Streamlining Identification of Ketosis-Prone Diabetes in Youth Using PEPPER: Advancing Atypical Diabetes Discovery

Highlight

• Introduction of PEPPER, a novel Python-based EMR search program designed to identify youth with ketosis-prone diabetes (KPD).
• PEPPER demonstrated 100% accuracy and significantly reduced chart review time compared to manual methods.
• Successfully identified 110 youths with type 2 diabetes and diabetic ketoacidosis (DKA) within six months of diagnosis, 21 of whom met criteria for atypical A-β+ KPD.
• This innovative tool promises improved discovery and classification of atypical diabetes phenotypes, potentially enhancing patient care and research efficiency.

Study Background

Ketosis-prone diabetes (KPD) is an atypical form of diabetes that combines features of both type 1 and type 2 diabetes. Its recognition is clinically significant because patients present with diabetic ketoacidosis (DKA) yet lack the autoimmune beta-cell destruction characteristic of classic type 1 diabetes, often showing preserved beta-cell function (A-β+). Identifying such cases is critical for proper management, prognosis, and understanding disease heterogeneity.

Traditional identification of KPD relies heavily on intensive manual chart review of electronic medical records (EMRs), which is time-consuming and resource-intensive. Automating initial screening steps could expedite diagnosis, facilitate large-scale epidemiologic and pathophysiologic studies, and improve clinical workflows. To address this gap, Ahmed et al. developed the Python-based Expeditious Program for Parsing Electronic Records (PEPPER) to automate EMR searches for youth with type 2 diabetes (T2D) presenting with DKA, a prerequisite for identifying A-β+ KPD.

Study Design

This study retrospectively analyzed the electronic medical records of 1,660 youth diagnosed with T2D. PEPPER was deployed to automatically screen for evidence of DKA occurring within six months after diabetes diagnosis. The algorithmic output was then compared with manual chart review to assess performance metrics including sensitivity, specificity, accuracy, and time efficiency.

Following initial identification, further manual review was performed to confirm full classification of A-β+ KPD according to criteria established by the Rare and Atypical Diabetes Network. These criteria include presence of DKA, absence of islet autoantibodies, and preserved beta-cell function. The key endpoints of the study were the accuracy of PEPPER in identifying DKA cases, the reduction in chart review time, and the number of patients ultimately classified as A-β+ KPD.

Key Findings

PEPPER identified 110 youth with T2D and documented evidence of DKA within six months of diagnosis from the 1,660-patient cohort. Of these, 21 met the strict A-β+ KPD criteria, demonstrating the program’s utility in isolating atypical diabetes cases.

In analyzing workflow efficiency, PEPPER dramatically reduced chart review time: the mean time per chart was 13.4 ± 3.9 seconds compared with 26.6 ± 9.4 seconds for manual review (p < 0.001). Importantly, PEPPER’s identification of DKA cases was 100% accurate against the manual review gold standard, underscoring its reliability.

These findings indicate that PEPPER effectively automates a critical step in the classification of atypical diabetes phenotypes. The tool not only mitigates the substantial manual labor traditionally required but does so without compromising the accuracy essential for clinical and research applications.

Expert Commentary

The development of PEPPER addresses a significant unmet need in diabetes research and care: rapid, accurate identification of patients with atypical presentations. The ability to parse large EMR datasets with high fidelity paves the way for broader epidemiologic characterization and improved phenotype-driven management of diabetes.

While promising, some limitations warrant consideration. PEPPER’s dependence on accurate clinical documentation means that errors or omissions in EMR coding could impact detection sensitivity. Additionally, while the study focused on youth with T2D, external validation in diverse populations and healthcare systems would strengthen generalizability.

Future integration of PEPPER with clinical decision support systems could further enhance diagnostic pathways by flagging potential KPD cases in real time. Moreover, coupling the algorithm with biomarker data may refine classification and guide individualized treatment.

Conclusion

The novel application of PEPPER represents a significant advance in leveraging computational tools to optimize EMR data mining for atypical diabetes phenotypes. By reducing manual effort by approximately half while maintaining perfect accuracy in DKA identification, PEPPER facilitates efficient, scalable discovery of ketosis-prone diabetes cases.

The study exemplifies how targeted informatic solutions can revolutionize clinical research methodologies, enabling deeper insights into diabetes heterogeneity. These advancements hold the potential to improve patient-centered care through more precise diagnosis and treatment of atypical diabetes forms.

Continued development, validation, and integration of such algorithms are vital next steps to realize their full translational impact.

Funding and ClinicalTrials.gov

The study was conducted by the RADIANT Study Group. Specific funding details were not provided in the abstract. No clinical trial registration was indicated.

References

• Ahmed M, Kubota-Mishra E, Siller AF, et al. A Novel Electronic Medical Record Search Method to Identify Patients With Ketosis-Prone Diabetes: Implications for Discovery of Atypical Diabetes. Diabetes Care. 2026 Jun 26. PMID: 42360321.
• Umpierrez GE, et al. Ketosis-Prone Diabetes: A Clinically Recognized Subtype of Diabetes Mellitus. Diabetes Care. 2006;29(4):876-882.
• Fitzpatrick SL, et al. Classification of Diabetes Beyond Type 1 and Type 2. Diabetes Spectrum. 2017;30(3):144-150.