Toward an “unconfounded” classification of epidemiologic research design