Clinical outcomes of basal insulin and oral antidiabetic agents as an add-on to dual therapy in patients with type 2 diabetes mellitus

While basal insulin remains the most effective antidiabetic agent and substantially reduces the risk of hypoglycemia, few studies have examined the comparative effect of basal insulin in the real-world setting. This study aimed to assess the outcomes of adding basal insulin compared with thiazolidinediones (TZDs) or dipeptidyl peptidase-4 inhibitors (DPP-4is) as a third antidiabetic agent in patients with type 2 diabetes mellitus (T2DM). A retrospective cohort study involving T2DM was conducted with health administrative data in Taiwan. Patients starting a third antidiabetic agent after receiving a metformin-containing dual combination were identified. The study endpoints included composite major adverse cardiovascular events (MACEs), all-cause mortality, and hypoglycemia. Propensity score matching and Cox modeling were used for analysis. After matching, the basal insulin and TZD groups contained 6,101 and 11,823 patients, respectively, and the basal insulin and DPP-4i groups contained 6,051 and 11,900 patients, respectively. TZDs and DPP-4is were both associated with similar risks of MACEs and hypoglycemia but a lower risk of all-cause mortality than basal insulin (TZDs: HR 0.55, 95% CI 0.38–0.81; DPP-4is: HR 0.56, 95% CI 0.39–0.82). Further studies are needed to elucidate the findings of increased all-cause mortality risk in patients receiving basal insulin, especially those with advanced diabetes.


Results
). There were 6,483 patients (4.7%) in the basal insulin group, 49,181 patients (35.6%) in the TZD group, and 82,446 patients (59.7%) in the DPP-4i group. The basal insulin group contained approximately 10%, 56%, and 34% of patients with NPH, insulin glargine, and insulin detemir, respectively. In each treatment group, approximately 89% of the patients received dual therapy consisting of metformin and sulfonylureas before receiving an intensification therapy agent. In the basal insulin group, previous dual therapy consisting of TZDs and DPP-4is was administered in 42 and 337 patients, respectively, and these patients were excluded from the following analyses of their respective drug classes. Figure 1 shows the process of study population selection and the final cohorts included in the outcome analysis after 1:2 propensity score (PS) matching.
Tables 1 and 2 present the cohort characteristics before and after PS matching. Before matching, the patients in the basal insulin group were younger, were male-dominated, were more likely to receive dual therapy as the initial treatment, had shorter DM duration, had lower HbA1c order rates in one year, and had more hospitalizations and emergency room (ER) visits than those in the TZD and DPP-4i groups. In addition, patients in the basal insulin group had lower prevalence rates of hypertension and dyslipidemia and used less related drugs, including angiotensin-converting enzyme (ACE) inhibitors/angiotensin receptor blockers (ARBs), calcium channel blockers (CCBs), and statins, than those in the other groups. After PS matching, no significant differences in patient characteristics were found between the treatment groups. The median daily dose of TZDs and DPP-4is were 0.7 and 0.9 defined daily dose (DDD) 25 , respectively. Cardiovascular events, all-cause mortality, and hypoglycemia. In the main analysis of the as-treated approach, the median follow-up time for the composite major adverse cardiovascular events (MACEs) was 0.35, 0.46, and 0.52 years, in the basal insulin, TZD, and DPP-4i groups, respectively. There was no significant difference in the risk of MACEs, hypoglycemia, or individual cardiovascular (CV) events among the treatment groups (Table 3). In contrast, the hazard ratio (HR) of all-cause mortality was significantly lower in the TZD (HR 0.55, 95% CI 0.38-0.81) and DPP-4i (HR 0.56, 95% CI 0.39-0.82) groups than in the basal insulin group.

Sensitivity and subgroup analyses. The results of the sensitivity analyses are presented in Supplemental
Tables S1-S4. With the intention-to-treat analytic approach, patients in the basal insulin, TZD, and DPP-4i groups were followed for 4. 16,4.52, and 4.09 years, respectively, and similar results were found for most of the outcomes, including MACEs and all-cause mortality. However, the risk of ischemic stroke was significantly higher in the TZD group (HR 1.25, 95% CI 1.03-1.50) than the basal insulin group. On the other hand, the risk of CV   (Fig. 2), we found that the reduced risk of all-cause mortality associated with TZD/ DPP-4i use was more prominent in patients whose diabetes had occurred for more than two years (TZD: HR 0.35 [95% CI 0.19-0.62]; DPP-4i: HR 0.32 [0.18-0.57]) (P value for interaction both <0.05). Moreover, females tended to have a higher risk of hypoglycemia than males when receiving TZD as the third antidiabetic agent (P value for interaction <0.05), although the increased risk of TZD compared to basal insulin in females was not significantly different (female HR 1.79 [95% CI 0.98-3.28] vs. male HR 0.71 [0.45-1.14]). No other subgroup differences were observed. The results of different types of basal insulins are presented in Supplemental Tables S5-S7, but the sample numbers were too small to conclude.

Discussion
The present study showed that the risks of MACEs and hypoglycemia were similar in type 2 DM (T2DM) patients treated with basal insulin compared to TZDs and DPP-4is as an add-on to dual OHA combination therapy. However, patients treated with basal insulin presented a higher risk of all-cause mortality than patients treated with TZDs and DPP-4is, especially those who endured long-term diabetes. The consistent results in the sensitivity analyses, particularly the intention-to-treat (ITT) analysis with a much increased follow-up time, further validated these associations. These results indicate that TZDs or DPP-4is as add-on therapies after the failure of   www.nature.com/scientificreports www.nature.com/scientificreports/ dual therapy could be a safer option than basal insulin as an add-on therapy, specifically for patients with more advanced DM.
The present study focused on comparing third antidiabetic agents because it has been revealed that the use of basal insulin is usually delayed; for instance, the results of a previous study found that Taiwanese patients received an average of 2.7 OHAs before starting basal insulin therapy 26 . Earlier studies on the treatment patterns of antidiabetic agents in Taiwan indicated that over 70% of dual therapy regimens involved a combination of metformin and sulfonylureas 14,15 . The resistance against the initiation of insulin may result from misconceptions about insulin, the inconvenience of injectable medications, and the easy accessibility of other oral treatment options, such as TZDs and DPP-4is 10 . While insulin has been demonstrated to preserve β-cells better than metformin and sulfonylureas by replacing endogenous insulin to induce β-cell rest 6-8 , similar effects were also found with TZDs and DPP-4is. TZDs are effective insulin sensitizers, and DPP-4is affect glucose regulation through multiple mechanisms by increasing incretin levels 12 . Both drugs not only preserve β-cells by reducing apoptosis but also improve β-cell function by maintaining or stimulating proliferation 27 . With similar pharmacological effects and uses in therapy, TZDs and DPP-4is are ideal comparators for evaluating the outcomes of basal insulin as a third-line therapy.
Several observational studies have explored the cardiovascular outcomes of insulin, but the findings have been inconsistent, especially among different active comparators [18][19][20][21]23 . When sulfonylureas were compared with insulin, the studies yielded comparable cardiovascular results between the treatments 18,19 . When DPP-4is were selected as a reference, the risk of macrovascular events appeared to be higher in patients receiving insulin treatment than in those receiving DPP-4i treatment. For instance, Nystrom et al. noted that the risk of fatal and non-fatal CVDs in the insulin arm was 1.39 times as much as in the DPP-4i arm for second-line treatment after metformin monotherapy 21 . Jil et al. also reported that the risk of composite outcome (non-fatal acute myocardial infarction (AMI), non-fatal stroke or all-cause death) was 1.6 times higher in patients receiving intensified treatment with insulin than in patients receiving intensified treatment with DPP-4is following dual-therapy failure 20 .  www.nature.com/scientificreports www.nature.com/scientificreports/ However, both studies included short-acting insulin in the analyses, which might obscure the effect of basal insulin. In contrast, one study compared the cardiovascular risk of NPH insulin to DPP-4is as a third antidiabetic agent and showed no significant increased risk among the treatment groups 23 . However, no study has directly compared the cardiovascular events associated with insulin and those associated with TZDs. In one study that compared various glucose-lowering agents as add-on medications to metformin and used sulfonylurea as the reference group, the risks of CVDs in the TZD and DPP-4i groups were significantly lower, while the risk of CVDs in the basal insulin group was similar to that in the sulfonylurea reference group 19 . www.nature.com/scientificreports www.nature.com/scientificreports/ Our findings showed similar risks of macrovascular events among the treatment groups, but the short exposure or follow-up times might have resulted in an insufficient effect or insignificant difference in the study drugs. In addition, our study population comprised relatively young patients with a short DM duration, and the reduced cardiovascular risk could be less pronounced in the young population. There remains debate as to whether intensive glycemic control leads to improved CV outcomes, no difference in risk or increased mortality. The inconsistent results in prior studies may originate from different study populations, the administration of early or late intensive glycemic control, and the lack of a long-term follow-up period in which cardiovascular benefits can be observed [2][3][4][5]24,28,29 .
Despite the inconsistent findings regarding CV events, nearly all the observational studies, including ours, found that insulin therapy was associated with an increased risk of all-cause mortality [18][19][20][21] . In addition, our study found that the increased risk of all-cause mortality associated with basal insulin therapy was even more prominent in patients who had more than two years of diabetic history. The results moderately corresponded with the ACCORD (Action to Control Cardiovascular Risk in Diabetes) trial 3 , which included an advanced DM population with an increased risk of mortality due to intensive glycemic treatment. The exact mechanism remains unclear; presumably, the increased risk could possibly be related to hypoglycemia and CV-related death. Hypoglycemia is known to provoke unstable hemodynamics and dysrhythmia and might result in an increased number of cardiovascular events and sudden death 30 22 . The latest American Diabetes Association guideline has recently revised the recommendations for choosing antidiabetic agents and suggests that a GLP-1 agonist, an injectable medication, is preferred over basal insulin to reduce the risk of hypoglycemia and weight gain 13 . While the present study showed that the risk of hypoglycemia was comparable among the treatment groups, the diagnosis of hypoglycemia may not be sensitive enough, and it is unlikely that the increased risks of all-cause mortality can all be attributed to the occurrence of hypoglycemia. More studies are warranted to investigate the mechanism of increased risk of death in patients treated with basal insulin.
Our study has several strengths. This study utilized a nationwide claims database, resulting in a large sample size to investigate the outcomes of basal insulin as a third antidiabetic agent for therapy intensification. In addition, financial issues do not affect the treatment choice or our study validity because the National Health Insurance in Taiwan did not impose specific reimbursement policy for the related use of TZD, DPP-4i, and basal insulin in diabetic patients. We also performed a series of sensitivity analyses to ascertain the robustness of the study results. The study results of MACEs, all-cause mortality, and hypoglycemia were consistent and therefore increased the credibility of our findings. Finally, this study provides additional insights into treatment strategies, revealing a safety concern associated with basal insulin, which is in line with the recommendation of the delayed use of basal insulin in the latest clinical guideline 13 .
There are several limitations in the present study. To ensure sufficient power in the outcome analysis, we utilized a broader definition for identifying patients with stable dual treatment (i.e., receiving metformin and one other OHA for more than 14 days in an outpatient setting) before starting the third antidiabetic agent. The definition might have resulted in enrolling patients who changed their treatment regimen frequently, but the results remained consistent in the sensitivity analysis, which only included patients who completed dual therapy for at least 90 days. This definition in this sensitivity analysis also aligned with the clinical guideline recommendation stating that a new antidiabetic agent should be added if an appropriate HbA1c level is not achieved after approximately three months of dual therapy 12 . Another limitation is the lack of laboratory results in the claims database. HbA1c is an important indicator of DM control. Although many proxies of disease severity (e.g., adapted Diabetes Complication Severity Index (aDCSI) score, time since first antihyperglycemic agents, disease burden, and HbA1c order rate) were adjusted to mitigate confounding effects, residual bias was still possible. Moreover, we failed to adjust for unmeasured confounders such as baseline body mass index (BMI), weight gain, and history of smoking in the analysis. Nonetheless, we reported the E-values, which indicate how strongly an unmeasured confounder must be related to the treatment and outcome to explain away a significant risk estimate. One of the strengths of the E-value approach is that it makes no assumptions specifying the nature of unmeasured confounder or the strength of the confounding association 31 . In our analysis of all-cause mortality, the E-values were respectively 3.04 and 2.97 for the comparison of basal insulin vs. TZDs and DPP-4is, and thus substantial unmeasured confounding would be needed to reduce the observed association to null. For example, several studies have assessed the impact of BMI and smoking on all-cause mortality [32][33][34] , and their risk estimates showed it might be implausible to have an unmeasured confounder associated with both treatment choice and all-cause mortality by a risk ratio of 3-fold. Last, similar to many other studies, we defined medication exposure only based on pharmacy dispensing records, and true medication adherence was unknown.
The present nationwide cohort study identified the safety outcomes of initiating basal insulin compared with TZDs and DPP-4is after the failure of dual therapy. No increased risk was found for cardiovascular events and hypoglycemia; however, an increased risk of all-cause mortality was observed in patients receiving basal insulin. Regarding safety concerns, as the latest clinical guideline suggested, there is no apparent benefit to initiating early basal insulin therapy when other OHA choices, such as DPP-4is and TZDs, are available. Further studies are needed to determine the mechanism of the increased risk in all-cause mortality in patients receiving basal insulin, especially those with advanced DM. Patients who received metformin and one of the other OHAs (sulfonylureas, meglitinides, TZDs, DPP-4is, or α-glucosidase inhibitors) for at least 14 days in the outpatient setting preceding the administration of a third antidiabetic agent were identified. Patients who were administered basal insulin, TZDs, or DPP-4is as a third antidiabetic agent after 2008 were included, and the first date of triple therapy was defined as the index date. The included patients were required to continue the triple therapy for at least two prescription cycles. Patients were excluded from the analysis if two or more antidiabetic agents were added at the same time after dual therapy or if the third antidiabetic agent was not a DPP-4i, a TZD, or basal insulin.

Methods
Exposure. The use of basal insulin (NPH, insulin glargine, and insulin detemir) as a third antidiabetic agent was compared to the use of TZDs (pioglitazone and rosiglitazone) and DPP-4is (sitagliptin, saxagliptin, vildagliptin, and linagliptin) separately. When comparing basal insulin and TZDs, we excluded patients with prior dual therapy containing TZDs. Likewise, when comparing basal insulin and DPP-4is, we excluded patients with prior dual therapy containing DPP-4is. The duration of each drug was calculated by days of drug supply, and when determining drug discontinuation, half of the supply days were considered a grace period to account for delayed refills 35 .
Outcome definition. The study outcome included MACEs, all-cause mortality, hypoglycemia, and individual cardiovascular events. MACEs were defined as hospitalization with a primary diagnosis of AMI (ICD-9-CM codes 410.x), ischemic stroke (433.x, 434.x), or CV death. All-cause mortality was determined based on the National Death Registry, and CV deaths were further defined considering the following CV causes: (1) heart-related; (2) hypertension-related; (3) cerebrovascular-related; (4) artery-, arteriole-, or capillary-related; and (5) 36 . Patients were followed until the occurrence of the study outcome, start of a new antidiabetic agent, discontinuation of study drug (basal insulin, TZD or DPP-4i), death, or end of 2015, whichever came first. To identify all related events, MACEs, all-cause mortality, and individual CV outcomes were followed for an additional 30 days after starting a fourth antidiabetic agent or discontinuation of the study drug.
Covariates. The study covariates, including age, sex, index year, clinical characteristics, and healthcare resource utilization (HCRU), were collected over a baseline period of one year prior to (and including) the index date. We included the following variables to assess and control for the baseline condition of diabetes: first antidiabetic agent, dual therapy medication other than metformin, time from the first antidiabetic agent to the index date (as a proxy of disease duration), duration of dual therapy, number of HbA1c tests in the past year, baseline proportion of days covered (PDC) by antidiabetic drugs (an estimate of medication adherence) 37 , aDCSI score (as a proxy of DM severity) 38 , and history of previous hypoglycemia. Patients' HCRU in the past year, in terms of the numbers of outpatient visits, ER visits, and hospitalizations, were considered as proxies of their general health condition and disease burden.
The following comorbidities present during the baseline period were identified (Supplemental Table S8): myocardial infarction (MI), other coronary arterial diseases (CAD), peripheral vascular disease (PVD), cerebrovascular disease, heart failure (HF), dysrhythmia, valvular heart disease (VHD), hypertension, dyslipidemia, chronic kidney disease, diseases that possibly require the frequent use of corticosteroids (autoimmune diseases, transplantation, asthma or chronic obstructive pulmonary disease), and mental disorders (depression, bipolar disorder, schizophrenia or anxiety). The data were adjusted for comedications, including α-blockers, β-blockers, ACE inhibitors, ARBs, renin inhibitors, CCBs, diuretics, nitrates, digoxin, lipid-lowering agents, antiplatelet agents, anticoagulants, atypical antipsychotics, and systemic steroids. Statistical analysis. Patient demographics, baseline characteristics, diabetes condition, comorbidities, and comedications were compared between the users of basal insulin and TZDs or DPP-4is. Continuous variables were expressed as means ± standard deviations, and categorical variables were expressed as numbers and proportions. A PS model was developed by including all the covariates in the logistic regression 39 . The comparison groups were 1:2 matched by PS (using nearest-neighbor matching with a caliper width equal to 0.2 times the standard deviation of the logit of the PS) 40 as well as a history of cardiovascular disease (i.e., presence of MI, CAD, PVD, cerebrovascular disease, HF, dysrhythmia, or VHD). The balance in the baseline covariates between the exposure groups before and after PS matching was evaluated by standardized differences. A standardized difference of less than 0.1 indicated a negligible difference 41 . Cox proportional hazards models were then used to estimate the HRs of each study outcome between the comparison groups of basal insulin and TZDs/DPP-4is.
A series of sensitivity analyses were performed to ascertain the consistency of the results. First, an ITT approach was applied for all study outcomes except hypoglycemia. In the ITT analysis, every patient was followed until the occurrence of the study outcome, death, or the end of 2015. Second, the dual therapy before the intensification therapy was confined to the combination of metformin and sulfonylureas. Third, the definition of dual therapy was restricted to combined use for at least 90 days to obtain results for a population in which the treatment strategy was in accordance with the clinical guideline 42 . Fourth, a strict definition of hypoglycemia was (2020) 10:5746 | https://doi.org/10.1038/s41598-020-62646-z www.nature.com/scientificreports www.nature.com/scientificreports/ used to include only the events recorded in the primary diagnosis. Fifth, to study the hypoglycemia outcome, one additional censoring criterion was applied; patient follow-up was censored when any of the first two antidiabetic agents was discontinued since the risk of hypoglycemia could change after altering the triple combination. Last, to evaluate the robustness of significant associations to potential unmeasured confounders, the E-values were calculated to present the minimum strength of the association that an unmeasured confounder would need to have with both the treatment and outcome to explain away the observed treatment-outcome association 31 .
Subgroup analyses were conducted to evaluate the differential effects of exposure among subpopulations with regard to (1) age (<65, ≥65 years); (2) sex (male, female); (3) history of cardiovascular disease (without, with); and (4) diabetes duration (<2, ≥2 years). Due to the limited number of patients with previous hypoglycemic events, we did not stratify the subgroups according to the presence or absence of hypoglycemia history. To explore the clinical outcomes of different types of basal insulins, we further divided basal insulins into NPH, insulin glargine, and insulin detemir.
All data analyses were performed using SAS, version 9.4 (SAS Institute, Cary, NC, USA). Two-sided P values less than 0.05 were considered statistically significant.

Data availability
The data that support the findings of this study are available from Taiwan's Ministry of Health and Welfare but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available.