Impact of Second-Line Combination Treatment for Type 2 Diabetes Mellitus on Disease Control: A Population-Based Cohort Study

We conducted a population-based cohort study including all adult patients with an active DM2 diagnosis and metformin prescription between 2015 and 2020 in Catalonia, Spain. The main inclusion criteria were: initial treatment of metformin monotherapy followed by a combination therapy; at least two measures of HbA1c; HbA1c measure > 7% 90 days before the second-line addition; and one or more HbA1c values at least 90 days after the start of the combination therapy. All International Classification of Diseases, Tenth Revision [18] and Anatomical Therapeutic Chemical classification system [19] codes included in the study are shown in Table 1 of the Electronic Supplementary Material (ESM).

The study data source is the Information System for Research in Primary Care (SIDIAP) [20], which captures clinical information of approximately 5.8 million people from Catalonia, Spain (around 80% of the Catalan population). This information is pseudonymized, originating from different data sources: (1) ECAP (electronic health records in primary healthcare of the Catalan Health Institute); including sociodemographic characteristics, residents in nursing homes/long-term facilities, comorbidities registered as International Classification of Diseases, Tenth Revision codes, specialist referrals, clinical parameters, unhealthy habits, sickness leave, date of death, laboratory test data, and drug prescriptions issued in primary healthcare, registered in the Anatomical Therapeutic Chemical classification system; (2) pharmacy claims data corresponding to the primary healthcare drug prescriptions; and (3) the database of diagnoses at hospital discharge [21, 22].

The primary outcome is the time (in days) to HbA1c normalization (≤ 7%) after initiating the hypoglycemic combined treatment. Different secondary outcomes were considered: (1) to assess the impact of combination therapy adherence on HbA1c levels; (2) impact on weight after initiation of the combination therapy; (3) distribution of combination therapy over time; (4) duration of the single metformin treatment (number of days between the first therapy and the combination therapy); and (5) duration of the combination therapy (number of days between the first and last prescription). The follow-up time for all the outcomes started at the day of the hypoglycemic combined treatment initiation and ended at one of the following censoring events: end of study period (720 days) or disenrollment from the database or death.

We have used the records from electronic prescriptions and dispensations in pharmacies for drug exposure. Prescriptions are recorded with the exact date but invoice data from the pharmacy are recorded by month-year and not linked to the prescription. To assess this linkage, we used the smooth algorithm, developed in our group, which automatically determines at a patient level the most probable treatment (monotherapy or combination) throughout the study period [23]. The smooth algorithm was used to model the drug exposure in both prescriptions and dispensations, to obtain the exact moment of addition of combination therapy, the period exposed to that treatment, and the compliance or adherence.

The adherence was calculated as an approximation of the medication possession ratio (MPR). It was calculated as the ratio between the total days covered by the packages dispensed at the pharmacy, the numerator, and the total days of active prescriptions for that treatment, the denominator. We used a threshold of 80% to classify patients into adherents and non-adherents.

Results are presented as percentages, means and standard deviations, or median and interquartile range (IQR). In addition, a bivariate baseline analysis was performed. Comparisons between groups in baseline variables were performed by the analysis of variance test for continuous variables and the χ² test for categorical variables (exact Fisher test with observed frequencies < 5). In the primary outcome, the time in days was calculated until the first time the HbA1c was below the threshold (≤ 7%). Those patients who did not achieve HbA1c control during the follow-up period were censored at the date of the last HbA1c sample.

To assess the impact of each combination treatment on HbA1c, a survival analysis was carried out using the Kaplan–Meier and restricted mean survival time (RMST) methods [24‐26]. Comparison between the survival curves was undertaken using the log-rank test. Regression models were used to estimate the RMST and its standard error using the delta method [27] and we compared the difference of each RMST with the combination of metformin with sulfonylurea, our reference treatment. The RMST was adjusted by the patient’s characteristics using relevant covariates and truncated to 720 days (the follow-up time). Differences between RMSTs were calculated using the identity link and reported with 95% confidence intervals (CIs) along with their p values. For the weight reduction analyses, we fitted a logistic regression model to estimate the odds ratio (OR) of losing more than 5% of weight after the combination treatment versus the same reference group. The ORs were adjusted by the same covariates, and the 95% CI and p values were calculated using classic methods. All analyses were performed with the R statistical package version 4.2 [28, 29] or above under a significant level of 0.05.

From the 43,068 patients starting with a metformin treatment, a total of 28,425 individuals had an addition of a combination therapy during the 720 days of follow-up, and at least two measures of HbA1c: one HbA1c ≥ 7% before the addition and at least another HbA1c result after 90 days (Fig. 1). The demographic and clinical characteristics of the population at baseline show differences between the study groups (Table 1). Overall baseline HbA1c was 8.63% [70.78 mmoL/mol] (standard deviation [SD] = 1.34), but higher in patients with a combination with insulin, 10.83% [94.86 mmoL/mol] (SD = 2.09). The mean age was 63.63 years (SD = 11.38), a total of 38.1% were women, and the average weight of the population was 83.64 kg (SD = 16.64). Patients with a combination therapy of GLP1 were more frequently observed in women, 52.1%, a higher weight, 107.26 kg (SD = 20.52), and were diagnosed with DM2 3.33 (SD = 3.99) years ago. Clinical characteristics were homogeneous between groups, the mean glomerular filtration rate (as CKDEPI) was 81.12 (SD = 12.92) mL/min/1.73 m², and the main comorbidities in the population were hypertension, 19,866 (69.9%), obesity, 18,311 (64.4%), and dyslipidemia, 17,814 (62.7%).

The most prescribed combination therapy were sulfonylureas, 13,216 (46.5%) and DPP4i, 12,000 (42.2%). The rest of the therapies were observed in 17.3% of the patients, with SGLT2i the least prescribed with only 837 (2.9%) patients. Distribution was constant over time, although the sulfonylurea prescriptions as a combination treatment showed a negative trend, with DPP4i becoming the most prescribed from 2017 (Fig. 1 of the ESM).

On average, all treatments had a positive impact on HbA1c with a mean reduction of 1.82% (SD = 1.44, Table 2). Independently of the baseline level, we observed an exponential decay of HbA1c during the first 3–6 months after the addition of second-line treatment. Then, the HbA1c reaches its minimum before entering a slow recovery phase but without returning to baseline values (Fig. 2 of the ESM).

More than a half of the patients, 65.4%, reached the HbA1c goal during the follow-up. In the group treated with GLP1, the percentage of controlled patients was higher, 236 (81.4%), followed by insulins, 61.7%, sulfonylureas, 65.8%, DPP4i, 65.4%, and other hypoglycemic agents, 63.2%.

The efficacy on controlling HbA1c was measured using the RMST. The combination that reached the HbA1c target earlier was the GLP1 with a RMST of 330 days (95% CI 304–357), whereas intensification with other hypoglycemic agents needed more time to reach the HbA1c target, 483 days (95% CI 469–498). In patients with a combination treatment of sulfonylurea (our reference group), the RMST was 455 days (95% CI 451–459), 447 days (95% CI 442–452) in DPP4i, 452 days (95% CI 435–469) in combinations with insulins, and 457 (95% CI 440–457) in SGLT2i (Table 2).

At 720 days, the adjusted difference of RMST to the combination treatment with sulfonylureas was statistically different in all treatments except for SGLT2i (p = 0.549). Patients receiving GLP1 and insulins reached the HbA1c goal earlier, − 126.26 (95% CI − 152.49 to − 100.03) and − 69.18 (95% CI − 88.07 to − 50.30) days, respectively. While those receiving a combination treatment with other hypoglycemic agents and DPP4i needed more days to reduce the HbA1c levels, 39.01 (95% CI 24.08–53.95) and 14.81 (95% CI 8.61–21.01) days (Fig. 2).

The MPR and frequency of adherent patients to each combination treatment were also calculated (Table 2). The mean MPR for the entire cohort was 87.76% (SD = 23.6), with a total of 80.2% adherents. The combination treatments with more adherent patients were DPP4i, 10,540 (87.8%) and sulfonylurea, 10,143 (76.7%), whereas patients prescribed with a combination with insulin were less adherent to the treatment, 465 (48.8%).

The effect of adherence on the primary outcome is shown in Table 3. At 2 years, the adjusted differences in RMST between adherent and non-adherent patients was significantly different (p < 0.001). Overall, adherent patients reached the HbA1c goal 42 days earlier (95% CI 34–49) than non-adherent patients. A similar effect was observed in combination therapies with sulfonylureas 54 (95% CI 44–64) days earlier, GLP1 66 days (9–122), DPP4i 54 days (40–68), and other hypoglycemic agents 50 days (18–82). In contrast, the time to HbA1c control was not statistically different in patients with the insulin and SGLT2i combination treatment.

We also performed the main analysis using only adherent patients and did not find significant changes. As seen with the entire population, the adjusted difference of RMST was significantly different in all therapies except for SGLT2i. The adherence is reducing the time to reach the HbA1c goal but is not changing the results in the comparison with sulfonylureas (Table 3).

During the first 2 years, we observed a weight loss (reduction > 5% respect to baseline) in 5370 (39.6%) of the patients. The combination of metformin and a sulfonylurea reached this reduction in 2434 (37.4%) patients, which was the same for other hypoglycemic agents and insulin combinations, although the percentages were higher in the other treatments. The odds of losing more than 5% of weight were significantly higher in combinations with GLP1, OR = 2.63 (95% CI 1.8–3.84), DPP4i, OR = 1.28 (1.19–1.38), and SGLT2i, OR = 2.27 (1.85–2.8) (Table 2).

The duration of previous single metformin treatment was a median 701 (IQR = 375, 1156) days and 75.6% were treated during at least 1 year before the addition. No significant differences between treatments were observed (Table 2). Regarding the exposition period after the second-line therapy, the median number of days was, overall, 679 days (IQR = 362, 720). The sulfonylurea prescriptions were longer, 733 days (IQR = 380, 720), while patients receiving SGLT2i maintained the combined treatment for less time compared with the other treatments, 423 days (IQR = 219, 720).

The average impact on weight was also positive regardless of treatment (Table 2). The overall mean weight reduction at 720 days was 4.10 kg (SD = 5.84) higher in GLP1 and SGLT2i combinations with a mean reduction of 10.40 (SD = 11.5) and 6.83 (SD = 7.2) kg, respectively.

Vlacho et al. found a trivial effect in HbA1c reductions between groups [30]. Although in our study comparative effect sizes have not been assessed, we highlighted that all groups, except for the SGLT2i, reached HbA1c control faster than those receiving the sulfonylurea combination. The N from our study (43,068) differs drastically from the study by Vlacho et al. (75,808) probably owing to differences in the study design and periods selected, as our study analyses more recent data, and this could explain the discordances found. The proportion of patients who reached the HbA1c goal was higher in our study but the proportion of patients with a decrease ≥ 0.5 at 180 days was lower. In both studies, the sulfonylurea group had a higher proportion of users reaching the target of HbA1c below 7%. Days to HbA1c normalization were similar in both studies, and significant weight reductions were found for DPP4i and SGLT2i groups [30].

Regarding adherence to second-line therapy, Sicras-Mainar et al. conducted an analysis in 2008–9 but this was not comparable to our study because of the population characteristics, methods, and differences between period studies [31]. In Vlacho et al., they also found high MPR levels but lower than in our study [3]. The sulfonylurea group had the lowest proportion of adherent patients (MPR > 80%), and the time exposed to second-line treatment or persistence in our study was higher in all groups. These differences can be related to divergences in the methodology but, independently on how this was measured, a higher adherence was associated with greater glycemic control [34‐36], lower ratios on some micro-macrovascular outcomes [37], and with lower healthcare costs [38, 39].

In our database, adherent patients reduced the time to reach HbA1c control with the GLP1 combination followed by insulins. This is in line with the clinical guidelines, as GLP1 is indicated for an uncontrolled obese population with diabetes while insulins are for general uncontrolled patients with diabetes [15].

Among the limitations, the GLP1 group in our context is a pharmacological treatment indicated in combination with metformin and reimbursed only for patients with obesity in the Spanish Health Service. It is a non-comparable group because of the selecting bias, but our aim was to show the general practice in our context. Insulin-treated patients may not be comparable either, as they are usually patients with poor glycemic control. The sulfonylurea group is the second-line gold standard treatment clinically, thus it was chosen as the reference group for the statistical comparison [15]. Another limitation of our study is that we were unable to consider hypoglycemic events because of poor recording in the Catalonia primary care system. Given the high prevalence of sulfonylureas in our study population, which are known to carry a risk of hypoglycemia, it is important to interpret our results with this limitation in mind. The strengths of our study are the large number of patients, complete sociodemographic data, and adherence to each treatment group.

We have used the smooth algorithm to determine the electronic drug records. It allowed us to automate the data management step and model different aspects of drug exposure. With the algorithm, we obtained all patients who had received the second-line combination treatment, determined treatment initiation, its duration, possible treatment changes or interruptions, and calculated the adherence.

Adherence in databases like SIDIAP is difficult to assess. We have pharmacy invoice records that are not linked to electronic prescriptions, and extra work is required to link both databases. We described a process to easily calculate an approximation to the MPR that is consistent with the official data reported in our region but different to the MRP levels reported by Vlacho et al. [3]. This is a common example of how, in pharmacoepidemiologic studies with similar characteristics, each research team has a unique method to assess drug exposure that may induce biases and less comparable results. We tried to overcome this by using an automatic algorithm that brings additional value to our study [23]. Future research should include a longer follow-up time and focus the analysis on adverse effects such as bleeding or ictus after the addition of second-line treatment.