Risk of injury due to alcohol: evaluating potential bias using the case-crossover usual-frequency method