Conditions for valid estimation of causal effects on prevalence in cross-sectional and other studies.