Emulating a target trial for assessing the comparative effectiveness of a newly marketed medication: the example of angiotensin-neprilysin inhibitor in patients with heart failure

Background: Assessing comparative effectiveness in the early phase of marketing comes with methodological challenges such as potential prescribing channelling bias or rapid changes in the new users’ characteristics. Sacubutril+valsartan, an angiotensin receptor-neprilysin inhibitor (ARNI), introduced in 2016 in Sweden, reduced by 21% the risk of hospitalization for HF versus angiotensin-converting enzyme inhibitors (ACEi) in patients with heart failure with reduced ejection fraction (HFrEF) in the pivotal randomised clinical trial (RCT). However, a study based on Swedish register data showed that between 34 and 76% of patients with HFrEF in clinical practice are eligible for ARNI treatment according to the criteria used in the RCT.

Objectives: To report on the emulation of a target trial for evaluating the effectiveness of ARNI versus RASi alone, in adults with HFrEF using register data from the Stockholm Creatinine Measurements (SCREAM) project.

Methods: We emulated a target trial based on routine healthcare data from the period 2016-2021 from the SCREAM database, to compare the 3-year hospitalization risk for HF between ARNI and ACEi. To mimic the main RCT exclusion criteria, we excluded patients with a baseline estimated glomerular filtration rate < 30ml/min/1.73m2, NT-proBNP < 400pg/ml and serum potassium >5.4 mmol/l. We used an active comparator new switcher design. To further minimize confounding, we used inverse probability of treatment weighting based on the propensity score including 70 covariates (PS-IPTW) with stabilized weights with 1% truncation and applied asymmetric PS trimming (2.5th/97.5th perc.). The analyses were replicated with historical controls using data from 2011-2015 instead of 2016-2021, i.e. before ARNI marketing, for the comparator ACEi group only. We used a Cox proportional model to derive hazard ratios (HR) with 95% confidence intervals (CIs) in an intention-to-treat analysis.

Results: A total of 834 ARNI and 655 ACEi switchers with HFrEF were included. The ‘unweighted’, adjusted ‘PS-IPTW with truncation’ and ‘PS-IPTW with trimming’ HRs were 1.33 [1.08-1.63], 1.26 [0.94-1.67] and 1.35 [0.98 - 1.85], respectively, away from the expected reduction of HF hospitalizations. When replicating the study by using data from 2011-2015 for ACEi, the three HR estimates dramatically changed: 0.95 [0.78-1.16], 0.74 [0.54-1.02] and 0.68 [0.50-0.92], respectively, in line with the findings from both randomized and comparable observational studies.

Conclusions: When assessing comparative effectiveness of newly marketed medications, using data from the period before marketing to define the comparator group might help to reduce variation in results between RCTs and observational studies designed to emulate them.