Describing the characteristics of patients with CKD and T2D and evaluating treatment patterns in different settings is an important first step to guide future research on new treatments to prevent CKD progression. In this multicountry, multicohort study to clinically profile the population with both CKD and T2D initiating a GLP-1 RA and examine patterns of use during the study period of 2012–2021, the median duration of the initial GLP-1 RA exposure episode ranged from 2.3 months (PHARMO) to 12.4 months (VID); variations in this observation were potentially influenced by the shortest duration of the study period in PHARMO. Between 7.6% (DNHR) and 41.3% (CDM) of GLP-1 RA initiators had an interruption of current use lasting 90 days or more during follow-up. At the 1-year time point, the percentage of patients who were currently receiving GLP-1 RA treatment was more than 70% across data sources, except in CDM (52%), potentially as a result of loss to follow-up or a change in healthcare coverage for the CDM cohort. Of note, rates of GLP-1 RA initiation increased throughout the study period, with 1.6% to 4.8% of patients in each data source initiating treatment in 2012 and 19.8% to 31.5% of patients in each data source initiating therapy in 2019, consistent with previous evidence showing increases in new users of GLP-1 RA over time [21, 22].

Durations of T2D and CKD are difficult to assess in secondary data, and the estimated median durations of T2D and CKD among GLP-1 RA initiators were variable across data sources. The median duration of T2D among GLP-1 RA initiators ranged from 4 to 12 years, with the lowest duration noted in CDM, which is composed of enrollees of a US commercial insurer, and the highest noted in the European databases, which are composed of EHRs from national or regional healthcare systems. The median duration of CKD ranged from 2 to 7 years across data sources. Importantly, the duration of chronic conditions such as T2D and CKD before index treatment is initiated is dependent on the length and completeness of pre-index data. Enrollee turnover is high in US commercial health plans [23], resulting in shorter pre-index time than in countries in which most residents have national insurance over their lifetime. Although the interpretation of database-specific median duration of T2D and CKD may be limited due to specific database characteristics and differences in health systems across countries, results show that median duration of T2D was consistently higher than median duration of CKD, indicating that, generally, T2D diagnoses seem to precede CKD diagnoses.

While markers of T2D varied across data sources, most patients across all data sources had HbA1c levels > 8%, and prior evidence suggests that those with higher HbA1c levels were more likely to be prescribed a GLP-1 RA [24]. Diabetes severity score was higher in J-CKD-DB-Ex and VID than in other data sources, and differences in health systems and access-to-care factors may contribute to the differences observed. Most patients across data sources were on another GLD in the 6 months before the index date, and metformin was the most commonly prescribed drug in all data sources, with the exception of J-CKD-DB-Ex, in which DPP-4 s were the most prescribed before index. Potential explanations for these differences might be regional differences in treatment guidelines and practices, including preferential use of DPP-4s in Japan [25,26,27]. Variability was observed in patterns of insulin use across data sources, suggesting different practice patterns for GLP-1 RA use. In particular, the use of insulin at baseline was observed in most individuals (85.4%) in J-CKD-DB-Ex, suggesting the use of a GLP-1 RA after the introduction of insulin during the study period.

The prevalence of obesity was quite variable, ranging from 15.5% (J-CKD-DB-Ex) to 90.1% (VID). This variation may reflect fundamental differences in the clinical characteristics of the study cohorts (e.g., less obesity among Japanese people with T2D), differences in the data sources (e.g., varying means of capturing data on exact body mass index [BMI] versus diagnoses of obesity), and differences in treatment patterns (e.g., with preferential prescribing of GLP-1 RA therapies for individuals with obesity and required documentation of obesity diagnoses for prior authorization of GLP-1 RAs in some settings). For instance, in the Danish cohort, proportionally fewer individuals were classified as having obesity (32%, based on hospital diagnoses only) than in a clinical cohort of individuals with newly diagnosed T2D in the Danish Centre for Strategic Research in Type 2 Diabetes (DD2) (52% with BMI ≥ 30, based on weight and height recordings) [28], highlighting that the prevalence of clinical characteristics can vary by how variables are captured in data sources. Moreover, in Spain, prescription of GLP-1 RA is only authorized in patients with a BMI greater than 30, which could account for the very high recorded prevalence of obesity in VID [29]. Broadly, our findings may point to the use of GLP-1 RA therapy for glycemic control and obesity, irrespective of CKD severity, during the study period of 2012–2021.

The greatest proportion of patients in each database cohort were classified as having stage 3 CKD based on eGFR values or a diagnostic code (range, 27.6% [CDM] to 48.1% [PHARMO]); however, between 35% and 60% of patients had CKD stage 1 or 2. CKD stage 3, based on either eGFR value or diagnosis code, was lowest in CDM, which contained limited information on laboratory measures such as eGFR and ACR, potentially resulting in an undercapture of cases of CKD. Moderate or severe CKD was comparably higher in the DNHR and J-CKD-DB-Ex cohorts than in the other data sources, with DNHR, J-CKD-DB-Ex, and VID also having the lowest percentage of patients missing ACR values.

The most common CKD-protective drug class used previously or recently in relation to the GLP-1 RA index date was either ACEi or ARB across all data sources, consistent with other studies that have noted antihypertensive agents to be most frequently prescribed for patients with CKD and T2D [30]; SGLT2i use was also common. ACEi or ARB and SGLT2i use were in line with CKD and T2D recommendations at the time of the study [5]. Although GLP-1 RAs were not approved therapies for CKD during the study period, they have been shown to have beneficial effects on kidney outcomes [8]. With regard to patterns of GLP-1 RA use, exenatide is not recommended for patients with low eGFR values (< 30 ml/min or stage 4 or 5 CKD) [9, 31, 32], but 6.3% of patients in VID and 11.6% of patients in CDM were prescribed this medication. We did not match the type of medication with patient eGFR values; thus, it is plausible that the patients in VID and CDM who were prescribed these medications had higher eGFR values. Broadly, our findings may point to the use of GLP-1 RA therapy for glycemic control in individuals with T2D and CKD, regardless of the severity of CKD, many of whom present with obesity.

The companion analysis to clinically profile and evaluate treatment patterns in SGLT2i initiators using the same data sources revealed some differences in cohorts treated with GLP-1 RAs versus those treated with SGLT2i [19]. While the GLP-1 RA cohorts and SGLT2i cohorts identified in these analyses are not mutually exclusive, some interesting patterns emerge. Obesity was more common in the GLP-1 RA cohorts than in the SGLT2i cohorts, reflective of the indication for GLP-1 RAs during the study periods, as was baseline use of insulin. The severity of T2D was also greater in the GLP-1 RA cohorts than in the SGLT2i cohorts, and the GLP-1 RA cohorts tended to have worse kidney function at baseline than the SGLT2i cohorts. These findings suggest later use of GLP-1 RAs relative to SGLT2i in the T2D treatment pathway. At the 1-year mark, a greater proportion of patients were currently using GLP-1 RAs versus users of SGLT2is at the same time point, which might suggest patients had better persistence to GLP-1 RA medications than SGLT2i medications. However, many other factors could explain the results, such as differences in patient profiles and characteristics, differences in side effects, and variations in patient preferences that were not measured in this study.

While the purpose of the current study was to describe the use of GLP-1 RAs in patients with CKD and T2D, future research should explore safety outcomes associated with the use of GLP-1 RA therapies in clinical practice. In addition, exploration of the clinical profiles of and treatment patterns among individuals initiating GLP-1 RA therapies for other disease states such as those diagnosed with obesity, atherosclerotic cardiovascular disease, and heart failure may enhance understanding of the populations that may benefit from these therapies.

Strengths of this analysis include the use of healthcare data from multiple countries and data sources, most population based, and the use of a common protocol and statistical analysis plan with data source-specific adaptations, as required. Nonetheless, limitations are noted. In DNHR, while capturing laboratory measurements and prescriptions from general practitioners (GPs), conditions usually managed by GPs (e.g., hypercholesterolemia) may be underestimated because GP data are not routinely captured. In PHARMO, GPs are not required to code all diagnoses using codes and can include this information in free-text notes instead, which were not used in the present study. Additionally, certain conditions (e.g., hypertension) may not have diagnostic codes assigned and might be recorded solely through laboratory values or clinical measurements, which may also lead to the underestimation of these conditions in this study where diagnostic codes alone were used to define a condition. BMI data were not available in J-CKD-DB-Ex; thus, there is the potential for underreporting of obesity in J-CKD-DB-Ex. Further, undercapture of medications in J-CKD-DB-Ex is plausible if medications were dispensed before entry into the J-CKD-DB-Ex cohort or at a hospital outside the catchment area of this registry. Undercapture of medications in CDM is plausible if patients paid out of pocket or sought care out of a network where a prescription was written. In the case of countries that do not have universal health coverage, including coverage for medications, there may also be the possibility of misclassifying new users of GLP-1 RAs. We were not able to assess across data sources for what condition(s) the GLP-1 RAs were prescribed nor the indications; the mode of administration also was not evaluated and may have influenced persistence patterns. As with all secondary database studies, the primary purpose of these data sources is for billing or routine health administration, not research; therefore, mechanisms such as healthcare use patterns can lead to missing data or misclassification. A prescription could refer to a written prescription for a medication or a dispensing of the drug. However, neither of these implies that an individual took the drug as intended or was exposed to the drug, including any refills. Finally, although formal comparisons across data sources were not conducted, differences in the observed results could reflect this inherent heterogeneity among the data sources in the type and nature of data captured as well as differences in healthcare systems, treatment guidelines, country-specific clinical practices, formulary policies, and application of diagnostic coding systems in the participating countries.