Prior event rate ratio adjustment produced estimates consistent with randomized trial: a diabetes case study.

Area of Science:

• Pharmacovigilance and Pharmacoepidemiology
• Biostatistics and Health Data Science
• Clinical Research and Drug Development

Background:

• Electronic health records (EHRs) are valuable for drug safety surveillance but are prone to confounding bias due to non-randomized treatment allocation in routine care.
• Assessing drug side effects using real-world data requires methods to mitigate unmeasured confounding.
• Second-line therapies for type 2 diabetes, such as thiazolidinediones and sulfonylureas, require robust safety evaluation.

Purpose of the Study:

• To evaluate the effectiveness of the Prior Event Rate Ratio (PERR) Pairwise method in reducing unmeasured confounding bias for drug side-effect estimates.
• To compare side-effect estimates derived from EHR data using the PERR Pairwise method with conventional statistical approaches and a randomized controlled trial.
• To assess the safety profiles of thiazolidinediones and sulfonylureas as second-line treatments for type 2 diabetes.

Main Methods:

• A case study utilizing the Prior Event Rate Ratio (PERR) Pairwise method was conducted.
• Primary care data from the Clinical Practice Research Datalink (n=41,871) were analyzed.
• Outcomes from the first-line metformin period were used to adjust for unmeasured confounding, with comparisons made to the A Diabetes Outcome Progression Trial (ADOPT) (n=2,545).

Main Results:

• Conventional Cox regression overestimated the risk of edema and identified a false association with gastrointestinal (GI) side effects for thiazolidinediones.
• The PERR Pairwise method yielded estimates for edema (1.43 [1.10, 1.83]) and GI side effects (0.91 [0.79, 1.05]) that were consistent with the ADOPT trial results (1.39 [1.04, 1.86] for edema and 0.94 [0.80, 1.10] for GI).
• Unmeasured confounding was suggested by increased risks of edema and GI side effects in patients prescribed thiazolidinediones after metformin therapy.

Conclusions:

• The Prior Event Rate Ratio (PERR) Pairwise approach shows promise for improving postmarketing surveillance of drug side effects using EHR data.
• Careful consideration of the underlying assumptions of the PERR Pairwise method is crucial for its effective application.
• This method can provide more reliable drug safety estimates compared to conventional regression techniques when dealing with observational data.