Fifty-four of the fifty-eight centers (93%) participated in this study (Figure 1). Each center prospectively registers data on all children with T1D at the time of diagnosis. This process follows agreed upon procedures established in 1997 by the Italian Insulin-Dependent Diabetes Registry (RIDI) (14), that are based on a standardized case report format. This study included all new cases of T1D over 0.5 and under 15 years of age presenting between January 1st, 2014 and December 31st, 2016. The diagnosis was confirmed with the presence of at least one of the antibodies against beta-cell antigens, islet cell antibodies (ICA), insulin antibodies (IA), glutamate dehydroxilase antibodies (GADA), islet antigen 2 antibodies (IA2A), and zinc-transporter protein 8 (ZnT8A). All other forms of diabetes were excluded. As the participating centers prospectively record all new cases of T1D that occur every year, the completeness of data collected on diabetes cases in the study period was higher than 95%. Moreover, detailed information on parents' socioeconomic status was mandatory for this study, while such information was not available for the analysis of the previous period (2004–2013).

Data collected consisted on date of birth, date of diabetes diagnosis, gender, region of residence at diagnosis, education, and occupation levels of single or both parents, ethnic minority status, first-degree relative(s) with T1D, venous pH, bicarbonate levels. Data were anonymously sent to the coordinating center between January 1st and May 31st, 2017 and a unique database was obtained for the analysis.

The International Society for Pediatric and Adolescent Diabetes (ISPAD) criteria for DKA was used as a standard reference (15) (overall: pH < 7.30 or serum bicarbonate <15 mmol/L; severe: pH < 7.1, serum bicarbonate <5 mmol/L). Age at diagnosis was grouped in 0.5–4, 5–9, and 10–14 years; residence at diagnosis was classified as North, Center, and South (including Sicily and Sardinia); ethnic minority status was defined as at least one parent born outside the country with a positive migration history, or using a list of ethnic categories (African, Asian, mixed). Parents' education level was classified as low (without a high school diploma: <13 schooling years), and high (high school diploma attainment or university studies: at least 13 schooling years). Parents' occupation was collected following the classification of the Italian National Institute of Statistics, which is totally cross-linkable with the International Standard Classification of Occupations. They are grouped in two levels: low (unoccupied, unskilled and semi-skilled workers, manual workers and craftsmen), and high (legislators, senior officials and managers, professionals, technicians and associate professionals, sales workers, small business and farm owners, administrators and higher executives).

All procedures were conducted in accordance with the ethical standards of each participating center. Parents' written informed consent was collected by each participating center. No ethics committee approval was required, as this study was based on data routinely recorded in clinical practice and anonymously transmitted to the coordinating center.

This analysis included two categories of DKA, overall (pH < 7.3) and severe (pH < 7.1). They were estimated as percent frequency and 95% confidence interval (95% CI) based on the binomial distribution and evaluated by gender, age group, geographical area of residence at diagnosis, parents' level of education and occupation, family history of type 1 diabetes, and ethnic minority status. The Chi-square test was used to compare the two groups. Diabetic ketoacidosis frequency between the two time periods were compared overall and by age groups using z-test for two proportions. The Kolmogorov-Smirnov test was used to compare the distributions of pH median values for a single year of age between the two time periods.

Two multiple logistic regression models were used to evaluate factors associated with DKA frequency. The dependent variable was DKA, considered as present vs. absent in the first model and as severe vs. absent in the second one. Gender, age class, geographic area of residence (Central Italy was the reference category), father and mother education and occupation level, and ethnic minority status were used as independent variables. Due to missing data for parents' level of education and occupation, a third category was added that included these missing values. The Likelihood Ratio test was used to detect the variables that were included in each model. The model's goodness of fit was evaluated by the Hosmer-Lemeshow test. All estimates were evaluated using the 95% CI.

All the statistical analyses were performed using the R.3.1.2 statistical package; a probability lower than 0.05 was used to assess statistical significance.

DKA frequency was significantly lower in 2014–2016 than 2004–2013 (40.3, 95% CI: 39.3–41.4, p = 0.002). Figure 2 shows a significant reduction of DKA frequency in children with 0.5–4 years diagnosed during 2014–2016 compared to 2004–2013 period (p = 0.007). Figure 3 shows that median venous pH values at diabetes presentation were significantly higher in 2014–2016 compared to the 2004–2013 period (p = 0.028).

Significantly higher DKA frequencies (both overall and severe) were found in children of 0.5–4 years of age, with both a low level of mother's education and parents' occupation (Table 1). The frequency of severe DKA was significantly higher in children when the father had a low level of education. No significant difference was found according to gender, residence at diagnosis, or minority status.

Table 2 shows factors associated with overall and severe DKA. The youngest children had a higher probability of overall and severe DKA at T1D diagnosis. The mother's high level of education, mother's or father's high level of occupation and the presence of at least one first-degree member with T1D decreased the probability of having DKA at diagnosis. Children living in Southern Italy had a higher probability of diagnosis with severe DKA than children living in Central Italy.

Few studies have analyzed the association between parental educational level and DKA at the diagnosis of diabetes, suggesting that education affects the chance of developing ketoacidosis (15, 16). In addition, the level of family income (17) and socio-economic status (17) can be a risk factor for ketoacidosis. Moreover, results from completed studies are not consistent. A low level of maternal education was associated with a higher risk of DKA in 401 children under 15 years of age in Lithuania (18) while in a series of 474 children under 17 years of age in Poland an association between delayed diagnosis of type 1 diabetes and level of maternal education (19) was not found. Likewise, in a study conducted in Korea, no association was found between the level of parental education and the risk of DKA in a group of 361 children, however, the children of parents with low education showed an increased risk of severe DKA (20).

In this study, the high level of maternal education has shown to be highly protective for both overall and severe DKA. It is possible that mothers, who provide more direct care of children than fathers in Italy, may notice diabetes symptoms earlier if they have a high level of education. In this study the father's educational level was not associated with the DKA at diagnosis of diabetes.

However, we observed that the father's high occupation level had a protective role toward overall DKA. The mother's occupation plays an important role in the prevention of both overall and severe DKA. It is plausible that a high level of occupation allows parents to have more exchanges with other people, and live in a favorable social network to obtain useful information that can lead to an early detection of the characteristic symptoms of diabetes.

A 2006–2016 study (7), demonstrated a significantly higher percentage of children from ethnic minority groups who presented with DKA at the diagnosis of T1D than peers from non-ethnic minority groups in Italy. This same result was not found in the case series analyzed during 2014–2016. This is in contrast with results from other studies in which children belonging to minority ethnic groups showed higher frequency of DKA (21). It is likely that the inclusion of parents' level of education and occupation in the multiple logistic regression analysis reduced the independent effect of ethnic minority status. Indeed, the access to the Italian healthcare system, that warrants services, medications, and devices free of charge, allows all residents to receive assistance by avoiding differences based on ethnic minority.

Other studies have reported a greater probability for DKA at diagnosis in younger children (7, 12, 22, 23). Our study confirmed that younger children had a higher probability of DKA than school age children and adolescents. However, when comparing these results with data reported over the past 10 years, we have observed a significant reduction in the percentage of DKA in the age group 0.5–4 years. Although this result is encouraging, it highlights that the probability is still too high and the presence of diabetes symptoms in younger children requires more attention from parents and pediatricians.

A few studies have highlighted the effectiveness of awareness campaigns to prevent DKA at diagnosis of diabetes during last 20 years (8, 10, 24–26). To date, due to methodological limitations of those studies, it is not possible to make a definitive conclusion on their effectiveness (8). To support the findings from the literature, it is difficult to associate the reduction of DKA frequency we have observed in this study and the implementation of the national awareness campaign promoted by ISPED. This is due to factors such as the lack of structured educational interventions, i.e., meetings with pediatricians were not simultaneously and systematically organized, variation in implementation by pediatric centers, i.e., some pediatric centers of the country started their own campaign earlier than others, and most importantly that the awareness campaign is an ongoing activity. To determine the effectiveness of this intervention in Italy, a large number of cases over a longer period of time need to be evaluated. In addition, several other non-structured and less conventional initiatives intended to prevent DKA at diagnosis were carried out during 2014 and 2015, such as messages by association of parents of children with diabetes launched through social media or sportive events, i.e., periodical marathons across Italy (http://www.weloveinsulina.it).

These initiatives could have contributed to the observed reduction of DKA and demonstrate the importance of non-conventional communication systems to reinforce efforts aimed to spread education on DKA prevention to both the general population and health care providers. However, it should be also noted that socioeconomic status and relation to social media is controversial; lower socioeconomic status might be related to an increased use of unproductive social media use and abuse. Beyond the reports of cumulative effects of various actions, the results from this study also demonstrate the need to explore education and awareness-raising campaigns tailored to the needs of families, in particular for those with a low level of parent education and the ability to access resources and for children that are a younger age. Ideally, the frequency of DKA should be reduced to a minimum level comparable to that found in children with islet-positive autoantibodies detected by the Bavarian screening program (27), who developed overt disease. The proportion of children reporting pH value <7.30 at type 1 diabetes diagnosis was <5%.

The Network of ISPED for DKA Study and Prevention involved 54 out of 58 Italian centers for pediatric diabetes. Strengths of this study included nationwide collaboration, which allowed the collection of a large number of cases of T1D at diagnosis under the age of 15 years. This collaboration has lasted from more than 30 years and is based on the Network of ISPED for DKA Study and Prevention of a national scientific society, involving pediatric diabetologists skilled on research methodology. In addition, data recording and collection was based on standard criteria, and used consistent definitions of DKA over time.

Limitations of the study included missing data for some crucial information. We included only cases of type 1 diabetes with information on DKA status; nevertheless, cases without information on pH and/or bicarbonate were <5% and they might not have negatively affected DKA frequency estimation. A high percentage of missing data was found in socioeconomic variables, and there was a consistent distribution of missing values for parents. To overcome this limitation and minimize the loss of information in the statistical analysis, we included a category for missing values both in parents' education and occupation.