Case data were abstracted from sample of 450 unique adults who had a ß-hydroxybutyrate value ordered and resulted from the ED drawn randomly from a total of 2413 orders. During the same time frame, 1648 patients had a primary diagnosis of diabetes mellitus, hyperglycemia or DKA in the EDs at the four sites. The number of ß-hydroxybutyrate orders is higher because patients with elevated values often have repeated testing. Table 1 presents key demographic information of the study population, divided into the four phenotypes described in the Methods section for all patients and then for patients with DKA. Salient points include the fact that 323/450, or 71% of the population identified as black, 47% identified as female sex and 326/450 or 72% of ß-hydroxybutyrate tests were ordered on patients with no history of DM, or only a history of type 2 diabetes (ie, non-type 1), suggesting that clinicians generally have a high index of suspicion for T2KPDM. Only 20 patients identified as Hispanic or Latino. Among non-type 1 patients, several potentially surprising findings emerged. The first was the high frequency of DKA, 122/326 or 37%. Second, in contrast to the suggestion of prior literature, the frequency of DKA was similar between black and white patients: 67/249 (26.9%) versus 18/59 (30.5%), respectively (unadjusted odds=1.2, 95% CI 0.6 to 2.2). Third, among non-type 1 patients, the frequency of DKA was not different between men and women: 49/186 (26.6%) versus 38/141 (26.9%), respectively (unadjusted odds=1.0, 95% CI 0.6 to 1.7). Fourth, again in contrast to prior literature, obesity, defined as a BMI >35 kg/m2, was not significantly associated with DKA (unadjusted odds=0.64, 95% CI 0.33 to 1.22).

View inline|View popup

Patient demographics and associations with DKA (BOHB>20 mg/dL)

Table 2 presents means and SD for continuous data, including age, BMI and laboratory values. These data had no missing values. Patients with type 1 were significantly younger (mean 35 years) compared with mean ages of all other phenotypes (p<0.001 by Tukey’s comparison for groups a–c vs d). Patients with type 2 DM and no prior ketosis had a higher BMI than all other phenotypes (p<0.001, Tukey’s). Regarding laboratory values, several differences were observed for key indices often associated with DKA severity: ß-hydroxybutyrate, glucose, bicarbonate and anion gap (AG). These laboratory values are also displayed in figure 1A–D as dot plots with median values. First, both the ß-hydroxybutyrate and glucose concentrations were significantly higher in type 1 patients compared with patients with no DM history or type 2 with no prior ketosis (p<0.001 Dunn’s post hoc), but not when compared with type 2 patients with prior history of ketosis (p=0.01). Bicarbonate concentrations were not different between type 1 and type 2 with prior ketosis (p=0.64, Dunn’s), but both phenotypes had lower bicarbonates compared with patients with no history of DM and type 2 without prior ketosis. The only significant difference for AG was that in patients with type 1, it was higher compared with those with type 2 DM without prior ketosis. We also recorded an ancillary data point, which was the frequency of laboratory-reported moderate to severe hemolysis on the initial sample, which occurred in 36% of specimens. The median (HbA1C) was higher in patients with DKA (median 11.3%) versus those without ketosis (median 9.5%, Mann-Whitney U p<0.001) and the median value was also significantly higher in patients with a history of non-adherence. Among patients with charted non-adherence, 43% had HbA1c concentrations>10%.

View inline|View popup

Mean values for age, body mass index and laboratory measurements

Table 3 presents the frequency of behaviors and social determinants, stratified by type of diabetes for all patients and then among patients with DKA, again, restricted to those without type 1 DM. The only statistically different single proportion between the non-type 1 groups was that non-adherence was highest in patients with prior T2KPDM (59% vs next highest 31%, 95% CI for difference of 28%, 14% to 40%). The data also suggest that charted non-adherence was more frequent in non-white patients, given that 101 (22%) of the 450 patients were white, but only 37/450 (8.2%) of patients were white and non-adherent—three times higher than the overall non-adherence rate of 27% in black patients. For association with DKA, three variables had unadjusted ORs with 95% CIs that did not cross unity: (1) history of non-adherence (odds=1.74, 95% CI 1.08 to 2.81), (2) arrival by emergency medical services (odds=0.54, 95% CI 0.33 to 0.86) and (3) private or Medicare insurance (odds=6.80, 95% CI 4.00 to 11.57). Other insurances included Medicaid only (29.5%), Medicaid and Medicare (11%) and none (3%). The frequency of charted evidence of medication non-adherence among patients with type 2 DM with private insurance who had a ß-hydroxybutyrate order was 50%.

View inline|View popup

Behavioral and social determinates of health

The final random effects model included 10 independent variables, and online supplemental table 1 details the model-derived ORs and 95% CIs. The only three variables that were significantly associated with DKA were a prior history of ketone-forming type 2 DM: having Medicare, private or commercial health insurance, and location of the patients’ residence. The final model included 308 (95%) of the 326 non-type 1 DM patients with complete data to analyze, resulting in an intraclass correlation coefficient=0.24.

If the variable ‘prior history of ketone-forming type 2 DM’ is removed from the multivariable equation, then the variable of non-adherence becomes significant (adjusted odds=1.96, 95% CI 1.05 to 2.79). Similarly, if the model is rerun, excluding patients with no prior history of DM (for whom non-adherence is less relevant), the variable of non-adherence is significant (odds 2.35, 95% CI 1.15 to 4.80; p=0.02). From this sequence, we created a putative causal diagram in figure 2 showing the central role of non-adherence as a mediator of the outcome of DKA.

Online supplemental table 2 presents the medications that were prescribed to the patients prior to presentation. online supplemental table 2 is not corrected for suspected lack of adherence. Notable differences included the 31% increase in frequency of insulin prescribed to patients with prior history of type 2 DM with ketosis (91%) compared with patients with type 2 DM but no history of ketosis (60%), 95% CI for difference of 31%=20.3% to 40.5%. On the contrary, patients with type 2 DM with ketosis were 16% less likely to be prescribed metformin (25%) compared with patients with type 2 DM and no prior ketosis (41%), 95% CI for difference of 16%=3.3% to 27.8%.

Online supplemental table 3 shows patient outcomes, including admission to intensive care, LOS and survival, none of which was significantly different between phenotypes. We also included two commonly coded diagnoses of ketoacidosis (alcoholic and diabetic) as outcomes, as these designations, although are system made (ie, result of physician documentation and billing coder decisions), may be important to future recognition and categorization. A potentially important finding in terms of social determinants was the observation that within the phenotype of patients with no prior diagnosis of diabetes, there was an almost equal frequency of diagnosis of alcoholic ketoacidosis (19%) and DKA (22%).

To assess for rigor of data abstraction, agreement statistics were computed from the duplicate abstraction of 50 charts by the two abstractors. Absolute agreement for first ß-hydroxybutyrate and glucose concentrations was 48/50 (96%) and agreement for age was 43/50 (86%). Weighted Kappa values for phenotyping were as follows: (a) no prior DM, kappa=0.92 (95% CI 0.77 to 1.07), (b) type 2 DM, no prior ketosis, kappa=0.83 (0.68 to 0.99), (c) type 2, prior ketosis, kappa=0.94 (0.82 to 1.05), (d) type 1, 0.78 (0.58 to 0.99) e. non-adherence=0.79 (0.61 to 0.99).