Considerations for conducting and interpreting long-term follow-up of intervention studies: avoiding spoiled milk

Abstract

Multiple early life exposures have been shown to influence the development of allergic disease in infancy and childhood. The timing of exposure to food allergens has been of particular interest. Two landmark trials, Learning Early About Peanut and the Enquiring About Tolerance studies, found early introduction of peanuts reduced peanut allergy in children, 2 a fundamental difference from previously accepted beliefs. The impact from cow’s milk is less clear, with the most recent trial suggesting avoidance of early cow’s milk supplementation decreased the risk of childhood asthma, while others have found an increased risk or no effect on childhood allergy. 5 In the current issue of Thorax, Hand and colleagues have further added to the ongoing cow’s milk debate by studying the effect of early avoidance on adult allergy. Investigating if early life exposures influence adult disease requires extended followup of participants. Trials are usually designed to investigate immediate or shortterm outcomes, but it is possible to proficiently plan and execute longterm followup of an intervention during infancy, such as in the Promotion of Breastfeeding Intervention Trial trial, where participants were followed up until adolescence. However, due to the nature of extended longitudinal followup, such trials are prone to considerable limitations that must be carefully considered in the design, analysis and interpretation. In the current study, Hand et al have shown great tenacity to continue the followup of individuals in an underrepresented, deprived area of South Wales, UK, recruited as part of the Merthyr Allergen Prevention Study between 1982 and 1984. In this prospective, randomised study of individuals selected to be at high risk of allergic disease, investigators originally set out to understand the effect of withholding cow’s milk protein during the first 3 months of life and the development of wheeze or asthma in infancy and childhood. In 1993, the study investigators reported that the intervention did not affect the incidence of atopy at 7 years old, although breast feeding was associated with a lower likelihood of wheeze, particularly in nonatopic children. In this design, the intervention of withholding cow’s milk also included a soy milk preparation, rather than a hydrolysed cow milk preparation for instance, which introduces confounding that limits interpretation of the endpoint. While such a design would be unlikely if planned today due to longterm consequences of phytooestrogen ingestion, importantly, the authors concluded in 1993 that ‘either cow’s milk protein is not particularly allergenic or soya milk has an equal effect’. In the present Hand et al study, the children were followed up through to young adulthood, providing detailed postal questionnaires on exposures and symptoms at the age of 15 and again at 23. Questionnaires are useful noncontact methods of data collection but rely heavily on recall rather than empirical measurement, and selfreporting is prone to bias, cognitive or otherwise. Additionally, questionnaires are notoriously poor with regard to response rates, particularly if there are competing interests on the participant’s time. At age 23, researchers received questionnaires from 61% of the cohort, while a more restricted 24% attended clinical assessment. This can increase the chances of attrition bias, where those with complete data may not reflect the original distribution of participant characteristics. Online questionnaires may be able to overcome some of these limitations in contemporary studies but can similarly create a selection bias based on access, in addition to any selection bias based on inclusion criteria. Alternatively, linkage to routinely collected data sources can increase a trial’s capacity to efficiently extend followup, but this approach has other challenges and problems. When investigators have collected all the necessary data, ideally blinded to any