A Bayesian hierarchical approach for multiple outcomes in routinely collected healthcare data

Area of Science:

• Health Services Research
• Biostatistics
• Pharmacovigilance

Background:

• Clinical trials have limitations in assessing rare outcomes and generalizability.
• Routinely collected healthcare data offers a valuable, albeit challenging, alternative data source.
• Existing statistical methods may not fully leverage the potential of real-world data.

Purpose of the Study:

• To propose and evaluate a Bayesian hierarchical model for analyzing routinely collected healthcare data.
• To assess the model's performance in detecting treatment outcomes and controlling error rates.
• To apply the model to real-world data on direct oral anticoagulants in Scotland.

Main Methods:

• Development of a Bayesian hierarchical model allowing population stratification into similar clusters.
• Application of the model to observational data from direct oral anticoagulant use in Scotland.
• Conducting a simulation study to rigorously assess the model's statistical performance.

Main Results:

• The Bayesian hierarchical model demonstrated effective outcome detection.
• The proposed method showed favorable error rates in simulation studies.
• The model was successfully applied to real-world data, providing insights into treatment performance.

Conclusions:

• Bayesian hierarchical modeling is a powerful tool for analyzing complex healthcare data.
• This approach enhances the evaluation of treatments beyond traditional clinical trials.
• The model offers a statistically sound method for leveraging real-world evidence.