This analysis has not only confirmed the long-term cyclical variation of childhood type 1 diabetes reported previously 
, but has also established effects over shorter periods. Having allowed for annual differences, the numbers of cases in each quarter of a year showed greater variation than would be expected under the Poisson distribution. In contrast, there was no evidence of extra-Poisson variation within quarters of a year. These results indicate that temporal clustering of cases occurs over periods of a few months, in addition to cyclical variation over years. This pattern of occurrence is consistent with the involvement of exogenous agents, such as an infection, that may exhibit epidemicity. We also checked the methodology using data for influenza in children, based on a similar total number of cases as for the diabetes dataset and again with confirmed diagnoses. The P-W method detected temporal clustering of influenza at levels of months, quarters of a year and flu seasons. Consequently, the methodology used here should have good power to detect clinically meaningful clustering.
Several studies, e.g. 
, have reported seasonality in dates of diagnosis of childhood type 1 diabetes, often with peaks in October to January and troughs in June to August for centres in the northern hemisphere. However, not all studies have reported the same pattern; indeed, a recent analysis from Denmark 
suggested that the seasonal pattern might change over time. The present study adds to the evidence for such an irregular temporal pattern (i.e. temporal clustering). The statistical method used here is well-placed to identify changing patterns of the form shown in because-unlike the usual analyses of seasonality-it does not assume that peaks or troughs occur at the same time each year.
Genetic factors, notably the HLA system, influence susceptibility to type 1 diabetes; however, the increases in disease incidence seen in many countries in recent years 
cannot be explained by genetic factors alone and highlight the role of the environment in disease evolution. Factors such as the increase in population obesity and associated insulin resistance have been viewed by some as a plausible explanation for the increase in diabetes incidence in young people, but cannot easily be reconciled with the present finding of temporal clustering 
. The finding of such an irregular pattern in incidence is consistent with the involvement of an infectious agent which itself displays an irregular pattern in the environment. Evidence for a role of infections in the aetiology of type 1 diabetes comes from epidemiological studies of, for example, birth order 
, interbirth interval 
, rural locality 
, population mixing 
, day care attendance 
and recorded neonatal illnesses 
, whereas studies of recorded infections in early life have been inconsistent 
. Experimental evidence also supports the role of an infectious agent or agents in the development of type 1 diabetes 
and the involvement of infections in northeast England is supported by the previous identification of space-time clustering in our locality 
. The evidence from the present study for temporal clustering over periods of a few months suggests that an infectious agent or agent may act as a final trigger in the development of the disease amongst susceptible individuals.
A number of candidate viruses have been implicated in the aetiology of type 1 diabetes including enteroviruses, rotavirus, mumps, cytomegalovirus, rubella and Ljungan virus 
. For example, a recent systematic review found a strong association between enterovirus infection and both type 1 diabetes-related autoimmunity and clinical type 1 diabetes 
. The type of clustering found in the present study suggests that the underlying pattern in the risk of such an agent being passed from a reservoir host may exhibit a natural “epidemicity”. The most common type of reservoir for zoonotic infections, especially in temperate regions, is wild rodent populations 
. Patterns of variation in the risk of transmission from them will reflect a combination of variation in infection prevalence and in host abundance. Wild rodents in the north of England are known to exhibit multi-annual cycles of abundance, though these typically have a period of 3–4 rather than 6 years 
, but it is also known that superimposed upon host abundance patterns, infection prevalence and the timing of peaks of prevalence may vary markedly from year to year, as a result of interplay between prevailing weather, host demography and infection dynamics 
. However, making a direct and detailed link between the intra- and multi-annual patterns observed here, and intra- and multi-annual patterns in the infection dynamics of wild rodents that are a potential natural reservoir, will require further research focused specifically on elaborating those dynamics.
The precise mechanism whereby an organism such as an enterovirus might affect the evolution of type 1 diabetes is unclear 
, although islet cell damage could be a key part of this process (reviewed in 
). The level of exposure could affect the immune system in a manner that is dependent on factors such as age and genotype. The hygiene hypothesis 
has been proposed as a potential explanation for the association between greater disease incidence and improved sanitation. This trend has been observed both between and within countries 
. High levels of exposure to infectious agents in the population as a whole may refine immune responses, with the interaction between micro-organism and individual potentially decreasing as well as increasing the likelihood of disease development 
. In particular, discussion of the hygiene hypothesis in relation to autoimmune diseases has highlighted the key role of timing of exposure, in that certain viruses might provoke autoimmunity when given late but be protective when given very early 
. Laboratory studies of type 1 diabetes provide support for this 
, whereas epidemiological studies of type 1 diabetes and recorded infections in the first year of life are inconsistent 
. Nevertheless, the lack of evidence for temporal clustering between quarters within years at ages under 5 years in our study might be explained by a protective effect of exposure to infectious agents at very young ages. Furthermore, the time between exposure to virus infections and disease onset is likely to vary between individuals and may be shorter amongst susceptible individuals (i.e. those exposed at an inappropriate stage of maturation). Thus, our findings of temporal clustering at older ages are consistent with virus infection or infections acting on the immune system of susceptible individuals and leading to clinically observable disease in some of these individuals shortly thereafter.
Suboptimal ascertainment with incomplete data collection and the potential for patients from one locality to be managed in centres outside the area studied are important considerations in a study such as this. We collected and cross-checked data from sources other than local paediatric clinic databases and also liaised with neighbouring regions to make sure that these factors would not be significant confounders in our analyses and that ascertainment was high. We remain unable to reconcile highly significant temporal clustering with aberrations in the way data have been collected or with cases of diabetes in young people being classified incorrectly.
In conclusion, the present study adds substantively to the growing body of literature that supports the involvement of infectious agents in the aetiology of type 1 diabetes in children. Specifically it suggests that the precipitating agent or agents involved might be an infection that occurs in “mini-epidemics.”