Predicting inpatient hypoglycaemia in hospitalized patients with diabetes: a retrospective analysis of 9584 admissions with diabetes. Issue 10 (12th July 2017)

• Record Type: Journal Article
• Authors: Stuart, K.; Adderley, N. J.; Marshall, T.; Rayman, G.; Sitch, A.; Manley, S.; Ghosh, S.; Toulis, K. A.; Nirantharakumar, K.
• Is Part Of: Diabetic medicine. Volume 34:Issue 10(2017)
• Page Start: 1385
• Page End: 1391
• Publication Date: 2017-07-12
• Subjects: Diabetes -- Periodicals; 616.462
• DOI: 10.1111/dme.13409
• Languages: English
• ISSNs: 0742-3071

Abstract

Abstract: Aims: To explore whether a quantitative approach to identifying hospitalized patients with diabetes at risk of hypoglycaemia would be feasible through incorporation of routine biochemical, haematological and prescription data. Methods: A retrospective cross‐sectional analysis of all diabetic admissions ( n =9584) from 1 January 2014 to 31 December 2014 was performed. Hypoglycaemia was defined as a blood glucose level of <4 mmol/l. The prediction model was constructed using multivariable logistic regression, populated by clinically important variables and routine laboratory data. Results: Using a prespecified variable selection strategy, it was shown that the occurrence of inpatient hypoglycaemia could be predicted by a combined model taking into account background medication (type of insulin, use of sulfonylureas), ethnicity (black and Asian), age (≥75 years), type of admission (emergency) and laboratory measurements (estimated GFR, C‐reactive protein, sodium and albumin). Receiver‐operating curve analysis showed that the area under the curve was 0.733 (95% CI 0.719 to 0.747). The threshold chosen to maximize both sensitivity and specificity was 0.15. The area under the curve obtained from internal validation did not differ from the primary model [0.731 (95% CI 0.717 to 0.746)]. Conclusions: The inclusion of routine biochemical data, available at the time of admission, can add prognostic value to demographic and medication history. The predictive performance of the constructed model indicates potential clinical utility for the identification of patients at risk of hypoglycaemia during their inpatient stay. What's new?: Hypoglycaemia is associated with worse outcomes in inpatients with diabetes. At present there is no targeted and validated prediction model in the UK for identifying those patients with an increased risk of hypoglycaemia. We performed a retrospective analysis of inpatient admissions with diabetes to develop a prediction model for hypoglycaemia incorporating routinely collected biochemical data. We found that the occurrence of hypoglycaemia could be predicted by a model incorporating background medication, ethnicity, age, admission type and laboratory measurements. Model performance indicates potential clinical utility in identifying patients at risk of hypoglycaemia during their inpatient stay, which could lead to improved patient management and outcomes.