New Clinical Model May Predict Abnormal Glycemic Control During Hospital Stay

Close up of nurse pinching patients finger to measure sugar level with a glucometer at the hospital
Researchers developed a clinical tool to identify patients at risk for persistent abnormal glucose levels during hospitalization.

A new practical clinical prediction model may help to identify patients with diabetes at high risk for persistent hypoglycemia and/or hyperglycemia during hospitalization, according to study results published in the Canadian Journal of Diabetes.

There are multiple potential factors that may significantly alter glucose levels in hospitalized patients with diabetes. Up to half of inpatients will develop hyperglycemia during hospital stays, while hypoglycemia affects up to 20% of inpatients.

The goal of the current study was to identify risk factors and develop a clinical tool to identify patients at risk for persistent abnormal glucose levels during hospitalization, defined as ≥2 days with capillary glucose levels <4 or >15 mmol/L during a hospital stay.

The study cohort included 594 patients with type 2 diabetes (mean age, 72 years; 57% men). Persistent adverse glucose levels were documented in 153 patients (26%). Most patients with hypoglycemia also experienced hyperglycemia (61%), while a minority of those with hyperglycemia also had hypoglycemia (17%).

The researchers identified several risk factors for persistent adverse glucose levels, including blood glucose levels of <4 or >15 mmol/L in the first 24 hours after admission (odds ratio [OR], 3.65; 95% CI, 2.09-6.37; P <.001); glycosylated hemoglobin (HbA1c) between 7.1% and 8.0% (OR, 2.15; 95% CI, 1.13-4.10; P =.020) or >8.0% (OR, 5.08; 95% CI, 2.63-9.87; P <.001) compared with HbA1c ≤7.0%; preadmission treatment with glucose-lowering medications, including sulfonylurea (OR, 3.50; 95% CI, 1.57-7.77; P =.002) or insulin (OR, 4.22; 95% CI, 1.84-9.70; P =.001), compared with diet-controlled diabetes; glucocorticoid treatment (OR, 2.27; 95% CI, 1.17-4.40; P =.015); modified Charlson comorbidity index; and length of hospital stay.

Two logistic regression models were used to predict persistent adverse glucose levels. An early-identification model included 4 clinical variables that were available within 24 hours of admission (admission dysglycemia, HbA1c, preadmission treatment regimen, and glucocorticoid treatment) and showed a receiver operating characteristics area under the curve of 0.806 (95% CI, 0.751-0.861). A hospital-based model that included all of the aforementioned variables in addition to modified Charlson comorbidity index and length of stay showed a receiver operating characteristics area under the curve of 0.872 (95% CI, 0.828-0.916).

While the hospital-stay model was a better predictor of persistent adverse glycemia than the early-identification model, not all 6 variables included in the model are readily available shortly after admission. As such, the researchers suggested that the early-intervention model was the more practical tool for predicting persistent hypoglycemia and/or hyperglycemia during a hospital stay, with a sensitivity of 84%, specificity of 66%, and positive predictive value of 53%.

The study had several limitations, including missing data on several clinical variables and inpatient diabetes management, inclusion of only patients with type 2 diabetes, and lack of external validation.

“Clinical prediction tools could become essential for promoting early identification and targeted management of individuals at risk for adverse glycemia in hospital, and ultimately assist to improve the care and outcomes of people with diabetes,” concluded the researchers.

Follow @EndoAdvisor


Kyi M, Gorelik A, Reid J, et al. A clinical prediction tool identifies patients with diabetes at risk for persistent adverse glycemia in hospital [published online June 10, 2020]. Can J Diabetes. doi:10.1016/j.jcjd.2020.06.006