Original Research

Predictors of Suboptimal Glycemic Control for Hospitalized Patients with Diabetes: Targets for Clinical Action

Journal of Clinical Outcomes Management. 2015 April;22(4)

References

1. Umpierrez GE, Hellman R, Korytkowski MT, et al; Endocrine Society. Management of hyperglycemia in hospitalized patients in non-critical care setting: an endocrine society clinical practice guideline. J Clin Endocrinol Metab 2012;97:16–38.

2. Moghissi ES, Korytkowski MT, DiNardo M, et al; American Association of Clinical Endocrinologists; American Diabetes Association. American Association of Clinical Endocrinologists and American Diabetes Association consensus statement on inpatient glycemic control. Diabetes Care2009;32:1119–31.

3. Coursin DB, Connery LE, Ketzler JT. Perioperative diabetic and hyperglycemic management issues. Crit Care Med 2004;32(4 Suppl):S116–25.

4. Pandit MK, Burke J, Gustafson AB, et al. Drug-induced disorders of glucose tolerance. Ann Intern Med 1993;118:529–39.

5. Donihi AC, Raval D, Saul M, et al. Prevalence and predictors of corticosteroid-related hyperglycemia in hospitalized patients. Endocr Pract 2006;12:358–62.

6. McHugh MD, Shang J, Sloane DM, Aiken LH. Risk factors for hospital-acquired ‘poor glycemic control’: a case-control study. Int J Qual Health Care 2011;23:44–51.

7. McCowen KC, Malhotra A, Bistrian BR. Stress-induced hyperglycemia. Crit Care Clin 2001;17:107–24.

8. Rady MY, Johnson DJ, Patel BM, et al. Influence of individual characteristics on outcome of glycemic control in intensive care unit patients with or without diabetes mellitus. Mayo Clin Proc 2005;80:1558–67.

9. Selvin E, Steffes MW, Zhu H, et al. Glycated hemoglobin, diabetes, and cardiovascular risk in nondiabetic adults. N Engl J Med 2010;362:800–11.

10. Kim KS, Kim SK, Lee YK, et al. Diagnostic value of glycated haemoglobin HbA(1c) for the early detection of diabetes in high-risk subjects. Diabet Med 2008;25:997–1000.

11. Quan H, Li B, Couris CM, et al. Updating and validating the Charlson comorbidity index and score for risk adjustment in hospital discharge abstracts using data from 6 countries. Am J Epidemiol 2011;173:676–82.

12. Kirk JK, Oldham EC. Hyperglycemia management using insulin in the acute care setting: therapies and strategies for care in the non-critically ill patient. Ann Pharmacother 2010;44:1222–30.

13. Draznin B, Gilden J, Golden SH, et al. Pathways to quality inpatient management of hyperglycemia and diabetes: a call to action. Diabetes Care 2013;36:1807–14.

14. Boord JB, Greevy RA, Braithwaite SS, et al. Evaluation of hospital glycemic control at US academic medical centers. J Hosp Med 2009;4:35–44.

15. Neubauer KM, Schaupp L, Plank J, et al. Failure to control hyperglycemia in noncritically ill diabetes patients despite standard glycemic management in a hospital setting. J Diabetes Sci Technol 2013;7:402–9.

16. Schnipper JL, Barsky EE, Shaykevich S, et al. Inpatient management of diabetes and hyperglycemia among general medicine patients at a large teaching hospital. J Hosp Med 2006;1:145–50.

17. Rohlfing CL, Wiedmeyer HM, Little RR, et al. Defining the relationship between plasma glucose and HbA(1c): analysis of glucose profiles and HbA(1c) in the Diabetes Control and Complications Trial. Diabetes Care 2002;25:275–8.

18. Nathan DM, Kuenen J, Borg R, et al; A1c-Derived Average Glucose Study Group. Translating the A1C assay into estimated average glucose values. Diabetes Care 2008;31:1473–8.

Analysis

Univariate analyses including t tests and chi-square tests were conducted to investigate the unadjusted association between variables and glycemic control. Good glycemic control was defined as 90% of all POC blood glucose tests between 70 and 180 mg/dL based on empirical distribution and organization targets. A predictive model of inpatient glycemic control was then developed using a backward stepwise multivariable logistic regression approach. The data were split into a model building and validation set. Variables were included that represented both baseline and transitional state during the hospital stay to account for potentially mediating effects and a sensitivity analysis was conducted with them in and out of the final model to assess impact. Standardized coefficients were calculated to rank order variables in the model allowing indication of the variables with the greatest predictive impact on the outcome. Further investigation of the optimal classification points for the variables was conducted to indicate best differentiation of good glycemic control. Significant variables from the multivariable logistic regression were included in an exploratory classification tree analysis that recursively partitioned data in order to improve the fit, with optimal splitting identified over all variables at all possible split points. Classification tree cut-off points were used to further develop models identifying odds ratios for various thresholds for the top three predictive variables. A series of logistic regression models were then run with differing cut points of top 3 predictors to isolate adjusted odds of good glycemic control. Analytic data set building and statistical analyses were completed using SAS 9.4.

Results

Patient Characteristics

Table 1 shows patient demographic and clinical characteristics for the entire sample and for the top quartile (76% or greater POC blood glucose values within target range) and bottom quartile (25% or less POC blood glucose values within target range). Unadjusted results show a significant difference across quartiles for all factors except age, gender, dementia, rheumatic disease and paraplegia. Patients in the bottom 25th percentile (ie, the poorest control) were more likely than the total population to have a higher admission blood glucose (198 mg/dL vs. 153 mg/dL), higher HbA1c (8.53 [70mmol/mol] vs. 7.35 [57mmol/mol), a medical (74% vs. 66%) and/or respiratory (18% vs. 12%) diagnosis, corticosteroid use (17% vs. 27%), an insulin order set use (80% vs. 70%) and higher mean blood glucose during hospitalization (206.3 vs. 157.1 mg/dL). Patients with poorest control were less likely that the total population to have a high risk of mortality (11% vs. 15%) and severity of illness (13% vs. 18%). They also had less ICU care (8% vs. 13%), and a shorter LOS (5.82 vs. 7.82 days).

Predictive Modeling

The final multivariable logistic regression model had a c-statistic of 0.88. Model variables are detailed in Table 2 and rank ordered by standardized coefficient. The predictive performance was found to be robust when we examined the performance by splitting the data and running the model on a validation data set. The factor most predictive of glycemic control was HbA1c (OR, 0.60 [95% confidence interval {CI}, 0.58–0.61]), followed by admission blood glucose (OR, 0.91 [CI, 0.91–0.92]) and treatment with corticosteroids (OR, 0.06 [CI 0.04–0.08]). Other statistically significant predictive factors in the model included renal disease, BMI, risk for mortality, facility, major diagnostic category, liver disease, payor status, gender, cerebrovascular disease and COPD (see Table 2 for odds ratios). Additional analyses were conducted excluding variables potentially assessed after admission (insulin management, steroid use, and ICU) and possibly associated with the outcome. The results remained the consistent and we present the full final model here.

Classification tree analysis resulted in the same top 3 predictors, but in a different order. The analysis also provided cut-off values that predict suboptimal glycemic control. Classification tree analysis showed admission blood glucose was the most influential predictor, with 164.5 mg/dL indicating the optimal cut-point for prediction of SGC, followed by HbA1c with an optimal cut-off point of 6.65% indicating prediction of SGC, followed by treatment with corticosteroids, with an optimal cut-off point of 24% of the LOS on corticosteroids indicating prediction of SGC.