We show that lagged duration dependence is non-parametrically identified in mixed proportional hazard models for duration data, in the presence of competing risks and consecutive spells.