(B-223) Challenges and Learnings from a Prescription Sequence Symmetry Analysis Replication Study to Inform a Reusable Negative Control Programming Tool

Background: The FDA’s 2023 efforts to accelerate the use of real-world evidence in decision-making highlighted negative controls as a tool for assessing confounding, reducing bias, and enabling robust conclusions from real world data (RWD).

Objectives: We performed a methods investigation to replicate a RWD study to inform our development of programming tools to support negative controls analysis in RWD.

Methods: The selected use-case was a retrospective study by Vouri et al, 2019 using MarketScan Commercial Claims and Encounters (CCAE) database between 2005 and 2017. Our replication study like Vouri evaluated the incidence and magnitude of a prescribing cascade involving the use of dihydropyridine calcium channel blockers (DH CCB) and loop diuretics using a prescription sequence symmetry approach (PSSA). Angiotensin-converting enzyme inhibitors and Angiotensin receptor blockers (ACEi/ARB) were employed as negative control exposures to assess the presence of the prescribing cascade as there is no expected association. Our analysis was conducted in Merative MarketScan CCAE from 2011-2017.

Results: We found that 1.49% of DH CCB initiators experienced the prescribing cascade, compared with 1.44% in Vouri. The magnitude of the prescribing cascade in our study versus Vouri was, respectively: adjusted sequence ratio (ASR) for DH CCB, [(1.86; 95%CI: 1.71-2.01)] vs (1.87; 95%CI: 1.84-1.90) and ASR for ACE/ARB, [(1.37; 95%CI: 1.31-1.44) vs (1.27; 95%CI: 1.24-1.29)]. Key challenges encountered in the replication included fewer years in our database, insufficient detail on the original study population and PSSA estimation formulas, lack of definitions for other hypertensive medications, and changes in hypertension treatment guidelines, and changes in data contributors, all of which could have impacted our results. One example was the difference observed patient counts (e.g. in 2012-2013, 133,557 DH CCB initiators in our study versus 389,015 in the original study). In the time periods captured by the database there were differences in our study and the original study (e.g. 2005-2017 vs 2011-2017). The resulting changes in data overtime may be due to transitions in health plan contributors as well as changes in data stewardship over the years.

Conclusions: The challenges encountered while conducting our study are valuable lessons which illustrate the nuances with replicating prior RWD studies as a validation process to inform decision making. Despite these challenges we were able to obtain similar results suggesting confidence in the programming tool we developed.