Trends in Pharmacological Treatment of Patients With New Onset Type 2 Diabetes: Usage Patterns in an Evolving Guideline Landscape

2.1 Data Source

This retrospective repeated, cross-sectional study included adults with first diagnosis of type 2 diabetes from January 1, 2014 to December 31, 2022 in the University of California Health Data Warehouse (UCHDW), which captures information from all six UC Health systems (Davis, Irvine, Los Angeles, Riverside, San Diego, San Francisco). The UCHDW database was operationalized by UC Health as “nonhuman subjects research.” Analyses are institutional review board exempt [15]. This study examined electronic health record data from health care records of patients from the 5 academic medical centers and 12 hospitals with associated clinics, comprising the UC Health system. Electronic health records are connected and translated into a standardized vocabulary of clinical codes using the Observational Medical Outcomes Partnership (OMOP) version 5.4. As of May 2024, the source UCDHW dataset contained records for more than 196 million visits for more than 16 million unique patients within the UC Health system. Medical visit records extend retrospectively to 2012.

2.2 Study Sample

We identified adult patients (ages 18 years old and above at time of diagnosis) with incident type 2 diabetes based on preliminary type 2 diabetes diagnosis in their health record by each year of the study interval 2014–2022. We used Systematized Medical Nomenclature for Medicine–Clinical Terminology (SNOMED CT) codes 313436004 (Type 2 diabetes mellitus without complication) and 44054006 (Diabetes mellitus type 2) to identify patients with type 2 diabetes. To validate accurate type 2 diabetes incident status, study patients were required to have evidence of health record data presence in the UCHDW at least 2 years prior to first type 2 diabetes diagnosis (pre-index period) and at least 1 year follow-up data after type 2 diabetes diagnosis (follow-up period) to measure medication use after new onset. Additionally, a data string search was performed to identify diseases related to or caused by “type 2 diabetes mellitus” and any patients with evidence for any such conditions before their designated study incident date were excluded (Figure 1).

We identified glucose-lowering medications listed on the ADA 2024 Standards of Care Table 9.2: Medications for lowering glucose [16]. We determined medication use using Anatomical Therapeutic Chemical (ATC) Level 3 and 4 codes maintained by the WHO [17] and RxNorm codes maintained by the National Library of Medicine [18]. Both code sets indicate prescription events.

Insulins were identified using RxNorm codes descended from ATC code “A10A” (Insulins and Analogues). We identified Metformin using RxNorm codes descended from RxNorm code “6803” (Metformin). We identified SGLT-2 inhibitors using RxNorm codes descended from ATC code “A10BK” (SGLT-2 inhibitors). We identified GLP-1 receptor agonists using RxNorm codes descended from ATC code “A10BJ” (Glucagon-like peptide-1 (GLP-1) analogues) and RxNorm codes descended from RxNorm code “1 991 302” to capture the oral GLP-1 receptor agonist, Rybelsus. We identified DPP-4 inhibitors using RxNorm codes descended from ATC code “A10BH” (Dipeptidyl peptidase 4 (DPP-4) inhibitors). We identified Thiazolidinediones using RxNorm codes descended from ATC code “A10BG” (Thiazolidinediones). Finally, we identified Sulfonylureas using RxNorm codes descended from ATC code “A10BB” (Sulfonylureas).

The ADA 2024 guidelines, “Use of Glucose-Lowering Medications in the Management of Type 2 Diabetes,” were referenced to identify four high-risk groups where CVD and CKD risk reduction is crucial: Patients with atherosclerotic CVD (ASCVD), those at high risk for ASCVD, HF, and CKD [16]. Study patients could be assigned into one or multiple groups at baseline using SNOMED codes. Logical Observation Identifiers Names and Codes (LOINC) codes were used to identify relevant laboratory measurements.

For the ASCVD group, conditions included ischemic stroke (422504002), myocardial infarction (22298006), ischemic heart disease (414545008), and disorder of coronary artery (414024009). CKD was identified using code 709044004 and included incidences due to underlying conditions. HF was identified using code 84114007.

The high risk ASCVD group included patients aged 55 or older with two or more of the following indicators: obesity (414916001, 238136002, or BMI [LOINC 39156-5] ≥ 30) [19], hypertension (59621000), smoking status (449868002, 428041000124106, 428061000124105, 77176002), hyperlipidemia (55822004), and albuminuria [LOINC 9318-7 ≥ 300] [20].

Baseline clinical characteristics were identified using LOINC codes 39156-5 (BMI ratio), 8480-6 (systolic blood pressure), and 8462-4 (diastolic blood pressure). Measurements must have been taken on the day of the index diagnosis or in the 30 days prior. If multiple measurements were recorded, the value closest to the time of incidence was taken.

2.3 Graphical Indicators of Guideline Changes in Plots

Medication categorical percentage use by year over the study period were plotted for patients with incident type 2 diabetes for patients overall and separately by the four predefined high-risk groups requiring CVD and CKD protection, established ASCVD, high ASCVD risk, HF, and CKD.

In 2017, language was added to the ADA Standards of Care to “consider [the SGLT-2] empagliflozin or [the GLP-1] liraglutide in patients with established CVD” [21]. Following the 2018 consensus report on management of hyperglycemia released by the ADA and European Association for the Study of Diabetes, guidelines were revised in 2019 to consider key comorbidities including ASCVD, CKD, and HF. GLP-1 receptor agonists also became preferable to insulins as the first choice of injectable therapy [5]. Since 2022, guidelines no longer explicitly state metformin as the preferred first-line treatment in concordance with statements recognizing alternative initial treatment approaches to metformin as acceptable [22]. To indicate these material changes to ADA guidelines for type 2 diabetes treatment over the study period, three vertical lines mark were added to all use plots at years 2017, 2019, and 2022.

2.4 Statistical Analysis

Annual percentage use of type 2 diabetes medication by category within the first year following diagnosis was determined for the years 2014–2022. The annual percentage response variable was defined as the proportion of patients using a specific glucose-lowering medication category out of the total number of incident patients receiving any glucose-lowering medication in that incident year. The x-axis was segmented into intervals representing the year of diagnosis for each patient.

The analysis was repeated stratifying by each of the four predefined high-risk groups requiring CVD and CKD protection, established ASCVD, high ASCVD risk, HF, and CKD, to examine trends specific to these subgroups.

The Mann–Kendall Trend Test [23] was used to determine statistical evidence of monotonic directional change in percentage over time for each of the glucose-lowering medication categories used within 1 year of type 2 diabetes diagnosis. All statistical analyses were performed using Python, version 3.8, using the pyMannKendall package (Python Software Foundation, Beaverton, OR) on a protected Databricks platform workspace (Databricks, San Francisco, CA) with an α = 0.05 significance level applied for all statistical hypothesis tests.

3.1 Glucose-Lowering Medication Use Trends

Throughout the 2014–2022 study period, metformin was consistently the most prescribed medication in incident patients with type 2 diabetes, varying between a low of 68.1% (2014) to a maximum of 78.3% (2019). The variation in metformin use was not statistically significant based on the trend test (p = 0.12). Overall, insulins remained the second-most utilized medication throughout the study period. While usage dropped from 40.8% in 2014 to 30.3% in 2022, the decrease was not significant based on the trend test (p = 0.07). The GLP-1 receptor agonist and SGLT-2 inhibitor categories both demonstrated evidence of statistically significant, increasing usage over time in incident patients with type 2 diabetes. Usage of GLP-1 receptor agonists grew from 1.7% of incident patients in 2014 to 20.2% in 2022 (p < 0.01). Usage of SGLT-2 inhibitors grew from 0.9% to 14.2% (p < 0.01) in the same 9-year period. The rate of adoption for both medication categories accelerated significantly after 2017 and 2019. By 2021, GLP-1 receptor agonists and SGLT-2 inhibitors had eclipsed all other medication categories except for metformin and insulins to become the third and fourth most frequently utilized medications at 20.2% and 14.2%, respectively, with continued increase observed at the end of the study period in 2022 (Figure 2).

In 2014, sulfonylureas were the third most prescribed glucose-lowering medication category, used in 20.5% of incident patients with type 2 diabetes taking medication. However, sulfonylurea utilization dropped every year throughout the study period to 5.7% in 2022 (p < 0.01). In 2014, at 7.8%, DPP-4 inhibitors were more frequently prescribed than GLP-1 receptor agonists and SGLT-2 inhibitors. However, the use percentage remained relatively stable over the study period and demonstrated no statistically significant trend (p = 0.12). By 2020, both GLP-1 receptor agonists and SGLT-2 inhibitors were more commonly prescribed in incident patients with type 2 diabetes than DPP-4 inhibitors. Throughout the study interval, thiazolidinediones were the least frequently used and remained below 5% of use (Figure 2).

3.2 Drug Use Trends Among High-Risk Patients Requiring CVD and CKD Protection

Among study patients, 21 664 (54.0%) had a history of established ASCVD, CKD, HF, or were at high risk for ASCVD. Under 2024 ADA guidelines, adults with previous HF should begin treatment with an SGLT-2 inhibitor with proven benefit. Adults with CKD, ASCVD, or high ASCVD risk should incorporate GLP-1 receptor agonists or SGLT-2 inhibitors.

Across all four high-risk categories analyzed, SGLT-2 inhibitors and GLP-1 receptor agonists demonstrated statistically significant growth in utilization over the study period, with each category below 10% in 2019 growing markedly thereafter (p < 0.01). The SGLT-2 inhibitors increased from below 10% in 2019 to 25.06%, 17.3%, 42.4%, and 24.1% for ASCVD, high ASCVD risk, HF, and CKD patients, respectively, in 2022 (Figure 3).

For the GLP-1 receptor agonists, compared to the SGLT-2 inhibitors, the increase was significant but less pronounced for the four high-risk groups. With utilization below 10% in 2019, the percentage of users rose to 15.2%, 17.5%, 9.3%, and 11.5% for ASCVD, high ASCVD risk, HF, and CKD patients in 2022, respectively (p < 0.01 for all four groups). While GLP-1 receptor agonists and SGLT-2 inhibitors both demonstrated similar increases in utilization for patients with high ASCVD risk, SGLT-2 inhibitors were more commonly used across incident patients of other risk categories. The preponderance of SGLT-2 inhibitor use was most evident in the HF risk category, with nearly half (42.4%) of incident type 2 diabetes patients in 2022 who had initiated any medication using a SGLT-2 inhibitor within 1 year (Figure 3).

Sulfonylurea usage decreased in patients in all high-risk categories (p < 0.01 for all four groups). Use of DPP-4 inhibitors decreased in incident patients over time in those with high ASCVD risk (p < 0.05), but did not exhibit a statistically significant monotonic trend in patients with ASCVD (p-value 0.47) and HF (p = 0.11) (Figure 3).

Among patients with HF or CKD, insulin was the most prevalent medication throughout the study period, exceeding metformin. Trend tests in insulin use were non-significant for both groups with p > 0.05. Despite an increase in usage within the HF group between 2014 and 2018, insulin use began to decline markedly starting in 2019. For patients with HF, insulin use decreased from 69.6% in 2014 to 50.4% in 2022. In patients with CKD, usage fell from 74.5.0% in 2014 to 52.9% in 2022 (Figure 3).

In patients with ASCVD, both insulin and metformin were utilized at similar rates during the study period. While each was used in more than 50% of incident patients in 2014, usage declined to below half for each by 2022, with metformin reduction statistically significant via trend test (p < 0.01) Insulin use oscillated in patients with ASCVD during the study period in each of the high-risk groups from 2014 to 2022 study period consistent with a non-significant trend test (Figure 3).