22.7 Patient Level Simulation/Microsimulation

In decision analysis, the most efficient calculation is to use expected values, as described in the section Building Healthcare Decision Trees and illustrated in this section. However, it is also possible to evaluate decision trees using individual-level simulation, sometimes referred to as patient level simulation or microsimulation.

Patient Level Simulation in decision trees approximates an expected value by “sampling” a representative distribution of paths through the model’s chance events. Microsimulation of complex models generally utilizes as many “trials” as time allows, in order to improve the EV estimation (ensuring even small probability paths are “sampled” proportionally). If run at a decision node, each trial is repeated for each strategy, to facilitate strategy comparison (e.g., CEA).

• Refer to the chapters on Patient Level Simulation for further details on Monte Carlo simulation, including details on running a Monte Carlo probabilistic sensitivity analysis.

• In survival/Markov models, microsimulation is particularly useful because it allows modeling any number of continuous and discrete state variables (whereas cohort models work well for more limited numbers of discrete states). The Patient Level Simulation provides details on Markov simulation topics.