Heterogeneity in Radiotherapeutic Parameter Assumptions in Cost-Effectiveness Analyses in Prostate Cancer: A Call for...

Cost-effectiveness analyses (CEAs) may provide useful data to inform management decisions depending on the robustness of a model's input parameters. We sought to determine the level of heterogeneity in health state utility values, transition probabilities, and cost estimates across published CEAs assessing primarily radiotherapeutic management strategies in prostate cancer.

We conducted a systematic review of prostate cancer CEAs indexed in MEDLINE between 2000 and 2018 comparing accepted treatment modalities across all cancer stages. Search terms included "cost effectiveness prostate," "prostate cancer cost model," "cost utility prostate," and "Markov AND prostate AND (cancer OR adenocarcinoma)." Included studies were agreed upon. A Markov model was designed using the parameter estimates from the systematic review to evaluate the effect of estimate heterogeneity on strategy cost acceptability.

Of 199 abstracts identified, 47 publications were reviewed and 37 were included; 508 model estimates were compared. Estimates varied widely across variables, including gastrointestinal toxicity risk (0%-49.5%), utility of metastatic disease (0.25-0.855), intensity-modulated radiotherapy cost ($21 193-$61 996), and recurrence after external-beam radiotherapy (1.5%-59%). Multiple studies assumed that different radiotherapy modalities delivering the same dose yielded varying cancer control rates. When using base estimates for similar parameters from included studies, the designed model resulted in 3 separate acceptability determinations.

Significant heterogeneity exists across parameter estimates used to perform CEAs evaluating treatment for prostate cancer. Heterogeneity across model inputs yields variable conclusions with respect to the favorability and cost-effectiveness of treatment options. Decision makers are cautioned to review estimates in CEAs to ensure they are up to date and relevant to setting and population.