Implications of study design

In deciding on which study to use to estimate the prevalence of food allergy statistical, practical and financial constraints must be considered. The ideal sample would include all the individuals in the population but this is clearly impossible and our studies must be conducted on a subset of the total population. It is this down-sizing that leads to important methodological problems due to the selection procedures. The different types of study described below represent different selection procedures and give rise to different problems. It is impossible to obtain a subset that completely represents the entire population from which it is derived.

Case series

Many reports about food allergy have been based on personal series derived from general clinics or tertiary clinics. Such series are unable to provide any information about incidence and prevalence in a population as there is no known denominator associated with the data. Nevertheless, such series provide useful qualitative information about food allergies in different populations. Thus the fact that allergy to royal jelly is the most common cause of food allergy diagnosed in tertiary clinics in Hong Kong, but is never seen in European tertiary clinics is highly relevant (Leung et al. 1997). Furthermore, case series are useful in identifying novel problems. The fact that sesame seed allergy was rarely seen in European allergy clinics several decades ago but today represents an important component of the clinical case load suggests that this problem is increasing (Kanny et al. 1996).

Case series, however, are fraught with methodological problems, most notably bias. A bias is any error in the design or conduct of a study that results in a result other than the true one, due to systematic (though unintentional) skewing of the data. Bias may be introduced either in the selection of subjects or in the collection of information (Sackett 1979). A study looking at the association of soy allergy with cows' milk allergy (Zeiger et al. 1999) provides a good example of selection bias in a case series. The prevalence of soy allergy in one clinic was more than ten times greater than in the other three centres. This particular centre was a tertiary paediatric allergy clinic that saw a highly selected population, more likely to include children with multiple food allergies.

There is also a potential for information bias in case series because data are often collected retrospectively either directly from subjects or from their clinical notes. Patients often do not have a good recall of events, leading to a form of information bias called recall bias. A good example of recall bias is a birth cohort study in which mothers were asked about the duration of breast feeding at 11 and 47 months of age (Huttly et al. 1990). At 47 months there was only 70% agreement with data obtained from the same mothers at 11 months. Although case series do not provide robust epidemiological data, they provide a window through which current clinical experience may be viewed. They often form the initial basis of many hypotheses that can subsequently be tested in more definitive studies where cases and control subjects are compared.

Case-control studies

Case-control studies are a natural extension of case series having the added advantage that they provide control subjects with which the cases can be compared. Similar to case series, they provide no data on incidence and prevalence as here again the denominator remains unknown. They are, however, useful in the early testing of hypotheses that relate to associations and risk factors for food allergy. However, they make the assumption that all the differences between subjects and controls represent risk factors for the disease being investigated. In practice measured differences may be brought about by important biases in selection and information. Furthermore, confounding factors may occur where an apparent association between an exposure and an outcome is partially or entirely due to another associated exposure.

An example of a case-control study, is one looking at the aetiology of peanut allergy. It was concluded that children sensitised to peanut had a higher level of peanut exposure in utero due to higher maternal consumption (Frank et al. 1999). This result, which has not been confirmed in cohort studies, probably occurred because of recall bias as the mothers of infants with peanut allergy, are likely to have spent more time considering their consumption of peanuts during pregnancy prior to filling in the study questionnaire. Despite these potential problems, case-control studies represent a rapid way of providing important evidence about a hypothesis that can be later tested using a more definitive approach.

Cross-sectional studies

Cross-sectional studies are population-based studies within a defined geographical region. This approach considerably reduces the potential for selection bias. Furthermore they allow the point prevalence of the condition studied to be estimated. However, such studies by their nature afford a single glimpse of the population at one specific time point. Therefore no data can be derived about changes in incidence and natural history of the condition over time. Such studies allow us to identify risk factors for food allergies as the population contains both cases and control subjects. Cross-sectional studies involve large numbers of subjects and require considerable resources. They may also be affected by bias and confounding factors.

Mailed questionnaires are often used in cross-sectional studies but response rates can be very low, even after reminders are sent. In the High Wycombe population study of food intolerance (Young et aJ. 1994), replies were received from only 52.7% of subjects. It can be argued that responders are likely to differ substantially from the non-responders, introducing an important selection bias. Such problems may be reduced with door-to-door interviews but other problems emerge, for example, subjects out at work may escape interview. Telephone-based interviews are becoming increasingly common (Munoz-Furlong et aJ. 1989) but these exclude subjects without a telephone and those who choose to be ex-directory, which will bias the sample. Evidence from cross-sectional studies also provides useful data on the prevalence of disease in a population and highlights potential causal factors. Evidence collected using this approach must eventually be substantiated by the results of cohort studies to decide if it has been affected by selection or information bias.

Cohort studies

Cohort studies are less affected by the problems inherent in other approaches for the single reason that subjects are included and exposures recorded before the outcome has occurred. This eliminates a major source of bias. Cohort studies, unlike cross-sectional studies, are not subject to the cohort effect as all the participants are born over a specified narrow time period. Furthermore, one is able to estimate the incidence and remission rates as well as prevalence and thus obtain a more complete picture of the natural history of a disease. Such studies provide the best quantitative and qualitative description of food allergy within a population but make the highest demands on time and resources. Nevertheless, cohort studies are not completely immune to methodological problems. Selection bias may operate slowly over a longer period of time. At the start of the study, there is likely to be a loss of participants due to failure to enrol while others may become lost to follow-up during the study. The loss of potential subjects at enrolment and during follow-up is likely to introduce important selection bias.

Cohort studies are important in identifying risk factors for food allergy. This risk is usually quantified using odds ratios or relative risks. Confounding can still occur where a third factor may account for a perceived association between a particular exposure and an allergic outcome. Where such confounding variables are suspected and identified, their effects can be eliminated by the application of statistical methods such as logistic regression analysis. An example is the association seen between prolonged breast feeding and food allergy. This is not a real association as it is confounded by eczema; infants with eczema are deliberately breast fed for longer periods and eczema is a known risk factor for food allergy.

Although cohort studies have their limitations, they generally provide the best form of evidence concerning the prevalence and natural history of a disease within a population. They are well suited to the study of the natural history of food allergy. They also provide pointers to potential causal factors which can be subsequently tested within the context of a randomised interventional study where allocation of the exposure is random and not subject to known or unknown confounders.

Interventional studies

In many ways interventional studies are very similar to cohort studies except that the investigator is able to allocate the exposure artificially, preferably at random. The randomised, double-blinded, placebo-controlled study (RDBPC study), where exposure allocation is random and known to neither subject nor investigator, is the gold standard for generating evidence. Unfortunately, although the data generated by a good study are invaluable, these studies are difficult to set up for financial, practical and ethical reasons. It is only ethical to randomise subjects between two or more interventions if there is no evidence to suggest that one is more beneficial than the other. It may be difficult to recruit a sufficiently large and representative study population to have sufficient power to be able to arrive at definitive conclusions. Even once a study population has been recruited problems may occur with loss to follow-up because of a perceived failure of the active or control intervention.

Most interventional studies in food allergy have focused on maternal dietary exclusion during pregnancy and/or lactation as well as modification of the infant diet. In general they have been unsuccessful. A problem that pervades all such studies is that elimination of a food from the diet may not be achieved to a sufficient degree or at an early enough time point to ensure a successful intervention. These 'no difference found' studies become very difficult to interpret, leading to 'newer and better' interventional studies.

Keep Your Weight In Check During The Holidays

Keep Your Weight In Check During The Holidays

A time for giving and receiving, getting closer with the ones we love and marking the end of another year and all the eating also. We eat because the food is yummy and plentiful but we don't usually count calories at this time of year. This book will help you do just this.

Get My Free Ebook


Post a comment