Article Text
PDF
Abstract
Objectives To analyse time to cardiovascular events and mortality in patients with type 2 diabetes (T2D) who received treatment intensification with insulin or a glucagon-like peptide-1 (GLP-1ar) analogue following dual therapy failure with metformin (MET) and sulphonylurea (SU).
Methods A retrospective cohort study was conducted in 2003 patients who were newly treated with a GLP-1ar or insulin following dual therapy (MET+SU) failure between 2006 and 2014. Data were sourced from The Health Improvement Network database. Risks of major adverse cardiovascular events (MACE) (non-fatal myocardial infarction, non-fatal stroke and all-cause mortality) were compared between MET+SU+insulin (N=1584) versus MET+SU+GLP-1ar (N=419). Follow-up was for 5 years (6614 person-years). Propensity score matching analysis and Cox proportional hazard models were employed.
Results Mean age was 52.8±14.1 years. Overall, the number of MACE was 231 vs 11 for patients who added insulin versus GLP-1ar, respectively (44.5 vs 7.7 per 1000-person-years adjusted HR (aHR): 0.27; 95% CI 0.14 to 0.53; p<0.0001). Insulin was associated with significant increase in weight compared with GLP-1ar (1.78 vs −3.93 kg; p<0.0001) but haemoglobin A1c reduction was similar between both treatment groups (−1.29 vs −0.98; p=0.156). In a subgroup analysis of obese patients (body mass index >30 kg/m2) there were 84 vs 11 composite outcomes (38.6 vs 8.1 per 1000 person-years; aHR: 0.31; 95% CI 0.16 to 0.61; p=0.001) in the insulin and GLP-1ar groups, respectively.
Conclusions In this cohort of obese people with T2DM, intensification of dual oral therapy by adding GLP-1ar analogue is associated with a lower MACE outcome in routine clinical practice, compared with adding insulin therapy as the third glucose-lowering agent.
Statistics from Altmetric.com
Introduction
The achievement of tight glucose control has been shown to reduce the risk of long-term vascular complications in patients with type 2 diabetes (T2D).1 ,2 Following the initiation of antidiabetic medication with metformin (MET), about 40%–60% of patients with T2D fail to achieve their glycaemic target, requiring intensification with a second-line agent, typically, with a sulphonylurea (SU).3 ,4 For many patients, failure to maintain optimal haemoglobin A1c (HbA1c) level despite up-titration to maximal doses of dual therapy (MET+SU) will necessitate the need for further intensification with a third-line agent. Although a variety of treatment options are available following failure of MET+SU dual therapy, limited data are available on the cardiovascular (CV) safety and diabetes-related outcomes on the most appropriate third-line antidiabetic therapy.5 ,6 Moreover, in the last 7 years, questions regarding the long-term CV safety of insulin (INS) have been raised.7 ,8 These epidemiological studies, however, have mainly investigated the use of INS as monotherapy or in combination with MET.9–11 Conversely, the CV benefits of the glucagon-like peptide-1 (GLP-1ar) analogues,12 ,13 a novel glucose-lowering therapy with favourable effects on weight reduction and low risks of hypoglycaemia, are an active area of clinical investigations (http://www.clinicaltrials.gov). GLP-1 analogues are hypothesised to have pleiotropic effects on the CV system based on evidence from experimental studies.14 Furthermore, since INS is known to be associated with weight gain and increased risk of hypoglycaemia, adding a GLP-1 analogue to MET+SU is an attractive alternative to lower HbA1c in patients with T2D. No comparative outcome data on GLP-1 analogue versus INS in patients with dual therapy failure are however available. Further work is therefore needed to explore the CV safety of INS compared with GLP-1 analogues when used as a third-line (injectable) therapy in patients with longer-duration disease and higher CV risk.
The aim of the present study therefore was to compare the real-world composite CV and mortality outcomes in UK clinical practice among patients with T2D following intensification of MET+SU dual therapy with either INS or a GLP-1 analogue.
Methods
Study design and data sources
This was a retrospective cohort study using the UK primary care database—The Health Improvement Network (THIN). THIN is the UK computerised longitudinal anonymised primary care records with information systematically entered by primary care physicians. It contains details of over 10.5 million patients derived from 532 general practices. THIN has been validated and shown to be demographically representative of the dynamics of the UK population in terms of demography, major conditions prevalence and mortality rate15 ,16 and has been used previously to evaluate diabetes-related outcomes in routine clinical practice.17 ,18 Ethics approval was provided by South East Research Ethics committee.
Study population
This comprised patients with T2D aged 18 and above, whose MET+SU dual therapy was intensified with either INS or GLP-1 agonist analogue from January 2006 to May 2014. We selected patients whose index date (treatment intensification with INS or GLP-1ar) was at least 90 days after the baseline date (registration into the database). Patients who started INS or GLP-1 analogue first before MET+SU commenced, previously on other antidiabetic medication, on more than triple therapy, with any form of CV, who died before intensification of dual therapy or those with type 1 diabetes were excluded.
Exposures and outcomes
Our exposures of interest were INS (either ultrashort/short-acting, premixed or long-acting) and GLP-1 agonist analogues (exenatide, liraglutide or lixisenatide) with a follow-up period of 5 years from index date. The study was exposure-based and participants were censored following the development of the primary outcomes, addition of another therapy, change of either GLP-1ar or INS, loss to follow-up or final records in the data (transfer out of practice) or at the end of the study.
The primary composite outcome was time to the risk of composite major adverse cardiovascular event (MACE) (including non-fatal myocardial infarction (MI), non-fatal stroke and CV death). These were identified by their appropriate read codes in the THIN database and must have occurred at least 180 days after the intensification of MET+SU with either INS or GLP-1 analogue.
Covariates
The study covariates were collected at least 180 days before intensification of MET+SU. These time-varying covariates included the baseline demographic parameters as age, sex, socioeconomic deprivation and smoking status; clinical measures as body weight, body mass index (BMI) and blood pressure (systolic and diastolic); biochemical parameters as baseline HbA1c, creatinine level, total cholesterol levels, low-density lipoprotein, high-density lipoprotein and triglycerides; and medications as statins, aspirin, antihypertensive drugs and oral antidiabetic drugs; comorbidities; the duration of diabetes treatment; and duration of MET+SU dual therapy before intensification.
Statistical analyses
Primary analysis was time to the composite outcome of non-fatal acute MI, non-fatal stroke or all-cause death in a propensity score (PS)-matched cohort. A PS model was used to adjust for allocation bias and was estimated using a logistic regression model in which the treatment status was regressed on the baseline covariates. We assessed the balance in baseline covariates between the treated (INS) and reference (GLP-1ar) subjects using standardised differences before and after matching. The mean and frequency distribution of measured baseline covariates between treatment groups with the same estimated PS was examined and summarised. Pairs of treated group and reference subjects were matched based on the estimated treatment probabilities; the average treatment effect on the treated was estimated by finding at least one match for each of the treated subjects from the reference group, at the nearest distance measured by the estimated PS. PS was considered as a prognostic covariate and included in a Cox proportional hazards regression model.
We used the log-rank test to compare the equality of the survival curves between the treatment groups in the full cohort, while the stratified log-rank test was used in the PS-matched group. Kaplan-Meier survival curves were estimated separately for INS-treated and compared with GLP-1ar treated participants in the PS-matched sample. From these survival functions, the absolute reduction in the probability of an event occurring within a 5-year follow-up was calculated. The marginal HRs were also estimated to enable us to quantify the adjusted hazard of an event occurring in the INS-treated group compared with the GLP-1 analogue group. Proportional hazards assumptions were confirmed through Schoenfeld residuals test. Point estimates with 95% CIs at the conventional statistical significance level of 0.05 were used in the regression models. Missing data among covariates were accounted for with multiple imputations using the chained equation model. Student's t-test was also used to estimate the mean changes between the baseline HbA1c and weight in both treatment groups throughout the follow-up duration. All analyses were conducted using Stata Software, V.13 (StataCorp. Stata Statistical Software: Release 13. College Station, TX: StataCorp LP. 2013).
Subgroup analyses
Cox proportional hazard models were fitted to adjust for baseline and time-varying demographics, comorbidities, medications and metabolic indices for those with BMI of 30 kg/m2 and above and 40 kg/m2 in order to explore the impact of obesity in influencing the primary outcomes.
Statistical significance was put at a p level of 0.05. To avoid the probability of type II error, the study was powered to 0.9 and the sample size of 412 in each treatment group was found to detect a true difference of 0.1 between the two treatment groups at 5% significance level. The study fulfilled the STROBE criteria for reporting observational studies.
Results
Cases and total follow-up
From the THIN database, we identified 2003 eligible patients in the UK Primary Care, whose MET+SU dual therapy was intensified with the addition of either INS (1584) or GLP-1ar (419). The flow diagram (figure 1) shows how our cohort was derived. The median treatment duration was 8.33 (interquartile range (IQR): 6.63–8.34) years. The median follow-up was 3.74 years (IQR: 2.10–4.93) representing a total follow-up period of 6614.12 person-years. The
Selection of study cohort.
Patients' characteristics
In the full cohort, the overall median age was 53.0 (IQR: 43.0–63.0) years; 50.2% were females. The mean BMI and HbA1c level were 31.8(7.9) kg/m2 and 9.7(2.9)%, respectively. One-on-one PS matching yielded 419 patients each in both treatment arms. The baseline characteristics in both treatment groups were compared between the full and matched cohort of patients with their standardised differences shown in table 1.
Baseline patient characteristics
Crude event rates
Survival analyses at 5 years were 95.8% vs 99.3% for INS and GLP-1ar intensified therapies, respectively (figure 2). The log-rank test and stratified log-rank tests used to compare the equality of the curves in both the full and matched cohorts, respectively, showed a significant difference between the treatment groups (p<0.001). Overall, there were 242 composite events with a crude incidence rate of 36.9 per 1000 person-years (95% CI 32.3 to 41.5). There were 231 vs 11 composite events in the INS versus GLP-1ar groups, respectively, with unadjusted incidence MACE rates of 44.5 vs 7.7 per 1000 person-years (table 2). Among the obese population (BMI≥30 kg/m2), there were 84 vs 11 composite MACE events, accounting for an unadjusted incidence rates of 38.6 vs 8.1 per 1000 person-years in patients intensified with INS versus GLP-1ar, respectively. Similarly, when stratified for morbid obesity (BMI≥40 kg/m2), 30 vs 7 events (29.6 vs 7.1 per 1000 person-years) occurred (table 2).
Comparison of number of events, incidence rate and HR between the treatment groups in the propensity score-matched cohort
(A) Kaplan-Meier survival analysis plot for matched cohort (stratified log-rank test p value<0.001). (B) Kaplan-Meier survival analysis plot for the full cohort (log-rank test p value <0.001). INS, insulin; MET, metformin; SU, sulphonylurea; GLP-1ar, glucagon-like peptide-1 analogues.
Table 3 shows the components of MACE—mortality, non-fatal MI and stroke. In the INS versus GLP-1ar treatment groups, there were 151 vs 5, 38 vs 3 and 42 vs 3 events of mortality, MI and stroke, respectively. Higher events of all the component outcomes were also reported in the INS group than in the GLP-1ar groups for all the components.
Comparison of number of events, incidence rate and HR between the treatment groups by the components of MACE
Risk of composite cardiovascular outcomes and mortality
Table 2 shows the comparison of number of composite CV events, crude incidence rate and HR between the treatment groups in the PS-matched cohort. In the unadjusted model, the risk of composite CV outcomes was 80% less (HR: 0.20, 95% CI 0.11 to 0.37) in patients whose dual therapies were intensified with GLP-1 analogue compared with INS. Following adjustment for gender, there was a slight reduction to 73% (adjusted HR (aHR): 0.27, 95% CI 0.14 to 0.53). Similar patterns were shown when stratified for obesity (BMI ≥30 kg/m2) and morbid obesity (BMI ≥40 kg/m2) with the risks being lower in the GLP-1 group (aHR: 0.31, 95% CI 0.16 to 0.61 and aHR: 0.31, 95% CI 0.13 to 0.75, respectively). We tested for violations of the proportional hazards assumptions using Schoenfeld residuals test, which tests the null hypothesis that the HR is constant over time. There is no evidence (p=0.42) to reject the assumption of proportional hazards for the treatment groups.
Of all the individual components of MACE, aHR was only significant for mortality. There was a 71% reduced risk of mortality (aHR: 0.21, 95% CI 0.08 to 0.51), similar to that of the composite outcome. The risks of stroke and MI were also 61% and 55% less in the GLP-1ar group compared with INS. However, this was not significant. This trend was also observed when stratified for obesity and morbid obesity (table 3).
Changes in HbA1c and weight
In figure 3, the trend in changes in HbA1c and weight per year in both treatment groups within the 5-year follow-up period is highlighted. There was no statistically significant change in the mean HbA1c levels in both treatment groups although reduction in HbA1c was seen more in the INS group throughout the follow-up period (−1.27% vs −1.0%, p=0.117). Conversely, the INS group recorded more weight gain than the observed weight loss in the GLP-1ar group (1.19 vs −3.35 kg, p<0.0001) during the study period.
(A) Trend in mean changes in HbA1c level (%) between the treatment groups (Met+SU+insulin vs Met+SU+GLP-1ar). There was no significant change in both treatment groups throughout the study period. (B) Trend in mean changes in weight (kg) between the treatment groups (Met+SU+insulin vs Met+SU+GLP-1ar). For all the years of the follow-up duration, all the p values were less than 0.05. GLP-1, glucagon-like peptide-1; HbA1c, haemoglobin A1c; MET, metformin; SU, sulphonylurea.
Sensitivity analyses comparing changes in weight and HbA1c between both treatment groups, using both complete and missing data, reported similar trend in both groups, showing that the imputation robustly addressed the missing data.
Discussion
This study showed that, among patients who are taking MET+SU, intensification of glucose-lowering therapy with GLP-1ar in routine clinical practice was associated with a significant 73% risk reduction in the risk of adverse composite CV events and mortality compared with intensification with INS therapy. HbA1c reduction was similar between the two groups but significant difference in weight response was observed between the two groups, that is, weight gain with INS and significant weight reduction with GLP-1 agonist.
Many trials comparing GLP-1ar with INS or other comparators including placebo have reported conflicting findings with those with placebo comparators showing CV benefits. Two recent meta-analyses, however, reported CV benefits of GLP-1ar.12 ,13 A similar recent observational study in a large cohort of 39 225 patients with T2D reported a similar reduced risk of heart failure, MI and stroke in three treatment groups comparing exenatide and exenatide+INS to INS only (61/56%, 50/38% and 52/63%, respectively).19 This collaborates with other reports showing the novel pleiotropic cardioprotective effects of GLP-1ar have also been described.20 A further possible explanation for the observed reduction in CV events with GLP-1ar compared with INS in our study could be due to the effects of GLP-1ar in reducing hyperglycaemia with limited increased risks of hypoglycaemia, as well as the beneficial effects of GLP-1ar in inducing weight loss.21 Both hypoglycaemia22 and weight gain,23 which are commonly associated with INS therapy, are known risk factors for adverse CV events. While further exploring the possible effect of obesity in our study cohort, we demonstrated a greater reduction of CV events with GLP-1ar compared with INS therapy in the subgroup of obese (BMI ≥30 kg/m2) and morbidly obese (BMI ≥40 kg/m2) patients with T2D.
The CV safety of INS is a controversial issue. Despite methodological adjustments, it is hard to exclude in observational studies all the potential bias. In the ORIGIN study, a randomised clinical trial with glargina INS in a high CV risk population, INS therapy was not associated with higher CV events. On the other hand, although preclinical studies and some meta-analyses suggest that GLP-1ar could have a protective CV effect, the ELIXA study (the only CV randomised clinical trial with a GLP-1ar published so far) showed that lixisenatide had a neutral CV effect compared with other antidiabetic therapies.
Our study showed comparable reductions in HbA1c levels in the patients on either GLP-1ar or INS. Clinical trials24 ,25 involving exenatide and liraglutide26 have reported similar HbA1c reduction compared with INS. Similarly, among patients on MET+SU, a recent randomised clinical trial reported similar HbA1c reduction between the GLP-1ar taspoglutide and INS glargine.27 INS therapy has been known to be associated with weight gain and this was consistent throughout the study period in contrast to GLP-1ar, which showed consistent decline in weight. The observed increase in body weight following INS therapy is in conformity with previous studies.4 ,28 ,29 Although the baseline BMI in our matched cohort was close to the morbid obesity range, our findings can be generalised to all patients with T2D because subanalyses in the obese and morbidly obese subgroups showed very similar findings.
The main strength of our study derives from the inclusion of a large cohort of patients with T2D receiving antidiabetic medications in a real-world population, which is largely representative of the UK population. This implies that our findings will be generalisable to the UK population and other countries that share similar demographics. Being derived from the UK primary care data, our findings mirror common clinical practice in the UK than the results of clinical trials. The large cohort from which the study participants were derived from provides adequate statistical power and also contains information on other time-varying covariates to adjust for possible confounders.
We adjusted for a large set of factors that could have differed at the baseline through PS matching. This would have been a major drawback in our study because GLP-1 analogues, being relatively newly introduced, had very fewer patients but more with CV risk factors as obesity, hypertension, hyperlipidaemia and greater weight than INS. A potential source bias was the inconsistency in the measurement of HbA1c levels according to guidelines (3–6 monthly). Due to this, many patients had no recordings for weight and HbA1c beyond the baseline. Some residual confounding in our study could be from our inability to measure and adjust for the dosage of the glucose-lowering therapies used in this study as well as the reliability of diabetes duration due to the ongoing issue of identifying incident versus prevalent diabetes. In addition, while there was a trend towards a lower diastolic blood pressure in the GLP-1 group compared with INS, this difference was not significantly different. Also, the classification of exposure into two broad drug groups could have possibly masked the effects of individual drugs and could have driven our study away or closer to the null hypothesis.
In summary, the evidence from this large cohort study, tracking outcomes in routine clinical practice, suggests that intensification of dual oral therapy by adding INS is associated with a higher risk of CV events, compared with adding a GLP-1ar therapy as the third glucose-lowering agent especially among obese patients with T2DM. This observation needs to be confirmed in a randomised clinical trial setting.
Key messages
What is already known on this subject?
Insulin therapy is widely used to manage hyperglycaemia in people with type 2 diabetes. Its use, however, is well recognised to be associated with weight gain and increased risk of hypoglycaemia—two known risk factors for cardiovascular events. More recently, concerns have been raised regarding the cardiovascular safety of insulin in people with type 2 diabetes.
What might this study add?
This study compares the cardiovascular safety of insulin with an alternative injectable glucose-lowering therapy, the glucagon-like peptide-1 (GLP-1) analogues in routine clinical practice. The later treatment is known to induce weight loss without any increased risk of hypoglycaemia.
How might this impact on clinical practice?
In people with type 2 diabetes who require intensification of glucose-lowering therapy following failure of metformin and sulfonylurea, GLP-1 analogues should be considered first before insulin treatment, especially in patients who are overweight. The use of GLP-1ar appears to be associated with a reduction in cardiovascular events and mortality compared with insulin.
References
Footnotes
Twitter Follow Uchena Anyanwagu at @ucheanyanwagu
Contributors II and RD conceived the idea for the study. UA and JM drafted the study proposal. II sought and obtained THIN Ethical Advisory board's approval. Data analyses were conducted by UA, JM and RM UA wrote the first draft. All authors reviewed and contributed to subsequent drafts and approved the final version of the manuscript.
Competing interests None declared.
Ethics approval South East Research Ethics Committee.
Provenance and peer review Not commissioned; externally peer reviewed.