We developed two strategies to identify patients with atypical diabetes mellitus using EMR data. Using a clinical questionnaire and EMR queries we found that 1%–6.6% of pediatric patients with diabetes mellitus have atypical forms of diabetes. T1D and T2D compromise the vast majority of diabetes in the USA,1 and the current literature suggests 1%–6.5% of children with diabetes mellitus are estimated to have atypical forms of diabetes mellitus, consistent with our observation.11 12

In a prior RADIANT study, Parikh et al recently developed a data mining framework using phenotypical characteristics of two diabetes mellitus cohorts: one an adult patient cohort in a predominantly Hispanic population in South Texas, and the other a multiethnic pediatric cohort at TCH, Houston. The frequency of atypical diabetes mellitus in the adult diabetes cohort was 11.5%, whereas it was 5.3% in the pediatric diabetes mellitus cohort, very similar to the present study. In addition, the Parikh et al study demonstrated the existence of two distinct clusters of atypical diabetes mellitus phenotypes in the pediatric cohort.13

Of the atypical cases identified in our study, half were children with T2D diagnosed before 10 years of age. All had negative islet autoantibodies. This group exemplifies an atypical form of diabetes mellitus, since it is rare for children to be diagnosed with T2D before 10 years of age, especially if they are also pre-pubertal.14–16 The incidence of T2D increases with age17 and several studies have shown an association between puberty timing and the development of T2D.18 Information about the stage of pubertal development was not available in all patients, therefore we cannot rule out the presence of precocious puberty that could have contributed to the earlier presentation of diabetes mellitus.

The other half of the suspected atypical cases were identified by their physicians due to a phenotype of MODY; of these, all had negative MODY gene panel testing (with one VUS). Among the atypical cases identified using strategy 2, the majority of the ‘unknown diabetes mellitus type’ group also comprised suspected MODY cases.

There are no well-defined clinical practice guidelines once a patient with atypical forms of diabetes mellitus has been identified. Most of the patients with clinically suspected MODY had received a clinical diagnosis of either T1D or T2D as documented in the EMR. Subtle clinical phenotypes such as overlapping T1D versus T2D are less likely to be studied further. Leveraging the EMR to identify patients with atypical diabetes mellitus can have direct impact both on clinical and research efforts.

From the clinical care perspective, system-wide identification and monitoring of these cases may improve clinical outcomes. From the research perspective, better identification and characterization of atypical forms of diabetes mellitus may improve enrollment in clinical trials such as RADIANT and accelerate discovery of new pathophysiologic pathways to diabetes mellitus and the possibility of targeted treatment. The questionnaire used in strategy 1 aimed to exclude the more described forms of diabetes mellitus as categorized by the American Diabetes Association: T1D, T2D, gestational diabetes mellitus and specific diabetes types due to other known causes (eg, monogenic diabetes, diseases of the exocrine pancreas and drug or chemical-induced diabetes).1 The use of questionnaires as screening tools has proven helpful in clinical practice to identify rare forms of diabetes mellitus. For example, Shields et al developed and validated a calculator based on clinical criteria to determine an individual’s probability of having MODY as a tool for rational genetic testing with improved sensitivity and specificity for identifying MODY patients.19

In strategy 1, our questionnaire was effective in identifying atypical diabetes mellitus cases. Out of the 100 patients, only 11 required a more detailed manual review. In the clinical practice, identification of atypical diabetes mellitus may be easier and also non-biased using such a questionnaire. A questionnaire with 19 questions might seem time consuming for a busy clinical practice; however, the first 12 questions were close-ended questions with only the remaining seven questions requiring more detailed clinical information and some of those questions were not applicable to all patients. If validated, this questionnaire could potentially become a screening tool to be applied by clinicians as a guide in patients that have mixed phenotype of diabetes mellitus and may help to determine if they need further testing or referral for investigation of atypical forms of diabetes mellitus. The questionnaire featured a straightforward design primarily comprising binary questions. It could potentially serve as a self-referral tool accessible to patients or their families online. Alternatively, it could be made available to patients and families in the waiting room through an EMR-connected tablet. The latter strategy has been deemed a valuable tool in primary and specialty care clinics.20 21

In strategy 2, obtaining queries of patients with diabetes mellitus categorized as ‘unknown diabetes mellitus type’ was possible mainly due to the Diabetes Flowsheet included in our EMR system that is filled out at the clinic visit only by the pediatric endocrinologist or the diabetes nurse practitioner taking care of the patient. This approach streamlined the identification of atypical forms of diabetes mellitus from a vast amount of data that would otherwise require a significant manual chart review effort to complete. Institutions that have not yet integrated diabetes mellitus flowsheets into their EMR systems may consider incorporating these tools.

Our study reported a prevalence of 5.6% of autoantibody-negative T1D cases, however the prevalence reported in children has been as high as 19%.22 The race/ethnicity proportions of patients identified by the two strategies were similar to those reported in other studies, which assessed populations composed mainly of under-represented racial and ethnic groups (primarily Hisp and AA).23 24 However, in the subset of autoantibody-negative T1D cases identified using strategy 2, NHW (50%) and AA (22%) children comprised the majority of cases. Wang et al reported no significant racial/ethnic difference in prevalence of autoantibody-negative diabetes in younger children with newly diagnosed diabetes mellitus. However, in children older than 14 years, autoantibody-negative diabetes was much more prevalent in under-represented racial/ethnic groups compared with NHW (57.1% in Hisp, 70% in AA and 25.9% in NHW), due in part to a higher rate of T2D in Hisp and AA.22 Individuals from under-represented racial/ethnic groups with T1D and T2D have increased morbidity and poor long-term control of diabetes mellitus overall,25–27 yet very little is known on the prognosis of the atypical forms of diabetes mellitus that appear to be highly prevalent in these groups. Our approach allowed us to identify a significant number of children from under-represented racial/ethnic groups who are less likely to be referred for further evaluation due to multiple barriers to participation in biomedical research.10 Hence, it is important for physicians to be familiar with the concept and forms of atypical diabetes mellitus in persons belonging to underrepresented racial/ethnic groups.

The strength of our study is the use of an unbiased approach by leveraging EMR to identify pediatric patients with atypical diabetes mellitus. The limitations include that determination of atypical nature of a particular case was entirely limited to medical documentation in EMR. Also, EMR systems are designed differently, and the diabetes flowsheet used in this study might not be adaptable across organizations, limiting the use of our current approach. However, EMR usually has a significant amount of data for each patient enabling it as a valuable tool for screening purposes.

In conclusion, EMR-based approach revealed that approximately 1%–6.6% of pediatric diabetes mellitus cases are atypical. More studies are required to validate our clinical questionnaire and to streamline the process of EMR-based identification of atypical cases.