Weight-Based, Insulin Dose - Related Hypoglycemia in Hospitalized Patients With Diabetes
Weight-Based, Insulin Dose - Related Hypoglycemia in Hospitalized Patients With Diabetes
O R I G I N A L A R T I C L E
T
reatment with insulin in hospitalized insulin therapy for hospital-ward patients Abstracting data from electronic medical
patients is a well-recognized risk (4–6), in which one approach to deter- records, we retrospectively selected a
factor for hypoglycemia. Although mining insulin dose is based on weight sample of adult inpatients who received
this clearly has been demonstrated in (i.e., total daily units per kilogram of insulin at Boston Medical Center, most
critically ill patients treated with contin- body weight). The initial daily insulin of whom had a diagnosis of diabetes,
uous intravenous insulin (1), evidence doses of basal-bolus protocols vary admitted between 1 July 2005 and 31
regarding subcutaneous insulin in non– widely, from 0.3 to 1.5 units/kg (5,7– December 2009. Diabetes was defined by
critical-care settings is lacking. Physicians 10). Rates of hypoglycemia in random- any one of the following: 1) administra-
may underdose insulin in fear of hypogly- ized trials using 0.4–0.5 units/kg/day of tion of an oral antidiabetes medication
cemia because there is uncertainty about insulin range from 3 to 33% of subjects (metformin, a thiazolidinedione, or a sul-
the dose-response relationship between (9,11). Furthermore, these studies ex- fonylurea) during hospitalization; 2) a di-
insulin and hypoglycemia in noncritically cluded patients with known risk factors abetes ICD-9 code of 250.xx; or 3) an A1C
ill inpatients with diabetes (2,3). for hypoglycemia, such as renal impair- .6.5%. Patients were excluded if they
Current guidelines on inpatient di- ment. Despite the use of relatively high were aged ,18 years, were admitted to
abetes management call for basal-bolus insulin doses in clinical practice, the risk an intensive care unit, had a primary di-
agnosis of hypoglycemia, had hypogly-
c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c cemia (any blood glucose ,70 mg/dL)
From the 1Section of Endocrinology, Diabetes, and Metabolism, Temple University School of Medicine, within 24 h of admission, or had no
Philadelphia, Pennsylvania; the 2Data Coordinating Center, Boston University School of Public Health, point-of-care (POC) capillary glucose val-
Boston, Massachusetts; the 3Department of Biostatistics, Boston University School of Public Health, ues. The sample was restricted to only the
Boston, Massachusetts; and the 4Section of Endocrinology, Diabetes, and Nutrition, Boston University
School of Medicine, Boston, Massachusetts.
first admission per patient, and data sub-
Corresponding author: Daniel J. Rubin, djrubin@temple.edu. sequent to the first hypoglycemic event
Received 26 December 2010 and accepted 24 May 2011. after 24 h were excluded. Only POC
DOI: 10.2337/dc10-2434 values were used to measure glucose
This article contains Supplementary Data online at http://care.diabetesjournals.org/lookup/suppl/doi:10. because of variability among different
2337/dc10-2434/-/DC1.
© 2011 by the American Diabetes Association. Readers may use this article as long as the work is properly glucose assays.
cited, the use is educational and not for profit, and the work is not altered. See http://creativecommons.org/ Case subjects were defined by a POC
licenses/by-nc-nd/3.0/ for details. glucose ,70 mg/dL after the first 24 h of
admission. The time period examined for as the first POC glucose value within 24 h Approximately 50% of the patients were
each case was 24 h prior to the first hypo- of admission. given metformin, 33% were given a sulfo-
glycemic event in order to include all rel- nylurea, and 13% were given a thiazolidi-
evant doses of insulin. Control subjects, Statistical analysis nedione. The sample was racially diverse:
defined by a POC glucose $70 mg/dL, Summaries of categorical variables in- 40.6% were African American, 34.1% were
were matched one to one on the basis of cluded counts and percentages, whereas white, 13.9% were Hispanic, and 11.4%
the hospital day of hypoglycemia, age for continuous variables means and SDs were of other race/ethnicity. The mean
(18–30, 31–64, or 65–80 years), sex, were used. Comparability between the A1C was 8.1 6 2.2%, and mean serum
and BMI (,18.5, 18.5–25, 25–30, or case and control groups for categorical creatinine was 1.5 6 1.7 mg/dL. Length
.30 kg/m2). Matching on hospital day variables was determined using the x2 test of stay averaged 7.6 6 7.0 days. Of the
of hypoglycemia provided a control time for categorical variables and the two- 1,990-patient sample, 1,418 (71.3%)
frame as well as controlled for variation in sample t test for continuous variables. g were not on insulin as an outpatient.
clinical status during hospitalization. Regression was used for length of stay Hypoglycemic patients were more
because this continuous, positive variable likely to be given glargine or NPH and
Variable definitions was not normally distributed. less likely to be given only SSI or no in-
The exposure was defined as the total Conditional logistic regression was sulin than control subjects (P , 0.001).
insulin dose per body weight (units/kg) used to examine the unadjusted and the SSI plus scheduled insulin was more
over the 24-h study period. Insulin doses adjusted association of insulin dose with common, whereas SSI without scheduled
were stratified into nonoverlapping ranges, hypoglycemia. The initial adjusted model insulin was less common, among case
as follows: 0, ,0.2, 0.2–0.4, 0.4–0.6, included all variables that were associated subjects than control subjects (P ,
0.6–0.8, and $0.8 units/kg. The exposure with hypoglycemia, as well an interaction 0.001). Hypoglycemic patients had a
was grouped by insulin regimen type as term for dose by regimen to assess effect lower albumin and hematocrit and higher
follows: 1) glargine plus any other insulin, modification by insulin regimen type on creatinine and length of stay than nonhy-
2) NPH plus any other insulin, 3) lispro the relationship between insulin dose and poglycemic patients (P , 0.001 for each
and/or regular insulin only, and 4) no in- hypoglycemia. We then performed back- comparison). Case subjects also had higher
sulin. For 11 patients who received both ward selection with the a level to keep comorbidity index scores than control
glargine and NPH in 1 day, group assign- variables in the model set at 0.2 (13). subjects (P , 0.001). There was no differ-
ment was on the basis of whichever insulin A P value , 0.05 was considered statisti- ence in mean A1C (8.0 6 2.1 vs. 8.1 6
dose was greater. For two patients who cally significant. All analyses were per- 2.3%, P = 0.44) or admission glucose
received the same total 24-h dose of glar- formed using SAS (version 9.2; SAS (199.8 6 109.5 vs. 193.6 6 96.9 mg/dL,
gine and NPH, group assignment was on Institute, Cary, NC). The Boston Univer- P = 0.20) between case and control sub-
the basis of the insulin given closest to the sity Medical Center Institutional Review jects. Higher insulin doses were associated
hypoglycemic event. Guidelines for insu- Board approved the protocol. with progressively higher blood glucose
lin dosing at Boston Medical Center have levels (Supplementary Table 2). Within
been published (10). RESULTS—Of 6,376 eligible patients each dose range, the mean glucose levels
The following known predictors of abstracted from hospital records, 1,012 of case subjects were lower than the mean
hypoglycemia and potential confounders (15.8%) had hypoglycemia. After match- glucose levels of control subjects.
were examined: weight, race, serum cre- ing, 995 case subjects (98.3%) and 995 Relative to insulin doses ,0.2 units/
atinine, albumin, liver function (aspartate nonhypoglycemic control subjects re- kg, the unadjusted odds of hypoglycemia
aminotransferase and alanine amino- mained (Supplementary Table 1). Com- increased with increasing insulin dose
transferase), A1C, white blood cell count, pared with 4,386 unmatched patients, the (Table 2). Adjusted for insulin regimen,
hematocrit, use of sliding-scale insulin matched control sample was older (aged SSI use, and albumin, creatinine, and he-
(SSI), fasting, length of stay, service (med- 63.3 vs. 60.6 years, P , 0.001), had lower matocrit levels, the higher odds of hypo-
ical or surgical), and disease severity de- weights (86.1 vs. 91.4 kg, P , 0.001), glycemia with increasing insulin doses
fined by the Charlson Comorbidity Index were less obese (BMI 31.2 vs. 33 kg/m2, remained (Table 3 and Fig. 1). Patients
(12). A higher comorbidity index repre- P , 0.001), had a higher proportion of who received insulin doses of $0.6
sents a greater burden of comorbid dis- African Americans (39.4 vs. 36.6%, P = units/kg were at increased odds of hypo-
ease. Use of SSI was categorized as only 0.04 for race), and a had lower proportion glycemia. In contrast, the adjusted odds
SSI, no SSI, or SSI plus scheduled insulin. of Hispanics (12.7 vs. 16.1%). The of hypoglycemia were not higher in pa-
Data were collected during hospitaliza- matched and unmatched control subjects tients who received 0.2–0.6 units/kg. The
tion, except for missing values that were were similar in terms of sex distribution association of insulin dose with hypogly-
substituted by the closest value up to and mean A1C. cemia did not vary by regimen type, as
3 months prior to admission. BMI values The median time from admission to indicated by a statistically nonsignificant
,10 or .100 kg/m 2 , weight ,20 or the first hypoglycemic event in case sub- interaction term in the multivariate model
.400 kg, and height ,100 or .300 cm jects was 2.7 days (range 1.0–33.7). Dur- (P = 0.524).
were considered erroneous and were de- ing the index 24-h period, glargine was In addition to insulin dose, there were
leted. No height was available for 195 pa- the most frequently administered insulin several significant predictors of hypogly-
tients (9.8%). For these individuals, the (31.7%), followed by short-acting insulin cemia, including creatinine and hematocrit.
average height of the sample for each sex (28.4%), no insulin (24.0%), and NPH in- The odds of hypoglycemia were threefold
was imputed (174.2 6 9.6 cm for male sulin (15.9%) (Table 1). SSI only was less greater among those who did not receive
subjects and 160.8 6 8.5 cm for female common than SSI plus scheduled insulin SSI relative to those who did. Patients who
subjects). Admission glucose was defined and no SSI at all (25.3, 37.0, and 37.7%). received SSI plus scheduled insulin or SSI
alone had higher blood glucose levels To investigate whether the risk of in the odds of hypoglycemia between
than patients who did not receive SSI hypoglycemia depends on the ratio of basal ratios .0.6 or 0 and 0.4–0.6. Fur-
(203 6 64, 185 6 52, and 128 6 33 basal insulin (NPH or glargine) relative thermore, there was no interaction of basal
mg/dL, P , 0.001). There was a trend to- to the total daily dose, a post hoc anal- ratio with the insulin dose–hypoglycemia
ward higher odds of hypoglycemia among ysis revealed that basal ratios ,0.4 confer relationship.
patients who received NPH compared lower odds of hypoglycemia than 0.4–0.6 Hypoglycemia was not more com-
with patients given glargine or short-acting (odds ratio 0.50 [95% CI 0.25–0.97], P = mon among patients given insulin with an
insulin. 0.040). In contrast, there was no difference oral diabetes medication compared with
Table 2—Distribution of insulin dose by case status threshold, patients given SSI are hyper-
glycemic by design. Patients who did not
All patients Case subjects Control subjects P* receive SSI may have been given excessive
scheduled insulin, resulting in more fre-
n 1,990 995 995 quent hypoglycemia. Together, these data
Insulin dose (units/kg) suggest that the insulin program with the
0 478 (24.0) 20.0 28.0 ,0.001 lowest risk of hypoglycemia consists of
,0.2 595 (29.9) 22.2 37.6 scheduled insulin plus SSI at total daily
0.2–0.4 285 (14.4) 15.0 13.7 doses ,0.6 units/kg.
0.4–0.6 263 (13.2) 16.5 9.9 Our findings are broadly consistent
0.6–0.8 161 (8.1) 11.4 4.8 with other studies of insulin use in non–
$0.8 208 (10.5) 15.0 5.9 intensive care unit inpatients that show
Data are n (%) or percentages. *P value by x2 test. that insulin is a risk factor for hypoglyce-
mia (14,15). These studies, however, did
those on insulin alone (49.4 vs. 54.6%, A1C .6.5%. A sensitivity analysis ex- not examine how the risk of hypoglyce-
P = 0.13). However, of the 201 patients cluding these patients yielded similar mia varies by insulin dose or type. The
who did not receive insulin but did re- results. lack of a clear difference in the odds of
ceive an oral agent, sulfonylureas were hypoglycemia between glargine and NPH
more common among hypoglycemic pa- is similar to data from a randomized trial
tients than in control subjects (27.4 vs. CONCLUSIONS—This retrospective, comparing weight-based analog insulins
16.4%, P , 0.001). Patients who received matched, case-control study of 1,990 with NPH and regular insulins that found
only SSI were more likely to be given an hospital-ward patients with diagnosed no difference in hypoglycemia rates be-
oral agent than those on SSI plus sched- or probable diabetes shows that higher tween the treatment groups (11). Also con-
uled insulin and those not given SSI weight-based insulin doses are associated sistent with other studies are our findings
(54.5, 28.1, and 38.3%, P , 0.001). with greater odds of hypoglycemia, in- that high creatinine and low hematocrit
Although hypoglycemia was propor- dependent of the types of insulin used. levels are independently associated with
tionately more common on medical serv- Adjusted for insulin regimen, SSI use, and hypoglycemia (16,17).
ices than on surgical services (Table 1, P = albumin, creatinine, and hematocrit levels, The frequent use of oral agents was
0.008), this variable was not significant in patients given at least 0.8 units/kg within surprising given that our medication
the multivariate model. There was no dif- a 24-h period are at threefold-higher odds guideline for hospitalized patients with
ference in mean blood glucose in medical of hypoglycemia than patients who re- diabetes encourages the discontinuation
versus surgical patients (173 6 62 vs. ceive ,0.2 units/kg. However, 0.6 units/kg of oral agents. However, we support the
168 6 59, P = 0.12). However, medical seems to be a threshold below which the judicious use of oral diabetes medications
patients received more insulin than surgi- odds of hypoglycemia are relatively low. In in patients who are eating regularly, who
cal patients (0.35 6 0.88 vs. 0.26 6 0.38 addition, patients who do not receive SSI are no longer acutely ill, and/or who
units/kg, P = 0.013). In addition, SSI only are at threefold-greater odds of hypoglyce- preparing for discharge, consistent with
was less common among medical patients mia than patients who receive SSI with some expert opinions (1,18). Our data
(20.9 vs. 36.0%, P , 0.001). The sample or without scheduled insulin. Because provide evidence for the safety of this ap-
included 77 patients (3.9%) without a SSI typically is administered in response proach in terms of hypoglycemia.
diagnosis of diabetes who had an inpatient to a glucose value above a predetermined A number of limitations must be
acknowledged. Generalizability is limited
by the inclusion of only one medical
Table 3—Predictors of hypoglycemia in the multivariate conditional logistic regression model center. Also, bias cannot be completely
accounted for by matching and multivar-
Odds ratio (95% CI) P iate analysis of retrospective data. The use
of POC glucose values at the exclusion
Insulin dose 0.005 of serum glucose values may under-
0.2–0.4 vs. ,0.2 1.08 (0.64–1.82) 0.777 estimate the occurrence of hypoglycemia.
0.4–0.6 vs. ,0.2 1.60 (0.90–2.86) 0.109 This effect, however, is likely to be small
0.6–0.8 vs. ,0.2 2.10 (1.08–4.09) 0.028 because almost all inpatients with diabe-
.0.8 vs. ,0.2 2.95 (1.54–5.65) 0.001 tes are ordered POC testing four times
Insulin regimen 0.062 daily, and a POC glucose is more likely to
Glargine vs. lispro/regular 1.58 (0.67–3.74) 0.297 be checked on a patient with hypoglyce-
NPH vs. lispro/regular 2.38 (0.97–5.87) 0.059 mic symptoms than a serum glucose.
NPH vs. glargine 1.51 (1.00– 2.27) 0.050 Another limitation is that a case-control
SSI ,0.001 design precludes the calculation of hy-
No SSI vs. SSI only 3.04 (1.23–7.51) 0.016 poglycemia rates; however, the odds ratio
SSI plus scheduled insulin vs. SSI only 0.83 (0.36–1.88) 0.648 provides a reasonable estimate of relative
No SSI vs. SSI plus scheduled insulin 3.68 (2.25–6.03) ,0.001 risk. An additional limitation is our in-
Albumin level 0.80 (0.60–1.08) 0.143 ability to classify the sample by diabetes
Creatinine level 1.14 (1.02–1.28) 0.018 type because of unavailable data. How-
Hematocrit level 0.96 (0.93–0.99) 0.013 ever, at least 90% of the sample is estimated
6. King AB, Armstrong DU. Basal bolus of an academic hospital. Endocr Pract ML. Hypoglycemia and clinical outcomes
dosing: a clinical experience. Curr Dia- 2010;16:512–521 in patients with diabetes hospitalized in
betes Rev 2005;1:215–220 11. Umpierrez GE, Hor T, Smiley D, et al. the general ward. Diabetes Care 2009;32:
7. Chen HJ, Steinke DT, Karounos DG, Lane Comparison of inpatient insulin regi- 1153–1157
MT, Matson AW. Intensive insulin pro- mens with detemir plus aspart versus 16. Kosiborod M, Inzucchi SE, Goyal A, et al.
tocol implementation and outcomes in neutral protamine hagedorn plus regular Relationship between spontaneous and
the medical and surgical wards at a Vet- in medical patients with type 2 diabetes. iatrogenic hypoglycemia and mortality in
erans Affairs Medical Center. Ann Phar- J Clin Endocrinol Metab 2009;94:564–569 patients hospitalized with acute myo-
macother 2010;44:249–256 12. D’Hoore W, Bouckaert A, Tilquin C. cardial infarction. JAMA 2009;301:1556–
8. Maynard G, Lee J, Phillips G, Fink E, Practical considerations on the use of the 1564
Renvall M. Improved inpatient use of basal Charlson comorbidity index with ad- 17. Kagansky N, Levy S, Rimon E, et al. Hy-
insulin, reduced hypoglycemia, and im- ministrative data bases. J Clin Epidemiol poglycemia as a predictor of mortality in
proved glycemic control: effect of struc- 1996;49:1429–1433 hospitalized elderly patients. Arch Intern
tured subcutaneous insulin orders and an 13. Budtz-Jørgensen E, Keiding N, Grandjean P, Med 2003;163:1825–1829
insulin management algorithm. J Hosp Med Weihe P. Confounder selection in environ- 18. Korytkowski MT. Treatment options for
2009;4:3–15 mental epidemiology: assessment of health safely achieving glycemic targets in the
9. Umpierrez GE, Smiley D, Zisman A, effects of prenatal mercury exposure. Ann hospital. ACP Hospitalist, 2009;(Suppl.)
et al. Randomized study of basal-bolus Epidemiol 2007;17:27–35 Chapter 3:15–23
insulin therapy in the inpatient man- 14. Wexler DJ, Meigs JB, Cagliero E, Nathan 19. Centers for Disease Control and Prevention.
agement of patients with type 2 diabetes DM, Grant RW. Prevalence of hyper- National Diabetes Fact Sheet: General In-
(RABBIT 2 trial). Diabetes Care 2007;30: and hypoglycemia among inpatients with formation and National Estimates on Dia-
2181–2186 diabetes: a national survey of 44 U.S. hos- betes in the United States, 2007. Atlanta,
10. Pietras SM, Hanrahan P, Arnold LM, pitals. Diabetes Care 2007;30:367–369 GA, U.S. Department of Health and Human
Sternthal E, McDonnell ME. State-of-the- 15. Turchin A, Matheny ME, Shubina M, Services, Centers for Disease Control and
art inpatient diabetes care: the evolution Scanlon JV, Greenwood B, Pendergrass Prevention, 2008