Dimple Patel (1), Nadia Gaudel-dedieu (2), Mei Zhang (1), Fatiha (2), David Delvart (2),Zhaoling Meng (1)
(1) Translational Informatics, Sanofi, Bridgewater, NJ, USA (2) Translational Informatics, Sanofi, Chilly-Mazarin, FRANCE
Objectives: Following a successful cardiovascular (CV) outcome study EMPA-REG (1), SGLT2 inhibitor (SGLT2i) empagliflozin was approved for reducing the risk of cardiovascular death in adults with T2DM and established cardiovascular disease. AstraZeneca’s CVD-REAL study showed SGLT-2i, as a class, significantly reduced death and hospitalisations for heart failure versus other type-2 diabetes (T2DM) medicines using real world evidence data (2). We analyzed two real world data: an Electronic medical record (EMR) and a Claim database to assess SGLT2i CV benefit compared to other T2DM medications (standard of care or SOC). We explored the impacts of patient characteristics, such as baseline CV risk and renal status, on the CV benefit. The results can be applied to leverage real world data, assess effectiveness of an approved drug and inform the clinical trial design for a new drug.
Methods: Quintiles/GE EMR (1995-2016) contains sparse patient-level data and prescriptions may or may not be filled, however also contains large number of laboratory data. Truven database (2008-2016) contains longitudinal patient-level claim data from inpatient as well as outpatient and facility settings, but limited laboratory data. T2DM patients with at least one SGLT2i prescription, without any SGLT2i prescription but 1 recorded activity in (GE EMR) and without any SGLT2i but other new antidiabetic medication prescription (Truven) in the time period were included. A one-to-one propensity score method was applied to select non-SGLT2i patients (SOC) to match SGLT2i patients’ baseline characteristics. CV endpoints of interest included MI, stroke, CV death (MACE), and hospitalization due to heart failure (HHF). Three patient populations for all patients, patients with prior CV risk (with any prior CAD, CD, PVD or events of interest), and patients with prior CV risk/renal impairment were analyzed. ICD 9 and ICD 10 codes were used to search for type 2 and CV events. Prxmatch search function and NDC codes were used to search for SGLT2 drugs. First SGLT2i date, non SGLT2i selected activity date (GM EMR) and the new non SGLT2 first prescription (Truven) in the time period were used as the real world evidence study start date. Time to first CV event (MACE or MACE + heart failure (HF)) was analyzed using the COX proportional hazards model. The CV treatment differences between SGLT2i and SOC were presented for all 3 populations.
Results: For GE EMR and Truven Claim, more than 70,000 and ~200,000 T2DM patients on SGLT2i and the same number matched SOC patients were included. For GE EMR, baseline variables age, sex, diabetes duration, race, HbA1c, eGFR, prior CV risk, BMI, SBP/DBP, LDL/HDL, etc. were included in the propensity score matching. For Truven Claim, only limited baseline variables such as age, sex, diabetes duration and prior CV risk were included in the matching due to the database limitation. For GE EMR, stroke, MI and HF only were analyzed since no death or consistent hospitalization information was available. For CV endpoints of stroke + MI and stroke + MI + HF, statistically signification CV risk reductions ~ 10% to 20% for SGLT2i versus SOC were observed with HF risk reduction having a larger effect size in all patient population. For prior CV risk population, a similar trend was observed without statistical significance. For prior CV risk and moderate renal impairment (baseline eGFR ≤ 60 mL/min/1.73 m2) population, the sample size was too small to conclude. For Truven, deaths in circulation system were used instead of CV death and CKD status was used for renal impairment due to lack of eGFR data. HFs were analyzed to keep the consistency with EMR data, even though hospitalization information is available. Consistent treatment benefits were observed for both MACE and MACT+HF endpoints for all 3 populations.
Conclusions: Though there are issues in real-world data collection such as lack of quality and content, the real world data can contribute in evaluating patients’ characteristics, treatment effects and population characteristic’s impact. The results can be used to inform the clinical study design for a similar treatment. Due the differences between the real world and clinical data, relative effect magnitudes not the exact effect sizes should be referenced and further researches are needed to account for potential bias and confounding.
References:
[1] Zinman B, et. Al. Empagliflozin, Cardiovascular Outcomes, and Mortality in Type 2 Diabetes. N Engl J Med 2015; 373:2117-2128.
[2] Kosiborod M, et. AL. Circulation. 2017;136:249-259. http://circ.ahajournals.org/content/136/3/249
Reference: PAGE 27 (2018) Abstr 8719 [www.page-meeting.org/?abstract=8719]
Poster: Methodology - Other topics