Long‐term clinical outcomes of oral antidiabetic drugs as fixed‐dose combinations: A nationwide retrospective cohort study

To compare treatment patterns and clinical outcomes of single‐pill fixed‐dose combination (FDC) and two‐pill combination (TPC) therapies using real‐world data.

therapies for metabolic diseases, because they have a lower price than two-pill combination (TPC) therapy and improve medication adherence. However, only a few studies have investigated the long-term clinical outcomes of FDC therapy with antidiabetic medications.
Prior evidence suggests that early FDC medication in the treatment pathway may have more advantages from the perspective of patients and physicians. 3,4 Improving compliance with oral antidiabetic drugs (OADs) and/or other medications for complications is key to prevent or remedy long-term co-morbidities associated with diabetes and minimize the economic burden. 5,6 A previous study reported a strong association between adherence and healthcare costs, where a 10% increase in adherence was associated with a 0.1% reduction in HbA1c levels. 7 Furthermore, poor adherence is associated with poor glycaemic control and subsequently increased morbidity, mortality and healthcare costs. 6,[8][9][10] To ease the complexity of treatment for type 2 diabetes, FDC may enhance medication adherence and thus glycaemic control by simplifying the treatment regimen for patients with type 2 diabetes. 4,11,12 Therefore, there is a clinical need to better elucidate the role of FDCs in patients with type 2 diabetes in routine care by comparing the treatment patterns of adherence and persistence and clinical outcomes versus TPCs using real-world data.    13,14 ( Figure S1).

| Exposure
Treatment regimens of MET+SU or MET+DPP4i, either as FDC or TPC, comprised more than 70% of total dual OAD prescriptions in South Korea. 15 Because of the reimbursement criteria and concerns regarding insufficient power, we did not investigate other dual OAD combinations (Tables S1 and S2). Indeed, the proportions of MET+DPP4i and MET+SU among the total dual combinations were 50% and 34% in 2015, respectively, according to the KDA annual report. 15,16

| Outcome
The primary outcome of interest was the clinical outcome, defined as a composite endpoint of all causes of death and hospitalization for stroke, AMI or HF. The individual components of the composite outcomes were the secondary clinical outcomes. A previous validation study found overall positive predictive values of 92.0% and 90.5% for AMI and stroke, respectively, in the NHIS-NSC database compared with the hospital's electronic medical records. 17 For clinical outcomes, patients were followed up from the index date until the earliest of outcome occurrence or end of the study period (31 December 2015), using an intention-to-treat analysis. An as-treated follow-up definition was used in a sensitivity analysis, with patients censored on the earliest outcome occurrence, treatment discontinuation or end of the study period.
For treatment pattern outcomes, we assessed adherence and persistence at 12 and 24 months using as-treated analysis after the index date. Adherence to FDC or TPC therapy was defined as the proportion of days covered (PDC), which was calculated as the duration of the prescription divided by the duration. Second, persistence of FDC or TPC therapy was defined as the duration between the index date and discontinuation of the medication of interest. Treatment discontinuation was defined as the absence of a refill prescription(s) within 150% of the previous prescription supply (e.g. if the previous prescription supply was for 30 days, the grace period would be 45 days). Moreover, all medications of interest were considered at the class level and not at the individual active ingredient level (e.g. for DPP4is, switching from sitagliptin to saxagliptin would not be considered a discontinuation).
In the sensitivity analysis, we varied the definition of the grace period to investigate the robustness of our study outcomes.

| Potential confounders
We assessed the sociodemographic characteristics of sex, age, area of residence, insurance quintile and health insurance type on the index date. The clinical characteristics of the Charlson co-morbidity index (CCI) score, co-morbidities (AMI, stroke, HF, hypertension, hyperlipidaemia, diabetic nephropathy, diabetic retinopathy and diabetic neuropathy) and use of co-medications (antiplatelet, statin, antihyper-

| Statistical analyses
To minimize any effects from measured confounders and obtain comparability between the FDC and TPC groups, matching (1:1 ratio) was performed based on the propensity score, which was estimated using a multivariable logistic regression model. 19 Balance in baseline covariates was determined using the absolute standardized difference estimate, with a value greater than 0.1 indicating an important imbalance.
We used caliper matching, where the maximum tolerated difference was 0.2 of the standard deviation of logit of the propensity score.
Moreover, a c-statistic of 0.6-0.8 was considered fit for propensity score matching. and FDC therapy, respectively, after 1 January 2003. After propensity score matching, 5143 patient-pairs were identified ( Figure 1).
All baseline covariates were well balanced and had an absolute standardized difference of less than 0.1 after propensity score match-  Table 4). The as-treated sensitivity analysis was comparable with that of the main analysis (Table S7). The results of additional sensitivity analyses wherein the grace period (50% and 300%) and the period for assessing the index medication were varied (at the time, within 60 days, and within 180 days) also remained consistent (Tables S8   and S9).
The results of our subgroup analyses for the association between the composite clinical outcome and medication regime stratified by select covariates revealed no statistically significant associations (all P values for interaction >.05) (Table S10-S14).

| DISCUSSION
Using South Korea's comprehensive and nationwide real-world data, we found significantly longer persistence, higher adherence and a which was not used in our study. In our sensitivity analysis that varied the grace period, we found an enhanced adherence rate with a PDC of 0.8 or higher at 12 and 24 months under less stringent conditions (Table S5). Although further investigations are warranted on the adherence and persistence of FDC therapy, our findings support the previously reported positive association between FDC therapy and increased adherence and persistence compared with TPC therapy among patients with type 2 diabetes.
Several studies have examined the association between clinical outcomes and FDC and TPC groups, focusing on the improvement of glycaemic control through regime simplification. [25][26][27][28] Although one study showed an association between non-adherence and increased mortality rates, it was probably caused by poor glycaemic control. 29 Another study that compared drug compliance and morbidities over a 3-year period failed to detect a statistically significant difference in morbidities between TPC and FDC. 11 Our findings for the composite clinical outcome and hospitalization for stroke associated with FDC versus TPC revealed statistically significant differences in favour of the FDC group. The FDC group had 2.55 and 1.84 fewer events per 1000 person-years, respectively, than the TPC group (Table 4). However, there was no significant difference in mortality rates between the FDC and TPC groups ( Figure S3), which is consistent with the findings of a previous study. 11  studies with more recent real-world data, including sodium-glucose cotransporter-2 inhibitors, are needed. Fourth, we measured adherence using the PDC based on the medication dispensed and assumed that medications were taken exactly as prescribed. Although this is not a direct measure of medication adherence, it is considered appropriate when using secondary databases. 38 Lastly, this was not a randomized study. Observational studies are usually associated with measurable and unmeasurable confounding potentials. Despite the use of multiple methodological approaches to mitigate confounding variables such as propensity score matching and subgroup analysis for measurable confounding, and active comparator and sensitivity analysis for unmeasurable confounding, residual confounding is possible owing to the nature of observational studies. Therefore, we calculated the E-value, 39 which indicated that our effect estimates were probable to be directed toward the null only when there was a very strong unmeasured confounder (Table S15).
This nationwide retrospective cohort study using real-world data representative of South Korea provided important real-world evidence suggesting that FDC therapy, compared with TPC therapy, could significantly reduce the risk of long-term clinical outcomes, especially hospitalization for stroke, accompanied by higher medication adherence and treatment persistence among patients with type 2 diabetes.