<p>Multiple endpoints often arise in clinical trials where the relevant outcome, such as quality of life or clinical benefit, is defined by multiple outcomes and the interest is to test globally the overall difference between the two groups. It’s increasingly needed to compare nonrandomized or imperfectly randomized groups in multiple endpoints for various disease areas, e.g., in neuro-protective, arthritis, and immunotherapies. For example, in our study on knee osteoarthritis, we wanted to compare an active treatment with an external control, one arm of a previous randomized trial on the same patient population, in multiple primary endpoints such as pain, stiffness, and functional scores to determine an overall benefit. In these cases, it is known that conventional estimators are biased when the treatment is not randomized. Robust causal estimates are then needed to correct the bias in nonrandomized studies while guarding against potential deviations from various assumptions. However, no such methods are available for multiple endpoints to our knowledge, partly due to the complexity of the estimand. We propose a robust causal semiparametric estimation method for multiple endpoints, and derive the asymptotic properties of the proposed estimator. Simulation studies are performed to evaluate the finite sample properties of the proposed method and compare it with the parametric and naive methods. We then apply the procedure to analyze the knee osteoarthritis clinical study.</p>