Towards a more policy-relevant epidemiology: Exploring assumptions about generalizing intervention effects