How Real-world Data Augments What We Know About SGLT-2 Inhibitors for Cardiovascular Risk Reduction

Quick Takes

  • This large analysis of routinely collected 'real-world' data, evaluated over 386,000 matched participants from a cohort of over 2.4 million patients from 14 countries with type 2 diabetes commencing a glucose-lowering therapy.
  • When compared with those commencing a DPP4 inhibitor, new-users of SGLT-2 inhibitors experienced fewer adverse cardiovascular outcomes including hospitalization for heart failure, myocardial infarction, stroke, and death.
  • Although residual confounding has likely amplified the observed estimates of treatment effect, these data from routine clinical care in a comparably younger, and lower risk population highlight an important role of "real world" observations in confirming effectiveness in broader, non-trial populations.

Multiple clinical trials demonstrate the efficacy of sodium-glucose co-transporter-2 (SGLT-2) inhibitors to reduce cardiovascular events in persons with diabetes.1-4 Use of these agents has been associated with consistent reductions in the combined rates of cardiovascular death, myocardial infarction (MI) and stroke (HR 0.88, 0.82-0.94) and prevention of heart failure hospitalization (HR 0.68, 0.60-0.76), benefits not seen with dipeptidyl peptidase-4 (DPP-4) inhibitors.5 SGLT-2 inhibitors are now endorsed for cardiovascular risk reduction in Society guidelines6,7 and consensus documents.8

However, while randomized clinical trials (RCTs) remain the gold standard for determining efficacy of a new treatment, multiple factors can impact the degree to which these results translate into effectiveness in populations treated outside of a trial. This includes differences in the population studied in the trial compared with the population treated, variability in adherence and persistence to therapy in routine use, and differences in background therapies.  Understanding the "real world" effectiveness of therapy requires an assessment of treatments outside of the controlled environment of a trial.

"Real-world data" are data generated not explicitly for research but rather for clinical care, including billing systems, pharmacy claims transactions, or electronic health record (EHR) data. These types of data allow for the study of treatment effectiveness and utilization in a broader population of patients and providers than evaluated in a randomized trial. Some object to the term "real-world", arguing that clinical trials are conducted in the same literal "world" as therapies are used. However, in an RCT, patients are carefully screened, data are prospectively collected specifically for the trial, and the participation in a trial itself selects for a certain type of motivated patient and provider.

The study by Kohsaka et al. in July's edition of Lancet Diabetes and Endocrinology is an excellent example of how real-world data can be leveraged for research. In this analysis, the authors report on the cardiovascular outcomes of patients treated with SGLT-2 inhibitors compared to those treated with DPP-4 inhibitors.9 This analysis is performed within the broader CVD-REAL observational platform which has obtained de-identified and routinely collected health system data of patients commencing an oral glucose-lowering drug in 14 countries across the globe. The data from CVD-REAL are derived from a variety of different healthcare sources including EHR, prescribing, administrative claims, and registry data. The platform acts as a distributed data network where data partners (health systems) use similar code lists and a common protocol to identify patients and ascertain endpoints. In this model, patient-level data remains with the participating data partner, and aggregate outputs are shared with the coordinating center. This approach satisfies local privacy and data provenance regulations, obviates central transformation, and avoids the need to share large quantities of data.

To compare the outcomes of patients treated with each agent, the authors first had to identify comparable cohorts. Unlike an RCT where patients are randomly assigned a therapy, in clinical practice physicians use several factors to determine which treatment to use, including patient clinical characteristics, patient access to therapy, and cost. Supplement table 6 in the manuscript shows significant differences in those treated with SGLT-2 inhibitors and DPP-4 inhibitors, with those on SGLT-2 inhibitors younger and less likely to have prior cardiovascular disease or chronic kidney disease. To account for these differences, the authors created a propensity score by modeling factors associated with initiation of a DPP-4 inhibitor or an SGLT-2 inhibitor. Next, they matched patients on this score to generate two cohorts of patients with the same propensity to receive an SGLT-2 versus DPP-4 inhibitor. For this analysis, 2 413 198 participants were identified as having type 2 diabetes mellitus and had newly commenced either an SGLT-2 inhibitor (n=230,721, 9.6%) or DPP-4 inhibitor (n=2,182,477. 90.4%), out of which a subset of 193,124 patients were selected for analysis. Table 1 of the manuscript shows how the propensity score approach created a population with more similar baseline characteristics than had they used the entire sample. Compared with DPP-4 inhibitor initiators, new users of SGLT-2 inhibitors had a significantly lower risk of hospitalization for heart failure (HR 0.69, 95%CI 0.61-0.77), death (HR 0.59, 0.52-0.67) and the HF/death composite (0.64, 0.57-0.72). The risk of MI (HR 0.88, 0.80-0.98) and stroke (HR 0.85, 0.77-0.93) were also modestly reduced.

While these results are directionally consistent with what was shown in the RCTs, the magnitude of benefit is larger. Is it possible that the RCTs under-estimate the "real-world" benefit? In this case, unlikely. If real-world cohorts are higher risk than in trials, one may expect a larger magnitude of benefit. Yet, the CVD-REAL cohort was actually lower risk overall (Table 1) compared to the RCTs with a lower mean age, a greater proportion of women and proportionally fewer patients with already established cardiovascular disease. These differences in the CVD-REAL population translated to lower event rates across all of the endpoints when compared with the RCTs (Figure 1). In particular, the rate of all-cause mortality, which is generally least affected by differences in event ascertainment and definition, was noticeably lower in CVD-REAL than any of the RCTs. More likely, these results indicate some degree of residual confounding.

Table 1: Characteristics of trials evaluating SGLT-2 inhibitors compared to DPP4i arm of CVD-REAL study.

Table 1

Figure 1: Event rates per 100 patient years by study, grouped by endpoint.

Figure 1

What residual factors may have led to differences in outcomes between the two groups? The significant difference in the rate of DPP-4 inhibitors and SGLT-2 inhibitors is likely a result of a combination of lower access to SGLT-2 inhibitors and lower provider familiarity with the medication class during the study period. While the propensity score accounted for many measured patient-level characteristics, the limitations of the data meant adjustment for laboratory results (especially eGFR and HbA1c) was not possible for all patients, and non-cardiac comorbidities were either incompletely assessed (e.g. frailty) or not available (e.g. smoking, cancer, chronic lung disease). Provider-level factors also likely affected the choice of therapy. Given the DPP-4 inhibitor class had been commercially available for over 5 years, and two outcome trials reporting cardiovascular safety had read out [EXAMINE for alogliptin10 and SAVOR-TIMI 53 for saxagliptin11] prior to the study period, clinicians may have been less inclined to prescribe an unfamiliar class of medication (SGLT2-inhibitors) to higher risk or more complex patients. Unmeasured confounders such as these have potential to impact any "real-world" comparative effectiveness analysis. Often, the degree of unmeasured confounding can be evaluated by comparing the treatment effect on a "falsification endpoint," an event not associated with therapy but that can indicate differences in the underlying health of the population, such as malignancy. Unfortunately, such an analysis is not presented for CVD-REAL. Other limitations of the study are acknowledged by the authors. For example, the abbreviated follow up of just over 1 year was much shorter than in RCTs.

One missed opportunity for CVD-REAL is lack of safety data for the two therapies. In the future, platforms such as CVD-REAL should be leveraged to evaluate the safety of new therapies in addition to treatment efficacy. Large postmarketing safety studies using real-world data may be able to identify rare events that may not be detected in clinical trials, or that occur in populations that may not have been represented in pre-licensure studies.

Nevertheless, the analysis from the CVD-REAL platform has a number of strengths. As mentioned, the new-user design is sound from a pharmacoepidemiological perspective. By identifying people at the moment, they start a therapy, the authors can adjust for patient characteristics at that time. More importantly, this design avoids "immortal time" bias, which happens when patients who are already on therapy are selected because by design, they could not have died between the time they started therapy and the time they are selected.12 The choice of participating health systems resulted in over 80% of participants enrolled from the Asia-Pacific region, almost four-fold the representation achieved in the RCTs. Furthermore, while there is no substitute for randomization, a placebo-controlled trial of an SGLT-2 inhibitor no longer has clinical equipoise, and thus the choice of a contemporary but neutral active comparator in the study design is both elegant and important.

Ultimately, while the results of this analysis may be influenced by some residual selection bias, the directionally consistent effects confirm in the "real-world" what has been seen in clinical trials; SGLT-2 inhibitor use is associated with lower risk of adverse cardiovascular outcomes. Importantly, the use of a distributed data network model has achieved this observation in what appears to be a younger and lower risk population, and across a broad geography and diverse patient population. While these types of observational data will always be limited in their ability to confidently estimate treatment effects, they provide critical insight into post-marketing prescribing patterns and effectiveness in more diverse patient populations and health systems. CVD-REAL has shown such observations are possible from a wide array of data sources and across a vast network of countries that tend to be underrepresented in clinical trials. Combined with the known treatment benefit seen in RCTs, the data from CVD-REAL reinforce the importance of utilization of SGLT-2 inhibitors in high risk persons with diabetes to reduce cardiovascular disease events.

References

  1. Neal B, Perkovic V, Mahaffey KW, et al. Canagliflozin and cardiovascular and renal events in type 2 diabetes. N Engl J Med 2017;377:644-57.
  2. Perkovic V, Jardine MJ, Neal B, et al. Canagliflozin and renal outcomes in type 2 diabetes and nephropathy. N Engl J Med 2019;380:2295-2306.
  3. Wiviott SD, Raz I, Bonaca MP, et al. Dapagliflozin and cardiovascular outcomes in type 2 diabetes. N Engl J Med 2019;380:347-57.
  4. 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.
  5. Zelniker TA, Braunwald E. Clinical benefit of cardiorenal effects of sodium-glucose cotransporter 2 inhibitors: JACC State-of-the-Art Review. J Am Coll Cardiol 2020;75:435-47.
  6. American Diabetes Association. 10. Cardiovascular disease and risk management: standards of medical care in diabetes-2020. Diabetes Care 2020;43:S111-S134.
  7. Cosentino F, Grant PJ, Aboyans V, et al. 2019 ESC Guidelines on diabetes, pre-diabetes, and cardiovascular diseases developed in collaboration with the EASD. Eur Heart J 2020;41:255-323.
  8. Das SR, Everett BM, Birtcher KK, et al. 2020 expert consensus decision pathway on novel therapies for cardiovascular risk reduction in patients with type 2 diabetes: a report of the American College of Cardiology Solution Set Oversight Committee. J Am Coll Cardiol 2020;76:1117-45.
  9. Kohsaka S, Lam CSP, Kim DJ, et al. Risk of cardiovascular events and death associated with initiation of SGLT2 inhibitors compared with DPP-4 inhibitors: an analysis from the CVD-REAL 2 multinational cohort study. Lancet Diabetes Endocrinol 2020;8:606-15.
  10. White WB, Cannon CP, Heller SR, et al. Alogliptin after acute coronary syndrome in patients with type 2 diabetes. N Engl J Med 2013;369:1327-35.
  11. Scirica BM, Bhatt DL, Braunwald E, et al. Saxagliptin and cardiovascular outcomes in patients with type 2 diabetes mellitus. N Engl J Med 2013;369:1317-26.
  12. Suissa S. Immortal time bias in pharmaco-epidemiology. Am J Epidemiol 2008;167:492-9.

Clinical Topics: Diabetes and Cardiometabolic Disease, Dyslipidemia, Heart Failure and Cardiomyopathies, Lipid Metabolism, Heart Failure and Cardiac Biomarkers

Keywords: Diabetes Mellitus, Metabolic Syndrome X, Diabetes Mellitus, Type 2, Sodium-Glucose Transporter 2, Cardiovascular Diseases, Dipeptidyl Peptidase 4, Electronic Health Records, Information Storage and Retrieval, Pharmacoepidemiology, Propensity Score, Privacy, Risk Factors, Sodium-Glucose Transporter 2, Hypoglycemic Agents, Consensus, Glucose, Follow-Up Studies, Random Allocation, Selection Bias, Treatment Outcome


< Back to Listings