Glucose Management Clinical Decision Support to Improve Outcomes in Academic and Community Hospitals

Introduction: Clinical decision support (CDS), use of personalized and intelligently filtered patients’ health records data, to improve care is increasingly promising. Hospital dysglycemia is associated to poor outcomes. A real-time CDS detecting glycemic gaps and providing care recommendations showed glycemic improvements in adults with diabetes and shorter length of stay in an academic hospital. We propose assessing the impact of this CDS tool across a health system.
Design and methods: A prospective intervention among hospitalized adults intermittently deployed CDS in the electronic health record to recognize gaps in care in type 1 (T1DM), 2 (T2DM) and stress hyperglycemia: hyperglycemia, hypoglycemia and insulin sliding scale monotherapy (ISSM). Notification of gaps and management recommendations to clinicians (intervention) vs regular practice (control) in an academic and three community hospitals were compared. Intensity Ratio (IR) of total and reoccurring gaps in care were determined using Zero-inflated Poisson regression. Binomial generalized estimating equations regression model estimated odds ratios (OR) of diagnoses suspected as hospital complications.
Results: There were 13,325 patients (20,190 admissions) whose average age was 67 years, and 56% were male. During intervention vs control, total ISSM use showed a reduced trend in T1DM [mean 0.04 v 0.04, IR 0.81 (0.64, 0.98), p 0.045] and significant reduction in T2DM [mean 0.14 vs 0.15, IR 0.94 (0.90, 0.98), p 0.006]. It showed significantly lower reoccurrence in T2DM [mean 0.55 vs 0.58, IR 0.94 (0.88, 0.99), p 0.027]. Reoccurrence of ISSM use after notifications of hyperglycemic events was significantly reduced in T1DM [mean 0.01 vs. 0.01, IR 0.55 (0.32, 0.78) p 0.005], and T2DM [mean 0.62 vs 0.64, IR 0.94 (0.90, 0.98), p 0.008]. Hypoglycemia notifications after hyperglycemia notifications was greater during intervention [mean 0.21 vs 0.19, OR 1.12 (1.04, 1.20), p 0.002]. However, no difference had been detected in counts of hypoglycemia notifications between study periods. ISSM use after notifications of hypoglycemia was marginally increased in T2DM [mean 0.03 vs 0.03, IR 1.22 (0.98, 1.46), p 0.048]. Diagnosis of hospital acquired pneumonia (HAP) was significantly reduced during intervention [8.5% vs 10.1%, OR 0.84 (0.76, 0.92), p <.001].
Discussion: Lower ISSM use in T1DM and T2DM, and lower reoccurrence in T2DM signals insulin use improvement during the intervention. This aligns with standard of care guidelines. Scheduled insulin therapy in T1DM is critically important, denoting this association may also signify a clinically meaningful improvement in safer practice patterns. Higher ISSM in T2DM after notification of hypoglycemia may reflect a change of management to a less optimal insulin choice as providers may be concerned for subsequent hypoglycemia. While we found a greater count of hypoglycemia notifications after hyperglycemia notifications, there was no difference in the total count of hypoglycemic events between groups. HAP diagnosis, lowered during the intervention, is more common among hospitalized patients with hyperglycemia, and merits further assessment.
Conclusions: Across health systems, academic and community hospitals should strive for improving practice and care outcomes. Diabetes imposes added burden to hospitalized patients, health providers and system at large. CDS can assist enabling best practice guidelines, and achieving meaningful and scalable clinical improvement.