A SIMPLIFIED METHOD FOR ASSESSING COST-EFFECTIVENESS OF PREDICTIVE BIOMARKERS IN ONCOLOGY

Purpose: Cost effectiveness of predictive biomarkers is currently assessed by state-transition modeling, involving derivation of transition probabilities and comprehensive data inputs. Our goal was to demonstrate the accuracy, accessibility, and flexibility of a simplified decision-analytic model.

Method: We used decision-analytic modeling to derive a “test incremental cost-effectiveness ratio” (TICER) for biomarker-guided treatment compared to no biomarker use. Our derivation incorporates critical data inputs, including clinical outcomes (median progression free survival and overall survival) obtained from clinical trials, quality of life metrics (biomarker guided treatment and standard therapy) obtained from clinical trials or meta-analyses studies, and costs (testing, treatment, progression) derived from existing cost-effectiveness analyses. Conditional sensitivity analyses were performed to evaluate the impact of the input parameters by sampling fifty sequential values within the 95% confidence interval of a parameter of interest. For each of these values, TICER was computed for 2,000 iterations, with all other parameters randomly sampled from their respective distributions. To protect against instability of the model when the denominator was small, the range of median TICERs for each parameter of interest was used a measure of sensitivity of TICER to that variable. This methodology was applied to three common predictive biomarkers; results were compared to results from comprehensive state-transition models.

Result: We considered the cost-effectiveness of several biomarker-guided treatment scenarios important in cancer.  Base case TICER for HER2 was $153,700/QALY (Quality Adjusted Life Year), for ALK was $222,00/QALY, and for OncotypeDX was $11,600/QALY, which are consistent with literature reported values ($180,000/QALY, $202,800/QALY, $8,900/QALY, respectively). Sensitivity analyses suggest that the TICER model is driven primarily by progression free survival (PFS), overall survival (OS) and health-related quality of life (HRQOL) values, while biomarker prevalence, costs of target therapy and biomarker testing have a lesser effect. Additional results include cost-effectiveness acceptability curves and treatment cost needed to meet a range of willingness-to-pay thresholds.

Conclusions: Our simplified approach is flexible to a variety of clinical scenarios, and it produces values that are consistent with existing cost-effectiveness analyses. Our model and sensitivity analysis engine was incorporated into a webtool, available at http://medicine.yale.edu/lab/pusztai/ticer/ and made available to users who are not experts in cost-effectiveness analysis. The simplicity and efficiency of this model potentially allows preliminary cost-effectiveness feedback at an early stage of biomarker development.