Premium
Risk of Injury from Drinking: The Difference Which Study Design Makes
Author(s) -
Cherpitel Cheryl J.,
Ye Yu,
Bond Jason,
Stockwell Timothy,
Vallance Kate,
Martin Gina,
Brubacher Jeffrey R.,
MacPherson Andrew
Publication year - 2014
Publication title -
alcoholism: clinical and experimental research
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.267
H-Index - 153
eISSN - 1530-0277
pISSN - 0145-6008
DOI - 10.1111/acer.12217
Subject(s) - crossover study , crossover , odds ratio , confidence interval , medicine , case control study , emergency department , odds , statistics , mathematics , logistic regression , computer science , alternative medicine , pathology , artificial intelligence , psychiatry , placebo
Background The magnitude of risk of injury from drinking, based on emergency department ( ED ) studies, has been found to vary considerably across studies, and the impact of study design on this variation is unknown. Methods Patients were interviewed regarding drinking within 6 hours prior to the injury or illness event, drinking during the same time the previous week, and usual drinking during the last 30 days. Risk estimates were derived from case–control analysis and from both pair‐matched and usual frequency case‐crossover analysis. Results The odds ratio ( OR ) based on case–control (2.7; 1.9 to 3.8) was larger than that based on pair‐matched case‐crossover analysis (1.6; 1.0 to 2.6). The control‐crossover estimate suggested the case‐crossover estimate was an underestimate of risk, and when this adjustment was applied to the case‐crossover estimate, risk of injury increased ( OR = 3.2; 1.7 to 6.0). Adjusted case‐crossover estimates compared with unadjusted showed the largest proportional increase at 7 or more drinks prior to injury ( OR = 7.1; 2.2 to 22.9). The case‐crossover estimate based on usual frequency of drinking was substantially larger ( OR = 10.7; 8.0 to 14.3) than that based on case–control or pair‐matched case‐crossover analysis, but less than either when adjusted based on control‐crossover usual frequency analysis ( OR = 2.2; 1.5 to 3.3). Conclusions The data suggest that while risk of injury based on case–control analysis may be biased, control data are important in providing adjustments derived from control‐crossover analysis to case‐crossover estimates, and are most important at higher levels of consumption prior to the event.