Creating synthetic populations in transplantation: A Bayesian approach enabling simulation without registry...

Area of Science:

• Transplantation research
• Computational biology
• Health policy analysis

Background:

• Current organ allocation simulations rely on historical registry data, which may be outdated or reflect past inequities.
• This limits the contemporary applicability and accuracy of simulation results for organ transplantation strategies.

Purpose of the Study:

• To develop and validate an alternative method for generating synthetic donor and candidate populations for transplant simulation.
• To overcome limitations associated with re-sampling historical transplant registry data.

Main Methods:

• Utilized hierarchical Bayesian network probability models to generate synthetic donor and candidate data.
• Developed two Bayesian networks modeling dependencies among 10 donor and 36 candidate characteristics.
• Estimated model parameters using data from the Scientific Registry of Transplant Recipients (SRTR).

Main Results:

• Generated synthetic populations closely matched observed data for categorical attributes (within 1% point) and continuous variables (median within IQR).
• Demonstrated the method's ability to capture complex joint variability among multiple characteristics.
• Showcased how altering population parameters can influence clinical outcomes in synthetic populations.

Conclusions:

• Generating synthetic populations offers a powerful alternative to historical data in transplant simulation.
• This method allows for customized population parameters, reflecting realistic or hypothetical future scenarios for policy evaluation.
• Enhances the adaptability and relevance of organ allocation strategy analysis.