BACKGROUND AND OBJECTIVES: Clinical trials increasingly use results of diagnostic tests as surrogate outcomes. Our objective was to answer the following questions: (1) is the parameter measured by the reference standard a valid surrogate? (2) How does the tests accuracy influence the estimate of the treatment benefit on surrogate? (3) Is it possible to correct the measured treatment effect given by results of inaccurate tests? METHODS AND SETTING: We reviewed the literature on asymptomatic deep venous thrombosis (DVT), detected by the reference standard and other imaging techniques, as surrogate for venous thromboembolism. The influence of test inaccuracy on the measurement of treatment benefit was calculated as a function of the patient baseline risk, the treatment effect model, and test performances. RESULTS: We show that: (1) asymptomatic DVT is correlated with clinical outcomes but is yet to be established as a surrogate; (2) inaccurate diagnostic test underestimates the treatment effect on surrogate; (3) the prevalence of the disease, the treatment effect model, and the accuracy of the test and the reference standard used to evaluate it need to be known to correct this underestimation. CONCLUSION: Even when the surrogate end point is valid, without a reliable study of the diagnostic test we cannot quantify the true treatment effect.
BACKGROUND AND OBJECTIVES: Clinical trials increasingly use results of diagnostic tests as surrogate outcomes. Our objective was to answer the following questions: (1) is the parameter measured by the reference standard a valid surrogate? (2) How does the tests accuracy influence the estimate of the treatment benefit on surrogate? (3) Is it possible to correct the measured treatment effect given by results of inaccurate tests? METHODS AND SETTING: We reviewed the literature on asymptomatic deep venous thrombosis (DVT), detected by the reference standard and other imaging techniques, as surrogate for venous thromboembolism. The influence of test inaccuracy on the measurement of treatment benefit was calculated as a function of the patient baseline risk, the treatment effect model, and test performances. RESULTS: We show that: (1) asymptomatic DVT is correlated with clinical outcomes but is yet to be established as a surrogate; (2) inaccurate diagnostic test underestimates the treatment effect on surrogate; (3) the prevalence of the disease, the treatment effect model, and the accuracy of the test and the reference standard used to evaluate it need to be known to correct this underestimation. CONCLUSION: Even when the surrogate end point is valid, without a reliable study of the diagnostic test we cannot quantify the true treatment effect.
Authors: V G De Gruttola; P Clax; D L DeMets; G J Downing; S S Ellenberg; L Friedman; M H Gail; R Prentice; J Wittes; S L Zeger Journal: Control Clin Trials Date: 2001-10