks reported to the CDC occur in long-term care facilities (LTCFs), including nursing homes, where older residents are at risk for more severe or prolonged infection. Because there is currently no publicly available norovirus vaccine, sound control measures are key to controlling norovirus outbreaks, but there is little evidence that standard control measures are effective in reducing the size and/or duration of LTCF norovirus outbreaks. Hence, studies leading to a better understanding of disease spread and prevention of additional cases, and thus more effective control measures, are needed. To this end, we aimed to quantify factors associated with norovirus transmission and to examine how transmission changes over the course of an outbreak. We show that vomiting and, to a lesser extent, diarrhea are critical in initiating and sustaining norovirus transmission in U.S. nursing home norovirus outbreaks. We also show that nursing home residents, rather than staff, are the primary drivers of transmission. Results suggest that control measures focusing on cases who vomit, particularly if those cases are residents, would be most effective at curtailing norovirus transmission in these settings.


Introduction 75
There are 49.2 million individuals over 65 in the U.S. population (15.2%) and this 76 population is growing [1]. With nearly half of this age group spending some part of their lives in 77 nursing homes [2], the number of older adults using paid long-term care services is expected to 78 grow substantially over the coming decade [3]. In the U.S. and other high-income countries, 79 gastroenteritis outbreaks are common in long-term care facilities (LTCFs), including nursing 80 homes [4][5][6][7]. Despite the perception that norovirus is a foodborne disease or the 'cruise ship 81 virus', the majority of all norovirus outbreaks reported to the CDC occur in LTCFs [6]. While 82 norovirus gastroenteritis is generally mild and self-limiting, older nursing home residents are 83 vulnerable to infection leading to hospitalization and death [8], with the vast majority of 84 norovirus-associated deaths in the U.S. occurring among persons aged 65 years and older [9]. 85 Norovirus is highly transmissible in nursing homes [10][11][12], but there is no vaccine or 86 specific antiviral therapy available to prevent or treat norovirus infection. As a result, rapid 87 implementation of standard control measures is the mainstay for curtailing transmission [13]. 88 Identifying factors associated with norovirus transmission is critical to better understanding 89 disease spread and preventing additional cases. Individual-level risk factors for susceptibility to 90 norovirus infection or severe disease in nursing home outbreaks have been identified, including 91 resident mobility, dependency on staff assistance [14], immunodeficiency [15], and statin use 92 [16]. But because transmission of norovirus from one person to another cannot be directly 93 observed (unlike symptoms and/or positive test results that follow transmission), it remains 94 poorly understood and the evidence base for the value of specific prevention and control 95 measures is lacking [10]. 96 Statistical algorithms can be used to infer outbreak transmission trees (i.e., who infected 97 whom) from case onset dates and independent estimates of the serial interval (i.e., the time 98 between symptom onset in primary cases and the secondary cases they generate) between 99 generations of case pairs [17]. Individual reproduction numbers (R i ), or the number of secondary 100 cases an individual generates, can then be calculated for all cases. We quantified the contribution 101 of specific symptoms and residents vs. staff in norovirus transmission by examining the 102 associations between these variables and individual case infectivity, which was characterized by 103 R i . Additionally, we examined how transmission changes over the course of an outbreak. Our overall aim was to inform implementation of effective norovirus prevention and control 105 measures to reduce the size and duration of norovirus outbreaks in nursing homes. We achieved 106 this aim by characterizing norovirus transmission in these settings.

109
Outbreak data 110 De-identified data from six separate and unique nursing home outbreaks from two 111 consecutive norovirus seasons (2014-2015 and 2015-2016) were provided by the South Carolina 112 Department of Health and Environmental Control (SCDHEC). All outbreaks were confirmed, 113 meaning they had at least two laboratory confirmed norovirus cases. Outbreak data were in the 114 form of line lists and included individual-level information on symptom onset dates, reported 115 symptoms (vomiting, diarrhea, and fever), age in years, sex, illness duration, hospitalization, 116 emergency department visit, and whether the case was a resident or staff. Probable cases were 117 defined as residents or staff who had at least one episode of vomiting and/or three or more loose 118 stools within a 24-hour period. Confirmed cases were probable cases with a laboratory confirmed 119 norovirus infection. As this was an analysis of anonymized data that had already been collected 120 through routine public health response, the Emory University Institutional Review Board (IRB) 121 determined that this study was exempt from IRB review. Transmissibility of a pathogen can be quantified by its basic reproduction number, R 0 , 125 defined as the average number of secondary cases generated by a single infectious individual in a population that is entirely susceptible, or its effective reproduction number, R E , defined as the 127 average number of secondary cases generated by a single infectious individual in a population 128 that has some level of immunity. R 0 or R E of 1 signifies the extinction threshold, below which 129 each infectious individual, on average, infects less than one other individual and the outbreak 130 cannot be maintained. R E can be converted to R 0 by dividing R E by the proportion susceptible in 131 the population. Estimates for the R 0 of norovirus vary widely, from 1.1 to 7.2, and depend on 132 differences in settings [18].

133
The primary outcome of interest in this study was individual case infectiousness, which 134 we measured by estimating the reproduction number, R Ei , for each case. Here, R Ei is defined as 135 the number of secondary cases generated by an individual case i. We estimated R Ei using a 136 maximum likelihood procedure to infer the number of secondary cases generated by each case  [17,19,20]. Briefly, this method calculates, in a statistically rigorous 144 manner, the probability that cases with earlier symptom onset dates infected cases with later 145 symptom onset dates, selects the probabilities that are greatest using the frequency distribution of 146 the serial interval, and then, using these probabilities, determines the number of secondary cases 147 produced by cases with each symptom onset date. Individual cases were assigned a R Ei based on their symptom onset date, and those with the same onset date within an outbreak were assigned 149 the same R Ei .

150
In preliminary analysis, we observed much higher R Ei for index cases compared to those 151 on subsequent days. To investigate whether this could indicate heightened infectiousness of 152 index cases or just the natural decline of the susceptible population, we also calculated R 0i by where C is the total number susceptible on day 1 and is cumulative incidence to day i. ∑ 0 157 Using this approach, we compared estimates of R 0i of index cases on day 1 to R 0i estimated from 158 cases with onset on days 2 to 4 of the outbreak (excluding days with no reported cases). 161 We used a linear mixed model to estimate the association between each case 162 characteristic and R Ei , while accounting for correlation between R Ei 's within each outbreak. The 163 outcome variable was the natural log of R Ei .

164
The following information was available for cases: symptom onset date, resident/staff 165 status, age in years, sex, illness duration, hospitalization, emergency department visit, and 166 presence of diarrhea, vomiting, and fever. Because information on fever, age, sex, emergency 167 department visit and hospitalization were missing for large percentages of cases (20%, 23%, 168 26%, 40% and 55%, respectively), we were unable to consider these variables as potential 169 exposure, confounder, or effect modifying variables in the regression model. Information on resident vs. staff, diarrhea (yes or no), and vomiting (yes or no) were rarely missing (1%, 1%, 171 and 0%, respectively) and were considered explanatory variables in our model. To account for 172 clustering induced by correlation of R Ei 's within the six outbreaks, outbreak number was 173 included in the model as a random intercept. The full model, with log R Ei as the outcome, 174 included the following explanatory variables: diarrhea, vomiting, resident. The model was 175 assessed for collinearity and no issues were found. We considered including 'time' in the model 176 and adjusting for it as a potential confounder, as R Ei inevitably declines over time. However, we 177 determined that time cannot be a confounder, since it cannot affect diarrhea, vomiting, or 178 resident vs. staff, our explanatory variables of interest. The final model is shown below: where log R Eij represents the estimated log R E of the j th case from the i th outbreak, b 0i represents 183 the random slope for the i th outbreak, and e ij represents residual heterogeneity of the j th case from 184 the i th outbreak not explained by the model. The residual heterogeneity, e ij , and random slope, 185 b 0j , are assumed to be independent and identically distributed (iid) with mean zero and their 186 respective variances. Cases from the same outbreak were assigned the same random effect, 187 whereas cases from different outbreaks were assumed to be independent. Final coefficient 188 estimates and 95% confidence intervals were exponentiated to show the relationships between 189 average R Ei (rather than log R Ei ) and the variables in the model.

190
In addition to regression analyses, we also used the continuously declined to a R Ei below 1 or increased again before declining to a R Ei below 1 (Fig   232   1). Of these index cases, at least one from each outbreak reported vomiting (Fig 2). While most 233 index cases also reported diarrhea, outbreak 6 began with a case that reported vomiting only.

Outbreak
No. In sensitivity analyses to examine the effect of using different norovirus serial intervals 284 (serial intervals shorter and longer than 3.6 days) when calculating R Ei , we found that 285 associations between vomiting and R Ei and, to a lesser degree, resident and R Ei increased as the 286 serial interval increased. The association between diarrhea and R Ei did not appear to change when 287 the assumption about serial interval length was changed (Fig 3). secondary cases produced by a single primary case comparing: cases with vomiting to cases with 296 no vomiting, cases with diarrhea to cases with no diarrhea, and resident-cases to staff-cases. vomiting and, to a lesser degree, diarrhea play a critical role in norovirus transmission in these 304 settings. Second, outbreaks tend to start with one or more cases who infect substantially more 305 individuals than later cases in the outbreak.