Determinants and risk factors of gastroenteritis in the general population, a web-based cohort between 2014 and 2017 in France

Background Although it is rarely fatal in developed countries, acute gastroenteritis (AGE) still induces significant morbidity and economic costs. The objective of this study was to identify factors associated with AGE in winter in the general population. Methods A prospective study was performed during winter seasons from 2014 to 2015 to 2016–2017. Participants filled an inclusion survey and reported weekly data on acute symptoms. Factors associated with having at least one AGE episode per winter season were analyzed using the generalized estimating equations (GEE) approach. Results They were 13,974 participants included in the study over the three seasons. On average, 8.1% of participants declared at least one AGE episode during a winter season. People over 60 declared fewer AGE episodes (adjusted OR (aOR) = 0.76, 95% CI [0.64; 0.89]) compared to individuals between 15 and 60 years old, as well as children between 10 and 15 (aOR = 0.60 [0.37; 0.98]). Overweight (aOR = 1.25 [1.07; 1.45]) and obese (aOR = 1.47 [1.19; 1.81]) individuals, those having frequent cold (aOR = 1.63 [1.37; 1.94]) and those with at least one chronic condition (aOR = 1.35 [1.16; 1.58]) had more AGE episodes. Living alone was associated with a higher AGE episode rate (aOR = 1.31 [1.09; 1.59]), as well as having pets at home (aOR = 1.23 [1.08; 1.41]). Conclusions Having a better knowledge of AGE determinants will be useful to adapt public health prevention messages.


Background
Acute gastroenteritis (AGE) is usually due to a viral infection involving the stomach and the small intestine, and its clinical picture associates diarrhoea and possible vomiting [1]. Although it is rarely fatal in developed countries, AGE still induces significant morbidity and economic costs [2,3]. A recent study in France estimated an annual number of cases of 21,000,000, corresponding to a yearly incidence rate of 0.33 case per person [4]. Moreover 95% of individuals consulting a general practitioner for an acute diarrhoea received a drug prescription [5] and more than 80% were prescribed a sick leave [6], leading to a substantial societal cost. In a study published in 2014, results showed that stool examinations were positive for at least one enteric virus in 65% (95% CI [57-73]) of patients presenting a Winter AGE, with a predominance of noroviruses (49%) [5].
Effective preventive measures are well-known, such as hand washing education, and current prevention strategies target fragile populations at risk for complications such as elderly [7] or young children [8]. Data available to adapt the prevention strategy have been collected mainly by healthcare professionals like the French Sentinelles network [9], the English General Practice Research Framework [10] or the Continuous Morbidity Registration of the Netherlands Institute of Primary Health Care [11]. However, since AGE is a benign pathology, with only 1 patient out of 3 consulting a physician [4], using only these data is bound to cause a significant bias in the analyzed results.
Through the years, few studies have tried to assess AGE incidence and risk factors at a population level [4,10,[12][13][14][15][16]. Most of them used a retrospective data collection by conducting a telephone survey of self-reported AGE in the month preceding the phone call [4,12,14,16]. There have been some prospective studies on AGE conducted on the general population, but focused mainly on incidence and not on risk factors [17,18].
The objective of this study was to characterize risk factors associated with the occurrence of AGE during winter on the general population in France. A better understanding of the profile of people having AGE among general population would help to develop more efficient and targeted public health actions.

Design and study population
We conducted an observational prospective study using data from the web-based GrippeNet.fr cohort [19]. Grip-peNet.fr is part of a European multicentric project (Influenzanet) for syndromic surveillance during winter in the general population through online systems [20]. Participation is voluntary and anonymous after an online registration on the website. Participants are asked to fill a preliminary survey at registration regarding sociodemographic characteristics, medical history, habits and lifestyle. Then, during each winter seasoncalled season thereaftertypically from November to April, they are invited on a weekly basis, by an email, to report and describe any symptoms on a predefined list that may have occurred in the past week. In every weekly survey, participants could select one, several or none of the symptoms suggested. The list contained the following symptoms: fever, chills, runny or blocked nose, sneezing, sore throat, cough, shortness of breath, headache, muscle/join pain, chest pain, feeling tired or exhausted (malaise), loss of appetite, colored sputum/ phlegm, watery/bloodshot eyes, nausea, vomiting, diarrhoea, stomach ache, other, no symptoms. If they indicate having any symptom, they are invited to complete further information on its nature, duration or intensity. Each week, it was specified that symptoms had to be reported only if they weren't related to any chronic condition. In case of diarrhoea, they must indicate the number of loose stool per day. The participants can report symptoms at any time and as often as they wish.
The representativeness of the GrippeNet.fr population was studied on 2011-2012 season [19]. Although it was not representative of the French general population in terms of age and gender (50-69 years old people and women being overrepresented), all age classes were represented. The GrippeNet.fr population was also found to be more frequently employed, with a higher education level of education than French population. No significant difference was found in terms of rate of studied chronic conditions, such as asthma and diabetes.

Inclusion criteria and study period
The study was conducted over three seasons Any individual who filled a preliminary survey was registered as a participant for the season in the Grippe-Net.fr study. In order to select only those who actively participated and provided information regularly, we included in our study participants who had filled at least one preliminary survey for the ongoing season, and reported at least three weekly surveys during a season, with at least one before, one during and one after the AGE French epidemic period. The AGE epidemic period of each season was defined by the French Sentinelles network [21]: from December the 1st, 2014 to

Case definition
AGE episodes and their duration were identified and reconstructed using symptoms reported by the participants in the weekly surveys.
To be close to the French surveillance data, we select the definition of AGE used by the French Sentinelles network: three or more daily watery (or nearly so) stools for less than 14 days [9]. For each season, a case was defined as a participant having at least one episode of AGE, and a control as a participant having none.
We also conducted a complementary analysis with an alternative definition proposed by an expert group mandated by the WHO (the International Collaboration on Enteric Disease): three or more loose stools or any vomiting in 24 h, but excluding those due to chronic illness causing diarrhoea or vomiting, or due to drugs, alcohol or pregnancy [22].

Participant's characteristics
Data were collected about gender, age, household composition and residential zip code, allowing us to differentiate urban from rural participants. Questions assessed social mixing in terms of daily contact with a group of people (more than 10 individuals), with children (more than 10), with elderly (more than 10, over 65-year-old), or patients, the use of public transportation and the presence of pets at home. Data were collected on participants' height and weight, making possible the body mass index (BMI) estimation. Smoking status was reported. Chronic conditions were evaluated by asking participants if they had taken drugs for any of the listed conditions: asthma, diabetes, heart condition, kidney condition, immunosuppression. They were also asked about the frequency of common cold or flu-like disease: never or almost never, rare (1 to 2 times a year), or often (3 or more).

Statistical analysis
The participant's characteristics were described by season. The same participant may have participated in one, two or three seasons. To account for this dependency, we used a logistic model using the generalized estimating equations (GEE) approach to identify variables associated with having at least one AGE episode during a season. Variables with a univariate p-value below 0.20 were included in the multivariable analysis. We then proceeded with a backward stepwise variables selection, using the Akaike information criterion, until reaching a final model including only variables with a p-value below 0.05. We conducted the same analysis with the alternate definition. All statistical analyses were performed using the R software [23] and geepack package [24].

Ethics approval
The GrippeNet.fr study was approved by the French Advisory Committee on Information Processing in Material Research in the Field of Health (authorization 11.565) and by the French National Commission for Computing and Liberties (authorization DR-2012-024). Performing ancillary studies was mentioned in the GrippeNet.fr protocol and in the inclusion survey.

Participation and population
During the 2014-2015 season, 3905 (59%) of the 6632 GrippeNet.fr registered individuals were considered active and were included in the study; 4906 (75%) of 6515 in 2015-2016; and 5164 (83%) of 6234 in 2016-2017 (Table 1). Overall they were 13,974 participants included in the study over the three seasons, and among them 3152 individuals were included to the study for the three seasons. During the study period, 385 participants (9.  Table 1. Socio-demographic characteristics, exposure and health status were similar for the three seasons ( Table 2). The participants were mostly women, 60% in each season (n = 2326 in the first one, n = 2942 in the second and n = 3113 in the third), they lived in urban area for 81% of them (according to the season, n = 3145, n = 3973 and n = 4161) and were on average 53 years old in the first season, 53.6 in the second and 53 in the third. Almost half of the participants had at least one social exposure: 10 to 11% with patients, n = 388 in the first season, n = 476 the second, n = 555 in the third, 10% with elderly, n = 401, n = 486 and n = 537), 31 to 32% with a group of people, n = 1230, n = 1533 and n = 1677) or 24 to 33% with children, n = 952, n = 1213 and n = 1223). Concerning health status, 811 participants (21%) of the first season, 1079 (22%) of the second and 1098 (21%) of the third season were treated for at least one comorbidity.

Risk factor analysis
With univariable analysis (Table 3), we identify several factors associated with the risk of having at least one AGE during a winter season (p < 0.05): season, age, household composition, main activity, having pets, common cold or flu-like disease frequency, being treated for at least one chronic condition, having a respiratory allergy, and BMI.
Regarding adjusted Odds Ratio (aOR) from the final multivariable model (

Discussion
This prospective study estimated the frequency of AGE episodes during winter in the French general population and identified risk factors associated with having at least one AGE episode during this period.
Depending on the definition used and the season, between 6 and 14% of participants presented at least one AGE episode during winter. In Sweden, a study collecting health status data on a weekly basis estimated that 35% of the participants reported having at least one AGE episode in 2013 [13]. The study however adopted a more inclusive AGE definition than ours, it considered a population where young children were overrepresented, and it focused on a whole year, all factors that may explain the higher reported incidence. Other previous studies in the general population were mostly designed to evaluate a weekly or a monthly incidence rate and did not follow individuals for a long period. Therefore, they do not allow the estimation of the number of individuals having at least one AGE episode during a winter period. Our results show a decrease in the number of AGE episodes observed over the three seasons. This decrease between the 2014-2015 and 2015-2016 seasons is reflected in the national surveillance data from the French GPs' Sentinelles network, which monitors cases of acute gastroenteritis consulting a physician during the winter period. Regarding these data, there were a slight re-increase incidence in 2016-2017, in contrast to the decrease in the number of cases observed in our study. This difference with our data is most likely due to a start of the 2015-2016 gastroenteritis epidemic before the start of the Grippenet.fr surveillance period, explaining that our study may not have accounted for AGE from the early beginning of the epidemic [25].
In our study, we identify some risk factors associated with an AGE episode, previously reported in other studies. We found that elderly people were associated to a lower risk. This association has been shown before, in France [4] and in other developed countries [12,14,15,26]. However, others studies reported that infants tend to have significantly more AGE episode [4,12,15] than adults. We only found this association when using a more inclusive definition for AGE episodes in the complementary analysis. This can mainly be explained by the large underrepresentation of children in the GrippeNet.fr cohort [19], causing a low statistical power. In some previous studies, female gender is found to be positively associated with having an AGE episode [12,14,27]. We found this GrippeNet.fr participants having filled at least three weekly surveys (one before, one during and one after the AGE epidemic period for the first two seasons, and one during and one after for the last season) association when using the more inclusive definition. This association is still discussed and not systematically highlighted. Surprisingly, having possible contacts "at risk" during the day (contacts with children, with elderly, with a large group of people and with patients) were never associated with an AGE episode, whereas such associations have been found in several previous studies [5,28,29]. The formulation of some survey questions might be too imprecise and induced misinterpretation from participants, for instance the word "contact" is not defined at any time in the survey, as well as the duration of such contact.
Others AGE risk factors studied here have been rarely analyzed in previous work but are expected. To our knowledge, having a chronic condition has not been studied before and identified as a risk factor for AGE episode in general population. Nonetheless, some chronic conditions such as chronic kidney disease [30] or immunosuppression [31] are known to be associated with an increased risk of acute infections. This association might explain the link between overall chronic conditions and AGE episodes. Association between AGE episode and BMI status has not been found before in general population. Although this association may be partly explained by the increased frequency of functional bowel disorders among overweight people [32], participants were supposed to report only new AGE episodes and not include persistent or chronic symptoms. Overweight is already known to be associated with an increased frequency of influenza [33] and further investigation would be needed to investigate its association with AGE.  Some risk factors associated with an AGE episode in our study are less expected, like having pets at home. This risk factor has been previously studied [5,34], but it has never been significantly associated with AGE episode. Most of the pets, such as cats, dogs, rodents or reptiles are known to carry some bacteria responsible for acute diarrhoea [35], which could partly explain this association, although the study took place during winter, when AGE are more frequently due to viral agent. The fact that individuals living alone had more episodes, was also surprising. We did not find this association in previous published literature. On the contrary, it was found that having AGE is positively associated with living with young children [4,5,28] or even living with more than 2 other people [4]. This discrepancy may partly be explained by the age difference: people living alone had an average age of 58 years, compared to 52 years for those living with other adults or children, (p < 0.001, Student-T test). Another explanation may be that people living alone have different eating habits than those living with other adults and children. A report published in 2008 by the Insee (French National Institute of Statistic and Economical Studies) showed that men living alone were more likely to buy pre-prepared dishes or to eat outdoors [36]. Eating at restaurants may increase the mixing of individuals and therefore the possible exposure to viral agents.

Strengths and limitations
This is the first study evaluating risk factors associated with an AGE episode in France on a prospective cohort, and those factors are evaluated on a very large sample. GrippeNet.fr is an online participatory study, so it is bound to induce some bias in the representativeness of the population followed, as previously shown [19]. This particular mode of recruitment is bound to cause underrepresentation of age group with limited Internet access, such as children or elderly people, notably those living in nursing homes. These two populations may be particularly exposed to AGE epidemics, as vaccination against rotavirus is not recommended as a common practice in France among children, and community life in retirement homes is more likely to lead to outbreaks. Nevertheless, all ages, gender and level of education category are represented in the cohort, thus allowing a study on risk factor. Another important limitation to acknowledge is the design of the GrippeNet.fr cohort that was originally developed to monitor acute winter infections, including gastroenteritis, but was mainly focused on ILI-like episodes. Several potential confounding factors of AGE episodes may not have been collected in the preliminary survey, such as ages of children in household, dietary choices and exposure, or type of pet. Also, communication to potential participants was mainly oriented on ILI-like episodes, and this may have led participants to be less thorough when reporting symptoms not directly related to influenza, like diarrhoea. Nevertheless, the rate of missing data is very low, and results show that French participants contribute very regularly [37], allowing a comprehensive data collection throughout the season and minimizing the risk of undetected AGE episode.   We only include the gathered variable "At least one comorbidity" in the final model, none of the individual one listed below