Background. There is still no consensus on the appropriate definition of an 'episode' of diarrhoea, even though it has been shown that the choice of definition has a major impact on reported incidence rates. Previous work has focused on the observed distribution of illness episodes in time but has not attempted to determine whether the patterns observed depart from those expected by chance. Methods, A simple theoretical model of the distribution of illness episodes is developed, based on the concept of a 'trigger event'. The model incorporates elements relating to the duration of symptoms, inter-individual variation in incidence rates and seasonality. Appropriate parameters for the model are derived from two empirical datasets. Results. It is shown that short intervals between one aetiologically distinct period of diarrhoea and the next will frequently occur by chance, especially in circumstances where high incidence rates and within-child clustering of illness prevail. The duration of symptoms will have no effect on the length of intervals between periods of illness, and seasonality is unlikely to have a major impact. Over 10% of all non-initial trigger events might be expected to occur during the course of a pre-existing period of diarrhoea, and would not therefore be identified in a study based on reported symptoms. Conclusions. The findings of previous studies, suggesting that 2 or 3 days without symptoms will generally mark a new episode of diarrhoea, are endorsed. Modelling the expected distribution of illness in time may help to highlight structural or analytical problems with empirical datasets.
