Glucagon-Like Peptide-1 Receptor Agonist Use and Renal Impairment: A Retrospective Analysis of an Electronic Health Records Database in the U.S. Population

The Practice Fusion electronic health record database furnished the study data, which covered the time period from 1 January 2012 through 30 April 2016. Practice Fusion’s web-based electronic health records (EHR) system contains data input by over 150,000 medical professionals primarily working in small, ambulatory practices and primary care practices and treating over 50 million patients in all 50 states [14]. In particular, Practice Fusion provided recent records on patient characteristics, diagnoses, medications prescribed, laboratory test results, and observational data, such as weight and blood pressure. All records were de-identified and fully compliant with Health Insurance Portability and Accountability Act regulations. The research does not contain any studies with human participants or animals performed by any of the authors.

This study focused exclusively on patients identified as having T2D over the time period from 1 January 2013 through 1 May 2015 (i.e., the identification window), based upon a validated algorithm designed for EHR data [15]. Specifically, all patients who received at least one diagnosis of diabetes (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM], code 250.xx) were initially considered. Patients were excluded if they received any of the following: (1) more diagnoses of type 1 diabetes (T1D; ICD-9-CM codes 250.x1, 250.x3) than of T2D (ICD-9-CM codes of 250.x0, 250.x2) in addition to a prescription for glucagon; (2) more diagnoses of T1D than of T2D with no record of receipt of a GLA other than metformin; (3) a negative laboratory test result for c-peptide or a positive test result for diabetes autoantibody; or (4) a prescription for a urine acetone test strip.

Patients were included in the analysis if they initiated therapy with a glucose-lowering agent (GLA) during the identification window, with the first such date identified as the index date. Patients were excluded if they: (1) received a prescription for their index class of medication during the 1-year prior to the index date (i.e., during the pre-period); (2) received a diagnosis of pregnancy at any time from the start of the pre-period through the 1 year following the index date (i.e., the post-period); (3) were younger than 18 years as of the index date; or (4) appeared to have dropped out of the database at any time from the start of the pre-period through the end of the post-period. Patients were also required to have at least two recorded serum creatinine laboratory test results, with the first such test occurring some time after the start of the pre-period through to the index date and the second serum creatinine test recorded in the post-period. The eGFR was estimated using the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation [16]. Patients were classified based on their eGFR categories (eGFR ≥ 90 mL/min/1.73 m²; eGFR < 90 and ≥ 60 mL/min/1.73 m²; eGFR < 60 and ≥ 45 mL/min/1.73 m²; eGFR < 45 and ≥ 30 mL/min/1.73 m²; eGFR < 30 and ≥ 15 mL/min/1.73 m²; or eGFR < 15 mL/min/1.73 m²) [17]. These criteria resulted in a sample of 40,299 patients—3225 of whom initiated therapy with a GLP-1 RA and 37,074 of whom initiated therapy with an alternative class of GLA. Figure 1 illustrates how each of the inclusion and exclusion criteria affected sample size.

The GLP-1 RA and other GLA cohorts were matched using Mahalanobis matching, with calipers defined by the propensity score [18]. Multivariable analyses were then used to examine differences between patients who were treated with a GLP-1 RA and those treated with an alternative GLA. The analyses examined changes in eGFR over the study period as well as the likelihood of having a ≥ 30% reduction in eGFR. A similar analysis was conducted for the subset of patients who also had a A1c laboratory test recorded during both the pre- and post-period, with the multivariable analysis examining the change in A1c. All multivariable analyses were controlled for patient demographics, observational patient characteristics, general health and comorbidities, prior medication use and resource utilization, as well as index A1c and eGFR values. Patient characteristics included age, sex, race, ethnicity, region of residence, and smoking status, while observational characteristics included body mass index and blood pressure. General health was proxied by the Charlson Comorbidity Index (CCI) [19, 20]. Diabetes-specific health was measured by the Diabetes Complications Severity Index (DCSI) [21]. Anxiety and depression were also studied, as these comorbidities have been shown to be both prevalent in patients with diabetes [22, 23] and associated with poorer patient outcomes, and as they are not included in either the CCI or the DCSI. Prior medication use focused on which classes of GLA drugs were prescribed in the pre-period, the number of classes of anti-hypertensive drugs prescribed, and whether any of the following medications were prescribed: angiotensin-converting enzyme inhibitors, angiotensin receptor blockers, hypertension fixed-dose combination therapy, non-steroidal anti-inflammatory drugs (NSAIDs), aspirin, statins, and ezetimibe. Prior resource utilization captured the number of visits to a cardiologist or endocrinologist.

Given the results of the multivariable analyses, differences in continuous outcomes were examined using t statistics, with a P value of < 0.05 considered, a priori, to be statistically significant. When examining the probability of having a ≥ 30% reduction in eGFR, odds ratios (ORs) and 95% confidence intervals (CIs) were recorded. All analyses were conducted using SAS version 9.4 (SAS Institute, Cary, NC, USA).