A prognostic tool to identify youth at risk of at least weekly cannabis use
Is part ofAI for disease surveillance and pandemic intelligence : intelligent disease detection in action ; pp. 37-48.
We developed and validated an 8-item prognostic tool to identify youth at risk of initiating frequent (i.e., at least weekly) cannabis use in the next year. The tool, which aims to identify youth who would benefit most from clinician intervention, can be completed by the patient or clinician using a computer or smart phone application prior to or during a clinic visit. Methodological challenges in developing the tool included selecting a parsimonious model from a set of correlated predictors with missing data. We implemented Bach’s bolasso algorithm which combines lasso with bootstrap and investigated the performance of the prognostic tool in new data collected in a different time period (temporal validation) and in another location (geographic validation). The tool showed adequate discrimination abilities, as reflected by a c-statistic above 0.8, in both validation samples. Most predictors selected into the tool pertained to substance use including use of cigarettes, e-cigarettes, alcohol and energy drinks mixed with alcohol, but not to mental or physical health.