170 related deaths in the United States. A group of experts convened by the Council for Agricultural Science and Technology concluded in 1994 that illness cases likely ranged between 6.5 and 33 million and that deaths might be as high as 9000 (CAST, 1994).

Most recently, Mead et al. (1999) compiled and analyzed information from multiple surveillance systems, including FoodNet. The analysis entailed three basic assumptions that concerned the degree of underreporting, the proportion of foodborne transmission for the individual pathogens, and the frequency of acute gastroenteritis in the general population. These investigators concluded that foodborne diseases cause approximately 76 million illnesses and 5200 deaths annually. Known pathogens were estimated to account for 14 million illnesses and 1800 deaths.

The same authors recognized two limitations in these estimates (Mead et al., 1999). First, separate calculation methods were necessary for estimates specific to bacterial, parasitic, and viral pathogens because of different surveillance information. Second, some rare infectious agents, such as Plesiomonas, Aeromonas, and Edwardsiella, as well as noninfectious agents, such as mushroom or marine biotoxins, metals, and other inorganic toxins, were not considered because of a lack of surveillance data. Mead et al. (1999) also discussed possible explanations for the discordance among estimates obtained by different authors. First, it is noted that the various figures often refer to different groupings of pathogens, i.e. either to known pathogens or to all causes of foodborne illnesses (known and unknown, infectious and noninfectious). Second, the single analyses used data from different sources. Finally, different rates of foodborne transmission were assumed in all cases.

7.2.2 Analytical methods for surveillance data: a Salmonella case study

Outbreak versus sporadic cases

While the cause of the majority of foodborne outbreaks is unknown, the number of outbreaks with definitive etiology remains relatively constant (Fig. 7.1). Although outbreaks are often the newsworthy effect of foodborne diseases, the number of cases that they cause is merely a fraction of all foodborne illnesses that occur each year. Hence, 'sporadic' cases - that is, cases of foodborne

Fig. 7.1 Number of foodborne outbreaks in the United States, 1988-1997.

illnesses that are not or cannot be linked to an outbreak - make up the majority of foodborne illnesses. This point is illustrated in the case of Salmonella isolates (Fig. 7.2). In the period from 1988 to 1997, more than two-thirds of the Salmonella isolates that were reported to CDC were from sporadic cases (range: 68% in 1996 to 93% in 1988). The proportion of sporadic cases for other foodborne pathogens is even higher than the one observed with Salmonella. Although the distinction between outbreak and sporadic cases is somewhat artificial, given the working definition of an outbreak used by CDC (i.e. two or more linked cases), the observation is meaningful in the context of microbial dose-response assessment. In fact, a high proportion of sporadic cases may be suggestive of a small attack-rate and thus of low-dose exposure. This conclusion is consistent with the postulate that the actual dose ingested in sporadic cases of

Fig. 7.2 Total Salmonella isolates and percent distribution between outbreak and sporadic cases in the United States, 1988-1997.

human salmonellosis frequently may be the 1% infective dose (ID1) (Blaser and Newman, 1982).

Host factors

The findings of epidemiologic studies probably offer the best opportunity to identify host factors that influence the risk of foodborne disease. A review of two dozen analytical studies, i.e. case-control and cohort studies, related to Salmonella infection is presented below (Table 7.3).

A common observation is that age of patients with Salmonella infections is distributed according to a bimodal distribution with peaks in children and the elderly. In a Belgian hospital-based study covering isolates for a 20-year period (1973-1992), S. Typhimurium and S. Enteritidis were mainly isolated in children of less than 5 years of age (Le Bacq et al., 1994). The age distribution was, however, less accentuated for S. Enteritidis than for S. Typhimurium. Both serovars were more likely to lead to bacteremia in middle and older age groups than in those younger than 5 years of age (Le Bacq et al., 1994), confirming a previous observation made in the United States (Blaser and Feldman, 1981). Another study reported on Salmonella isolates obtained by a Hong Kong hospital for the period 1982-1993 (Wong et al., 1994). Among both intestinal and extraintestinal isolates, S. Typhimurium, S. Derby and S. Saintpaul predominated in infants. In patients older than 1 year of age, S. Derby and S. Typhimurium remained the most common intestinal isolates, while S. typhi, S. Typhimurium, and S. Enteritidis were the most common extraintestinal isolates. In a British population-based study, the highest age-specific isolation rates for S. Enteritidis were observed in children aged less than 2 years, and for S. Typhimurium, in those under 1 year (Banatvala et al., 1999).

Table 7.3 Risk factors for foodborne non-typhoidal salmonellosis reported in case-control and cohort studies

Factor category

Demographic and socioeconomic factors

Genetic factors

Reported factors

Age Gender

Race and ethnicity Nutritional status

Social/economic/environmental factors Travel abroad

HLA-B27 gene

Health factors Immune status

Previous exposure Concurrent infections Underlying diseases Concurrent medications

In children younger than one 1 year of age, the peak incidence is generally observed in the second and third months (Ryder et al., 1976; Davis, 1981; CDC, 1983). The study from Hong Kong showed, however, a peak at 12 months of age (Wong et al., 1994). In a study on Peruvian children, the IgG and IgM titers against Salmonellae serogroups AO, BO, and DO were higher at 12 months of age than at 2 or 3 months of age, which was interpreted as an indication of acquired immunity (Minh et al., 1998).

It should be pointed out that association with age might be spurious. It is likely that children and the elderly with diarrhea are more frequently cultured than other age groups (Banatvala et al., 1999). Further, age influences the relative exposure to specific serovars. This may explain an increased risk of infection with resistant Salmonella serovars, which has been observed in infants (Lee et al., 1994). Moreover, age association may reflect behavioral characteristics. For instance, eating snow, sand, or soil - a behavior more likely in children - was found to be associated with S. Typhimurium O:4-12 infection (Kapperud et al., 1998b).


In terms of number of isolates, men seem to be more likely to become infected with Salmonella than women. A male-to-female ratio of 1.1 has been reported on various occasions (Blaser and Feldman, 1981; Le Bacq et al., 1994; Wong et al., 1994). The significance of such a finding does not appear to have been addressed. Several factors, such as proportion of the two genders as well as different age distributions for males and females within a country or hospital catchment area may play an important role. In the evaluation of single studies, it should be pointed out that the occurrence of other factors, e.g. use of antacids or pregnancy, tend to be gender specific, and gender may thus have the effect of a confounder.

Race and ethnicity

The potential role of race and ethnicity has seldom been considered. An association with black race and Hispanic origin was reported for resistant Salmonella infections (Riley et al., 1984; Lee et al., 1994). In the former case, the association was explained by differences in the distribution of infecting serovars among ethnic groups, which in turn depended on varying food preferences or methods of food preparation.

Nutritional status

An association between altered nutritional status and acute gastroenteritis has been shown in AIDS patients (Tacconelli et al., 1998). Apart from this report, no direct reference to the role of nutritional status was found in the recent literature.

Social/economic/environmental factors

Isolation rates of several Salmonella serovars have been compared among groups of different socioeconomic strata on the basis of the Townsend score, an index for deprivation (Banatvala et al., 1999). While isolation rates for S. Typhimurium were not related to the Townsend score, highest isolation rates of S. Enteritidis were observed in less-deprived areas. It was hypothesized that populations living in the less-deprived areas more frequently ingested vehicles, such as raw eggs, harboring S. Enteritidis.

Sanitation deficiencies have been associated with high rates of enteric disease, but direct reference to the potential role of Salmonella spp. is scarce. In the 1950s, lack of sanitation, poor housing, limited water supply, and poor personal hygiene were associated with high Shigella rates in Guatemala (Beck et al., 1957). A similar observation was made in the United States where, in areas of inadequate sanitary facilities, poor housing, and low income, Shigella infections were the major causes of diarrheal diseases. In particular, there were nearly twice as many cases of diarrhea among persons living in dwellings having outhouses than among those whose houses had indoor toilet facilities (Schliessmann et al., 1958). In certain Guatemalan villages, the habits of the people and the density of the population were found to be more important determinants of diarrheal disease than the type of housing (Bruch et al., 1963). In a study conducted in Panama, six representative types of dwellings were considered as an index of social and economic influences on the prevalence of specific enteric pathogens among infants with diarrheal disease (Kourany and Vasquez, 1969). Each dwelling type differed characteristically from one another, but five of the six types were considered substandard and their occupants were of low socioeconomic status. Infection rates for enteropathogenic Escherichia coli, Shigella, and Salmonella among infants from the various groups of substandard dwellings ranged from 6.0 to 10.2%, in contrast to the zero infection rate observed in infants from the better housing type. It is worth noting that the literature on sanitation and housing was mainly published in the 1950s and 1960s. It is possible that improved waste-water management and drinking water quality consequent to economic development has diminished the importance of those factors in some countries.

A French study on sporadic S. Enteritidis infections in children investigated the influence of diarrhea in another household member in the 3-10 days before a child showed clinical symptoms. The strength of the association with such a factor appeared stronger for cases in infants (1 year of age or less) as compared with cases in children between 1 and 5 years of age (Delarocque-Astagneau et al., 1998). On the basis of this observation as well as other results of the study, it was postulated that S. Enteritidis infection in children of less than 1 year of age is mainly related to exposure to a household contact, while children between 1 and 5 years of age are more likely to contract infection by consuming raw or undercooked egg products or chicken.

A seasonal pattern in isolations, which generally shows increased rates during warmer months, has been documented. For instance, in a British study, increased isolation rates for S. Enteritidis, S. Typhimurium, S. Virchow, and S. Newport were observed in summer (Banatvala et al., 1999). The French study mentioned above noted that the association between S. Enteritidis infection and prolonged storage of eggs was stronger during the summer (Delarocque-Astagneau et al., 1998).

Travel abroad

Travel abroad is a risk factor for Salmonella gastroenteritis that has been consistently demonstrated in both North America and northern Europe. For California residents, Kass et al. (1992) demonstrated an association between sporadic salmonellosis and travel outside the United States within 3 weeks prior to the onset of illness. Possible variations in sporadic salmonellosis were cited in a Swiss study (Schmid et al., 1996) where travel abroad within 3 days prior to disease onset was found to be associated with both S. Enteritidis and serovars other than Enteritidis, although to a greater extent for the latter case. While most patients with S. Enteritidis infection were more likely to have traveled within Europe, the majority of non-Enteritidis infections might have originated outside Europe. Individuals of a British region with Salmonella infection were more likely to have reported travel abroad in the week before the onset of illness (Banatvala et al., 1999). Frequency of overseas travel between patients with S. Enteritidis or S. Typhimurium infection was no different, but it was among patients with other serovars. Indication of how travel abroad may lead to increased risk of salmonellosis was reported in a study of Norway residents (Kapperud et al., 1998a). This study suggested that about 90% of the cases from whom a travel history was available had acquired their infection abroad. The study failed to show an association with either foreign travel among household members or consumption of poultry. However, consumption of poultry purchased abroad during holiday visits to neighboring countries was the only risk factor considered in the study that remained independently associated with disease. Only cases of S. Typhimurium allowed for a separate analysis, which showed an association with both poultry purchased abroad and foreign travel among household members.

Genetic factors

As far as acute gastroenteritis caused by Salmonella is concerned, no genetic factors related to the host have been reported. Reports concerning race and ethnicity probably should be considered in light of eating habits. There are, however, genetic determinants for chronic disease sequelae associated with Salmonella exposure. For example, a putative association of the gene Human Leukocyte Antigen B27 (HLA-B27) for patients with spondyloarthropathies, in particular reactive arthritis and Reiter's syndrome, has been described. The HLA-B27 gene has a very high prevalence among the native peoples of the circumpolar arctic and sub-arctic regions of Eurasia and North America, and in some regions of Melanesia. In contrast, it is virtually absent among the genetically unmixed native populations of South America, Australia, and among equatorial and southern African Bantus and Sans (Bushmen) (Khan, 1996). Fifty percent of Haida Indians living on Queen Charlotte Islands of the Canadian province of British Columbia have the HLA-B27 gene, which is the highest prevalence ever observed in a population. The prevalence among Americans of

African descent varies between 2 and 3%, while 8% of the Americans of European descent possess the gene (Khan, 1995).

Immune status

The host immune status is, as in any other infectious disease, a very important factor in determining both infection and clinical illness. In general terms, its importance does not seem to have been the direct subject of any formal work, and has thus to be indirectly assessed though other factors, e.g. age and underlying conditions.

Concurrent infections and underlying conditions

Persons infected with Human Immunodeficiency Virus (HIV) tend to have recurrent enteric bacterial infections. Such infections are often severe and associated with extraintestinal disease (Smith et al., 1988; Angulo and Swerdlow, 1995). The following six risk factors for enteric salmonellosis have been identified in HIV-infected patients: increasing value on the prognostic scoring system Acute Physiology and Chronic Health Evaluation (APACHE II); altered nutritional status; previous antibiotic therapy; ingestion of undercooked poultry/eggs or contaminated cooked food; previous opportunistic infections; and stage C HIV infection (Tacconelli et al., 1998).

The risk represented by other underlying conditions was evaluated in a large nosocomial foodborne outbreak of S. Enteritidis that occurred in 1987 in New York (Telzak et al., 1991). Gastrointestinal and cardiovascular diseases, cancer, diabetes mellitus, and alcoholism, as well as use of antacids and antibiotics, were the factors considered. 0f these, diabetes was the only condition that was independently associated with infection after exposure to the contaminated meal. Although people with diabetes were more likely to develop symptomatic illness than those without, the difference was not statistically significant. Decreased gastric acidity and autonomic neuropathy of the small bowel (which leads to reduced intestinal motility and prolonged gastrointestinal transit time) are the two biologically plausible mechanisms for the increased risk of S. Enteritidis infection among diabetics. Among patients with sporadic salmonellosis in Northern California, diabetes mellitus and cardiac disease were the only two health conditions (out of a total of 14) that were associated with clinical illness (Kass et al., 1992). Nongastrointestinal medical conditions and, to a larger extent, a recent history of gastrointestinal disorder, were associated with sporadic S. Typhimurium 0:4-12 infection in Norway (Kapperud et al., 1998b). It was, however, noted that physicians are more likely to recommend a stool culture for patients with preceding illness. In a British epidemiologic study, cases of Salmonella infection were more likely to report a long-term illness (including gastroduodenal conditions) than controls (Banatvala et al., 1999).

Concurrent medications

Although the use of gastric acidity reducers and antimicrobial medication are often considered risk factors for enteric diseases, the evidence found in the literature concerning their association with human salmonellosis is inconsistent. While some studies have shown an association with antacid use (Banatvala et al., 1999), others have failed to do so (Telzak et al., 1991; Kapperud et al., 1998a,b). A similar situation is found for the use of antibiotics in the weeks/days preceding the infection or disease onset: some studies have demonstrated an association (Pavia et al., 1990; Kass et al., 1992; Bellido Blasco et al., 1998) but other have not (Telzak et al., 1991; Kapperud et al., 1998a,b; Banatvala et al., 1999). Having a resistant Salmonella infection has been associated with previous antibiotic use (Lee et al., 1994). An association between serovars other than S. Enteritidis and intake of medications other than antacids was shown in Switzerland (Schmid et al., 1996). Regular use of medications was a risk factor for S. Typhimurium 0:4-12 infection in Norway (Kapperud et al., 1998b). In the same study, use of antacids and antibiotics were not risk factors.

Application of analytical methods to FoodNet surveillance data to explore typical risk factors for salmonellosis

Surveillance data have essentially been the object of descriptive rather than analytical investigations. For instance, annual FoodNet reports describe the temporal trend of foodborne diseases by contrasting yearly rates and interpreting the potential effect of demographic covariates through frequency tables (CDC, 2000a,b,c,d). A more analytical approach would harness the multivariate and longitudinal characteristics of the FoodNet data and could provide additional epidemiologic insights.

Common goals of surveillance data are to estimate incidences (rates of illness or infection per population at risk) and to establish the potential relationship between incidence and a set of available explanatory variables (e.g. site, age). In applying higher-order statistical methods to these goals, specific challenges are likely to emerge. First, surveillance data are less specific or precise than those from controlled research studies (Buehler, 1998), and may not be amenable to the assumptions constraining statistical analyses. For example, a quantitative approach would have to respect two constraints specific to surveillance data: (1) the discrete (rather than continuous) count characteristic of the dependent variable; and (2) the likely correlation among measurements repeated annually (i.e. autocorrelation). Additionally, exposure may not be well characterized by the available explanatory variables (place, time, covariate). Finally, the interrelationship among these effects may be complex.

Surveillance data often come in the form of discrete counts of events, in this case, the number of cases of infections that often are foodborne. When frequencies are counted, an adequate assumption is that the counts follow a Poisson distribution (Stokes et al., 2000). Such an approach has previously been applied in epidemiology. For instance, Shahpar and Li (1999) performed an age-period-cohort analysis to characterize the temporal trends and birth cohort patterns of death rates from homicide in the United States. Other recent examples are the analyses of (1) mortality trends for multiple sclerosis in Italy (Tassinari et al., 2001); (2) age-incidence relationships in cervical cancer in

Sweden (Hemminki et al., 2001); and (3) age, sex, geographic and socioeconomic effects in hospital admissions for anaphylaxis in the United Kingdom (Sheikh and Alves, 2001). A useful characteristic of the Poisson log-linear model is that, similar to logistic regression, the exponentiation of the parameter coefficients leads to measures of relative risk, i.e. the incidence rate ratio (IRR).

In capturing the temporal trend of health events, most surveillance systems collect data over consecutive time periods. Similar to time-series data, observations within a specific site (one cluster) are likely autocorrelated. Therefore, drawing valid statistical inferences requires respecting the longitudinal structure of the data in the analysis (Diggle et al., 1994). If ignored, inefficient estimates of the regression coefficients (i.e. imprecise estimates) and incorrect inferences about those coefficients would result. The Generalized Estimating Equations (GEE) method is an extension of the Generalized Linear Model that provides a semi-parametric approach to longitudinal analysis (Liang and Zeger, 1986).

In this study, we seek to extend the usefulness of surveillance data by applying an investigative analytical method to FoodNet data or other active surveillance data sets. We use Poisson regression analysis as an analytical tool to model rates of foodborne illnesses as a function of age, gender, site, and year. Parameters are estimated through the GEE method. Specific outcomes are incidence rate ratios of salmonellosis for the different levels of two covariates (age and gender). Such relative risks can be employed to characterize interindividual variability in susceptibility which is useful within the framework of microbial risk assessment.


Counts of Salmonella infections corresponding to the years 1996, 1997, 1998, and 1999 were extracted from the relative annual FoodNet reports (CDC, 2000a,b,c,d). Specifically, the frequency tables describing the distribution of age and gender stratified by site were consulted.

FoodNet surveillance does not necessarily cover an entire state and not all states were covered for the entire 4-year period. Data for the whole 4-year period were available for five sites: California, Connecticut, Georgia, Minnesota, and Oregon. Data for the years 1998 and 1999 were available for Maryland and New York. The level of data aggregation was not only different for the seven sites, but it also changed over the 4-year period. The 1996 data cover the entire states of Minnesota and Oregon, and selected counties in California, Connecticut, and Georgia (CDC, 2001a). Twelve Georgia counties and one county in Connecticut were added in 1997. In 1998, the surveillance became statewide for Connecticut, and selected counties in Maryland and New York were added. Finally, the remaining counties in Georgia and eight counties in New York were added in 1998. From 1996 to 1999, the total population in catchment areas went from 14.3 to 25.9 million.

State- and year-specific censuses stratified by age and gender were obtained from the US Census Bureau (US Census Bureau, 2002). As counties under surveillance are not specified in the FoodNet reports, state censuses were reduced proportionally to the site-specific populations listed in those reports. It was thus assumed that the age and gender distribution in each site was equal to that at the state level. By combining infection counts and census figures, two data sets - one with counts stratified by eight age categories, the other with counts stratified by gender - were obtained.

After calculation of the annual incidence rates, data were explored qualitatively using graphs in which two of the three explanatory variables (age, gender, and site) were contrasted. Poisson regression was implemented in SAS/STAT version 8.01 (SAS Institute, Cary, NC) with the PROC GENMOD software procedure. The GEE method was used to estimate model parameters. Specification of the REPEATED statement in which the variable identifying clusters was the crossing of age group and site resulted in the implementation of the GEE method (independent covariance structure). Since there are no readily specified procedures to assess goodness-of-fit within the GEE framework, goodness-of-fit of the final model was investigated through analysis of Anscombe residuals (Cameron and Trivedi, 1998). Plots of the Anscombe residuals against the observed number of cases and levels of the explanatory variables were used to assess the goodness-of-fit of the final models. The normality of the residual was checked through the Kolmogorov-Smirnov test and normal probability plots.

Univariate analyses were used to screen explanatory variables to be included in the multivariate models. Specifically, only those variables with a significance level smaller than 0.25 were considered further. This arbitrary threshold was chosen in accordance with standard epidemiological practices (Hosmer and Lemeshow, 1989).

Multivariate analyses started with the model specifying, in addition to the three main effects, all first-order interactions of the retained variables. Through backward selection, the interaction or main term that was the least significant (at a >0.05 level) was subsequently eliminated. The procedure stopped when no term or effect in the model exceeded the 0.05 significance level. Incidence rate ratios were calculated through exponentiation of the parameter estimates. Additional, more technical discussion of the analytical framework used in this study is provided in the Appendix.


Graphical results: Graphical representation of the incidence rates offers insight into potential interactions among covariate (age/gender), place (state), and time (year). Figures 7.3-7.7 systematically contrast two variables by stratifying for the third one. In each figure, the upper and lower series of graphs essentially show the same information, where the levels of the x-axis variable in the upper series become the lines in the lower set of graphs, and vice versa. Horizontal lines imply that the x-axis variable has no influence on the infection rates; vertical distance among the lines reflects the effect of the other variable. Lines that are parallel between two subsequent levels of the x-axis variable suggest a lack of interaction between the two variables. Such parallelism should be evident in both series of graphs. If a similar pattern of lines emerges within each

Fig. 7.3 Age and state effects on Salmonella isolation rates. Each line connects the rates of age category (upper set of graphs) or a specific state (lower set of graphs).

series of graphs, one would infer that infection rates do not change at different levels of the stratifying variable. The effect of age is presented first.

Age: Rates of Salmonella infection are the highest for children less than 1 year of age (Fig. 7.3). Frequencies for the age groups 1-9 and 20-29 appear to be

Fig. 7.4 Age and year effects on Salmonella isolation rates. Each line connects the rates of age category (upper set of graphs) or a specific year (lower set of graphs).

higher than those of the remaining age categories. The lower series of graphs shows that infection rates vary among states only for infants. For each of the four surveillance years, Georgia has the highest rates among infants, followed by Maryland and California. For the other age groups, the infection rates are fairly constant across sites. Figure 7.4 confirms the highest risk is for infants, but also

Fig. 7.5 State and year effects on Salmonella isolation rates. Each line connects the rates of a state (upper set of graphs) or a specific year (lower set of graphs).

suggests that such risk can vary across surveillance years in an unpredictable manner (lower set of graphs, e.g. constant decline for California, decline and surge for Georgia, increase for New York). The dependence of Salmonella infection rates in infants on the variable state and - to a lesser extent - year is again evident in Fig. 7.5. However, this figure clearly shows that, for all other

Fig. 7.6 Gender and state effects on Salmonella isolation rates. Each line connects the rates of gender (upper set of graphs) or a specific state (lower set of graphs).

age groups, the frequencies are largely unaffected by those two variables. In summary, Figs 7.3 to 7.4 show that Salmonella infection rates are higher in infants than in other age groups. However, location (state) and time (year of surveillance) influence the specific risk in infants, which would suggest interaction between the considered variables.

Fig. 7.7 Gender and year effects on Salmonella isolation rates. Each line connects the rates of gender (upper set of graphs) or a year (lower set of graphs).

Gender. Figure 7.6 displays the influence of gender on Salmonella infection rates. The influence of gender on Salmonella rates does not follow a clear pattern. Depending on the state, the frequencies for females can either be higher than, equal to or lower than the frequency in males. The same consideration is true when the combined effects of age and surveillance year are combined (Fig.

Table 7.4 Annual incidence per 100 000 for reported cases of infections with Salmonella

Crude rate Age-state adjusted rate

(model with year effect) (model with all main effects)

Table 7.4 Annual incidence per 100 000 for reported cases of infections with Salmonella

Was this article helpful?

0 0
Arthritis Relief and Prevention

Arthritis Relief and Prevention

This report may be oh so welcome especially if theres no doctor in the house Take Charge of Your Arthritis Now in less than 5-Minutes the time it takes to make an appointment with your healthcare provider Could you use some help understanding arthritis Maybe a little gentle, bedside manner in your battle for joint pain relief would be great Well, even if you are not sure if arthritis is the issue with you or your friend or loved one.

Get My Free Ebook

Post a comment