Discussion
The present study shows that DKA remains an uncommon acute metabolic complication of diabetes, with an overall incidence of <4 confirmed/probable cases per 1000 patient-years. Half of the first episodes of DKA during follow-up in our study were in participants with type 1 diabetes and about half of the remainder were in those with type 2 diabetes. Participants with LADA and secondary diabetes made up the balance of cases, but there were no episodes of DKA among the small number with monogenic diabetes. Multivariable analysis confirmed that chronically poor glycemic control and limited pancreatic beta cell reserve were strong independent predictors of first incident DKA, but secondary diabetes was also in the model. These three variables were also predictors of the frequency of DKA episodes, together with insulin treatment at study entry.
The reported incidence of DKA complicating type 1 diabetes from a recent systematic review is up to 560 cases per 10 000 person-years19 and our overall incidence and 95% CIs of 178.6 (85.7 to 328.5)/10 000 person-years were within this range. Similarly, published rates for type 2 diabetes are up to 20/10 000 person-years,20 which is also consistent with our estimate (13.3 (5.7 to 26.1)/10 000 person-years). Sodium glucose cotransporter 2 inhibitors (SGLT2Is) are a class of blood glucose-lowering agent introduced into Australia after the follow-up period used in the present study. They are associated with an increased incidence of DKA in type 2 diabetes in most observational studies if not in clinical trials,21 with approaching half of the cases presenting with a serum glucose lower than the 13.7 mmol/L conventionally required for diagnosis (so-called ‘euglycemic’ DKA).22 However, inconsistencies between studies including some population-based data showing no significant increase in risk23 24 suggest that factors influencing prescribing, such as recent warnings of high-risk situations for SGLTI-associated DKA and measures for its prevention such as in the perioperative period,25 may vary geographically.
We found that insulin treatment at study entry was independently predictive of the frequency of DKA. Since all participants were insulin-treated at the time of DKA hospitalization, this implies that longer duration insulin treatment may have a role in DKA risk in type 2 diabetes. We26 and others27 have found a relationship between duration of insulin therapy and risk of severe hypoglycemia in type 2 diabetes. Taken together, these observations highlight the potential for long duration diabetes treatment to lead to acute complications at the extremes of glycemia.
The overall incidence of first DKA complicating LADA in the present study (121.5/10 000 person-years) was between our estimates for type 1 and 2 diabetes (178.6 and 13.1/10 000 person-years, respectively), paralleling the intermediate pancreatic beta cell function that has been reported previously.28 Although we had relatively few participants with secondary diabetes (which was due mostly to chronic pancreatitis), the incidence of DKA in this group was the highest at 446.5/10 000 person-years and secondary diabetes was an independent predictor of both incident first confirmed/probable DKA and the frequency of DKA. There are no previously published comparative data, but the potential for DKA to complicate this form of diabetes has been recognized for some time,11 including the risk of death29 as in the case of one of our participants. As in the present study, patients with monogenic diabetes, with the exception of maternally inherited diabetes and deafness,30 have a very low risk of DKA because there is not the usually progressive loss of pancreatic beta cell function characterizing other forms of diabetes.10
In a recent US review of electronic hospital data, 11 of 32 (34%) of admissions were not validated by endocrinologist review of biochemical and other data as DKA despite DKA being coded as the diagnosis.6 This is similar to the incorrect coding rates of 39.0% for initial episodes and 41.7% for recurrences in the present study, although we also had seven participants (17.1%) with incomplete data for the initial episode. In most of these latter cases there was no documentation of plasma or urinary ketone concentrations, but other details required for diagnostic confirmation such as the plasma lactate (especially in metformin-treated participants with renal impairment31) may not have been available. These considerations highlight the necessity of medical record validation to correctly ascertain cases of DKA for healthcare quality assessment and planning purposes. In a similar vein, coding of type of diabetes in cases of DKA in the US study was incorrect in one in six cases.6 There are data suggesting that individuals with type 2 diabetes and DKA require higher daily total insulin doses, larger replacement fluid volumes and greater potassium supplementation to resolve their DKA than are used in type 1 diabetes.5 Accurate ascertainment of the type of diabetes could, therefore, influence the nature of initial management, although the effect of more intensive fluid resuscitation and insulin administration on prognosis in DKA complicating type 2 diabetes remains unknown.
The present study had limitations. The number of DKA episodes was restricted by the FDS2 cohort size and follow-up period but it was similar to those in comparable published studies with accurate ascertainment of cases6 and we employed more stringent p values than usual to allow for the low numbers of DKA hospitalizations. Apart from mortality and changes in diabetes management, we did not record details of clinical outcomes including after discharge. As acknowledged, the time period of the study meant that DKA associated with SGLT2Is was not captured, with our definition of DKA excluding euglycemic cases. Lower serum glucose thresholds for DKA diagnosis have been recommended, such as >11 mmol/L in the case of International Society for Paediatric and Adolescent Diabetes guidelines,32 but their use would not have changed case ascertainment in the present series. The strengths of the study are the community-based nature of the FDS2 cohort, and the use of detailed FDS2 and WADLS data, together with individual hospital case records, to determine diabetes type and ascertain DKA events.
The present study provides relatively robust estimates of the incidence of validated first DKA events in a representative cohort of Australians with well-characterized diabetes and its sequelae. The incidence rates of DKA complicating type 1 and 2 diabetes (which represented around three-quarters of all confirmed/probable cases) were generally within the ranges reported by other authors, while that for LADA was between those for type 1 and 2. These data are consistent with the strong independent inverse association between serum C-peptide (which can be measured as a marker of pancreatic beta cell function) and first DKA attendance. Participants with secondary diabetes, who were few in number but contributed disproportionately to the overall incidence, may be a group in which the risk of DKA could be underestimated.