Ethelberg, Steen*; Olesen, Bente*; Neimann, Jacob§; Schiellerup, Peter*; Helms, Morten*; Jensen, Charlotte*; Böttiger, Blenda†; Olsen, Katharina E. P.*; Scheutz, Flemming*; Gerner-Smidt, Peter*; Mølbak, Kåre‡
Diarrhea among small children is a significant health problem worldwide. In industrialized countries where mortality is very low, diarrhea is nonetheless an important source of morbidity. Episodes are in general self-limiting and resolve within days or a few weeks, although they may in rare instances lead to serious sequelae or even death. Also, due to its high frequency, the disease has a significant socioeconomic impact as a result of both medical costs and lost working days for parents.
Diarrhea may have both infectious (viral, bacterial, or protozoan) and noninfectious causes (eg, malabsorption, toxins, antibiotic treatment, or as a symptom of other illnesses). Several diarrhea-causing viruses have been recognized within the last few decades but are not yet thoroughly characterized in terms of pathophysiology and epidemiology. For instance, norovirus, estimated to be the most frequent cause of diarrhea in the Western world,1,2 causes diarrhea by an unknown mechanism, cannot be grown experimentally, cannot easily be assayed in foods, and in most countries is very rarely considered in routine analyses of stool samples. The epidemiology of bacterial diarrhea is far better understood. An increase in incidence has been observed in Europe and North America during the last decades, primarily as a result of an increase in the number of Salmonella and Campylobacter cases.3–5
Diarrhea among small children often differs from adult diarrhea in terms of infectious agents, transmission routes, and clinical picture. Although a number of studies of risk factors for gastroenteritis caused by bacterial agents in the developed world have been performed (in particular for Campylobacter6 and Salmonella5), there are limited data on risk factors for viral agents7 and for all diarrheal disease among children.
We studied diarrhea among children under 5 years of age by randomly selecting patients submitting stool samples to the major diagnostic laboratory in Denmark and selecting age-matched control children from the Danish population register. The parents of the children were interviewed about previous exposures, and stool samples from both cases and controls were examined for infectious agents. The microbiologic and clinical findings have been published elsewhere.8 Based on the interview part of the study, the present report describes an analysis of risk factors for diarrhea in general by major etiologic categories.
A detailed description of the study design has been given elsewhere.8 Briefly, cases were randomly selected among children less than 5 years of age with a stool sample submitted for examination for infectious gastroenteritis at the Statens Serum Institut between March 2000 and December 2002. This facility diagnosed all patient stool samples from certain Danish counties in this period. When parental consent for a case had been given, controls were randomly selected from the eligible population using the Danish Civil Register. Controls were matched on sex, week of birth, and county of residence. A control-to-case ratio of 2 was aimed for but varied from one to 4. Controls who had experienced diarrhea, vomiting, or fever in combination with abdominal pains within 1 month before the interview were excluded from the study. Parents of participating children were interviewed by phone using a standard questionnaire. Parents of controls were in addition asked to submit a fecal sample from the child. The stool samples (one from each case and control) were examined for diarrhea-causing viruses, parasites, and bacteria, including diarrheagenic Escherichia coli as described.8
The questionnaire was designed to give information on risk factors for diarrhea in general. It included questions relating to prior diagnosis of common childhood diseases other than gastroenteritis, medication, contact with other symptomatic persons, foreign travel, contact with animals, playing in water at various locations, consumption of different types of milk and water, daycare, whether the family bought organic products, and socioeconomic status of the parents.
The dataset consisted of 1,288 children (422 cases and 866 controls). The age distribution was 24% less than 1 year, 43% age 1, 19% age 2, 8% age 3, and 6% age 4; 41% were girls. This age and sex distribution was similar to the overall age and sex distribution of children submitting samples to the laboratory. The median time between the date the stool sample was received in the laboratory and completion of the questionnaire was 12 days for cases and 4 days for controls. For cases, the median time between the first onset of symptoms and completion of the questionnaire was 25 days. Cases and controls were included in the analysis if correctly interviewed even if all microbiologic analyses had not been performed. The percentages of cases/controls that were examined for bacteria, viruses, and parasites were 100%/84%, 90%/75%, and 87%/69%, respectively. Because the results of the interviews were not evaluated until the data collection had ended, only 313 cases and 469 controls were part of matched sets.
For the etiology-specific analyses, we excluded 8% of children, including 5 cases positive for both viral and bacterial agents, 27 controls positive for bacterial agents, 24 controls positive for viral agents, 13 cases and 15 controls positive for pathogenic parasites, and 20 cases in the unknown etiology group from whom a second sample was positive within 40 days of entering the study. The final dataset thus consisted of 801 controls and 383 cases; 73 cases (19%) positive for a bacterial agent, 88 cases (23%) positive for a viral agent, and 222 cases (58%) with an unknown etiology. The virus group comprised cases with rotavirus (n=49), norovirus (15), sapovirus (11), adenovirus (14), and astrovirus (2). The bacteria group comprised cases with Salmonella (n = 20), Campylobacter (14), Yersinia enterocolitica (10), and E. coli that was shiga toxin-producing (STEC; 11), enteropathogenic (EPEC; 8), enterotoxigenic (ETEC; 2), enteroinvasive (EIEC; 1), and enteroaggregative (EaggEC; 9). Attaching and effacing E. coli that did not belong to recognized EPEC serotypes were not counted among the pathogenic bacteria.8
Variables associated with gastroenteritis with a P value of less than 0.10 in a univariate analysis were fitted in a multivariate logistic regression model and thereafter manually eliminated one at a time. We compared models using Wald statistics and the log-likelihood ratio test with a significance level of 0.05. Only 61% of the children were part of matched case–control pairs. We therefore performed unmatched analyses not to lose statistical power, instead controlling for the matching variables age, sex, and season in all multivariate analyses. Season of the year (4 seasons) of disease onset (cases) or interview (controls) was entered into the model as dummy variables. All variables were dichotomous, except age (in years).
To determine etiology-specific risk factors, we analyzed the reduced dataset described here using multinomial logistic regression.9 The variables selected in the general model were included. In addition, specific analyses were performed for each of the 3 pathogenic groups. These models included only cases belonging to the respective group, and the same set of controls was used in all analyses. We used the same model-building strategy as described here for the 3 (binomial) pathogen-specific analyses, which were subsequently examined for interactions.
Odds ratios (ORs) and 95% confidence intervals (95% CIs) were computed. We also calculated the population attributable risk percent (PAR%) as described by Miettinen.10
Risk Factors for Diarrhea
Table 1 shows the risk factors associated with diarrhea. The parents were asked if the child had been in contact with persons who had experienced diarrhea or vomiting within the 2 weeks before onset/interview. The 141 cases and 150 controls who answered yes were then asked about the number and age of contacts, whether they had shared bathrooms, whether the contacts were known to have submitted a stool sample for microbiologic examination, and finally whether the contact took place at home, in daycare, or in other settings. No association with disease was found with the number, age, and bathroom use of contacts. However, contacts of cases had more often submitted a stool sample (OR = 3.5, 95% CI = 1.5–8.0). When the place of contact was substituted into the model, the OR of “at home” was 0.96 (0.61–1.5), whereas the OR of “in daycare” was 3.1 (2.1–4.8) and the OR of “other places” was 2.2 (1.4–3.6).
Regarding daycare, the children fell into 4 groups: those not in daycare (30%), those using a private daycare provider (36%), and those using the public kindergartens for children under age 3 years (20%) and age 3 years and older (14%). Of children not in daycare, 60% were less than 1 year old and 32% were 1 year old. Use of a private daycare provider carried an increased risk of diarrhea in both univariate and multivariate analyses (Table 1); among children in private daycare, 10% were less than 1, 58% were 1, and 29% were 2 years old. No increased risk was associated with the number of children in the daycare group or, after adjustment for age, eating lunch prepared in daycare as opposed to prepared at home.
Parents were asked about foreign travel in the 30 days before onset/interview. For foreign travel in the 7 days before onset/interview, the OR for diarrhea increased from 1.9 (1.2–3.0) to 4.0 (2.0–7.9). Twelve cases had disease onset within 7 days of returning from a foreign destination, whereas another 16 cases had onset of disease while traveling. Cases had primarily traveled to Southern Europe and Northern Africa. When disregarding travel to North European countries, the multivariate OR rose to 6.5 (2.9–15).
Only 19 children had been hospitalized in the 2-week period before onset/interview. The reasons for admission were not known.
Fifty-five percent of the children had been in contact with a dog during the 2 weeks before onset/interview and such contact increased the risk of diarrhea (Table 1). There were further questions about the age of the dog, where it normally slept relative to the child, whether it was given dog treats made of dried meat, and whether it had had diarrhea within the last month. Of these variables, only the last one was associated with disease. Contact with a cat with diarrhea was not associated with disease.
The parent answering the questionnaire provided information about employment and education status; this was the mother in 94% of the instances. Low educational status was defined as 10 years or less of basic schooling. Both low educational attainment and unemployment increased the likelihood that the child had diarrhea (Table 1). No association with diarrhea was found with single-parent households (6%), number of persons in the household, number of siblings of the child, and the type of housing.
A number of questions concerned diseases other than diarrhea. Parents were asked if the child had been diagnosed with asthma (7% of children), pseudocroup (9%), atopic dermatitis (16%), urticaria (5%), bronchitis/pneumonia (26%), hayfever (1%), eczema (8%), or allergy toward milk (2%). Of these, the first 4 were associated with diarrhea in the multivariate analysis (Table 1). Parents were also asked whether the children within 3 months before onset/interview had had otitis media (12%), a cold (76%), cystitis (1%), or had been treated with antibiotics (18%); or had received medication for asthma (12%), allergy/urticaria (3%), or constipation (4%). None of these variables was associated with diarrhea. Combining the 3 variables of asthma, urticaria, and atopic dermatitis into one “atopy” variable in the model resulted in an OR of 1.86 (1.4–2.5) and negligible changes in the estimates of the remaining variables.
Two variables in the model concerned the consumption of milk. Having had formula milk or readymade porridge containing milk powder was associated with diarrhea (Table 1). In contrast, being breastfed was associated with a decreased risk of diarrhea. Of the children being breastfed, 75% were less than 1 year of age and 23% were 1 year. Among the children less than 1 year old, the OR with breast-feeding was 0.44 (0.26–0.75) after adjustment for sex, season, and age (in months).
Finally, more controls than cases reported having had contact with cows. No association was found for any other variables concerning contact with other specific animals. Deleting the 2 variables associated with a decreased risk of diarrhea from the model changed the OR estimates of the remaining variables only marginally.
Risk Factors for Pathogen-Specific Categories of Diarrhea
Based on the risk factors from the general model, a pathogen-specific multinomial multivariate analysis was performed (Table 2). In the group with unknown etiology, which contained the majority of patients, most of the variables still appeared to be important risk factors. For the children with a bacterial infection, foreign travel was the most important risk factor, and unemployment and low level of education of the parent were also risk factors. For the children with a viral infection, contact with other symptomatic persons was the most important risk factor.
To examine the full set of variables in a pathogen-specific context, 3 separate analyses were performed, one for each pathogenic group and each using only the cases with the given pathogen (ie, bacterial, viral, or none found, Table 3). The results for all cases combined and the 3 separate analyses were generally in good agreement. We excluded 3 determinants from the 3 smaller models, namely previous hospitalization (comprised only 19 children in total), having a diagnosis of atopic dermatitis, and being breastfed. Also, some additional variables were selected in the models presented in Table 3. In the model of bacterial diarrhea, 3 new risk factors appear, namely prior diagnosis with bronchitis, if the child never ate organic foods (as opposed to sometimes, frequently, or always, when available), and having had a cold within the 3 months before onset/interview. In the model of viral diarrhea, attending a public daycare institution with children aged 6 months to 2 years was found to be a risk factor along with contact with other ill persons. When the place of contact was substituted into the model, the OR of “at home” was 0.82 (95% CI = 0.35–1.9), whereas the OR of “in daycare” was 5.8 (3.1–11) and the OR of “other places” was 4.1 (2.1–8.0).
In the model of diarrhea of an unknown etiology, the risk factor “contact with a dog with diarrhea” was found to interact with age. The OR was 5.7 (2.1–15) for children younger than 3 years old and 0.9 (0.2–4.7) for children 3 years or older. No statistical interactions were detected in the models of viral or bacterial diarrhea. Entering age counted in months rather than years into either of the models had only very marginal effects on the OR and CI estimates.
This case–control study examined risk factors for diarrhea among children less than 5 years old in Denmark, an industrialized Northern European country of 5.3 million inhabitants. The children were enrolled continuously; cases came from a major microbiologic laboratory and controls from a population register. Cases were enrolled through both general practitioners and hospitals, and enrollment was presumably not biased by socioeconomic factors because the medical system in Denmark is public and financed by taxes.
Because the study concerned diarrhea of all pathogenic etiologies, the questionnaire did not focus on specific food items or categories of food, and this omission may limit some of our conclusions. The questionnaire instead focused on underlying illness, general underlying factors, and certain exposures such as contact with animals, water, and symptomatic persons. In addition to an overall analysis for risk factors for diarrhea, we conducted pathogen-specific analyses. These provided information about determinants for 3 broadly defined pathogenic groups chosen because the wide range of pathogens found in the stool samples made it not feasible to analyze individual pathogens. Thus, viral and bacterial pathogens were grouped. No specific analysis was made for parasites because these were only found in small numbers, and additionally the 7-day exposure window may have been too short to capture the time of infection. Clearly, this use of broad etiology groups may lead to an oversimplification because different types of agents within groups may not share risk factors. On the other hand, several of the risk factors for general diarrhea appeared to follow these etiology groups; for instance, viral diarrhea was closely associated with contact with other ill persons and bacterial diarrhea with foreign travel.
Contact with other symptomatic persons was the major risk factor for both general diarrhea (OR = 2.2) and viral diarrhea (4.1). Among the viral cases, contact with other ill persons carried a PAR of 37%. The contact took place outside of the home, primarily when the children were in daycare. In a recent Dutch study, contact with persons with gastroenteritis was also found to be the primary risk factor for infections with norovirus, sapovirus, and rotavirus.7
Breast-feeding appeared to protect against viral diarrhea, whereas there was a slightly increased risk associated with intake of formula milk, although only in the multinomial model. The protective value of breast-feeding against diarrhea is well described primarily in developing countries,11,12 but has rarely been documented in industrialized countries.13 The effect of breast-feeding was most pronounced in the most relevant age group infants (younger than 1 year), but the effect was strong enough to be seen in the analysis of the entire dataset. The specific association with viral diarrhea may well result from the fact that viral diarrhea primarily occurs among very young children.
For bacterial diarrhea, recent foreign travel—a well-known risk factor—was the most important determinant. Two parent-associated risk factors were also found, namely unemployment and a short education. This indicates a relationship between low socioeconomic status and bacterial diarrhea. Never buying organic foods was associated with an increased risk of bacterial diarrhea and may also be a proxy of low socioeconomic status. Person-to-person transmission played a much smaller role than for viral diarrhea but did occur, as indicated by the slightly raised OR in the multinomial model. Unexpectedly, experiencing a cold shortly before the onset of diarrhea appeared to be protective for bacterial diarrhea. This finding was not related to seasonality of bacterial diarrhea or case-management and has no straightforward explanation. Taken together, the analysis for bacterial and viral diarrhea corroborate the notion that pathogenic gastrointestinal bacteria are predominantly transmitted by food or water, whereas person-to-person transmission is a major route for the transmission of viral diarrhea among children.
The third and last group of cases contained those with an unknown etiology, ie, cases in which the stool samples tested negative. We assume that this group contains some children with diarrhea not caused by infections, but also a number of children with undetected infections. Based on a single specimen, fecal diagnostics are of modest sensitivity. Besides, not all known pathogens were included in the diagnostic panel. Also, viral pathogens were likely underdiagnosed, because some 10% of cases were never tested for viral pathogens, and the analysis for noro- and sapovirus (RNA viruses) was suboptimal due to repeated freezing and thawing of the material before analysis.8 The mixed composition of the group of cases with an unknown etiology is reflected in the risk factors (Table 3). Most appear to represent a carryover from the viral and bacterial groups. This is evident, for instance, for contact with symptomatic persons and foreign travel. However, the risk factor pattern is partly distinct from that of the virus and bacteria group, and some risk factors such as urticaria appear to be specific for the children with an unknown etiology. Overall, it appears as if a sizable proportion of the children with an unknown etiology had diarrhea of a noninfectious etiology, perhaps as many as one third of these children (one sixth of all the case children).
The questionnaire contained questions about prior diagnosis of a number of diseases not generally associated with diarrhea. Surprisingly, several of these were found to be risk factors for diarrhea, primarily for diarrhea of an unknown etiology. This indicates that the gastrointestinal disturbances may be part of the symptomatology of these diseases. This is not unlikely because diarrhea frequently accompanies a number of different diseases and ailments in small children. Furthermore, diarrhea may be more directly associated with atopic disease. Atopic dermatitis has been found to be associated with an increased rate of infections, suggesting impaired immunity as the cause of both,14–16 and therefore the risk of intestinal infections may also be increased. Also, children with atopic disease have an altered gut microflora,17,18 and it is possible that they thereby have an increased risk of diarrhea or of becoming colonized with pathogenic bacteria. Finally, selection bias may play a role, because children with diagnosed diseases might be more likely to see a doctor than unaffected children and as a result also more likely to have a stool sample taken. We also note that asthma is difficult to diagnose in small children, and several of the children diagnosed with asthma may in fact have experienced asthmatic bronchitis. It is unlikely that the associations are related to medications taken for the mentioned diseases, because the questionnaire contained questions about medication, and these were not associated with diarrhea.
Having had contact with a dog with diarrhea was found to be a risk factor among the younger children. This seems to be a plausible risk factor, given the very close contact that dogs and small children may have. Pet dogs have previously been found to be a risk factor of campylobacteriosis in general19,20 and specifically among children.21 The fact that no pathogen was found in the stool samples from these children may be because the diarrhea was caused by pathogens found in dogs but not normally examined for in humans. One candidate for such a pathogen would be Campylobacter upsaliensis.22 Having had recent contact with cows was found to be a protective factor, although it affected only a small number of cases. It is likely that recent contact with cows is simply a proxy for a rural lifestyle. Children in the countryside may have fewer diarrheas for a number of speculative reasons, including less contact with other children or a selection imposed by living relatively further away from medical care.
Places of daycare were where cases most frequently had contact with other persons (children) with diarrhea, contact being the major risk factor identified in this study. Increased focus on hygiene measures in daycare centers may therefore help to reduce the incidence of diarrhea. Even simple measures such as teaching children to wash their hands frequently and routinely washing their toys may help.23,24 The increased risk of private daycare relative to other types of daycare or no daycare seen for the children with diarrhea of unknown etiology may in part be explained by referral of children with various diseases to private daycare.
Concerning the methodology, we found the pathogen-specific analyses to give a substantial amount of additional information compared with the analysis for diarrhea in general. We analyzed the data using both a united multinomial and separate logistic regression models for each pathogenic group. The multinomial analysis supplemented the general model and gave information on the importance of each of the pathogenic groups for each determinant found to be part of the general model. On the other hand, model building using a multinomial approach only was not feasible, and to allow for all putative risk factors to be considered in a pathogen-specific analysis, models were built on the basis of datasets, including only the cases of the relevant pathogenic group.
In conclusion, this study shows that the etiology of diarrhea in industrialized countries is multifactorial, like it is in developing countries, but with somewhat different risk factors. The study also shows that there are different sets of risk factors dependent on the pathogenic group, and that the analyses became much more meaningful when the etiology was taken into account. Some of the risks identified may be addressed; in particular, the transmission of disease from child to child in daycare centers may be reduced.
We thank Carsten Struve, Andreas Munk Petersen, and a number of telephone interviewers from the Danish Zoonosis Center for help with the data collection and Christine Stabell Benn for fruitful comments on the association between diarrhea and asthma and atopy.
1. Koopmans M, Vinje J, Duizer E, et al. Molecular epidemiology of human enteric caliciviruses in The Netherlands. Novartis Found Symp. 2001;238:197–214.
2. Mead PS, Slutsker L, Dietz et al. Food-related illness and death in the United States. Emerg Infect Dis. 1999;5:607–625.
3. The Increasing Incidence of Human Campylobacteriosis. Geneva: World Health Organization; 2001.
4. Tauxe RV. Salmonella enteritidis and Salmonella typhimurium DT104: successful subtypes in the modern world. In: Scheld WM, Craig WA, Hughes JM, eds. Emerging Inections 3. Washington, DC: ASM Press; 1999:37–54.
5. Pegues DA, Ohl ME, Miller SI. Salmonella, including Salmonella typhi. In: Blaser MJ, Smith PD, Ravdin JI, Greenberg HB, Guerrant RL, eds. Infections of the Gastrointestinal Tract. Philadelphia: Lippincott Williams & Wilkins; 2002:669–697.
6. Friedman CR, Neimann J, Wegener HC, et al. Epidemiology of Campylobacter jejuni infections in the United States and other industrialized nations. In: Nachamkin I, Blaser MJ, eds. Campylobacter. Washington, DC: ASM Press; 2000:139–154.
7. de Wit MA, Koopmans MP, van Duynhoven YT. Risk factors for norovirus, Sapporo-like virus, and group A rotavirus gastroenteritis. Emerg Infect Dis. 2003;9:1563–1570.
8. Olesen B, Neimann J, Bottiger B, et al. Diarrhea in Danish children under 5 years of age: a case–control study. J Clin Microbiol. 2005;43:3636–3641.
9. Hosmer DW, Lemeshow S. Applied Logistic Regression. New York: John Wiley & Sons, Inc; 2000.
10. Miettinen OS. Proportion of disease caused or prevented by a given exposure, trait or intervention. Am J Epidemiol. 1974;99:325–332.
11. Feachem RG, Koblinsky MA. Interventions for the control of diarrhoeal diseases among young children: promotion of breast-feeding. Bull World Health Organ. 1984;62:271–291.
12. Molbak K, Gottschau A, Aaby P, et al. Prolonged breast feeding, diarrhoeal disease, and survival of children in Guinea-Bissau. BMJ. 1994;308:1403–1406.
13. Howie PW, Forsyth JS, Ogston SA, et al. Protective effect of breast feeding against infection. BMJ. 1990;300:11–16.
14. Benn CS, Melbye M, Wohlfahrt J, et al. Cohort study of sibling effect, infectious diseases, and risk of atopic dermatitis during first 18 months of life. BMJ. 2004;328:1223.
15. Bohme M, Lannero E, Wickman M, et al. Atopic dermatitis and concomitant disease patterns in children up to two years of age. Acta Derm Venereol. 2002;82:98–103.
16. Rystedt I, Strannegard IL, Strannegard O. Recurrent viral infections in patients with past or present atopic dermatitis. Br J Dermatol. 1986;114:575–582.
17. Bjorksten B, Naaber P, Sepp E, et al. The intestinal microflora in allergic Estonian and Swedish 2-year-old children. Clin Exp Allergy. 1999;29:342–346.
18. Bjorksten B, Sepp E, Julge K, et al. Allergy development and the intestinal microflora during the first year of life. J Allergy Clin Immunol. 2001;108:516–520.
19. Eberhart-Phillips J, Walker N, Garrett N, et al. Campylobacteriosis in New Zealand: results of a case–control study. J Epidemiol Community Health. 1997;51:686–691.
20. Neal KR, Slack RC. Diabetes mellitus, anti-secretory drugs and other risk factors for campylobacter gastro-enteritis in adults: a case–control study. Epidemiol Infect. 1997;119:307–311.
21. Tenkate TD, Stafford RJ. Risk factors for campylobacter infection in infants and young children: a matched case–control study. Epidemiol Infect. 2001;127:399–404.
22. Hald B, Pedersen K, Waino M, et al. Longitudinal study of the excretion patterns of thermophilic Campylobacter spp. in young pet dogs in Denmark. J Clin Microbiol. 2004;42:2003–2012.
23. Uldall P. Forms of care and children's infections. 1. Occurrence and causal factors [in Danish]. Ugeskr Laeger. 1990;152:2345–2348.
24. Uldall P. Forms of care and children's infections. 2. Consequences and possibilities for intervention [in Danish]. Ugeskr Laeger. 1990;152:2349–2351.
© 2006 Lippincott Williams & Wilkins, Inc.