A total of 87,924 individuals initiated a combination of GLTs (Fig. 1). Among 77,145 patients without previous MI or stroke, 11,081 (14%) used SU + insulin and 16,910 (22%) used metformin + insulin. Their baseline characteristics are presented in Table 1. Since some patients used more than one combination of drugs during the study period, the number of patients using each combination adds up to more than the total number of patients included (n = 25,404).
Compared with SU + insulin, patients receiving metformin + insulin were younger, had less comorbidity, had a longer duration of monotherapy treatment and had more often used metformin prior to initiating insulin combination therapy. Among individual SU + insulin combinations, patients receiving glibenclamide + insulin and glipizide + insulin were older than average, whereas patients receiving repaglinide + insulin were younger and had less comorbidity. Patients receiving gliclazide + insulin more often used cardiovascular pharmacotherapy.
SU + insulin was used by 16% of patients receiving a combination therapy in the initial phase of the study, but its use declined throughout the study period (Fig. 2). In 2009, SU + insulin was used by 2,857 (5.1%) of patients using dual combination therapies. The median treatment duration for SU + insulin combination therapy was 92 days (interquartile range [IQR] 44–216 days), and 788 (7.0%) patients died while receiving this therapy. Among patients using SU + insulin, 3,346 (30%) had a limited time of use (<2 months) before receiving another therapy, while after 1 year of treatment 1,899 (18%) patients were still alive and were continuing to use SU + insulin. When stopping SU + insulin, most patients changed to insulin monotherapy (68%), triple (or more) therapy (13%) or SU monotherapy (11%), while a few patients changed to metformin + insulin (3%), metformin monotherapy (<1%) or metformin + SU (<1%), or stopped claiming GLT (5%).
Metformin + insulin was increasingly used throughout the study period. The median duration of treatment was 465 days (IQR 127–1,109 days), and 774 (4.6%) patients died during the therapy. When stopping metformin + insulin, most patients changed to insulin monotherapy (55%), triple (or more) therapy (17%) or metformin monotherapy (20%), while only a few patients changed to SU + insulin (1%) or SU + metformin (1%), or stopped claiming GLT (5%).
Mortality and cardiovascular endpoints
Crude incidence rates of mortality, cardiovascular death and the combined endpoint according to treatment are presented in Table 2. SU + insulin was associated with mortality rates that were two to five times higher than combinations of metformin + insulin. Among individual combinations of SUs + insulin, glibenclamide, tolbutamide and glipizide were associated with the highest incidence rates for all endpoints, whereas the use of gliclazide was associated with lower rates.
RRs from multivariable analyses are presented in Fig. 3. All combinations of SU + insulin were associated with a significantly higher risk of mortality when compared with metformin + insulin. When pooling combinations of SUs + insulin, higher RRs were observed for all endpoints compared with metformin + insulin.
Using glimepiride + insulin as a reference, mortality was not statistically different (RR [95% CI]) for combinations of insulin with glibenclamide (RR 1.13 [0.91, 1.41]), gliclazide (RR 0.84 [0.64, 1.09]), glipizide (RR 1.24 [0.96, 1.58]) and tolbutamide (RR 1.20 [0.83, 1.74]), and the differences in the risk of cardiovascular death and the combined endpoint were not statistically significant among individual SUs.
The results were similar in the sensitivity analyses (see electronic supplementary material [ESM] Table 1): condition (a) using only the initial combination therapy (censoring patients if changing combination therapy), condition (b) excluding patients who had used insulin prior to combination therapy, condition (c) including only patients who had used solely SU prior to combination therapy, condition (d) using a continually updated Charlson score, and condition (e) including in the model a registered diagnosis of heart failure (ICD-10: I50.x, I42.x, I110 or J819), renal disease (ICD-8: 582–586, 588, T558–559, Z992; ICD-10: N03, N04, N19-N19, R34, I12.x–I13.x), chronic obstructive pulmonary disease (ICD 8: 490–492; ICD-10 J42, J44), peripheral vascular disease (ICD-8: 443; ICD-10: I70 and I74) and use of loop diuretics (ATC: C03C) instead of Charlson score.
For the propensity score model, the C statistic was 0.78. The baseline characteristics were similar in the propensity score matched populations (ESM Table 2). Multivariable analysis on this matched population confirmed the results of the main analysis: the use of SU + insulin was associated with increased mortality (RR 1.70 [1.48, 1.95]), cardiovascular death (RR 1.35 [1.07, 1.70]) and the combined endpoint (RR 1.25 [1.05, 1.49]).
At baseline, 267 (1%) patients had previously been hospitalised with hypoglycaemia (Table 1).
A total of 888 (3.5%) patients were hospitalised with hypoglycaemia during the follow-up, the majority only once (83%), but some two (10.6%), three (3.4%) or more (3%) times.
Metformin + insulin was associated with lower incidence rates of hypoglycaemia compared with combinations of SU + insulin, while differences between SUs were less evident (Table 3).
In the multivariable analysis, the RR of having a first hypoglycaemic event after baseline was significantly higher among users of SU + insulin compared with users of metformin + insulin (RR 2.06 [1.62, 2.61]).
Hypoglycaemia prior to baseline was associated with an increased risk of all-cause mortality (RR 2.16 [1.94, 2.40]), cardiovascular death (RR 2.18 [1.85, 2.56]) and the combined endpoint (RR 1.82 [1.56, 2.12]). However, including hypoglycaemia in the multivariable model did not markedly attenuate the risk associated with SU + insulin compared with metformin + insulin in terms of mortality (RR 1.82 [1.64, 2.02]), cardiovascular death (RR 1.36 [1.15, 1.62]) or the combined endpoint (RR 1.25 [1.09, 1.43]).