Ouwens et al. and Jansen have presented methods for (network) meta-analysis of survival data by using a multidimensional treatment effect as an alternative to the synthesis of constant hazards ratios, which allow for a better fit to the data and the expected survival of competing interventions for cost-effectiveness analysis. However, results may be sensitive to the assumed underlying survival function.
To estimate the expected progression-free survival (PFS) for fulvestrant 500 mg versus alternative hormonal therapies for postmenopausal women with advanced breast cancer who relapsed previously by means of a network meta-analysis of currently available randomized controlled trials using alternative underlying survival functions.
Eleven randomized controlled trials were included that evaluated fulvestrant 500 mg (n = 3), fulvestrant 250 mg (n = 5), fulvestrant 250 mg loading dose (n = 3), anastrozole 1 mg (n = 3), megestrol acetate (n = 4), letrozole 2.5 mg (n = 3), letrozole 0.5 mg (n = 3), and exemestane (n = 2). PFS percentages and numbers at risk were derived from Kaplan-Meier curves and combined by means of Bayesian network meta-analysis on the basis of the difference in the shape and scale parameters of the Weibull, log-normal, and log-logistic parametric survival functions.
The log-normal distribution provided the best fit, suggesting that the proportional hazard assumption was not valid. Based on the difference in expected PFS, it was found that fulvestrant 500 mg is more efficacious than fulvestrant 250 mg, megestrol acetate, and anastrozole (−5.73 months; 95% credible interval [CrI]−10.67,−1.67). Expected PFS for fulvestrant 500 mg ranged from 10.87 (95% CrI 9.21, 13.07) to 17.02 (95% CrI 13.33, 22.02) months for the Weibull versus log-logistic distribution.
Fulvestrant 500 mg is expected to be more efficacious than fulvestrant 250 mg, megestrol acetate, and anastrozole 1 mg and at least as efficacious as exemestane and letrozole 2.5 mg in terms of PFS among postmenopausal women with advanced breast cancer after failure on endocrine therapy. The findings were not sensitive to the distribution, although the expected PFS varied substantially, emphasizing the importance of performing sensitivity analyses.