Re-Thinking the Role of Statistics in Informing Heart Team Decisions: A Consensus Distribution Approach

To build a heart team consensus distribution, a biostatistician would ask each member of the team to independently provide a probability distribution, for example a survival curve, of an adverse outcome (blue lines) for different treatments a single patient may decide to undergo. Those individual predictions are then combined with a statistical algorithm. A common method to aggregate predictions into a consensus is a linear pool14, a weighted average of the heart team’s probability distributions (red line). Past work has shown consensus distributions outperform individual forecasts 15. The heart team consensus and their individual predictions can be compiled and sent to all members of the team and the patient to better inform optimal treatments. We stress that the consensus prediction would belong to a single patient, not an aggregate of patients like a traditional statistical model.