Development and validation of a risk predictive model for student harmful drinking-A longitudinal data linkage study

Background: This study aimed to develop a predictive model to quantify the risk of student harmful drinking associated with emergency department (ED) visits and/or campus-wide incidents reported to campus authorities in a U.S. public university. Methods: Six-year (2010/11-2015/16) student enrollment data were linked to subsequent harmful drinking events defined as either alcohol intoxication associated with ED visits or alcohol-related incidents reported to authorities within 1 year following the annual (index) enrollment. Multivariable logistic regression analysis was used to develop a risk predictive model based on the first 3-year student cohort (n = 93,289), which was then validated in the following 3-year student cohort (n = 85,876). Results: A total of 2609 students in the derivation cohort and 2617 students in the validation cohort had at least 1 harmful drinking event within 1 year following the index enrollment, providing an incidence of 2.8% and 3.1%, respectively. Student demographics (gender, age, ethnicity, parental tax dependency), academic level, Greek life member, transfer students, first-time enrolled students, having been diagnosed with depression or injury, and violence involvement were statistically significant predictors. C-statistics of the model were 0.86 in both cohorts, with excellent calibration and no evidence of over- or under-prediction observed from calibration plots. Conclusions: By linking routinely collected student data, a robust risk predictive model was developed and validated to quantify absolute risk of harmful drinking for every student. This model can provide a useful tool for clinicians or health educators to make real time decision to plan target interventions for students at elevated risk.