Mengman Wei, Qian Peng
5 min
This paper analyzes what drives adolescents to start using substances like alcohol, nicotine, and cannabis for the first time, using a massive longitudinal dataset from the ABCD Study (Adolescent Brain Cognitive Development), which tracks ~11,000 U.S. kids from ages 9-10 over four years. Early initiation is a red flag for later addiction and mental health issues, so the authors blend genetic risks (via Polygenic Risk Scores, or PRS) with changing environmental factors (like impulsivity or parental monitoring) in advanced survival models. They find genetics nudge kids toward earlier use, but environment—especially impulsivity and weak parental oversight—plays a huge modifiable role. Causal analyses pinpoint actionable targets like boosting monitoring to delay onset.
Think of initiation not as a yes/no, but 'how long until first use?' Four outcomes: alcohol, nicotine, cannabis, any substance. Time is measured in months from baseline via interview age. Data is structured as start-stop intervals—like episodes in a TV show—where each kid contributes rows like [start age, stop age), censoring at study end if no event. This handles dropouts naturally. Univariate models show broad links (e.g., poor sleep, school issues speed things up), but multivariable prunes to core signals.
Cox models estimate hazard ratios (HR): how much a predictor multiplies the instantaneous risk of initiation. Key twist: time-varying predictors (e.g., impulsivity score changes yearly). Solution? Counting process format: stack visits into a long panel, repeat fixed traits (like PRS) per row, align by age, forward-fill missings to avoid 'time leakage' (no peeking at future). LASSO screens high-dimensional env vars first. Stratify by site, cluster errors by kid. Results: PRS for nicotine (HR~2.4-3.0), alcohol, cannabis, SUD all predict earlier use; env standouts include high impulsivity (UPPS scales), low parental monitoring, caffeine use.
PRS are time-invariant genetic summaries (from GWAS) for disorders like AUD, CUD, nicotine dependence, SUD. Higher PRS = earlier initiation across substances, strongest for nicotine PRS even on non-nicotine outcomes (e.g., HR=2.98 for any substance). This suggests broad genetic liability. PRS held in multivariable models controlling demographics/ancestry, proving they add unique signal beyond baselines.
To go beyond association, they use MSM with IPTW (inverse probability treatment weighting) on select modifiables: parental monitoring (protective, OR=0.33-0.64), UPPS lack of planning/sensation-seeking (risky, OR=1.5-3.9), caffeine (risky). Weights balance confounders over time, approximating causal effects under no unmeasured confounding. Robust to specs, highlighting interventions: improve monitoring, curb impulsivity/caffeine to delay initiation.
Data prep shines: quality filters, median imputation + availability flags, no future leakage. ABCD's scale/repeated measures enable this. Big picture: genetics set the stage, but time-varying env factors like parenting/impulsivity are levers for prevention. Targets adolescent SUD trajectories, informing public health before disorders lock in.
Early initiation of alcohol, nicotine, cannabis, and other substances predicts later substance use disorders and related psychopathology. We integrate time-varying environmental factors with polygenic risk scores (PRS) in a longitudinal framework to identify determinants of substance initiation in adolescence. Using data from the Adolescent Brain Cognitive Development (ABCD) Study with repeated assessments over approximately four years, we defined time-to-event outcomes for first use of alcohol, nicotine, cannabis, and any substance. We constructed high-dimensional panels of time-varying environmental covariates across family, school, neighborhood, behavioral, and health domains, alongside time-invariant covariates and PRS for alcohol, cannabis, nicotine, and general substance use disorders. Time-varying Cox models with clustered standard errors were applied. Univariate analyses showed broad associations between earlier initiation and multiple environmental domains, including impulsivity, sleep disturbance, parental monitoring, caffeine use, and school functioning. In multivariable models, a smaller set of predictors remained robust, particularly impulsivity traits, parental monitoring, and selected health and lifestyle factors. PRS were positively associated with earlier initiation, with the strongest and most consistent effects for nicotine-related genetic risk. Secondary analyses using marginal structural models suggested that higher parental monitoring is protective, whereas higher impulsivity and caffeine exposure are associated with increased risk. These results demonstrate that integrating dynamic environmental exposures with genetic liability can identify key risk factors for adolescent substance initiation and highlight actionable targets for prevention.
Alex: But to claim monitoring causes a delay, not just tags along—how do they strengthen that?
Sam: They built multivariable survival models with genes and basics, where only a few like lower monitoring raised risk. For causal clues, they used inverse probability weighting: it balances groups so kids with or without monitoring look similar on other traits at each time, like adjusting a scale to compare apples and oranges by size alone.
Alex: Like giving extra weight to unusual cases, such as high-risk kids who get monitored?
Sam: Yes. Weights come from past traits, stabilized and clipped for reliability. Weighted models show monitoring with protective odds around 0.7 to 0.8 across substances—two to three times lower risk—after adjustments. Caffeine shows riskier odds; discrimination or rule-breaking peers push odds up, but monitoring counters.
Alex: Huh, so genes matter but don't override levers like monitoring.
Sam: Right. Polygenic scores link to earlier starts, strongest for cannabis and nicotine, but dynamic factors add heft.
Alex: In full multivariable setups, what patterns emerge across substances?
Sam: Peer rule-breaking stands out most consistently, raising risk across alcohol, nicotine, cannabis, and any substance—up to nearly 30% higher for cannabis. Alcohol shows impulsivity, family conflict, stress, sleep. Cannabis and nicotine tie to externalizing behaviors like rule-breaking.
Alex: Externalizing—like actually breaking rules, not just feeling worried?
Sam: Yes, parent- or peer-reported acting out. Protective signals appear too, but some tie to measurement quirks like missing items, so the paper urges caution there.
Alex: Limits? No study's perfect.
Sam: Questionnaires mean reporting errors could weaken signals. Unmeasured influences might linger; causal claims assume no hidden confounders and enough examples per group. Four-year self-reports cut some stories short. The paper stresses these.
Alex: Still highlights modifiable spots like peers, monitoring, and school.
Sam: Exactly. This maps genetic and environmental predictors, isolating targets beyond fixed risks. A solid step.
Alex: That's a grounded take. Thanks, Sam—this has been a clear dive into early timelines. Thanks for listening to ResearchPod.