Gravar-mail: Double robust and efficient estimation of a prognostic model for events in the presence of dependent censoring