The following systematic searches were performed to identify sources of published data for the rates of type 1 diabetes in childhood:

• Medline was accessed using OVID restricted to human studies published since 1980 and using (exp registries OR exp incidence OR exp prevalence) AND exp diabetes mellitus, insulin-dependent AND exp <country name> with the /ep [Epidemiology] sub-heading. If a country was not indexed in Medline then it was included in the search as a text word.
  
• PubMed using the Boolean search terms (incidence OR prevalence) AND diabetes AND <country name>.
  
• Published abstracts from recent international meetings including those in the Institute for Scientific Information (ISI) Proceedings were also searched.

The titles and abstracts of all articles were reviewed and those likely to provide incidence or prevalence rates were obtained. The reference lists of articles were also scanned to check for further relevant publications. No restrictions were placed on the language of published articles.

1. Criteria
The following criteria were used, although not necessarily in the order shown, to select the most suitable studies in countries with a number of available studies:

• More recent studies, preferably covering periods into the 1990s.
• Studies with widest coverage within the country. 
• Studies providing rates for the target age range of 0-14 years.
• Studies providing sex-specific rates for the 0-4, 5-9 and 10-14 year age groups.

If necessary the numerators and denominators of rates from a number of registers within a country were combined to obtain pooled rates.

2. From incidence to prevalence
The majority of studies found by the literature search provided incidence rates rather than prevalence rates. An estimate of the number of cases in each country was obtained by multiplying the population projections in each of six age/sex subgroups (males or females aged 0-4, 5-9 or 10-14 years) by the corresponding estimated prevalence rate.

Prevalence rates in each age group were obtained by averaging cumulative incidence rates for the five individual years in the age group. For example, the prevalence in the 5-9 age group was obtained as an average of: 

Prevalence (age 5) = 5* (0-4 year incidence rate) + 0.5*(5-9 year incidence rate)
Prevalence (age 6) = 5* (0-4 year incidence rate) + 1.5*(5-9 year incidence rate)
Prevalence (age 7) = 5* (0-4 year incidence rate) + 2.5*(5-9 year incidence rate)
Prevalence (age 8) = 5* (0-4 year incidence rate) + 3.5*(5-9 year incidence rate)
Prevalence (age 9) = 5* (0-4 year incidence rate) + 4.5*(5-9 year incidence rate)

In a few countries that reported age-specific rates pooled for boys and girls, the rates were taken to apply to both boys and girls.

The incidence rate is not uniform in the 0-14 year age group but rather it tends to be lower in young ages and increases to a peak usually in the 10-14 year age group. For countries in which age-specific rates were not available, a single multiplier to convert incidence rates to prevalence rates was derived as the median multiplier for the 65 countries for which age- and sex-specific incidence rates were available. Equal-sized populations in each age-sex subgroup were assumed in this calculation. The resulting prevalence to incidence ratio of 6.2 was therefore employed to convert incidence rates to prevalence rates in all countries in which age-specific incidence rates were unavailable. Using an assumption that the mean age at onset of diabetes occurring before the 15th birthday was 8.5 years, a similar conversion factor of 6.5 was derived in the second edition of the Diabetes Atlas, as the mean duration of diabetes in the 0-14 year age range.

This method of estimating prevalence from incidence assumes that the effects of mortality are minimal. In developed countries, which tend to have high quality incidence data, mortality rates amongst diabetic children are low and any adjustment for mortality is unlikely to have much impact. In less developed countries, which often have poorly estimated incidence rates based on small numbers, the application of an adjustment for mortality was not felt to be justified. In many African countries estimates of numbers of cases were derived directly from reported prevalence rates (usually extrapolated from other countries), rather than indirectly through incidence rates and in this situation no adjustment for mortality was required.