Treatment effects on patient-important outcomes can be small, even with large effects on surrogate markers

OBJECTIVE: Surrogate markers are often used in clinical trials if too much time or expense is involved to observe the effect of treatment on patient-important outcomes. We wished to estimate the ultimate effects of treatment when randomized trials have addressed only a surrogate marker, and additional, independent studies evaluate the association between the surrogate and the final outcome. STUDY DESIGN AND SETTING: We show how to calculate the overall effect of treatment on a final outcome, together with its standard error and confidence interval. The methods are illustrated with data on the effect of therapy on hepatitis B seroconversion, a surrogate marker, and its association with patient-important outcomes (cirrhosis and liver cancer). RESULTS: We find that the effect of treatment on the final outcome may be small even if there are strong associations between treatment and the surrogate and between the surrogate and the patient-important outcome. CONCLUSION: Apparently, robust treatment effects on surrogates are likely to lead to small and uncertain effects on patient-important outcomes. We should be cautious in advising patients to adopt a therapy when compelling evidence is restricted to its impact on surrogate outcomes, particularly if that therapy may be toxic or otherwise cause harms.