Table 1.

Steps to minimise confounding in the EMPRISE study design

Minimising confounding Key aspects of the EMPRISE study design
PS matching	Patients are 1:1 matched with a “nearest neighbour” based on 140 predefined baseline characteristics (“covariates”) Covariates include key factors relating to disease severity (such as # antidiabetic medications), comorbidities (such as CVD history) and many other clinical and demographic characteristics
Appropriate comparator choice	The most commonly prescribed DPP-4 inhibitor is the chosen active comparator to empagliflozin, owing to the similar position of DPP-4 inhibitors to SGLT2 inhibitors in the treatment pathway Using a comparator with a similar position is designed to maximise the similarity of disease severity between cohorts
No overlap between comparators	Patients are excluded if they had received any SGLT2 inhibitor or DPP-4 inhibitor in the year preceding cohort entry, and follow-up is terminated if a patient switches to the comparator Minimises the potential for immortal time bias
Sequential enrolment	PS matching is performed independently for each enrolment Ensures that study arms are balanced not just across the full cohort, but also for temporally matched populations
“As-treated” approach	Follow-up captures only outcomes occurring during treatment exposure + 30 days Minimises bias from confounding events not related to treatment
Assessing balance between cohorts Data used to independently confirm robustness of PS matching approach
Baseline laboratory scores	A range of laboratory scores at baseline are available for a subset of the population, including Hb1Ac, cholesterol and creatinine levels These scores are not used for PS matching, and so can provide an independent indication of equivalence between study arms
Sensitivity analyses In each case, the conclusions regarding HHF benefit with empagliflozin were unchanged
High-dimensional PS matching	PS matching with 100 additional covariates
Alternative comparator	The sitagliptin cohort is replaced with a cohort composed of patients receiving any DPP-4 inhibitor
Subgroup analyses	Subgroup analyses include:  With/without CVD at baseline  With/without HF at baseline  Gender  Empagliflozin dose
Alternative HHF definition	Broadening the definition of HHF from hospitalisation with HF in the primary discharge position to hospitalisation with HF in any discharge position
Control outcome	An outcome with an expected null finding (flu vaccination)

The EMPRISE study design used several approaches to minimise confounding [16], although undetected bias from residual confounding cannot be excluded. CVD cardiovascular disease, DPP-4 dipeptidyl peptidase-4, HF heart failure, HHF hospitalisation for HF, PS propensity score, SGLT2 sodium–glucose transporter 2