Risk Factors for Sporadic Shiga Toxin–producing Escherichia coli Infections in Children, Argentina

These infections can be prevented by avoiding known risk factors.

Argentina had the highest rates of HUS globally, 10.4 and 12.2 cases/100,000 children <5 years of age in 2001 and 2002, respectively (17). In Argentina, HUS is the leading cause of acute renal failure among children; in 1 study, after at least 10 years of follow-up, ≈20% of Argentine children had low creatinine clearances (5). HUS is responsible for 20% of kidney transplants among children and adolescents in Argentina (18). In studies in the 1990s, evidence of STEC infection was found in 59% of Argentinean HUS case-patients, and E. coli O157 was the predominant serogroup (19,20). In 2000, HUS became reportable in Argentina, and sentinel sites began screening for STEC on all routine stool cultures. Given the high rate of HUS, the lack of defi nitive treatment, and the high morbidity, primary prevention of STEC infections is needed to lower the incidence of childhood kidney disease. However, controlled epidemiologic studies to identify risk factors associated with STEC infection have not been conducted in Argentina. To evaluate risk factors for sporadic STEC infection, we conducted a case-control study in 2 sites, Mendoza and Buenos Aires cities and their surroundings.

Risk Factors for Sporadic Shiga
Toxin-producing Escherichia coli Infections in Children, Argentina 1

Case Ascertainment
Patients were enrolled from January 2001 through December 2002 from the public tertiary-care pediatric hospitals, Hospital "Dr. Humberto Notti" in Mendoza, which serves an urban and semirural area, and Hospital "Dr. Juan P. Garrahan" in Buenos Aires, which serves an urban area. Study personnel detected STEC cases through daily review of the hospitals' laboratory records and detected diarrheaassociated HUS cases through biweekly discussions with nephrologists.

Defi nitions
A case of STEC infection was defi ned as illness in a previously healthy child <16 years old who was evaluated at a participating institution and who had either culture-confi rmed O157 STEC diarrhea, culture-confi rmed non-O157 STEC diarrhea, or defi nite diarrhea-associated HUS. For convenience, we included only children permanently residing within 15 km of each institution. Local investigators stated that the characteristics of the population within each of these areas were similar to that within 50 or 100 km of the respective hospitals. We chose these 2 areas so that the population serviced would be similar to the Argentine population living in areas with a high incidence of HUS.
Defi nite HUS was defi ned as acute microangiopathic hemolytic anemia, thrombocytopenia, and renal impairment as determined by 1) hematocrit <30% with microangiopathic changes on peripheral blood smear (e.g., schistocytes, burr cells, helmet cells, or red cell fragments); 2) platelet count <150,000/mm 3 ; and 3) serum creatinine concentration >2 standard deviations above the upper limit of normal for age and sex (21); or abnormal urinary sediment by dipstick, i.e., hematuria (>2+) or proteinuria (>2+). HUS was considered diarrhea-associated when a diarrheal illness preceded HUS by <3 weeks. We excluded children with a family history of HUS, secondary HUS (e.g., drug-associated), or HUS associated with pneumococcal infection. Probable HUS was defi ned as an illness that met only 2 of the laboratory criteria in a patient with culture-confi rmed STEC infection.

Case-Control Study
For each patient enrolled, 2 age-and neighborhoodmatched control children without gastrointestinal illness in the 2 weeks before the matched case's illness onset were identifi ed. We neighborhood-matched control children to control for socioeconomic status, which was not a factor of interest, was a possible confounder, and was quite variable in the study areas. In urban areas, a trained inter-viewer sought controls by walking, starting at the casepatient's home, going to the third house from the nearest corner, and continuing to every house on the block, then to the block facing the case-patient's residence, then to other blocks in a clockwise fashion, until 2 eligible and consenting controls were interviewed. In nonurban areas, the interviewer randomly chose a cardinal direction and then sought controls beginning from the third residence from the case patient's house in that and the opposite direction until controls were found. Informed consent was obtained from the adult primary caregiver, who was interviewed with a standardized questionnaire, administered in person. Most eligible controls were enrolled; information on potential controls who were excluded or chose not to participate was not kept.
The questionnaire (available upon request from the corresponding author) had 89 major questions and was divided into 3 major sections: characteristics of and treatment for the illness (19 questions), exposures (55 questions), and demographics (15 questions). Exposure questions were divided into sections that dealt with human contacts, care and feeding of young children, water sources and treatment, beef, other meats, fruits and vegetables, meat handling at home, animal contact, swimming, and travel. Many exposure questions had several parts (e.g., if beef consumed, was it pink; if chicken consumed, indicate if at home, restaurant, or other location). Almost all responses were measured discretely (i.e., categorically). Responses to each question were treated independently, except for those about fruits and vegetables, which were combined and dichotomized at the median; this analysis method was checked against other choices and found robust. All exposure questions for case-patients and controls were about the 7 days before the onset of illness. Interviews of case-patients were conducted a median of 12 days (range 3-41) after diarrhea onset; control interviews were conducted a median of 15 days (range 1-41) after diarrhea onset in the matched case. Controls were age-matched to case-patients by using the following groups: <12 months; 1-5 years ± 1 year; 6-9 years ± 2 years; 10-15 years ± 3 years.

Laboratory Methods
Fecal samples were plated onto sorbitol-MacConkey agar directly, and after enrichment at 37°C for 4 h in trypticase soy broth were supplemented with cefi xime (50 ng/ mL) and potassium tellurite (25 mg/mL). The confl uent growth zone and colonies were screened for stx1, stx2, and rfbO157 genes by a multiplex PCR (22,23). Isolates with stx1 or stx2 genes were identifi ed by standard biochemical methods. Stx-positive colonies were serotyped (24) and characterized at the Argentina National Reference Laboratory (11,25). Data analysis used 3 steps: an initial univariate analysis, a second univariate analysis adjusted for highly signifi cant factors of prior or secondary interest, and a fi nal multivariable model-building analysis. Four factors were ultimately chosen in the second univariate analysis as a fi xed set from which to explore further model-building, based on epidemiologic sensibility, strong association, and stability of subsequent adjusted associations. Single-and multiple-variable conditional logistic regression models were used to evaluate associations between the outcome and exposure variables. At each step, risk factors that were statistically signifi cant (p<0.05) and had biologic plausibility were selected for further modeling. In multivariable model-building, we pursued forward, backward, and the best subset selection strategies, as well as manual strategies. Standard methods were used to assess model fi t, including residual analyses. Maximum likelihood parameter estimates from these models were used to calculate point estimates and confi dence intervals for odds ratios, referred to henceforth as matched odds ratios. Exact analysis was used where small sample size would make asymptotic analysis suspect, and Mantel-Haenszel odds ratios were computed when maximum likelihood estimates did not exist.
Exploratory and sensitivity analyses were performed on subsets of the data defi ned by location and subcategory of disease status. Because the subsets by site (Buenos Aires and Mendoza) and by serogroup (O157, non-O157) were small and most factors examined had already been demonstrated as risky or protective in the larger dataset, we used a p value of 0.10 to assess signifi cance by site and serogroup. For similar reasons, we did not perform multivariable analyses on these subsets. Three patients with mixed STEC infection and 2 stx-positive HUS patients without STEC isolated were excluded from the analysis by serogroup. Data were analyzed with Epi Info 2000 (Centers for Disease Control and Prevention, Atlanta, GA, USA) and SAS 9.0 (SAS Institute, Inc., Cary, NC, USA) software. The study was approved by the hospitals' ethics committees as well as the institutional review boards of the Ministry of Health of Argentina and CDC.

Case and Control Characteristics
Among 157 eligible case-patients, 150 (96%) were enrolled; the parents of 1 child refused, and interviewers could not contact the parents of 6. The hospital "Dr. Juan P. Garrahan" in Buenos Aires enrolled 54% of the cases, and the hospital "Dr. Humberto Notti" in Mendoza enrolled 46%. Among the 150 enrolled cases, 17 met both entry criteria of culture-confi rmed STEC infection and defi nite HUS, 82 met only the criterion of culture-confi rmed STEC infection, and 51 met only the criterion of defi nite HUS. In addition to the patients with defi nite HUS, 10 patients with culture-confi rmed STEC infection had probable HUS.

Risk Factors Overall
Analysis of single variable associations, when the fi xed adjustment factors were controlled for, identifi ed dietary habits and animal exposures linked to illness (Table 1). General dietary habits linked to STEC illness included eating at a social gathering, eating any meal prepared at home, and drinking from a baby bottle left at room temperature for >2 hours. Many beef-related exposures were signifi cantly associated with STEC infection (Tables 1, 2). Eating beef outside the home and eating undercooked beef (described as uncooked, red and juicy, or pink) anyplace was associated with illness. Eleven percent of case-patients but only 5% of controls consumed jugo de carne (liquid squeezed from a tender, usually lightly cooked piece of beef, and spoon-fed); case-patients with this exposure ranged from 7 months to 9 years old. Living in or visiting a place with farm animals, contact with farm animals (including horses, pigs, poultry, and cattle), and contact with cattle manure were associated with illness. Risky exposures that suggest person-to-person transmission from young children included contact with a child <5 years of age, attending daycare or kindergarten, and contact with a child <5 years of age with diarrhea. Wearing diapers was also linked to illness. No signifi cant differences between case-patients and controls were found in the distribution of most variables that relate to socioeconomic status (e.g., number of bedrooms, water supply, garbage disposal, educational level of parents). However, case households were more likely to have nonparental income. Four protective factors were identifi ed, all related to beef ( Table 2). These were the child eating meatballs at home; the child eating empanadas (fried or baked pastries with ground beef fi lling) at home; the child eating meat pie at home; and the respondent always washing hands with soap and water after handling raw beef. The factors controlled for in the adjusted univariate analysis were eating more than the median number of fruits and vegetables, male sex, having a nonparent respondent, and the respondent always washing hands after handling raw beef; all were protective (Tables 1, 2).

Risk Factors by Site and Etiology
We performed a univariate adjusted analysis for all variables by site (Tables 1, 2). For every variable that was signifi cantly risky or protective in the combined analysis, the OR went in the same direction in the site-specifi c analysis (Buenos Aires 81 cases, Mendoza 69 cases), although the association was not always statistically signifi cant. Many dietary habits, most of which were beef associated, were signifi cantly associated with illness in Buenos Aires, whereas fewer reached statistical signifi cance in Mendoza. Consuming jugo de carne was signifi cantly associated with illness in both sites; however, 19.1% of case-patients from Mendoza consumed this item compared with only 4.9% from Buenos Aires.
Risk and protective factors were also analyzed separately for patients with culture-confi rmed O157 or non-O157 STEC infection (Table 3). These 2 groups were similar in age, sex, and site distribution. The risk and protective factors among these 2 groups were similar to those of all study participants. Among patients with O157 STEC infection, illness was signifi cantly associated with eating at a social gathering, with many meat-related dietary habits, exposures related to farm animals and their environment, wearing diapers, and having a nonparental household income. Protective factors included several related to eating beef at home and buying beef less than once a week. Among the smaller group of patients with non-O157 STEC infection, the only risk factors significantly linked to illness were drinking from a bottle left at room temperature, drinking formula (a factor not identifi ed in the full group), eating a piece of beef outside the home, teething on undercooked beef at home, contact with a child <5 years of age with diarrhea, wearing diapers, and living in an overcrowded condition. Eating meat pie at home was protective for this group.

Discussion
This fi rst study of risk factors for sporadic STEC infections in Argentina demonstrates a broad range of factors associated with transmission. Undercooked beef in many forms was the most risky food. The presence of E. coli O157 in beef purchased in Argentina has also been demonstrated microbiologically (25). Beef, especially undercooked ground beef, is well recognized as a vehicle for E. coli O157:H7 infections (26). Our results also suggest that many STEC infections are acquired in the home as a result of breaches in kitchen hygiene in relation to beef; washing hands after handling raw beef, especially with soap and water, was protective. The protective effect of consuming some beef products at home is further evidence of the important role of the food preparer. Few beef-related factors were signifi cantly risky in Mendoza, suggesting that this population may consume less undercooked beef; however, consumption of jugo de carne, a risky food, was much more common in Mendoza.

768
Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 14, No. 5, May 2008 †For STEC O157, the denominator (number of respondents) for case-patients varied from 56 to 58, except for contact with a child <5 y with diarrhea, in which the number was 51. The denominator for their controls varied from 112 to 116, except for this same factor, in which the number was 104. For non-O157 STEC, the denominator (number of respondents) varied from 37 to 38, except for contact with a child <5 y with diarrhea, in which the number was 31. The denominator for their controls varied from 73 to 75, except for this same factor, in which the number was 67. ‡All significant associations except these were also significant associations in the total dataset with 150 cases. §The term "meat" includes ground beef. ¶Liquid squeezed from a tender, usually lightly cooked piece of beef, and spoon-fed.
Whereas ground beef consumed as hamburgers is frequently implicated in North America, the spectrum of major risky beef items we identifi ed in Argentina is wider. Other beef-based items linked to sporadic STEC infections in this study included ground beef, pieces of beef, pieces of tender "teething" beef, beef milanesa (breaded beef), steak, jugo de carne, and salami. Salami and other types of beef are uncommon causes of STEC outbreaks (27). Among 183 foodborne E. coli O157 outbreaks reported in the United States from 1982 to 2002, 41% were linked to ground beef, but only 6% to other beef items (28). These fi ndings highlight the importance of conducting studies locally to determine local risk factors and corresponding control measures. The risk we demonstrated from contact with farm animals and their environment supports studies from other areas that this is an important mode of transmission (9,(29)(30)(31)(32). The variety of animals to which exposure conferred risk, including some which have never been directly implicated as a source of STEC infections, suggests widespread contamination of farm environments. Our data also indicate that person-to-person spread is an important mode of STEC transmission in Argentina, as evidenced by the increased risk for illness from contact with a young child with diarrhea and from attendance at a daycare or kindergarten. The neighborhoodmatched study design limited the likelihood of identifying socioeconomic factors, but the fi nding that case households were more likely than controls to have nonparental household income suggests that they were poorer.
To our knowledge, this is the largest study of risk factors for sporadic non-O157 STEC infection. Most exposures that were risky for the 150 case-patients also had high ORs for the non-O157 STEC case-patients, suggesting that similar exposures are risky; however, few of the risks were statistically signifi cant in the subgroup. This fi nding may partially refl ect the small size (38 cases) and diversity (14 serogroups) of the non-O157 STEC subgroup. Others have also reported outbreaks and sporadic cases of non-O157 STEC infections caused by cattle-related items (1,9). However, a study from Belgium of both O157 and non-O157 STEC infection found that consumption of fi sh but not beef was risky (33). Our fi nding that drinking infant formula was risky only in the non-O157 subgroup merits further study. Powdered infant formula is a known source of invasive infections in infants (34). With larger studies, strains that are less likely to be pathogens can be excluded and serotype-specifi c risk factors can be examined.
To our knowledge, others have not reported male sex as a protective factor (or female sex as a risk factor) for STEC infection. However, others have reported female sex as a risk factor for E. coli O157-associated hemolytic anemia (35). The reason for this sex difference is not known. We do not know why having a nonparent responder to the questionnaire was associated with a lower risk for STEC infection. However, it suggests a setting in which help with childcare is available from family members or paid care providers.
Our fi nding that eating a wider variety of fruits and vegetables was protective against STEC infection merits further investigation. A varied diet may increase resistance to disease by providing bowel fl ora that help to protect against colonization with pathogens by providing compounds that block bacterial adhesions, as has been postulated for urinary tract infections (36), or by some other mechanism (37).
Our study had several limitations. First, we included as STEC cases children with diarrhea-associated HUS who did not have laboratory confi rmation of STEC infection. However, other data indicate that almost all diarrhea-associated HUS cases in children are due to STEC infection (7). To our knowledge, Shigella dysenteriae type 1, the only other known cause of diarrhea-associated HUS (38), has not been isolated from ill persons in Argentina in recent decades. Inclusion of these HUS cases provided power needed for the analysis; a subanalysis examining these cases alone indicates that they did not introduce any extraneous associations into the analysis of the full dataset (data not shown). Second, we analyzed multiple exposures, which can lead to fi nding associations by chance alone. However, the factors we identifi ed are plausible biologically and supported by other evidence. Third, features inherent to the study design may have led to fi nding risk and protective factors that were not representative of the Argentine population. Although the study population included those with urban, suburban, and rural residences, our inclusion of patients only within 15 km of study hospitals likely resulted in exclusion of some very rural segments of the population. Fourth, our neighborhood matching of controls may have led to overmatching on some fi xed environmental features, but decreased the chance of identifying risk factors that were surrogates for differences in socioeconomic factors. We matched for neighborhood to control for socioeconomic status both to decrease the number of signifi cant factors that were diffi cult to change and because we expected most of the causal pathways to relate to food and animal exposures. Fifth, we studied 2 geographically separated populations. Our subanalysis indicated that some risky factors were more prominent in 1 location. Studies of other Argentine populations may identify risky practices important in those populations.
We considered using the multivariable model as the complete basis for describing our results. However, we were able to create many multivariable models of similar strength that varied in the factors that remained signifi cant. In all multivariable models, some important factors dropped out, but those factors varied. Because no one multivariable model adequately described the fi ndings, we chose to present both the multivariable model that had the strongest individual predictors and the adjusted univariate analysis. The latter retains some important factors amenable to intervention, such as eating undercooked beef at home.
Measures are needed to decrease the likelihood of persons in Argentina consuming food contaminated with STEC. Effective safety practices at all stages of the food chain must be ensured. In particular, the contamination of beef by STEC O157 should be reduced. Major efforts to educate the Argentine public and the food industry could help to reduce these serious illnesses. Social research is needed to better understand practices involving giving meat for teething and jugo de carne to young children. Ensuring that beef is well cooked is a key message. Education is needed to explain the risks related to exposure to farm animals and the ability of people to protect themselves by washing hands.
Evidence indicates that measures instituted by industry, in response to government regulations, recalls, and outbreak investigations, are critical in decreasing STEC infections. After a large outbreak due to ground beef in 1993 in which 4 children died (39), the US Department of Agriculture declared E. coli O157:H7 an adulterant in ground beef; retail beef from lots known to contain the organism must now be recalled. In 2002, a recall of >18 million pounds of ground beef with E. coli O157 contamination (40), and a new USDA directive (41), galvanized the US beef industry to institute more aggressive pathogen control measures, including testing of all lots of beef trimmings or ground beef for E. coli O157 in plants (R. Huffman, American Meat Institute Foundation, pers. comm.). Implementation of prevention measures by industry, government, and consumers could result in a decrease in the incidence of STEC infections in children, and thereby decrease the incidence of childhood kidney disease from HUS in Argentina with its associated human and economic costs.