A GENERIC MULTI-STATE SIMULATION MODEL FOR ASSESSING DIAGNOSTIC TESTS

Purpose: To date, assessing diagnostic tests in a single self-consistent model remains challenging due to three issues.  First, diagnostic tests are often used repeatedly over the course of a patient's life cycle. Second, it is therapeutic treatments rather than diagnostic tests that directly affect interest (final) outcomes. Third, diagnostic states and health states are dynamically interdependent. This study intends to provide a generic simulation-based framework in which all the three issues are solved.

Method: We designed a unique discrete-time two-stage decision model (Figure 1) that simulates an agent's diagnostic test decisions and latent health state transitions. The model maps health state prior, diagnostic test characteristics, therapeutic treatment characteristics, and discounting factors into endogenous random variable sequences (i.e., diagnostic decisions, diagnostic states and health states).  The model accommodates stochastically repeated diagnostic tests and can flexibly characterize the interdependency between diagnostic test decisions and latent health state transitions.  Analytical derivation reveals that when the demand for a diagnostic test is perfectly price inelastic (e.g., the test is fully covered by Medicare) a study diagnostic test's cost-effectiveness cutoff cost is linearly related to the comparison diagnostic test's cost. This insight provides a computational shortcut for identifying cost-effectiveness cutoff costs in many applications.

Result: In a demonstration, we applied the model to the cost-effectiveness analysis of 12-core prostate cancer biopsy in comparison with 6-core biopsy for a 65-year-old man referred for prostate biopsy and followed until death.  Using input parameters extracted from SEER-Medicare datasets, we found that the incremental cost-effectiveness ratio of 12-core biopsy is $15,947 per life year gained. A 12-core biopsy remained cost-effective with test sensitivity as low as 0.65 and test cost as high as $6,291 (on average, a 12-core biopsy costs $2,573).

Conclusion: Our model can evaluate the effects of a diagnostic test on both intermediate outcomes (e.g., the number of diagnoses, choice of therapeutic treatments) and final outcomes (e.g., overall cost, QALYs, deaths avoided). In addition, various sensitivity analyses on patient characteristics (e.g., health state prior, initial age), therapeutic treatment characteristics (e.g., effectiveness parameters, costs), diagnostic test utilization parameters (e.g., propensity of re-testing), and diagnostic test characteristics (e.g., ROC curve, costs) can be implemented in the model for clinical policy design.