1398-P: Emulating Target Trials Using Instrumental Variables to Inform Care—Add-on Therapy to Metformin

Introduction & Objective: Guidelines now rely on studies using routinely-collected clinical data to emulate a ‘target trial’, a hypothetical randomized controlled trial (RCT) one would wish to do. A key question remains what drug to add to metformin to improve glycemic control in people with type 2 diabetes (T2D). To inform care for people excluded from RCTs, we developed a target trial comparing add-on therapy with dipeptidyl peptidase-4 inhibitors (DPP4) vs. sulfonylureas (SU), a relevant choice in many settings. We used an instrumental variable as a ‘natural randomizer’.

Methods: We used anonymized data from the UK Clinical Practice Research Datalink (2011 to 2015) to identify people with T2D on metformin monotherapy prescribed <60 days before starting 2^nd-line DPP4 or SU. We recreated the inclusion criteria of a ‘real’ RCT (Nauck et al. 2007) that tested non-inferiority of sitagliptin vs. glipizide on HbA1c at 1 year. We then broadened the eligibility to include people over age 78 years or with HbA1c >10% (85.8 mmol/mol), which we analyzed separately. We used an instrumental variable (IV), a GPs’ preference for prescribing either DDP4 or SU, to test the effect of treatment on HbA1c at 1 year, and to minimize confounding. Our study did not estimate complications or adverse effects.

Results: We identified 13,240 people for the target trial: 6,497 met and 6,743 would not have met the criteria for the real RCT (1172). The mean baseline HbA1c by subgroup was 68.2 (s.d. 8.3) and 79.4 (22.3) mmol/mol. The difference in HbA1c at 1 year for DPP4 vs. SU (negative reflects lower values for DPP4) for people in the target trial who met the real RCT criteria was -0.3 (-2.3, 1.6) compared with 0.2 (-0.9, 1.3) for the actual trial. For those in the target trial who would have been excluded from the real RCT, the difference was -3.4 (-6.6, -0.1).

Conclusion: Emulating a target trial provides evidence when none exists. If IV assumptions hold, then DPP4 appear more effective than SU in the subgroup of patients excluded from a key trial.