Patient characteristics
A total of 13,423 unique admissions met the initial inclusion criteria, of which 816 were patients who started BB after admission, 400 patients who switched BB after admission, and 171 patients who received nonselective BB other than carvedilol were excluded (patients may have been excluded for more than one reason). This left a final sample size of 1020 carvedilol, 886 SBB, and 10,216 non-BB recipients. Study population characteristics are shown in Table
1. BB recipients were older, less likely to be male or be on a surgery service, more likely to be on cardiac service, and had higher creatinine, hospital length of stay and frequency of basal insulin use (carvedilol 27%, SBB 33%, no BB 4.9%, p < 0.0001 by ANOVA), sulfonylurea/glinides, other cardiac medications and heart failure. Carvedilol users were less likely to be Caucasian than SBB or non-BB users while SBB were more likely to receive other diabetes medications.
Table 1Baseline characteristics by BB status
Age, years | 60 (15) | 64 (13) | 65 (13) | < 0.0001 |
Male, N (%) | 4986 (49) | 381 (37) | 383 (43) | < 0.0001 |
Race, N (%) | | | | < 0.0001 |
Caucasian | 7422 (73) | 628 (62) | 689 (78) | – |
African | 2293 (22) | 357 (35) | 174 (20) | – |
Asian | 121 (1.2) | 10 (1.0) | 4 (0.45) | – |
American Indian | 20 (0.20) | 1 (0.10) | 0 (0) | – |
Middle Eastern | 28 (0.27) | 2 (0.20) | 2 (0.23) | – |
Other | 332 (3.3) | 22 (2.2) | 17 (1.9) | – |
Body mass index, kg/m2 | 33.1 (9.4) | 33.7 (9.6) | 33.4 (9.6) | 0.13 |
Surgery service, N (%) | 2837 (28) | 72 (7.1) | 140 (16) | < 0.0001 |
Cardiac service, N (%) | 1330 (13) | 401 (39) | 352 (30) | < 0.0001 |
Hemoglobin A1c, % | 7.98 (2.53) N = 1704 | 7.71 (2.14) N = 186 | 7.78 (2.30) N = 113 | 0.30 |
Admission creatinine, μmol/L | 1.40 (1.50) | 2.24 (2.14) | 1.63 (1.46) | < 0.0001 |
Hospital length of stay, days | 4 (3–7) | 6 (4–10) | 6 (4–11) | < 0.0001 |
Basal insulin at admission, N (%) | 500 (4.9) | 272 (27) | 292 (33) | < 0.0001 |
Heart failure, N (%) | 461 (4.5) | 814 (80) | 459 (52) | < 0.0001 |
Sulfonylurea/glinide | 25 (0.24) | 40 (3.9) | 31 (3.5) | < 0.0001 |
Other glucose lowering agents (non-insulin, non-sulfonylurea/glinide) | 67 (0.66) | 95 (0.93) | 61 (6.9) | < 0.0001 |
ACE inhibitor/ARB | 171 (1.7) | 515 (50) | 278 (31) | < 0.0001 |
Statin | 340 (3.3) | 692 (68) | 395 (45) | < 0.0001 |
Aspirin | 361 (3.5) | 698 (68) | 397 (45) | <0.0001 |
Mean glucose (mmol/L) | | | | |
Admission | 175 (95) | 175 (92) | 168 (87) | 0.11 |
24 h | 167 (61) | 167 (62) | 165 (59) | 0.75 |
72 h | 164 (47) | 164 (47) | 162 (46) | 0.51 |
Hypoglycemia, N (%) |
< 3.9 mmol/L (70 mg/dL), first 24 h | 497 (4.9) | 138 (13.5) | 151 (17) | < 0.0001 |
< 3.9 mmol/L (70 mg/dL), during hospital stay | 1194 (12) | 393 (38) | 441 (50) | < 0.0001 |
< 2.2 mmol/L, (40 mg/dL), during hospital stay | 132 (1.3) | 47 (4.6) | 43 (4.9) | < 0.0001 |
Events < 3.9 mmol/L (70 mg/dL)/day during hospital stay | 0 (0.0) | 0 (0.17) | 0 (0.025) | < 0.0001 |
Glucose coefficient of variation (%) |
24 h | 0.22 (0.13) | 0.26 (0.15) | 0.26 (0.15) | < 0.0001 |
72 h | 0.24 (0.11) | 0.29 (0.13) | 0.28 (0.12) | < 0.0001 |
Admission glucose and mean glucose at 24 and 72 h were similar in carvedilol, SBB, and non-BB recipients (Table
1). Glucose coefficient of variation at 24 and 72 h was higher in carvedilol and SBB patients than non-BB recipients. Unadjusted frequency of Hypo
1day, Hypo
T, and Hypo
Severe were more common in carvedilol and SBB patients than non-BB recipients.
Relationship between BB and hypoglycemia
The unadjusted odds of all 3 definitions of hypoglycemia were increased for patients receiving carvedilol or SBB compared to no BB use (Table
2). Furthermore, the unadjusted odds of Hypo
1day and Hypo
T, but not Hypo
Severe were higher in SSB compared to carvedilol recipients.
Table 2Relationship between beta blocker use and hypoglycemia
Odds ratio for glucose < 3.9 mmol/L (70 mg/dL) within 24 h |
Unadjusted model |
Carvedilol vs. none | 3.01 | 2.48–3.71 | < 0.0001 | – | 2.56 | 1.95–3.37 | < 0.0001 | 0.91 | 0.65–1.27 | 0.58 |
SBB vs. none | 4.02 | 3.30–4.89 | < 0.0001 | 4.96 | 3.89–6.33 | < 0.0001 | 0.62 | 0.44–0.88 | 0.008 |
SBB vs carvedilol | 1.32 | 1.03–1.70 | 0.03 | 1.94 | 1.39–2.70 | 0.0001 | 0.68 | 0.46–1.02 | 0.06 |
Adjusted model |
Carvedilol vs. none | 1.45 | 1.05–2.01 | 0.0245 | < 0.0001 | 2.15 | 1.38–3.37 | 0.0007 | 1.18 | 0.74–1.90 | 0.49 |
SBB vs. none | 1.78 | 1.31–2.40 | 0.0002 | 4.29 | 2.96–6.20 | < 0.0001 | 0.82 | 0.51–1.32 | 0.42 |
SBB vs. carvedilol | 1.22 | 0.87–1.72 | 0.25 | 1.99 | 1.28–3.09 | 0.0023 | 0.70 | 0.41–1.18 | 0.18 |
Odds ratio for glucose < 3.9 mmol/L (70 mg/dL) overall |
Unadjusted model |
Carvedilol vs. none | 4.72 | 4.10–5.42 | < 0.0001 | – | 3.39 | 2.82–4.09 | < 0.0001 | 1.05 | 0.72–1.51 | 0.81 |
SBB vs. none | 7.49 | 6.48–8.66 | < 0.0001 | 6.30 | 5.25–7.56 | < 0.0001 | 0.91 | 0.64–1.29 | 0.58 |
SBB vs. carvedilol | 1.59 | 1.32–1.90 | < 0.0001 | 1.86 | 1.46–2.35 | < 0.0001 | 0.87 | 0.58–1.30 | 0.48 |
Adjusted model |
Carvedilol vs. none | 2.56 | 1.94–3.37 | < 0.0001 | <0.0001 | 6.30 | 4.59–8.65 | < 0.0001 | 1.03 | 0.65–1.63 | 0.91 |
SBB vs. none | 2.61 | 2.05–3.33 | < 0.0001 | 8.69 | 6.57–11.5 | < 0.0001 | 0.86 | 0.56–1.31 | 0.48 |
SBB vs. carvedilol | 1.02 | 0.77–1.35 | 0.89 | 1.38 | 1.02–1.86 | 0.03 | 0.84 | 0.52–1.36 | 0.47 |
Odds ratio for glucose < 2.2 mmol/L (40 mg/dL) overall |
Unadjusted model |
Carvedilol vs. none | 3.69 | 2.62–5.18 | < 0.0001 | – | 2.63 | 1.50–4.60 | 0.0007 | 1.04 | 0.65–1.64 | 0.88 |
SBB vs. none | 3.90 | 2.74–5.54 | < 0.0001 | 6.12 | 3.91–9.58 | < 0.0001 | 0.45 | 0.25–0.80 | 0.007 |
SBB vs. carvedilol | 1.06 | 0.69–1.61 | 0.80 | 2.33 | 1.23–4.43 | 0.0098 | 0.43 | 0.23–0.81 | 0.009 |
Adjusted model |
Carvedilol vs. none | 1.68 | 1.03–2.74 | 0.04 | < 0.0001 | 3.21 | 1.49–6.91 | 0.0029 | 1.11 | 0.58–2.10 | 0.76 |
SBB vs. none | 1.70 | 1.06–2.75 | 0.03 | 6.11 | 3.37–11.1 | < 0.0001 | 0.56 | 0.26–1.20 | 0.13 |
SBB vs. carvedilol | 1.10 | 0.59–1.75 | 0.97 | 1.90 | 0.90–4.02 | 0.09 | 0.50 | 0.22–1.14 | 0.10 |
For all adjusted models, there was a strong interaction between basal insulin use and BB type and this interaction term was therefore included in all models. All logistic regression models were also adjusted for age, gender, race, body mass index, surgery service, admission glucose, admission creatinine, heart failure, and basal insulin at admission. The adjusted odds of Hypo1day (OR 4.66, 95% confidence interval [CI] 3.61–6.01, p < 0.0001), HypoT (OR 12.1, 95% CI 9.86–14.95, p < 0.0001), and HypoSevere (OR 3.56, 95% CI 2.41–5.26, p < 0.0001) was increased in basal insulin nonusers compared to non-uers. The adjusted odds of Hypo1day were greater for carvedilol (odds ratio [OR] 1.45, 95% CI 1.05–2.01, p = 0.0245), and SBB (OR 1.78, 95% CI 1.31–2.40, p = 0.0002) compared to no BB. There was no difference in odds for SBB compared to carvedilol (OR 1.22, 95% CI 0.87–1.72, p = 0.25). The p-value for interaction between BB category and basal insulin use was < 0.0001.
For HypoT, the fully adjusted odds of hypoglycemia were greater for carvedilol (OR 2.56, 95% CI 1.94–3.37, p < 0.0001) and SBB (OR 2.61, 95% CI 2.05–3.33, p < 0.0001) vs. no BB. Further, the odds were similar for SBB vs. carvedilol (OR 1.02, 95% CI 0.77–1.35, p = 0.89). The p-value for interaction between BB category and basal insulin use was < 0.0001.
For HypoSevere, the fully adjusted odds of hypoglycemia were greater for carvedilol (OR 1.68, 95% CI 1.03–2.74, p = 0.04) and SBB (OR 1.70, 95% CI 1.06–2.79, p = 0.03) compared to no BB. The odds were similar for SBB vs. carvedilol (OR 1.10, 95% CI 0.59–1.75, p = 0.97). The p-value for interaction between BB category and basal insulin use was < 0.0001.
Due to the interaction by basal insulin use, separate models were created for basal insulin users and non-users. For Hypo
1day the fully adjusted odds of hypoglycemia were greater for basal insulin non-users but not for basal insulin users (Table
2). Moreover, there were greater odds of hypoglycemia associated with SBB vs. carvedilol in basal insulin non-users (OR 1.99, 95% CI 1.28–3.09, p = 0.0003) but not in users (OR 0.70, 95% CI 0.41–1.18, p = 0.46). Likewise, for Hypo
T the fully adjusted odds of hypoglycemia were greater for basal insulin non-users but not for basal insulin users. Further, there was a greater odds of hypoglycemia associated with SBB vs. carvedilol in basal insulin non-users (OR 1.38, 95% CI 1.02–1.86, p = 0.03) but not in users (OR 0.84, 95% CI 0.52–1.36, p = 0.47). Similarly, for Hypo
Severe the fully adjusted odds of hypoglycemia were greater for basal insulin non-users but not for basal insulin users. The odds of hypoglycemia associated with SBB vs. carvedilol was not significantly different in basal insulin non-users (OR 1.90, 95% CI 0.90–4.02, p = 0.09) or users (OR 0.50, 95% CI 0.22–1.14, p = 0.10).
Due to the observation that patients with heart failure were more likely to receive carvedilol compared to SBB, a sensitivity analysis was conducted among heart failure subgroups (Additional file
1: Table S1). In fully adjusted models, both carvedilol and SBB were associated with increased odds of Hypo
1day, Hypo
T, and Hypo
Severe in patients without heart failure, but not in patients with heart failure. There was no significant difference in odds of any of the hypoglycemia measures in carvedilol vs. SBB recipients in either subgroup.
Relationship between hypoglycemia and hospital mortality
Hypo
1day was associated with increased adjusted odds of in-hospital mortality in non-BB recipients (OR 2.10, 95% CI 1.15–3.86, p = 0.016) but not among carvedilol (0.79, 95% CI 0.15–4.10, p = 0.78) or SBB (OR 1.67, 95% CI 0.54–5.16, p = 0.37) recipients (Table
3).
Table 3Relationship between hypoglycemia and mortality in beta blocker subgroups
Odds ratio for mortality in patients with vs. those without hypoglycemia (< 3.9 mmol/L) within 24 h of admission |
Unadjusted model | 2.40 | 1.56–3.71 | < 0.001 | 0.83 | 0.25–2.81 | 0.77 | 1.60 | 0.74–3.46 | 0.23 |
Adjusted model | 2.10 | 1.15–3.86 | 0.016 | 0.79 | 0.15–4.10 | 0.78 | 1.67 | 0.54–5.16 | 0.37 |
Odds ratio for mortality in patients with vs. those without hypoglycemia (< 3.9 mmol/L) overall |
Unadjusted model | 2.21 | 1.60–3.04 | < 0.0001 | 1.18 | 0.54–2.60 | 0.68 | 2.17 | 1.08–4.37 | 0.03 |
Adjusted model | 1.80 | 1.16–2.80 | 0.009 | 1.55 | 0.49–4.95 | 0.46 | 4.89 | 1.76–13.56 | 0.002 |
Odds ratio for mortality in patients with vs. those without hypoglycemia (< 2.2 mmol/L) overall |
Unadjusted model | 4.63 | 2.52–8.52 | < 0.0001 | 1.76 | 0.40–7.67 | 0.45 | 6.42 | 2.73–15.1 | < 0.0001 |
Adjusted model | 3.74 | 1.48–9.46 | 0.005 | 1.94 | 0.36–10.3 | 0.44 | 10.6 | 3.27–34.3 | < 0.0001 |
In contrast, Hypo
T was associated with higher adjusted odds of in-hospital mortality in both non-BB (OR 1.80, 95% CI 1.16–2.80, p = 0.009) and SBB (OR 4.89, 95% CI 1.76–13.6, p = 0.002), but not carvedilol (OR 1.55, 95% CI 0.49–4.95, p = 0.46) recipients (Table
3).
In a separate model that included only BB recipients, there was no interaction between hypoglycemia (< 3.8 mmol/L) and BB type on the odds of mortality. However, data are not shown due to unstable estimates.
Likewise, HypoSevere was associated with increased adjusted odds of in-hospital mortality in both non-BB (OR 3.74, 95% CI 1.48–9.46, p = 0.005), and SBB (OR 10.6, 95% CI 3.27–34.3, p < 0.0001), but not carvedilol (OR 1.94, 95% CI 0.36–10.3, p = 0.44) recipients.
There was no significant interaction by basal insulin use in these models. However, these models demonstrated unstable estimates and therefore were not presented.