fixed-sequence, crossover clinical trial in food allergy

Question

I want to conduct a fixed-sequence, crossover clinical trial to compare the effect of stopping allergic food at one time period with that of another time period when they eat the allergic food. When they enter the study, they were eating the allergic food. So the sequence is eat-stop-eat. There is only one group in the study. No randomization. Kindly advice me the statistical methods. Is it necessary to study the third group which follows the advice to stop allergic food only partially.

Answer 1

In a crossover clinical trial there is no control group, as the subjects act as their own controls, as you are presumably aware.

A fairly standard method of analysing such trials is with a repeated measures model, such as repeated measures ANOVA or a linear mixed effects model.

Is it necessary to study the third group which follows the advice to stop allergic food only partially.

This question makes me wonder why you want to have a third group but then not use it in the analysis? The protocol of the trial should dictate the statistical analysis plan, so if you have a third group, you should use it. It would still require a model for repeated measures, so it doesn't add too much complexity compared to a two-group design. Obviously it does add the complexity of recruiting and monitoring an additional group.

This question should be addressed based on the research question(s) and the resources available. A good reason to include the 3rd group is that you would then be able to have different sequences that participants are randomised to, which would mitigate some of the problems with fixed-sequence designs such as "Order effects" where, for example, the act of ceasing to consume the allergic items may influence the reaction when the participants resume eating it (these effects are intrinsic to the sequence and cannot be eliminated through randomisation); carryover effects (even with a washout period, there may still be lingering effects from the initial phase that could impact reactions in the next stages; and Time-related bias (where changes in environmental conditions, individuals' overall health, or behavioural modifications over time could all have an impact on the results).

As with all crossover studies, be wary of carryover effects, and implement a suitable washout period.