Clinical prediction tool to identify adults with type 2 diabetes at risk for persistent adverse glycemia in hospital

Objectives
Given the high incidence of hyperglycemia and hypoglycemia in hospital and the lack of prediction tools for this problem, we developed a clinical tool to assist early identification of individuals at risk for persistent adverse glycemia (AG) in hospital.

Methods
We analyzed a cohort of 594 consecutive adult inpatients with type 2 diabetes. We identified clinical factors available early in the admission course that were associated with persistent AG (defined as ≥2 days with capillary glucose <4 or >15 mmol/L during admission). A prediction model for persistent AG was constructed using logistic regression and internal validation was performed using a split-sample approach.

Results
Persistent AG occurred in 153 (26%) of inpatients, and was associated with admission dysglycemia (odds ratio [OR], 3.65), glycated hemoglobin ≥8.1% (OR, 5.08), glucose-lowering treatment regimen containing sulfonylurea (OR, 3.50) or insulin (OR, 4.22), glucocorticoid medication treatment (OR, 2.27), Charlson Comorbidity Index score and the number of observed days. An early-identification prediction tool, based on clinical factors reliably available at admission (admission dysglycemia, glycated hemoglobin, glucose-lowering regimen and glucocorticoid treatment), could accurately predict persistent AG (receiver-operating characteristic area under curve = 0.806), and, at the optimal cutoff, the sensitivity, specificity and positive predictive value were 84%, 66% and 53%, respectively.

Conclusions
A clinical prediction tool based on clinical risk factors available at admission to hospital identified patients at increased risk for persistent AG and could assist early targeted management by inpatient diabetes teams.