Prognostic tool predicts likelihood of smoking initiation in adolescents
A 12-item prognostic tool touts good predictive ability for identifying adolescents likely to initiate smoking over a 1-year period, as reported in a recent study from Canada.
The tool implicitly acknowledges that the first puff is a sentinel event that can rapidly lead to nicotine dependence and sustained smoking. It includes diverse factors based on a broad socioecological understanding of the risk, with the questions being short, easily interpretable and mostly requiring yes-or-no responses.
“Growing evidence on how quickly nicotine dependence symptoms can manifest after the first puff supports treating the first puff as a clinical emergency necessitating intervention to prevent long-term smoking,” according to the authors.
“If the predictive ability is replicated in other settings, this tool can be used to help clinicians select who should be counselled and, because several items in the tool are amenable to prevention, how they should be counselled,” they said, adding that preventing initiation should be a top priority in paediatric practice given the burden of smoking.
Good predictive ability
The study included 842 adolescents (mean age 13.8 years; 47.8 percent male), among whom 370 eventually initiated smoking. Data were partitioned into a training set for model-building and a testing set to evaluate the performance of the model. [Pediatrics 2018;doi:10.1542/peds.2017-3701]
In the training set, smoking initiation was significantly associated with 12 out of the 58 candidate predictors. These 12 included variables related to age, stress, depression, self-esteem, alcohol consumption and the presence of peers who smoke.
The 12-item model achieved good predictive ability in the testing set, with a c-statistic of 0.77 and sensitivity of 0.80. At a cutoff of 0.11, the model yielded a sensitivity and specificity of 0.80 and 0.55, respectively, for predicting the risk of smoking initiation within 1 year.
“Our validation suggests that the model performed satisfactorily outside the training data set, but its predictive validity remains to be established in external populations,” the authors said.
However, they also noted that the prognostic tool may not be generalizable to other populations, especially if the prevalence of the items tapped differs significantly. The threshold used to designate high or low risk depends on smoking prevalence, and the cutoff of 0.11 may only be meaningful if the adolescent smoking prevalence is approximately 16 percent, which is the current rate in Canada, slightly higher than in the US. [Burkhalter R, et al. 2012/2013 Youth Smoking Survey Microdata User Guide, fifth edition. University of Waterloo: Waterloo, 2013; https://www.surgeongeneral.gov/library/reports/preventing-youth-tobacco-use/full-report.pdf]
“If smoking prevalence differs substantially, our model can still be used to provide guidance on the relative importance of each predictor and allow clinicians to flag adolescents with several risk factors in the model,” the authors continued.
“By presenting the tool … we hope to lay the groundwork for its use and validation in the many clinical settings in which it could be deployed. We view this tool not as a static entity but as a way to address a gap in current clinical practice that can be iteratively improved over time,” they said.
Patient data to inform delivery of care
In a linked commentary, Dr Jonathan Klein from the University of Illinois, US, noted that while the study’s statistical analysis and use of modelling are elegant, the presence of certain limitations, acknowledged by the authors themselves, raises concerns that may hinder the model’s use for either research or practice. [Pediatrics 2018;doi:10.1542/peds.2018-2298]
Nevertheless, the model shows how patient data could inform care delivery and points to a future where health systems are able to use electronic health records integrated with computer-based decision support to provide targeted, personalized care, according to Klein.
“If the known associations between multiple risky behaviours, patient attitudes and previous care encounters are fully integrated into electronic health record prompts, individual care could be more fully informed by all available data,” he added.
“Eventually, one hopes we will have tools that can be used to address multiple healthy and risky behaviours, embedded in real-time interactive learning health systems thus able to improve both individual health, clinical care delivery and population health,” Klein said.