What does this research mean for the field?

Strict adherence to diabetic ketoacidosis management protocols, as evaluated by an artificial intelligence auditing tool, is strongly associated with a significantly shorter hospital length of stay. Novelty: ClaimNovelty.INCREMENTAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

To assess the link between adherence to DKA management protocols and hospital length of stay.

June 7, 2026

2033-P: Artificial Intelligence–Audited Protocol Adherence Is Strongly Associated with Shorter Length of Stay in Diabetic Ketoacidosis

Key Points

To assess the link between adherence to DKA management protocols and hospital length of stay.
Retrospective cohort study of adults admitted with DKA at Barwon Health from November 2024 to October 2025.
Used an AI auditing tool to evaluate adherence to protocol across first seven decision points.
Analyzed association of DKA severity and specialist involvement with length of stay.
Total of 82 admissions evaluated; 92.7% had Type 1 Diabetes Mellitus.
Patients without prompt insulin initiation faced a longer mean LOS (9.4 vs 3.7 days).
Complete adherence resulted in a mean LOS of 1.7 days (n = 25), compared to 5.5 days with deviations (n = 57).

Abstract

Introduction and Objective: Diabetic ketoacidosis (DKA) requires protocolised management, yet real-world adherence is often inconsistent beyond initial stabilisation. We assessed the association between dynamic adherence to local DKA protocol and hospital length of stay (LOS) and examined the impact of DKA severity and specialist involvement (Endocrinology and ICU)). Methods: We conducted a retrospective cohort study of adults admitted with DKA at Barwon Health between November 2024 and October 2025. An Artificial Intelligence (AI)-based auditing tool retrospectively evaluated adherence across the first seven protocol decision points. The primary outcome was hospital length of stay (LOS). Results: A total of 82 admissions were included; 92.7% were patients with Type 1 Diabetes Mellitus. Correct initial insulin dosing occurred in 100% and prompt initiation in 89% of cases. Patients without prompt initiation had a longer mean LOS (9.4 vs 3.7 days). DKA severity was distributed as mild (32.9%), moderate (26.8%) and severe (29.3%) with increasing severity associated with longer LOS (3.1, 3.5 and 5.6 days, respectively). Greater protocol adherence correlated with shorter LOS (r = -0.43). Patients with complete adherence had a mean LOS of 1.7 days (n = 25) compared with 5.5 days in those with any deviation (n = 57). Specialist involvement was common and was associated with longer LOS, consistent with higher clinical complexity. Conclusion: In this retrospective cohort, strict dynamic adherence to DKA protocol steps was strongly associated with a significantly shorter hospital LOS. Delays following initial stabilisation were dominant contributors to prolonged hospitalisations. Specialist involvement did not reduce LOS, likely reflecting confounding by indication. AI-enabled auditing tools may provide a scalable approach to improving protocol reliability and support earlier corrective intervention in DKA management. Disclosure I. Shahid: None. T. Ramasamy: None. A. Anderson: None. K. Cukier: None.

Mark Helpful

Bookmark

Relay