Multiple endpoints are often needed to characterize the benefit of a new treatment. This article develops an adaptive design where estimates of the correlation and variance parameters are updated using interim data based on the modified O’Brien test. In particular, we will a conditional likelihood approach where the likelihood of all endpoints conditional on the relevant test statistics evaluated at maximum information time is considered because of the superior power preserving property in the single endpoint setting. Heuristically, the conditional approach is tackles the multiplicity problem by requiring the design operating characteristics being close to those of a non-adaptive design. We show the adaptive design preserves the alpha level and statistical power to a large extent. We illustrate the method with the trials in breast cancer and Parkinson disease.