EUROPEAN PROGRAMME FOR INTERVENTION EPIDEMIOLOGY TRAINING

The European Programme for Intervention Epidemiology Training (EPIET) provides practical experience in infectious disease epidemiology. EPIET aims to create a network of professionals throughout Europe trained to use a standard approach in intervention ep


Introduction
Outbreaks provide unique opportunities to study specific hypotheses because of the large number of cases and the number of persons exposed to the etiological agent. In the field of food borne infectious disease one can look at the protective or facilitating effect of specific behaviours or practices. For example we can study the impact of gastric acidity on the risk of food borne diseases during an epidemic. For instance, subjects who have achlorhydria have an increased risk of gastro-enteritis. Thus one can hypothesise that factors that influence the secretion of gastric juice may increase or reduce the risk of illness among those exposed to the etiological agent during an outbreak. It has been shown experimentally that acute alcohol intake increases the secretion of gastric juice; on the other hand H 2 antagonist drugs block acid secretion.
The following case study deals with the effect of ingesting alcohol at the same time as food contaminated with a specific infectious agent. It is based on a study performed during an outbreak investigation in the US in 1988.
In August 1988 an outbreak of hepatitis A was recognised in a tourist city (Panama City) on the Gulf of Mexico in Florida and reported to the Florida Department of Health and Rehabilitative Services (1). The field epidemiologist seconded to the Florida State from the Centers of Disease Control conducted the investigation of this outbreak. The initial descriptive epidemiological information indicated that at least 50 cases of confirmed acute hepatitis A had occurred among Florida residents and tourists to Panama City who travelled there from neighbouring states (Georgia, Alabama, Tennessee) and that the outbreak might be related to raw oysters consumed in Panama city seafood restaurants. To further investigate this suspicion, it was decided to conduct a study among persons who attended Panama City seafood restaurants. To discuss the study protocol the field epidemiologist contacted his scientific supervisor at CDC. In addition to reviewing the selection of controls and the questionnaire, the scientific supervisor proposed to use this outbreak as an opportunity to assess the effect of alcohol on the risk of hepatitis A.

Question 1a
State clearly the study questions you wish to answer. Formulate any hypotheses you would like to test.
The investigators decided to conduct a case control study

Question 1b
Which questions would you include in your case-control study questionnaire?
A case was defined as a person who had acute hepatitis A in July and August 1988 and had attended a seafood event in Panama City 10 to 50 days before onset of illness. Cases were asked to nominate up to 3 controls who had attended the same seafood event. Cases and controls were interviewed by telephone using a standardised questionnaire. In addition to clinical, demographic, food exposures (type of event, foods, seafood, amount, raw, undercooked or cooked...) the following information was collected: type of alcohol (beer, wine, cocktails...) and the amount (number of glasses) consumed during the implicated meal.
Sixty-one cases were identified. Fifty-three of the cases and 64 controls were included in the case-control study (tables 1,2). Raw oysters were implicated as the vehicle of the outbreak, with a dose-response effect. In addition, cases tended to be younger than controls (means 27.5 and 32.2 years, respectively, p = 0.02) and more often male (81% and 39%, respectively, p <0.001). The genome of the hepatitis A virus was recovered by polymerase chain reaction on oysters from the growing waters from which the implicated oysters had been harvested. The next part of the exercise deals with the evaluation of the hypothesis of a protective effect of concomitant alcohol consumption on the risk of hepatitis A (2).

Question 2 How would you proceed to test this hypothesis?
In order to assess the protective effect of alcohol consumption on the risk of hepatitis A, cases and controls need to have been exposed to the hepatitis A virus, that is, to the vehicle of the outbreak. Therefore the analysis was restricted to cases and controls who had consumed raw oysters. The consumption of alcoholic drinks was coded as a dichotomous variable (Alcohol yes/no). The distribution of cases and controls according to the consumption of alcohol is shown in table 2.

Question 3
Comment on these results.

Question 4
What do you do next? Table 2 suggests that the consumption of alcohol may facilitate the occurrence of hepatitis A. At this stage of the analysis one must evaluate the role of chance (p value or confidence interval), bias (selection, information and misclassification bias) and potential confounding factors. The upper and lower limits of the 95% confidence intervals include one (0.6-3.9).

Question 5
Discuss the role of variables that might confound the association between alcohol and hepatitis A.
Confounding is a distortion of the association between an exposure and an outcome. It arises when a third variable is associated with • the outcome independently of the exposure of interest, and • the exposure of interest, but not as a consequence of it.
In this study raw oyster consumption (a dozen or more versus less than a dozen) is associated with hepatitis A (table 3). For this variable to meet the definition of a confounder for the association between alcohol and hepatitis A, one should be able to verify that there is a potential association between alcohol and raw oyster consumption. : smoking, alcohol intake, food habits, blood pressure, physical exercise...). In 1988, questions on raw oyster consumption were added to the core questionnaire of this survey in Florida (sample size = 1481) in order to assess the prevalence of this "risky behaviour" (3). An association between consumption of raw oysters and strong alcohol (drinks with alcohol concentration >10% [wine, whisky, cocktails]) was found.

Question 6
Do you think that the variable "amount of raw oyster consumption" could confound the association between alcohol consumption and hepatitis A?

Question 7: How would you find out whether the amount of oyster consumption is a confounder in this study? Draw the relevant table.
In the subsequent analysis the authors have stratified the effect of alcohol according to the amount of raw oysters eaten (1 -11 versus ≥12). Table 4 gives the distribution of alcohol consumption by cases and controls according to the amount of raw oysters eaten.

Question 8
Describe the results of the stratified analysis.
The authors considered that there was no effect modification since the values of the odds ratios were similar in both strata.
In addition we see that the value of the crude OR (table 2) did not fall between the stratified odds ratios, which suggests a confounding effect of the variable amount of raw oysters eaten.
To take into account the confounding effect one should calculate a measure of association between alcohol consumption and hepatitis A adjusted for the amount of raw oysters consumed. An adjusted OR can be obtained by different ways: the exact method, the direct method and the Mantel-Haenszel method. These three methods yield quite similar results. However, the Mantel-Haenszel estimation is the most simple, can take into account small numbers in any cell (even 0) and can easily be calculated by hand. The formula is given below

Σ [(a i x d i ) / T i ] OR MH = ---------------------Σ [(b i x c i ) / T i ]
where a i , d i , b i , c i are the number of subjects in the different cells and Ti the total of each i strata Table 5 Cases and controls according to alcohol consumption and amount of raw oysters eaten.

Question 9a
Calculate the adjusted odds ratio for alcohol.

Question 9b
What do you conclude?
After taking into account the confounding effect of the amount of raw oysters consumed there is no longer a positive association between alcohol and hepatitis A (adjusted OR = 0.6 compared to a crude OR = 1.6, difference between adjusted and crude OR=60%). Thus, adjusting for the amount of oysters consumed suggests a protective effect of alcohol for the occurrence of hepatitis A.
The 95% confidence interval of the adjusted OR is given below: Alcohol: OR MH = 0.56 95% CI 0. 1-2.2 Age and gender may also have some confounding effect on the relation between alcohol consumption and hepatitis A among oyster consumers. This concern is supported by the case control study results (cases younger and more often male than controls) and the data of the 1988 BRFS survey where alcohol consumers were younger and more often male than nondrinkers (46.5 versus 52.8 years, p = 10 -5 and 50.2% versus 36.6%, p = 10 -6 respectively).

Question 10
How would you take into account the possible confounding effect of these two variables in the analysis?
To take into account the effect of age and sex in the analysis multiple logistical regression must be used. In the original study alcohol consumption was categorised into "weak alcohol" and "strong alcohol". Four independent variables: weak alcohol (yes/no), strong alcohol (yes/no), ≥ 12 oysters (yes/no) age (continuous variable) and male gender (yes/no) were included in the model. The final model presented by the authors is shown in table 6.

Epilogue
This case study confirms that confounding can artificially create an association that does not exist but also illustrated that it can hide a true association. Therefore during the preparation of a study protocol, investigators should carefully review the potential confounders in order to include them in the data collection and to take them into account in the final design (e.g. matching) and/or analysis.
Is the negative association between alcohol and hepatitis A causal?
• The adjusted measure of association was strong.
• Alcohol was consumed during the same meal but the questionnaire did not investigate whether the alcohol was taken before, at the same time or after the oysters. • A bias or further confounders cannot be ruled out as a possible explanation.
• Is a negative association biologically plausible? For bacteria (Salmonella, Shigella and Campylobacter sp) epidemiological and experimental data clearly suggest that alcohol taken at the same time as an inoculum reduces the risk of illness by increasing gastric acidity (5). However, no such data are available for the hepatitis A virus.
In conclusion only some causality criteria are present. No definite conclusions can be drawn about the causal nature of the negative association between hepatitis A and strong alcohol consumed at the same time as the contaminated food.