| To Register SMDM Homepage | |
|
|
|
||||
Purpose: It has been suggested that Bayesian approaches offer superior analytical methods for decision-makers. We sought to demonstrate the advantages of Bayesian methodology to quantify the long-term health outcomes of patients with high-risk (stage III) melanoma under alternative treatment strategies.
Methods: A continuous-time Markov model was designed with 5 health states (no evidence of disease, local-regional recurrence, distant recurrence, death of disease and death of other causes) to assess quality-adjusted life years (QALYs) for patients treated with IFN-a compared to no adjuvant treatment following surgical resection. Patient data were retrospectively collected from a tertiary cancer center. A subgroup analysis was performed based on 4 age cohorts: < 40, 40-49, 50-59, and > 59 years old. The survival time in each state was assumed to have an exponential distribution. Transition probabilities between health states were estimated according to Bayes Rule: Posterior distribution is proportional to Prior distribution *Likelihood. An informative gamma prior was used based on previously published studies. A sensitivity analysis was performed using different priors with the same mean and different variances. Health state utilities were incorporated from published data. The models were implemented in Splus.
Results: For high-risk melanoma patients, the model predicted that treatment with IFN-a resulted in 6.94 QALYs compared to 5.12 QALYs for patients in the control group. The incremental effectiveness for patients treated with IFN-a was 1.82 QALYs with a 95% credible interval of 0.48 to 3.37. Table 1 summarizes the effectiveness (QALYs) and incremental effectiveness for each age cohort. The results of the sensitivity analysis were very robust to the selection of prior distributions.
Table 1. QALYs for various age cohorts
| Age | Control | IFN-a | Incremental Effectiveness | 95% Credible Interval |
| < 40 | 8.47 | 10.40 | 1.93 | -3.60, 8.52 |
| 40-49 | 5.42 | 11.21 | 5.79 | 1.58, 12.22 |
| 50-59 | 6.54 | 5.25 | -1.29 | -4.01, 1.50 |
| > 59 | 4.17 | 4.68 | 0.50 | -1.08, 2.67 |
Conclusions: Bayesian approaches produce results that incorporate variability in patient histories and uncertainties in model parameters. In this case study, adjuvant IFN-a was effective overall and in patients less than 50 years of age. This approach provides a solid theoretical framework for decision-making.
See more of Poster Session - Public Health; Methodological Advances
See more of The 26th Annual Meeting of the Society for Medical Decision Making (October 17-20, 2004)