Comparing effects of GLP-1 receptor drugs and metformin in patients new to diabetes treatment

In Denmark, all citizens have equal access to healthcare services, including both primary and hospital care. Danish residents are at birth or upon immigration assigned a unique 10‐digit personal identification number, making linkage of the national registers possible.17 The following seven registers were used for this study: (1) The Danish National Prescription Registry which holds information on dosage, dates, and anatomical therapeutic chemical (ATC) codes on all prescriptions dispensed from pharmacies since 199518; (2) The Danish National Patient Registry, which holds information on all outpatient contacts and hospital admissions since 1977, coded according to The International Classification of Disease (ICD)19; (3) The Danish Civil Registration System, which holds information on the date of birth, sex, emigration/immigration, and vital status20; (4) The Income Statistics Registry which has data on yearly income21; (5) The Population Education Register which holds information on highest attained education22; (6) The Clinical Laboratory Information System Research Database, which holds information on individual‐level biomarker results and date and time of sampling17; and (7) The Danish Adult Diabetes Registry, which collects information on patients with diabetes from the primary and secondary health care sector including biomarker measurements.23

Between January 1, 2018, and December 31, 2021, we identified incident (first‐line) users in monotherapy with GLP‐1 RA or metformin, having an HbA1c level above 42 mmol/mol (a list of ATC codes used for definitions is available in Table S1). Exclusion criteria included polycystic ovarian syndrome (defined as female sex and age under 40 years at the time of first metformin prescription), immigration within the last 10 years, age over 95 years, an eGFR measurement below 30 mL/min/1.73 m² within the year before the index date, an ICD code indicating heart failure within the last 10 years (Table S2), missing prebaseline HbA1c data, an HbA1c level below 42 mmol/mol, or a diagnosis of gestational diabetes (DO244). The reason for excluding all patients with heart failure was due to our inability to determine the presence of congestive heart failure stage IV, which is contraindicated for GLP‐1 RA use. To ensure that the incident users in the study period were indeed new users, we reviewed all prescriptions dating back to the year 2000. Further, to ensure exclusive treatment with GLP‐1 RA or metformin, we incorporated a 90‐day grace period from the date of the initial prescription to evaluate the prescriptions of other glucose‐lowering drugs, only including those who claimed solely their initial prescription or more of the initial treatment GLP‐1 RA within this period. Patients were thus included 90 days after their first prescription (index). A patient was considered to be under treatment from the first prescription until the date of the subsequent prescription. If no additional prescription was obtained or the last prescription concluded, a treatment duration of 3 months was assumed. Metformin initiators were matched with GLP‐1 RA initiators in a ratio of 1:1 without replacement on sex, age (5‐year intervals), and latest HbA1c measurement within 1 year before initiation of antidiabetic treatment (42–47, 48–52, 53–57, 58–64, 65–74, >75 mmol/mol) to address concerns regarding exchangeability between the two groups. Only GLP‐1 RA initiators were included for whom a matched metformin initiator was available.

The following outcomes were evaluated: the 1‐year risk of add‐on glucose‐lowering therapy and the 1‐year risk of nonadherence. In patients with a baseline and second measurement during a 1‐year follow‐up, the changes in HbA1c, eGFR, and cholesterol levels were likewise evaluated.

Age was calculated for the calendar year at treatment initiation; education level was defined as the highest achieved education level before treatment initiation and classified into four groups based on the International Standard Classification of Education (ISCED).25 Income was included as the equivalized income in the year before treatment initiation and categorized into quartiles for the main analysis.26 Comorbidities within 10 years before the treatment initiation were identified through the Danish National Prescription Registry and the Danish National Patient Registry (a list of ATC and diagnosis codes used for definitions is available in Table S2). Medicine use was identified through the Danish National Prescription Registry and defined as any redemption of a prescription 180 days before treatment initiation (a list of ATC codes is available in Table S3). Biomarkers were identified through the Clinical Laboratory Information System Research Database and the Danish Adult Diabetes Registry. Baseline values were included as the last value before treatment initiation, no longer than 1 year before.

Baseline characteristics were summarized at the time of treatment initiation. Patients were followed from the index date until outcome, death, emigration, or December 31, 2022, whichever came first. All analyses were conducted in subgroups of HbA1c, that is, 42–48 mmol/mol indicating prediabetes and ≥ 48 mmol/mol indicating diabetes. The 1‐year probability of the outcomes, add‐on glucose‐lowering therapy, and nonadherence were calculated, using the empirical distribution function, as the number of patients with the respective outcome within x days after the index date divided by the number of patients at risk at that particular date. Biomarker measurements were analyzed in posttreatment selected patients who had a follow‐up measurement within 1 year, that is, excluding patients who died without follow‐up measurement and patients who were alive but did not have a follow‐up measurement registered. The baseline values and the absolute changes in biomarker measurements were summarized by interquartile ranges. Multiple logistic regression was used to estimate the standardized 1‐year risk, risk difference (average treatment effect among GLP‐1 RA initiators), and risk ratio (RR) using the g‐formula,27 where the reference population included all patients in the subgroups. Multiple linear regression was used to estimate the average changes of the biomarkers. All regression models were adjusted for additive effects of the baseline patient characteristics including co‐morbidity and co‐medicine but not for the matching variables (sex, age, and HbA1c levels). In the analysis of biomarkers, the baseline level of each specific biomarker was included in the regression analysis, except for HbA1c We repeated all analyses in subgroups of sex (male, female), age groups (<55 years, ≥55 years), previous cardiovascular disease, defined as previous stroke, ischemic heart disease, or peripheral vascular disease (present, absent), and hypertension (present, absent). We further performed the following supplementary analyses: The 1‐year probability of having a biomarker measurement during follow‐up was calculated as the number of patients with a measurement within x days after the index date divided by the number of patients at risk at that particular date. To describe the selection of patients for the biomarker analyses, we used multiple logistic regression to estimate the odds of having an HbA1c measurement during follow‐up. R version 4.2.1 was used for data management and analysis.28

Compared with metformin initiators, GLP‐1 RA initiators had higher levels of education and higher incomes. The degree of urbanization was similar in the two treatment groups. Similar distributions were found for comorbidities. In regard to medical therapy before treatment initiation, GLP‐1 RA initiators more often redeemed prescriptions for mineralocorticoids, beta‐blockers, calcium channel blockers, renin‐angiotensin system inhibitors, loop diuretics, and thiazides, but less often for acetylsalicylic acids and statins, compared with metformin initiators. Similar distributions for baseline biomarkers were found (Table 1). Among those initiating GLP‐1 receptor agonists, 57.5% started with liraglutide, 42.0% with semaglutide, and 0.5% with dulaglutide.

The 1‐year absolute risks of add‐on glucose‐lowering therapy stratified according to baseline HbA1c are shown in Figure S1A. After standardizing with respect to the relevant confounders, estimating the average treatment effects among the treated, GLP‐1 RA initiation in those with prediabetes was associated with a 1‐year risk of add‐on glucose‐lowering therapy of 0.99% (95% confidence interval [CI]: 0.45–1.53), whereas metformin initiation was associated with a risk of add‐on glucose‐lowering therapy of 3.68% (95% CI: 2.55–4.81), resulting in a RR of 0.27 (95% CI: 0.10–0.44, p‐value<0.001). In patients with diabetes, GLP‐1 RA was associated with an RR of 0.67 (95% CI: 0.37–0.98, p‐value = 0.037) (Figure 2). Among those initiating add‐on therapy, 40.7% of metformin users added SGLT2 inhibitors, 36.0% added GLP‐1 receptor agonists (RAs), 14.0% added dipeptidyl peptidase 4 (DPP‐4) inhibitors, 6% added insulin, and 3.3% added sulfonylureas. For GLP‐1 RA initiators, 84.8% added metformin, 13.0% added SGLT2 inhibitors, and 2.2% added insulin, with no additions of DPP‐4 inhibitors or sulfonylureas. For nonadherence, the 1‐year absolute risks of nonadherence stratified according to baseline HbA1c are shown in Figure S1B. After standardizing with respect to the relevant confounders and estimating the average treatment effects among the treated, GLP‐1 RA initiation in patients with prediabetes was associated with a 1‐year risk of 55.68% (95% CI: 52.84–58.51) of nonadherence, whereas the risk associated with metformin initiation was 34.79% (95% CI: 32.09–37.50), yielding a RR of 1.60 (95% CI: 1.45;1.75, p < 0.001) for GLP‐1 RA initiation compared to metformin initiation. For patients with diabetes, this RR was 0.88 (95% CI: 0.70–1.06, p = 0.185) (Figure 2). Among those who did not adhere to their initial treatment within the 1‐year follow‐up period, 90.5% of metformin initiators and 94.2% of GLP1‐RA initiators did not start another glucose‐lowering therapy afterward. For the risk of add‐on therapy, the magnitude but not the direction changed in subgroups of sex, age, cardiovascular disease (CVD), and hypertension. Likewise, in all subgroups, the risk of nonadherence was increased for GLP‐1 RA initiation, compared with metformin initiation (Figure S2).

Investigating the differences in biomarkers, we included only those with a measurement during follow‐up. Among GLP‐1 RA initiators, these percentages were 73% (HbA1c), 36% (eGFR), 31% (LDL cholesterol), and 62% (triglycerides). For metformin initiators, the corresponding percentages were 86%, 51%, 42%, and 71%. There were the following unadjusted mean differences (SD) in biomarkers for treatment with GLP‐1 RA versus metformin: HbA1c; −6.43 mmol/mol (8.95) versus −2.95 mmol/mol (8.39), eGFR; −2.40 mL/min/1.73 m² (10.03) versus 0.61 mL/min/1.73 m² (9.06), LDL cholesterol; −0.22 mmol/L (0.71) versus −0.31 mmol/L (0.93), and triglycerides; −0.37 mmol/L (1.25) versus −0.21 mmol/L (1.19), respectively. In groups defined by baseline values, larger improvements were seen on the absolute scale in patients with a high baseline value compared to patients with a low baseline value (Figures S3–S6). After adjusting for relevant confounders, GLP‐1 RA initiation was associated with a greater average reduction in HbA1c, eGFR, and triglycerides compared with metformin initiation in both patients with prediabetes and patients with diabetes. A small increase in LDL cholesterol compared to metformin initiation was observed (Figure 3). The overall direction and magnitudes of average changes associated with GLP‐1 RA initiation compared to metformin initiation, on HbA1c, eGFR, LDL cholesterol, and triglycerides did not change in subgroups of sex, age, CVD, and hypertension (Figure S7).

A larger proportion of the patients initiating metformin died compared to patients initiating GLP‐1 RA, 16 and ≤5, respectively. There was no loss to follow‐up, and none emigrated during follow‐up. GLP‐1 RA users had relatively lower odds of having an HbA1c measurement during follow‐up (odds ratio: 0.41, 95% CI: 0.35–0.50), compared with metformin users, when adjusting for sex, age, education, income, degree of urbanization, co‐morbidities, and co‐medicine (Table S5). The baseline characteristics by status of prediabetes or diabetes revealed similar distributions as Table 1 (Table S6). The tendencies in the timing of biomarker measurement were similar regardless of treatment, although metformin initiators had a higher probability of having their biomarker levels measured in general (Figure S8).

There are several strengths of this study, including its large sample size and the nationwide coverage of data. Additionally, this study utilized information on morbidities and medication from officially maintained registers for administrative purposes, providing a reliable source of data, without relying on self‐reporting. One potential limitation of this study is the observational design, which entails a risk of residual confounding. Specifically, we lacked information on factors, such as body mass index and indication for use, which could have influenced the results. Additionally, other potential confounders, such as factors affecting the physician's prescription preferences, were not accounted for and we cannot rule out confounding by indication. Selection bias may have affected the generalizability of our findings. In particular, many individuals were excluded due to a lack of information on their HbA1c levels before treatment initiation. This may in part be attributed to the fact that some GLP‐1 RA products are used to treat obesity, which is a target group beyond the scope of this paper. Furthermore, GLP‐1 RA initiators may be a selected affluent group, since the cost of this therapy is higher than metformin. However, in Denmark, medication is subsidized by the government; thus, the maximum copayment of any patient amounts to approximately 595 Euro per year (in 2023),38, 41 thereby minimizing this risk slightly. We note that the effect estimates should be interpreted considering the matched population which imposed a selection. Furthermore, those who died during follow‐up could have been misclassified, that is, as nonadherent; however, mortality was rare. We evaluated the biomarker outcomes in the period 3 to 12 months after treatment initiation and analyzed only patients who had a second measurement in that period. The posttreatment selection of patients for the biomarker analyses was thus affected by the timing of follow‐up measurements, reflecting disease progression and patient behavior, limiting the interpretation of the changes in biomarker values. Hence, our results have a hypothesis‐generating character. Notably, nonadherent patients with subsequent blood chemistry measurements may not reflect the treatment's potential benefits accurately, while patients receiving add‐on treatment could contribute with blood chemistry measurements affected by other treatments.

In patients with prediabetes, GLP‐1 RA initiation was associated with a lower risk of requiring additional glucose‐lowering therapy, compared with metformin; however, metformin initiation was associated with substantially higher levels of adherence. In patients with diabetes, GLP‐1 RA initiation was also associated with a lower risk of requiring additional glucose‐lowering therapy, but no difference in the risk of nonadherence, compared to metformin initiation.