Assessing performance of sequential analysis methods for active drug safety surveillance using observational data.

Area of Science:

• Pharmacovigilance
• Biostatistics
• Real-world data analysis

Background:

• Sequential methods are standard in clinical trials.
• Growing interest in applying sequential methods for post-market drug safety surveillance using real-world data.
• Limited application and understanding of sequential approaches for marketed drugs with real-world data.

Purpose of the Study:

• Compare the performance of conditional sequential sampling procedure (CSSP) and a log-linear model with Poisson distribution (LLMP).
• Evaluate methods for monitoring drug safety using longitudinal administrative health claims data.
• Determine suitability of sequential approaches for real-world data in pharmacovigilance.

Main Methods:

• Utilized two large longitudinal US administrative health claims databases.
• Compared CSSP (a group sequential method) with LLMP (using SAS PROC GENMOD and alpha-spending function).
• Assessed performance on 50 known drug-outcome associations.

Main Results:

• Neither CSSP nor LLMP identified all known drug-outcome associations.
• LLMP demonstrated superior ability and reduced time for identifying associations compared to CSSP.
• LLMP exhibited better computational performance but yielded more false positives than CSSP.

Conclusions:

• LLMP offers a potentially valuable alternative or complement to CSSP for drug safety monitoring.
• LLMP's flexible confounding control and ease of implementation are advantageous.
• Further research is needed to optimize sequential methods for real-world data in pharmacovigilance.