Odds ratio or relative risk for cross-sectional data?