Evaluating Potential Recall Bias in Case-crossover Methods for Risk of Injury Associated with Alcohol in Emergency Room Samples from Two Countries

Estimating relative risk of injury related to alcohol consumption is challenging given the transient, rather than accumulative, causal mechanism of alcohol intoxication on injury. The traditional case-control method is both costly and prone to error due to difficulty in correctly identifying the control population. The case-crossover approach was thus employed recently for this purpose. Using injury patients as their own controls, risk of injury associated with drinking is estimated comparing alcohol consumption before the injury event with drinking during a control period. The most commonly used control period is the same time one week before. While this allows controlling for the day of the week when injury occurs which is correlated with drinking, the approach is subject to potential recall bias. Current analysis examines this potential bias using samples of ER injury patients from two countries: Vancouver, Canada (n=443) and Santo Domingo, Dominican Republic (n=501). Drinking during the same time last week is compared to the same time yesterday, presumably less likely to be subject to recall bias. A higher proportion of injury patients reported drinking yesterday compared to last week from the Dominican site (12.1% vs 8.6), but not from the Canadian site (13.4% vs 14.3%). Similar results were seen after controlling for day of the week with injury patients from the Dominican site more likely to report drinking yesterday than last week (OR=1.46). Our findings suggest relative risk estimates for injury related to alcohol using last week as the control period might be biased upward for some countries.