Using decision trees to determine participation in bundled payments in sepsis cases

The purpose of this research is to determine and develop a valid analytical method that can be easily implemented by providers to evaluate whether they should join the bundled payments for care improvement (BPCI) advanced bundled payment program, and analyze the projected impacts of BPCI advanced payment on their margins.

We have developed a decision tree model that incorporates the types of sepsis encountered and the resultant typical complications and associated costs.

The initial cost of a sepsis episode was $30,386. Since Medicare requires that there is a 3% cost reduction under BPCI, we applied the model with a 3% cost reduction across the board. Since the model considers probabilities of the complications and readmission, there was actually a 3.36% reduction in costs when the 3% reduction was added to the model. We applied 2-way sensitivity analysis to the intensive care unit (ICU) long and short costs. We used the unbundled cost at the high end, and a 10% reduction at the low end. Per patient episode cost varied between $28,117 and $29,658. This is a 5.2% difference between low and high end. Next, we looked at varying the hospital bed (non-ICU) costs. Here the resultant cost varied between $28,708 and $29,099. This is only a 1.34% difference between low and high ends. Finally, we applied a sensitivity analysis varying the attending physician and the intensivist reimbursement fees. The result was a cost that varied between $29,191 and $29,366 which is a difference of only 0.595%.

This is the precise environment where decision tree analysis modeling is essential. This analysis can guide the hospital in just how to allocate resources in light of the new BPCI advanced payment model.