Do quantity-frequency data underestimate drinking-related health risks?