Epidemiology/Health services research••

Cardiovascular and mortality benefits of sodium-glucose cotransporter-2 inhibitors and glucagon-like peptide 1 receptor agonists as third-step glucose-lowering medicine in patients with type 2 diabetes: a retrospective cohort analysis

•,,,,,.

...

Abstract

Introduction Studies have found that sodium-glucose cotransporter 2 inhibitors (SGLT2) and glucagon-like peptide 1 receptor agonists (GLP1) have cardiovascular benefits for patients with type 2 diabetes (DM2) and atherosclerotic cardiovascular disease (ASCVD), chronic kidney disease (CKD), or heart failure (HF). The literature does not provide evidence specifically for patients with these conditions who are adding one of these medicines to two glucose-lowering medications (ie, as “third-step” therapy). We explored the effects of different third-step medications on cardiovascular outcomes in patients with diabetes and these comorbid conditions. Specifically, we compared third-step SGLT2 or GLP1 to third-step dipeptidyl peptidase-4 inhibitors (DPP4), insulin, or thiazolidinediones (TZD).

Research design and methods We assembled a retrospective cohort of adults at five Kaiser Permanente sites with DM2 and ASCVD, CKD, or HF, initiating third-step treatment between 2016 and 2020. Propensity score weighted Poisson models were used to calculate adjusted rate ratios (ARRs) for all-cause mortality, incident major adverse cardiovascular event (MACE), and incident HF hospitalization in patients initiating SGLT2 or GLP1 compared with DPP4, insulin, or TZD.

Results We identified 27 542 patients initiating third-step treatment with one or more of these conditions (19 958 with ASCVD, 14 577 with CKD, and 3919 with HF). ARRs for GLP1 and SGLT2 versus DPP4, insulin, and TZD in the patient subgroups ranged between 0.22 and 0.55 for all-cause mortality, 0.38 and 0.81 for MACE, and 0.46 and 1.05 for HF hospitalization. Many ARRs were statistically significant, and all significant ARRs showed a benefit (ARR <1) for GLP1 or SGLT2 when compared with DPP4, insulin, or TZD.

Conclusions Third-step SGLT2 and GLP1 are generally associated with a benefit for these outcomes in these patient groups when compared with third-step DPP4, insulin, or TZD. Our results add to evidence of a cardiovascular benefit of SGLT2 and GLP1 and could inform clinical guidelines for choosing third-step diabetes treatment.

What is already known on this topic

Current guidelines recommend that patients with type 2 diabetes add a third agent when their blood glucose is not controlled with two medications, but the treatment of diabetes has become significantly more complex with the addition of new classes of medications. Providers are looking for guidance on how to properly sequence medications as well as how to describe the potential benefits and risks to patients of each added medication. Randomized controlled trials have shown that sodium-glucose cotransporter 2 inhibitors (SGLT2) generally reduce the risk of mortality, major adverse cardiovascular event (MACE), and heart failure (HF) hospitalizations in patients with diabetes as well as atherosclerotic cardiovascular disease (ASCVD), chronic kidney disease (CKD), or HF; and that glucagon-like peptide 1 receptor agonists (GLP1) generally reduce the risk of mortality and MACE in patients with diabetes and ASCVD; but there is not clear evidence about whether these medications have the same benefits specifically as third-step treatments in patients who have diabetes as well as ASCVD, CKD, or HF.

What this study adds

In patients with diabetes as well as ASCVD, CKD, or HF who initiate third-step treatment, this study suggests that third-step SGLT2 and GLP1 are generally associated with lower rates of mortality, MACE, and HF hospitalization when compared with third-step dipeptidyl peptidase-4 inhibitors, insulin, or thiazolidinediones.

How this study might affect research, practice, or policy

This study adds evidence that could be used to inform clinical guidelines specifically for patients who are initiating third-step treatment and have diabetes as well as one of these comorbid conditions.

Introduction

Multiple studies have demonstrated cardiovascular benefits associated with the use of sodium-glucose cotransporter 2 inhibitors (SGLT2) and glucagon-like peptide 1 receptor agonists (GLP1) among adults with type 2 diabetes (DM2) and atherosclerotic cardiovascular disease (ASCVD).1 2 Additionally, studies of SGLT2 have demonstrated slowing of diabetic renal disease among all diabetics as well as a reduction in heart failure (HF) hospitalizations among those with HF. Subsequently, the American Diabetes Association (ADA) and European Association for the Study of Diabetes issued consensus statements3 4 that patients with DM2 with these comorbid conditions preferentially receive an SGLT2 or GLP1 as a glucose-lowering medicine and to prevent worsening of these comorbid disease processes.

The use of metformin as a first-line medicine for the treatment of DM2 has been uniformly recommended for decades, due to its effectiveness,5 well-known safety profile, and low cost. Though the ADA recently changed its consensus statement6 to include additional medications as first-line glucose-lowering options, large healthcare systems,7 including Kaiser Permanente (KP), recommend metformin as a safe, effective, and cost-effective initial step for glucose lowering among adults with DM2 who need a medication to reach a target blood glucose range.

The evolution of evidence on which large healthcare systems base treatment recommendations has led to re-evaluation of glucose-lowering medicines through rigorous guideline development.7–10 KP updated its guidelines in 2020 to suggest that clinicians specifically consider SGLT2 or GLP1 for second-step therapy, in certain patient populations demonstrated to receive benefit, with suggested use of SGLT2 and GLP1 among individuals with ASCVD, and for SGLT2 additional suggested use among those with chronic kidney disease (CKD) or HF. Recommendations were similar to those of ADA and others, based on evidence of reduced mortality, major adverse cardiovascular events (MACEs), progression of renal disease, and/or HF hospitalizations, with consideration of the low risk for medication side effects and complications. The trials had included mostly patients taking one other glucose-lowering medication at the time of enrollment, providing a design that allowed evaluation of the additive benefit of an SGLT2 or GLP1 as a second step of the regimen. However, there was insufficient evidence to guide clinicians when considering third-step SGLT2 or GLP1 treatment (added to two other glucose-lowering medications) for individuals who have ASCVD, CKD, or HF.

There are benefits and risks of prescribing two glucose-lowering medications in combination as compared with prescribing them sequentially. The ADA advises clinicians to consider introducing combination glucose-lowering medications early in the course of diabetes for some patients, based on the benefits of dual therapy. Specifically, ADA consensus statements emphasize the benefits of obesity management and reduction of very high glycated hemoglobin (HbA1c) as rationale for early introduction of combination therapy including GLP1 and GIP (glucose-dependent insulinotropic polypeptide)-GLP1.6 However, it is important to consider the potential for interactions with other medicines and drug-specific side effects experienced with each added therapy.11 12 The provision of uniform evidence-based, stepwise recommendations to guide clinicians within large integrated healthcare organizations is the underlying impetus for the current study.

Previous observational studies and meta-analyses that assess head-to-head cardiovascular outcomes for SGLT2 and GLP1 medications have focused on patients with a specific step of treatment without providing results specific to patients with ASCVD, CKD, or HF,13–15 or focused on patients with ASCVD, CKD, or HF without specifying third-step treatment.14 16–21 We undertook this analysis to gather evidence from real-world use of third-step GLP1 or SGLT2 (compared with third-step treatments not expected to have cardiovascular benefits: dipeptidyl peptidase-4 inhibitors (DPP4), thiazolidinediones (TZD), or insulin) in patients with ASCVD, CKD, or HF, to explore whether the benefits seen in first-step and second-step treatment continue in third-step treatment, and to support a recommendation in a future KP guideline.

Research design and methods

This retrospective, observational study includes adult members enrolled in KP Colorado, Georgia, Northern California, Northwest, or Southern California health plans who initiated GLP1, SGLT2, DPP4, TZD, or insulin as third-step diabetes treatment in 2016–2020 and also had a diagnosis of ASCVD, CKD, or HF. We used inverse probability of treatment weighted (IPTW) Poisson regression models to calculate the relative rates of all-cause mortality, incident MACE, and incident HF hospitalization. The study followed the Strengthening the Reporting of Observational Studies in Epidemiology guideline22 for reporting observational studies.

Data sources

Data were extracted from existing health plan and care delivery data systems, including membership records, external claims (including hospitalizations), ambulatory and hospital-based electronic health records (EHRs), and administrative data repositories. We calculated baseline patient characteristics using data back to January 1, 2015.

Study population

We used an active comparator new user design. Patients became potentially eligible for study entry in the first month they were dispensed GLP1, SGLT2, DPP4, TZD, or insulin as a third-step treatment during the analysis period (January 1, 2016, to December 31, 2020), if they had no earlier dispense of the third-step treatment during the lookback period (January 1, 2015, to December 31, 2015), and if they also had ASCVD, CKD, or HF. We considered a medication to be a third-step treatment if it was initiated following a month where the treatment combination included exactly two other medication categories. The previous month’s two-medication combination was allowed to include medications that other patients used as third step, to include the variety of pathways that real-world patients take to third-step treatment.

Patients were identified based on outpatient dispenses of diabetes medications between January 1, 2015, and December 31, 2020. Medications were grouped into 10 categories: metformin, sulfonylureas, insulin, GLP1, DPP4, SGLT2, amylin analogs, alpha-glucosidase inhibitors, meglitinides, and TZD (see online supplemental table A for generic and category names). If a patient received only one dispense in a category for the entire period (2015–2020), this dispense was excluded from the analysis.

The calendar month is the minimum unit of time in the analysis. Time periods associated with dispenses were calculated based on the sold date and days’ supply, accounting for stockpiled supply. A patient’s treatment in a month was the combination of medication categories with periods of dispensed supply that overlapped with any days in the month.

We required patients to have at least 6 months of membership preceding third step. Patients were assigned to third-step treatment groups based on the first time these criteria were met. If third-step treatment criteria were met for the first time for multiple medication categories in the same month, the patient was excluded from the analysis.

Patients were included in the ASCVD, CKD, or HF subgroups if they had an indication of the condition in the same month as the start of third-step treatment, or earlier, based on diagnosis and procedure codes associated with these conditions in the Agency for Healthcare Research and Quality Healthcare Cost and Utilization Project Clinical Classifications Software23 and Clinical Classifications Software Refined24 tools (see online supplemental table B for category lists). Patients were included in the models for more than one subgroup if they satisfied the criteria for more than one of ASCVD, CKD, or HF.

Demographic data were extracted for each patient, including birth date, death date, gender, and self-reported race/ethnicity. The first indications of pregnancy or bariatric surgery were identified based on diagnosis and procedure codes extracted from inpatient and outpatient encounter data between 2015 and 2020. Patients with type 1 diabetes were excluded using the first two criteria of Klompas’ optimized algorithm.25

The presence of comorbidities in the Charlson Comorbidity Index (CCI)26 plus hypertension and hyperlipidemia were identified based on a 2-year lookback of inpatient and outpatient encounter diagnoses from the month that the patient started third-step treatment (see online supplemental table C for hypertension and hyperlipidemia diagnosis code lists). Outpatient HbA1c results were extracted from the EHR. Baseline HbA1c was the latest result in the 6 months up to and including the start of third-step treatment. HbA1c change around third step was calculated as the difference between the first result in the 2–6 months following the start of third-step treatment and the baseline value (if both exist).

Outcome definitions

The primary outcomes in this analysis are all-cause mortality, incident MACE, and incident HF hospitalization. MACE was calculated as a composite of all-cause mortality, acute myocardial infarction (AMI), and ischemic stroke. AMI, stroke, and HF hospitalization were identified based on diagnoses associated with inpatient hospital stays. HF diagnosis codes were from known association of HF coding algorithms with all-cause mortality26; AMI and stroke diagnoses are listed in online supplemental table D.

Statistical analysis

Baseline patient characteristics are described as frequency and proportion for categorical variables, median with IQR for count variables, and mean with SD for continuous variables.

We used Poisson regression with IPTW to calculate adjusted rate ratios (ARRs) for all-cause mortality, incident MACE, and incident HF hospitalization after patients started third-step GLP1, SGLT2, DPP4, TZD, or insulin. A separate model was fit for each outcome (mortality, MACE, HF hospitalization) within each patient subgroup (ASCVD, CKD, HF).

The exposure time followed an intent-to-treat approach, starting in the month that the patient initiated third-step treatment (for mortality) or the following month (for MACE and HF hospitalization). Exposure time ended at the earliest of the end of the analysis period; the first membership gap; the first indication of pregnancy or bariatric surgery; death; 24 months since the start of third-step treatment; or the outcome event. Patients with MACE or HF hospitalizations in the year before starting third-step treatment were excluded from the models for those outcomes.

The outcome models used average treatment effect IPTW to control for channeling bias. Propensity scores (PS) were calculated with multinomial logistic models that predicted the probability of being in the patient’s treatment group. The covariates in the PS model included the patient’s health plan site, race/ethnicity, gender, age, baseline HbA1c, and indicators for the presence of the individual Charlson comorbidities, hypertension, hyperlipidemia, second-step metformin, and second-step sulfonylurea. Weights were stabilized27 and Winsorized at the 95th percentile to reduce the impact of individuals with large weights and to obtain more accurate statistical inference and CIs. To quantify the degree of balance after weighting, absolute standardized mean differences (ASMD) were calculated28 for each variable in the PS model for each pair of treatment groups. A variable was considered balanced across the treatment groups if the maximum pairwise ASMD after weighting was less than 0.25.29

The outcome models included the treatment group as an independent variable, as well as the patient’s site, race/ethnicity, gender, age, baseline HbA1c, CCI, and indicators for whether the second-step treatment included metformin, sulfonylurea, TZD, GLP1, DPP4, SGLT2, or insulin.

Explorations of the sensitivity of the results to patient characteristics and the definition of the follow-up period were conducted in the largest patient group (all-cause mortality outcome in the ASCVD patient subgroup). These included models for patients with specific second-step combinations (metformin+sulfonylurea, metformin+insulin); a model that excluded data from 2020 (ie, the COVID-19 pandemic); and a model with exposure time starting after month 3 (to explore the effect of possible immortal time between the first and second dispenses).

As a further sensitivity analysis, we calculated the E-value30 to explore the strength of unmeasured confounding that would be necessary to shift ARRs to 1.0 (a null result) or to shift 95% CIs to include 1.0.

Analyses were conducted with SAS Enterprise Guide, V.7.1 and 8.2 (SAS Institute). Rate ratios were considered statistically significant if the 95% CI did not include 1.0. Additional methodological details are available in online supplemental file 1.

Results

Study population

We identified 36 319 adult members with DM2 who initiated third-step DPP4, GLP1, insulin, SGLT2, or TZD and had an indication of ASCVD, CKD, or HF during the analysis period (figure 1). Of these, 27 542 satisfied all criteria to be included in the analysis, with 19 958 (72.5%) in the ASCVD subgroup, 14 577 (52.9%) in the CKD subgroup, and 3919 (14.2%) in the HF subgroup. These groups are not mutually exclusive: 26.4% of patients are in two subgroups, and 6.6% are in all three subgroups. (See online supplemental table E for the effects of inclusion/exclusion criteria within each treatment group, and online supplemental table F for details of the overlap between patient subgroups.)

Figure 1

Request permission

Flowchart of patient inclusion and exclusion criteria. ASCVD, atherosclerotic cardiovascular disease; CKD, chronic kidney disease; DPP4, dipeptidyl peptidase-4 inhibitors; GLP1, glucagon-like peptide 1 receptor agonists; HF, heart failure; SGLT2, sodium-glucose cotransporter 2 inhibitors; TZD, thiazolidinediones.

Patient characteristics in the three subgroups are shown in table 1. Patients in the HF subgroup, when compared with the ASCVD and CKD subgroups, were less likely to be non-Hispanic Asian/Pacific Islander or Hispanic race/ethnicity, more likely to be non-Hispanic white race/ethnicity, more likely to have a low or missing baseline HbA1c, more often initiated third-step insulin, less often initiated third-step TZD, and had shorter median exposure time. Patients in the ASCVD subgroup had lower median CCI than patients in the CKD or HF subgroups.

Table 1

•

Patient characteristics in the ASCVD, CKD, and HF subgroups

Within each subgroup, patients in the GLP1 and SGLT2 third-step treatment groups generally had shorter median exposure times, smaller fractions in the oldest age group, less use of second-step metformin+sulfonylurea, and more use of second-step metformin+insulin when compared with patients in the other treatment groups (see online supplemental table G). Mean HbA1c change around third step was approximately a reduction of 1 unit across treatment groups, consistent with the literature for treatments other than insulin3 and in the insulin group likely due to the short time window. A recent network meta-analysis demonstrated similar HbA1c reductions for different combinations of dual therapy added to metformin.31

Outcomes

All patients in each subgroup were included in the model for the mortality outcome, but patients were excluded from the models for other outcomes if they had an event in the year before starting third-step treatment or if they had only 1 month of exposure time. Overall patient counts and unadjusted rates for the mortality, MACE, and HF hospitalization outcomes are displayed in table 2. Patients in the HF subgroup had approximately double the rate of mortality and MACE when compared with the ASCVD and CKD subgroups and had approximately five times the rate of HF hospitalization.

Table 2

•

Overall patient counts and unadjusted rates of all-cause mortality, MACE, and HF hospitalization in each patient subgroup

PS were calculated separately for patients in each outcome model in each subgroup (see online supplemental figures H,I, for example distributions of predicted probabilities and weights for the mortality outcome in the ASCVD subgroup). After weighting, the maximum pairwise ASMD for all variables was less than 0.25 except for the second-step sulfonylurea indicator (all models), one site (ASCVD and HF subgroup models), the “75 and up” age group (CKD subgroup models), one Charlson comorbidity (HF in the CKD subgroup mortality model and peripheral vascular disease (PVD) in the HF subgroup mortality model), and an additional site (HF subgroup HF hospitalization model) (see online supplemental table J). However, the outcome models included second-step medication indicators, site, age group, and the CCI (which includes HF and PVD) as covariates so there was a secondary control for these variables when modeling outcomes.

The ARRs from the Poisson models for all-cause mortality, MACE, and HF hospitalization in the ASCVD, CKD, and HF patient subgroups are displayed in figure 2 (see online supplemental table K for unadjusted rate ratio and ARR values).

Figure 2

Request permission

Adjusted rate ratios with 95% CIs from Poisson models of all-cause mortality, MACE, and HF hospitalization. ASCVD, atherosclerotic cardiovascular disease; CKD, chronic kidney disease; DPP4, dipeptidyl peptidase-4 inhibitors; GLP1, glucagon-like peptide 1 receptor agonists; HF, heart failure; MACE, major adverse cardiovascular event; SGLT2, sodium-glucose cotransporter 2 inhibitors; TZD, thiazolidinediones.

In the all-cause mortality and MACE models, GLP1 and SGLT2 are associated with a benefit (ARR less than 1.0) in all three patient subgroups when compared with the other third-step treatments. The ARRs for mortality are statistically significant except in the case of GLP1 versus DPP4 in the HF subgroup. The ARRs are often statistically significant in the MACE models (with the exceptions of GLP1 vs DPP4 and TZD in the ASCVD subgroup; GLP1 and SGLT2 vs DPP4 and TZD in the CKD subgroup; and GLP1 vs DPP4 in the HF subgroup).

In the HF hospitalization models, GLP1 and SGLT2 are associated with a benefit in all comparisons except in the case of GLP1 versus TZD in the ASCVD subgroup, where the ARR is 1.05 but not statistically significant (95% CI 0.69 to 1.58). In many of the other cases, the benefit is statistically significant (with the exceptions of GLP1 vs DPP4 and insulin and SGLT2 vs TZD in the ASCVD subgroup; and GLP1 and SGLT2 vs TZD in the CKD and HF subgroups).

In all models, the ARR between GLP1 and SGLT2 is not statistically significant.

The sensitivity analyses conducted for the all-cause mortality outcome in the ASCVD subgroup produced results qualitatively similar to those in the primary analysis (see online supplemental table L). The ARRs between GLP1 and SGLT2 versus the other third-step medications are all 1.0 or below, and in most cases the ARR from the primary model falls within the 95% CI of the sensitivity analysis model (and vice versa). In particular, the model that excluded the first 3 months of third-step treatment suggests that the possibility of immortal time between a patient’s first and second dispenses has little effect on the results. Two sensitivity models produced ARRs that suggest a noticeably larger benefit (ARR even smaller than 1.0) than in the primary model: the model excluding the year 2020 (for ARRs between GLP1 and other third-step treatments), and the model requiring second-step metformin+sulfonylurea (for SGLT2 vs TZD). We expect that these differences are due to small patient and event counts, and the fact that the ARR still shows a beneficial effect for GLP1 and SGLT2 is consistent with our primary results.

The ARRs for the covariates in the models are shown in online supplemental table M, and E-values calculated for the outcomes from the primary models are shown in online supplemental table N. In most cases, the E-values for significant ARR point estimates are larger than the largest significant ARR associated with covariates in the models (online supplemental table O), suggesting that an unmeasured confounder would need to have a stronger association with both treatment and outcome than the other covariates in the model, to explain away the observed effect. The exceptions (where the E-values are smaller than the largest significant covariate ARRs) are the ARRs for GLP1 and SGLT2 versus DPP4 in the CKD subgroup mortality outcome model; the ARRs in the CKD subgroup HF hospitalization outcome model; and the ARRs for SGLT2 versus DPP4 and insulin in the HF subgroup HF hospitalization model.

Discussion

In this retrospective cohort analysis of patients with DM2 as well as ASCVD, CKD, or HF who initiated third-step glucose-lowering treatment, we found that SGLT2 and GLP1 are generally associated with a benefit (compared with third-step DPP4, TZD, and insulin) in the rates of all-cause mortality, MACE, and HF hospitalization.

To our knowledge, this is the first head-to-head comparison of third-step treatments in patients with diabetes as well as ASCVD, CKD, or HF. Several previous studies have addressed cardiovascular benefits of GLP1 or SGLT2 compared with other medications, often including results for a specific step of diabetes treatment or for patients with additional conditions, but we are unaware of previous results for patients that are both initiating third-step treatment and have ASCVD, CKD, or HF. In the following paragraphs, we first compare our findings with results from the literature for a specific step of treatment (where additional conditions were not required, or where it was not the third step); then with results for patients with additional conditions (that were not required to be at a specific step of treatment); and finally with results that did not specify additional conditions or a step of treatment. Results from these studies are generally consistent with ours in that they show a benefit of GLP1 or SGLT2 compared with other medications (although statistical significance varies by study).

Previous observational studies with results for a specific step but not additional conditions showed a MACE and mortality benefit for third-step GLP1 versus insulin13; a MACE and mortality benefit for second-step GLP1 versus DPP414; a MACE benefit for third-step GLP1 versus DPP414; and a mortality benefit for fourth-step SGLT2 versus insulin.15 One study14 also showed a MACE benefit for second-step GLP1 versus DPP4 among patients with ASCVD.

Previous observational studies with results for patients with ASCVD that did not specify the step of treatment showed a mortality, MACE, and HF hospitalization benefit for GLP1 versus DPP416; a MACE benefit for SGLT2 versus DPP417; a mortality benefit for SGLT2 versus GLP118; and a MACE and HF hospitalization benefit for SGLT2 versus GLP1.18 19 Previous observational studies with results for patients with HF that did not specify the step of treatment showed an HF hospitalization benefit for SGLT2 versus DPP417; and a mortality, MACE, and HF hospitalization benefit for SGLT2 versus insulin and GLP1 versus insulin.20 One study with results for patients with CKD starting second-step SGLT2 found a mortality and MACE benefit versus not starting SGLT2.21

Among previous studies that did not include results for specific steps of treatment or patients with additional conditions, meta-analyses of randomized controlled trials (RCTs) showed mortality and HF hospitalization benefits for GLP1 versus DPP4 and SGLT2 versus DPP4,32 33 and MACE benefits for GLP1 versus DPP4 and SGLT2 versus DPP4.33 Observational studies with results comparing SGLT2 versus DPP4 showed an SGLT2 benefit for mortality and HF hospitalization17 34 35 and an SGLT2 benefit for MACE.17 34 Three of these studies included results showing a benefit of SGLT2 versus GLP1 for mortality and HF hospitalization32 33 35 which is consistent with our results in the ASCVD subgroup but not in the CKD or HF subgroups. One possible explanation for this discrepancy is that our models do not include the severity of the conditions defining the patient subgroups (ASCVD, CKD, or HF). For example, since SGLT2 is known from RCTs and observational studies to have more of a benefit than GLP1 for HF,36–38 patients with more severe HF may have been channeled into the SGLT2 treatment group in a way that our PS model could not correct. Unfortunately, we did not have the resources to assess disease severity with chart review.

Our analysis has several limitations. The use of International Classification of Diseases codes as surrogates for the presence of disease is subject to coding errors (such as miscoding or missing codes) and does not consider disease severity. For example, our PS models account for comorbidities (such as the presence of CKD in other subgroups), but not their severity (with variables such as estimated glomerular filtration rate (eGFR)). Because SGLT2 is contraindicated for patients with low eGFR,3 patients with more severe CKD may have been channeled out of the SGLT2 treatment group in a way our PS model could not correct. Our requirement that patients have at least two dispenses of third-step treatment was intended to focus the analysis on long-term outcomes rather than short-term safety issues but led to the possibility of immortal time if a patient’s first and second dispenses were in different calendar months. However, the sensitivity analysis excluding the first 3 months of treatment from the exposure time suggests a relatively small effect of this bias that does not qualitatively alter our findings. Our definition of third-step treatment includes patients who add a third medication, as well as patients who change one of the components of their second-step treatment, so we cannot distinguish between these two groups. We were unable to account for patients who obtained GLP1 from weight loss clinics outside of KP, although we expect this would be rare due to the associated expense. Our models use a limited set of covariates that do not include socioeconomic status, conditions outside of the CCI, or medications for conditions other than diabetes (eg, antihypertensives and lipid-lowering medications).39 We did not include the length of time that patients had DM2, ASCVD, CKD, HF, or uncontrolled HbA1c before third step. Extended duration of comorbid disease or uncontrolled HbA1c each could increase the baseline mortality rate. Despite our use of IPTW to mitigate the risk of confounding, there is always the threat that unmeasured confounders could affect an observational analysis in unknown ways. A small number of baseline covariates were not well balanced by the PS model, but we included the variables in the outcome models in an attempt to control for the differences. We did not model the decision to choose a new medication. Our treatment definition combines generic medications in the same class that could have different associations with the outcomes and does not include information about dose. Changes in clinician or patient behaviors during the period of the study could have affected the results, although we did not see evidence of a qualitative change in our sensitivity analysis that excluded data from 2020. All patients were part of KP’s integrated care model, so these results may not generalize to other models of care (especially those that do not focus on stepwise treatment guidelines) but given the diversity and size of the population, our results have broad implications.40

Our analysis has several strengths. Although an RCT is the gold standard in evidence-based medicine,41 it is unlikely that an RCT will be undertaken that explores head-to-head comparisons of third-step treatments in patients with diabetes as well as additional conditions. This observational analysis of thousands of patients in a real-world clinical setting contributes evidence that is unlikely to be found in a different way. Our study population includes patients with characteristics that are often under-represented in RCTs,42 43 including patients who are older, non-white, and have multiple conditions besides diabetes.

Conclusions

This study of patients with DM2 as well as ASCVD, CKD, or HF who initiated third-step diabetes treatment adds to evidence elsewhere in the literature that SGLT2 and GLP1 medications are associated with lower rates of all-cause mortality and MACE in these patient subgroups when compared with other common third-step treatments; and that SGLT2 is associated with lower rates of HF hospitalization. Our models also suggest that third-step GLP1 is associated with lower HF hospitalization rates, which is not supported by RCTs but may be explained by the lack of HF severity in our models. This work could be used to support guideline recommendations specifically for the use of SGLT2 and GLP1 as third-step treatments in these patient subgroups.

Supplementary PDF

Contributors: All authors contributed to the study design. TAM and JK conducted data collection and analysis, and all authors contributed to the interpretation of the results. TAM, JK, and JLA wrote the initial draft of the manuscript, and EGL and JPM contributed revisions. All authors critically reviewed the manuscript and approved the final version. JLA is the guarantor of this work, has access to the data, and takes responsibility for the decision to submit this work.
Funding: The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.
Competing interests: JPM provided paid consulting services to an expert panel with the American Academy of HIV Medicine. All other authors declare no competing interests.
Provenance and peer review: Not commissioned; externally peer reviewed.
Supplemental material: This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.

Data availability statement

No data are available.

Ethics statements

Patient consent for publication:

Ethics approval:

This study involves human participants and was approved by the Kaiser Permanente Southern California Institutional Review Board (ID #12799). The requirement for informed consent was waived for this pre-existing, data-only study.

Zelniker TA, Wiviott SD, Raz I, et al. SGLT2 inhibitors for primary and secondary prevention of cardiovascular and renal outcomes in type 2 diabetes: a systematic review and meta-analysis of cardiovascular outcome trials. Lancet 2019; 393:31–9.
doi:10.1016/S0140-6736(18)32590-X•Google Scholar
Kristensen SL, Rørth R, Jhund PS, et al. Cardiovascular, mortality, and kidney outcomes with GLP-1 receptor agonists in patients with type 2 diabetes: a systematic review and meta-analysis of cardiovascular outcome trials. Lancet Diabetes Endocrinol 2019; 7:776–85.
doi:10.1016/S2213-8587(19)30249-9•Google Scholar
Davies MJ, D’Alessio DA, Fradkin J, et al. Management of hyperglycemia in type 2 diabetes, 2018. A consensus report by the American Diabetes Association (ADA) and the European Association for the study of diabetes (EASD). Diabetes Care 2018; 41:2669–701.
doi:10.2337/dci18-0033•Google Scholar
Buse JB, Wexler DJ, Tsapas A, et al. 2019 update to: management of hyperglycemia in type 2 diabetes, 2018. A consensus report by the American Diabetes Association (ADA) and the European Association for the study of diabetes (EASD). Diabetes Care 2020; 43:487–93.
doi:10.2337/dci19-0066•Google Scholar
Diabetes Prevention Program Research Group. Diabetes prevention program research group. long-term effects of lifestyle intervention or metformin on diabetes development and Microvascular complications over 15-year follow-up: the diabetes prevention program outcomes study. Lancet Diabetes Endocrinol 2015; 3:866–75.
doi:10.1016/S2213-8587(15)00291-0•Google Scholar
American Diabetes Association Professional Practice Committee. 9. pharmacologic approaches to Glycemic treatment: standards of care in diabetes – 2024. Diabetes Care 2024; 47:S158–78.
doi:10.2337/dc24-S009•Google Scholar
National Institute for health and care excellence type 2 diabetes in adults: management guideline (NG28). 2022;
Available: here [Accessed 28 Aug 2023]
Google Scholar
Evans C, Emptage N, Rushkin M, et al. Cardiovascular benefits and harms of step therapy in the treatment of type 2 diabetes in adults with and without atherosclerotic cardiovascular disease. 2018;
Available: here [Accessed 22 Mar 2023]
Google Scholar
Perkovic V, Jardine MJ, Neal B, et al. Canagliflozin and renal outcomes in type 2 diabetes and nephropathy. N Engl J Med 2019; 380:2295–306.
doi:10.1056/NEJMoa1811744•Google Scholar
Zinman B, Wanner C, Lachin JM, et al. Empagliflozin, cardiovascular outcomes, and mortality in type 2 diabetes. N Engl J Med 2015; 373:2117–28.
doi:10.1056/NEJMoa1504720•Google Scholar
Ayalew MB, Spark MJ, Quirk F, et al. Potentially inappropriate prescribing for adults living with diabetes mellitus: a scoping review. Int J Clin Pharm 2022; 44:860–72.
doi:10.1007/s11096-022-01414-7•Google Scholar
Caughey GE, Roughead EE, Vitry AI, et al. Comorbidity in the elderly with diabetes: identification of areas of potential treatment conflicts. Diabetes Res Clin Pract 2010; 87:385–93.
doi:10.1016/j.diabres.2009.10.019•Google Scholar
Anyanwagu U, Mamza J, Mehta R, et al. Cardiovascular events and all-cause mortality with insulin versus glucagon-like Peptide-1 analogue in type 2 diabetes. Heart 2016; 102:1581–7.
doi:10.1136/heartjnl-2015-309164•Google Scholar
Svanström H, Ueda P, Melbye M, et al. Use of Liraglutide and risk of major cardiovascular events: a register-based cohort study in Denmark and Sweden. Lancet Diabetes Endocrinol 2019; 7:106–14.
doi:10.1016/S2213-8587(18)30320-6•Google Scholar
Wong CKH, Tang EHM, Man KKC, et al. SGLT2i as fourth-line therapy and risk of mortality, end-stage renal diseases and cardiovascular diseases in patients with type 2 diabetes mellitus. Diabetes Metab 2021; 47:101196.
doi:10.1016/j.diabet.2020.09.005•Google Scholar
Longato E, Di Camillo B, Sparacino G, et al. Better cardiovascular outcomes of type 2 diabetic patients treated with GLP-1 receptor agonists versus DPP-4 inhibitors in clinical practice. Cardiovasc Diabetol 2020; 19.
doi:10.1186/s12933-020-01049-w•Google Scholar
Filion KB, Lix LM, Yu OH, et al. Sodium glucose cotransporter 2 inhibitors and risk of major adverse cardiovascular events: multi-database retrospective cohort study. BMJ 2020; 370.
doi:10.1136/bmj.m3342•Google Scholar
Patorno E, Htoo PT, Glynn RJ, et al. Sodium-glucose cotransporter-2 inhibitors versus glucagon-like Peptide-1 receptor agonists and the risk for cardiovascular outcomes in routine care patients with diabetes across categories of cardiovascular disease. Ann Intern Med 2021; 174:1528–41.
doi:10.7326/M21-0893•Google Scholar
Longato E, Di Camillo B, Sparacino G, et al. Cardiovascular outcomes of type 2 diabetic patients treated with SGLT-2 inhibitors versus GLP-1 receptor agonists in real-life. BMJ Open Diabetes Res Care 2020; 8.
doi:10.1136/bmjdrc-2020-001451•Google Scholar
Staszewsky L, Baviera M, Tettamanti M, et al. Insulin treatment in patients with diabetes mellitus and heart failure in the era of new antidiabetic medications. BMJ Open Diabetes Res Care 2022; 10.
doi:10.1136/bmjdrc-2021-002708•Google Scholar
Gonzalez Perez A, Vizcaya D, Sáez ME, et al. Cardiovascular and renal outcomes among patients with type 2 diabetes using SGLT2 inhibitors added to metformin: a population-based cohort study from the UK. BMJ Open Diabetes Res Care 2023; 11.
doi:10.1136/bmjdrc-2022-003072•Google Scholar
von Elm E, Altman DG, Egger M, et al. The strengthening the reporting of observational studies in epidemiology (STROBE) statement: guidelines for reporting observational studies. J Clin Epidemiol 2008; 61:344–9.
doi:10.1016/j.jclinepi.2007.11.008•Google Scholar
Agency for healthcare research and quality healthcare cost and utilization project clinical classifications software for ICD-9-CM. 2017;
Available: here [Accessed 12 Mar 2021]
Google Scholar
Agency for healthcare research and quality healthcare cost and utilization project clinical classifications software refined. 2021;
Available: here [Accessed 12 Mar 2021]
Google Scholar
Klompas M, Eggleston E, McVetta J, et al. Automated detection and classification of type 1 versus type 2 diabetes using electronic health record data. Diabetes Care 2013; 36:914–21.
doi:10.2337/dc12-0964•Google Scholar
Quan H, Sundararajan V, Halfon P, et al. Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data. Med Care 2005; 43:1130–9.
doi:10.1097/01.mlr.0000182534.19832.83•Google Scholar
Xu S, Ross C, Raebel MA, et al. Use of stabilized inverse propensity scores as weights to directly estimate relative risk and its confidence intervals. Value Health 2010; 13:273–7.
doi:10.1111/j.1524-4733.2009.00671.x•Google Scholar
Austin PC. An introduction to propensity score methods for reducing the effects of confounding in observational studies. Multivariate Behav Res 2011; 46:399–424.
doi:10.1080/00273171.2011.568786•Google Scholar
Stuart EA, Lee BK, Leacy FP, et al. Prognostic score-based balance measures can be a useful diagnostic for propensity score methods in comparative effectiveness research. J Clin Epidemiol 2013; 66:S84–90.
doi:10.1016/j.jclinepi.2013.01.013•Google Scholar
VanderWeele TJ, Ding P. Sensitivity analysis in observational research: introducing the E-value. Ann Intern Med 2017; 167:268–74.
doi:10.7326/M16-2607•Google Scholar
Lautsch D, Alsumali A, McLeod E, et al. Comparative efficacy of dual and single initiation of add-on oral antihyperglycemic agents in type 2 diabetes uncontrolled on metformin alone: a systematic literature review and network meta-analysis. Diabetes Ther 2021; 12:389–418.
doi:10.1007/s13300-020-00975-y•Google Scholar
Zheng SL, Roddick AJ, Aghar-Jaffar R, et al. Association between use of sodium-glucose cotransporter 2 inhibitors, glucagon-like peptide 1 agonists, and dipeptidyl peptidase 4 inhibitors with all-cause mortality in patients with type 2 diabetes: a systematic review and meta-analysis. JAMA 2018; 319:1580–91.
doi:10.1001/jama.2018.3024•Google Scholar
Fei Y, Tsoi MF, Cheung BMY, et al. Cardiovascular outcomes in trials of new antidiabetic drug classes: a network meta-analysis. Cardiovasc Diabetol 2019; 18.
doi:10.1186/s12933-019-0916-z•Google Scholar
Sütő G, Molnár GA, Rokszin G, et al. Risk of morbidity and mortality in patients with type 2 diabetes treated with sodium-glucose cotransporter-2 inhibitor and/or dipeptidyl peptidase-4 inhibitor: a nationwide study. BMJ Open Diabetes Res Care 2021; 9.
doi:10.1136/bmjdrc-2020-001765•Google Scholar
Patorno E, Goldfine AB, Schneeweiss S, et al. Cardiovascular outcomes associated with canagliflozin versus other non-gliflozin antidiabetic drugs: population based cohort study. BMJ 2018; 360.
doi:10.1136/bmj.k119•Google Scholar
Palmer SC, Tendal B, Mustafa RA, et al. Sodium-glucose cotransporter Protein-2 (SGLT-2) inhibitors and glucagon-like Peptide-1 (GLP-1) receptor agonists for type 2 diabetes: systematic review and network meta-analysis of randomised controlled trials. BMJ 2021; 372.
doi:10.1136/bmj.m4573•Google Scholar
Brown E, Heerspink HJL, Cuthbertson DJ, et al. SGLT2 inhibitors and GLP-1 receptor agonists: established and emerging indications. Lancet 2021; 398:262–76.
doi:10.1016/S0140-6736(21)00536-5•Google Scholar
Blanco CA, Garcia K, Singson A, et al. Use of SGLT2 inhibitors reduces heart failure and hospitalization: a multicenter, real-world evidence study. Perm J 2023; 27:77–87.
doi:10.7812/TPP/22.137•Google Scholar
Schneeweiss S, Seeger JD, Maclure M, et al. Performance of comorbidity scores to control for confounding in epidemiologic studies using claims data. Am J Epidemiol 2001; 154:854–64.
doi:10.1093/aje/154.9.854•Google Scholar
Davis AC, Voelkel JL, Remmers CL, et al. Comparing Kaiser Permanente members to the general population: implications for generalizability of research. Perm J 2023; 27:87–98.
doi:10.7812/TPP/22.172•Google Scholar
Feinstein AR, Horwitz RI. Double standards, scientific methods, and epidemiologic research. N Engl J Med 1982; 307:1611–7.
doi:10.1056/NEJM198212233072604•Google Scholar
National Academies of Sciences, Engineering, and Medicine, Policy and Global Affairs. Improving Representation in Clinical Trials and Research: Building Research Equity for Women and Underrepresented Groups. Washington, DC, The National Academies Press 2022;
Google Scholar
Heiat A, Gross CP, Krumholz HM, et al. Representation of the elderly, women, and minorities in heart failure clinical trials. Arch Intern Med 2002; 162:1682–8.
doi:10.1001/archinte.162.15.1682•Google Scholar