After duplicates were removed, the search identified 1,441 papers, of which 1,333 were excluded as clearly irrelevant and a further 63 after full-text assessment. There was complete agreement between researchers throughout this process. The most common reasons for exclusion were failure to use a definition of DKA that included either pH or bicarbonate, duplication of data or no data for children with new-onset diabetes (Fig. ). Two studies were excluded based on quality alone. The first because we were unable to adequately interpret the numerical data after contacting the author, and the second because of a large amount of missing data. A further 12 papers were identified through citation searching. One paper compared incidence of type 1 diabetes at presentation between south-east Sweden and Lithuania and so is reported as two studies [17
]. Another [11
] reported the mean frequency of DKA across 24 centres in Europe. After contacting the authors we were able to obtain data for 11 of these centres individually. Results from three of these centres had been reported in other included papers in greater detail, leaving eight studies for inclusion from that paper. This review is therefore based on 65 studies.
The 65 included studies provided data on over 29,000 children (range 10 to 3,947) from 31 countries across five continents (two from Asia, 21 from Europe, three from the Middle East, two from North America, two from Oceania, and one from South America) (Table ). Notably, many large countries, such as South American countries, India, China and Japan, are not represented and there are no studies for the whole continent of Africa. All included studies were cohort studies. Most recruited children retrospectively from hospital or clinic records, with study lengths ranging from 1 to 17 years, and the periods of study covering a 30 year period from 1978 to 2008. Most included children up to age 15 (n
29), 16 (n
5) or 18 years (n
13), with eight studies including young people between 18 and 21 years and one only those under 6 years. There was a wide range of definitions of DKA, all of which included either pH ≤7.2 to <7.36 or bicarbonate <15 to ≤21 mmol/l.
Characteristics of included studies
Frequency of DKA across countries
Table shows the frequency of DKA at diagnosis in each study, together with the background incidence of type 1 diabetes, latitude, GDP and expenditure on healthcare as a percentage of the country’s GDP. The frequency of DKA varied sixfold, from 12.8% in Sweden to 80% in the United Arab Emirates and from 16% to 67% when only those studies defining DKA by pH
7.3 were included. Ranking of countries according to frequency of DKA (Fig. ) demonstrated that the highest frequencies were seen in the United Arab Emirates (80%), Romania (67%), Taiwan (65%) and Saudi Arabia (59%) and the lowest in Sweden (14%), Canada (18.6%), Finland (22%) and Hungary (23%).
Fig. 2 Plot of the frequency (±95% CI) of DKA at diagnosis of type 1 diabetes per study, grouped in countries in descending order of the average frequency of DKA per country. aStudies defining DKA as pH<7.3
Multivariable linear regression
Three studies had incomplete data (Table ) and were, therefore, excluded from the regression modelling. Initial linear modelling showed that GDP and latitude were strongly collinear and therefore it was only possible to include one. Latitude was chosen as it explained more of the variation.
The final model (Table ) shows that latitude and background incidence of type 1 diabetes were significantly associated with frequency of DKA, with frequency decreasing progressively with distance from the equator and in areas with a higher background incidence of type 1 diabetes. Although not reaching statistical significance, there was also an inverse association between expenditure on healthcare as a percentage of GDP and frequency of DKA. No significant associations were found for study size, period of study, design or method of case identification. These associations were the same when latitude was replaced with GDP, showing that GDP is also inversely associated with frequency of DKA.
After adjusting for study size, period of study, design, method of case identification, expenditure on healthcare as a percentage of GDP, and latitude, the frequency of DKA decreased by approximately 10% as the annual background incidence increased from 10 to 30 cases per 100,000, with greater changes at lower background rates. R2 for the model was 0.56, indicating that a large amount of the variation was explained by these factors.