Asymptotically efficient estimation in response adaptive trials