Article type
Year
Abstract
Background: A surrogate is any measure short of an obviously patient relevant outcome in a clinical trial. Surrogates are valuable because they often increase trial efficiency by reducing its size or duration. However, widely agreed upon criteria for validation of surrogates do not exist. Operationally, the effect of an intervention on the surrogate should reliably predict the effect on the patient relevant outcome in order to supply confidence for use in trials of the surrogate alone.
Methods used previously to establish the predictive capacity of surrogates include meta-regression of surrogate/outcome data from trials and formally examining the two dimensional data. Examples of this exist for HIV/AIDS and colorectal cancer [1,2].
Objectives: To develop a general method to quantitatively describe the predictive capacity of surrogate markers, enabling transparent and comparative determinations of their validity.
Methods: A preliminary systematic review of published randomized trials on pharmacologic treatment in non-operative non-small cell lung cancer since 1977 was undertaken. 48 trials were identified. All used common metrics for tumor response and time-to-progression. Each contributed a log hazard ratio of treatment to control for the outcome, overall survival, a log odds ratio for the surrogate, proportion responding, and a log hazard ratio for the surrogate, time-to-progression. Regression models were fit, using the log hazard ratio for survival as the outcome, and each of the surrogates, individually, as predictors. Models were shown to conform to standard regression assumptions of normally distributed errors and constant variance. Trials were weighted by size and an adjusted R2 determined, i.e., the amount of variance in the outcome accounted for by the surrogate.
Results: Analyses for non small-cell lung cancer are reported here; those for metastatic breast cancer, advanced colorectal cancer, and chronic myeloid leukemia will also be reported at the Colloquium. The adjusted R2 for the relation of the log hazard ratio for survival as a function of the log odds ratio for response was 0.41 with a significant b coefficient (p<0.001). The adjusted R2 for survival versus time-to-progression was 0.42, again with a significant b (p<0.001). When both surrogates were used, adjusted R2 increased minimally to 0.44 while now both b coefficients became non-significant (p=0.216 and p=0.162).
Conclusions: This analysis of the predictive potential of tumour response and time-to-progression on survival in non small-cell lung cancer shows these surrogates to be only modestly predictive. The lack of additional benefit using both surrogates simultaneously reflects collinearity. However, this mode of analysis is in its early stages and definitive conclusions and interpretations should be deferred until more surrogates in more settings have been
investigated.
Acknowledgements: This work was supported by the Commonwealth Department of Health, Australia.
References: 1. HIV Surrogate Marker Collaborative Group. Human immunodeficiency virus type 1 RNA level and CD4 count as prognostic markers and surrogate end points: A meta-analysis. AIDS Res Human Retroviruses. 2000; 16(12):1123-1133. 2. Buyse M, Thirion P, Carlson R, et al for the Meta-analysis Group in Cancer. Relation between tumour response to first-line chemotherapy and survival in advanced colorectal cancer: a meta-analysis. The Lancet. 2000;356:373-8.
Methods used previously to establish the predictive capacity of surrogates include meta-regression of surrogate/outcome data from trials and formally examining the two dimensional data. Examples of this exist for HIV/AIDS and colorectal cancer [1,2].
Objectives: To develop a general method to quantitatively describe the predictive capacity of surrogate markers, enabling transparent and comparative determinations of their validity.
Methods: A preliminary systematic review of published randomized trials on pharmacologic treatment in non-operative non-small cell lung cancer since 1977 was undertaken. 48 trials were identified. All used common metrics for tumor response and time-to-progression. Each contributed a log hazard ratio of treatment to control for the outcome, overall survival, a log odds ratio for the surrogate, proportion responding, and a log hazard ratio for the surrogate, time-to-progression. Regression models were fit, using the log hazard ratio for survival as the outcome, and each of the surrogates, individually, as predictors. Models were shown to conform to standard regression assumptions of normally distributed errors and constant variance. Trials were weighted by size and an adjusted R2 determined, i.e., the amount of variance in the outcome accounted for by the surrogate.
Results: Analyses for non small-cell lung cancer are reported here; those for metastatic breast cancer, advanced colorectal cancer, and chronic myeloid leukemia will also be reported at the Colloquium. The adjusted R2 for the relation of the log hazard ratio for survival as a function of the log odds ratio for response was 0.41 with a significant b coefficient (p<0.001). The adjusted R2 for survival versus time-to-progression was 0.42, again with a significant b (p<0.001). When both surrogates were used, adjusted R2 increased minimally to 0.44 while now both b coefficients became non-significant (p=0.216 and p=0.162).
Conclusions: This analysis of the predictive potential of tumour response and time-to-progression on survival in non small-cell lung cancer shows these surrogates to be only modestly predictive. The lack of additional benefit using both surrogates simultaneously reflects collinearity. However, this mode of analysis is in its early stages and definitive conclusions and interpretations should be deferred until more surrogates in more settings have been
investigated.
Acknowledgements: This work was supported by the Commonwealth Department of Health, Australia.
References: 1. HIV Surrogate Marker Collaborative Group. Human immunodeficiency virus type 1 RNA level and CD4 count as prognostic markers and surrogate end points: A meta-analysis. AIDS Res Human Retroviruses. 2000; 16(12):1123-1133. 2. Buyse M, Thirion P, Carlson R, et al for the Meta-analysis Group in Cancer. Relation between tumour response to first-line chemotherapy and survival in advanced colorectal cancer: a meta-analysis. The Lancet. 2000;356:373-8.