A statistical correction was incorporated for the number of effective tests

By collecting samples immediately after and 30 min. after TSST onset, we were able to collect saliva samples when cortisol levels were expected to increase post-TSST and increased our chances of capturing peak cortisol response in line with previous guidelines . Samples were frozen at -80oC and later thawed and assayed in duplicate using commercially available enzyme immunoassays in the SPIT lab, with low mean inter- and intra-assay coefficients of variation . Samples were assayed again if the optical-density intra-assay coefficient of variation was over 10%. Adolescents also provided emotion ratings throughout the session. They rated the degree to which they felt happy, sad, and angry at four times: at baseline immediately prior to task onset, and at 15, 30, and 60 min. following task onset. Importantly, at 15 min. following task onset, participants completed two reports: they reported how they felt during the TSST, as well as how they felt at that moment. Participants completed two reports at this time point in order to assess emotion felt during the TSST without interrupting the task itself, and because emotion would be expected to change most between baseline and during the task, as opposed to immediately afterward when participants may feel relieved that the task is finished. This resulted in a total of five emotion ratings, all of which were included in analyses. Adolescents reported each form of emotion on a scale from 1 to 10 . Brief and single-item measures of emotion have been commonly used in assessment of emotion responses to stress tasks and throughout the day . Models tested the association between adolescents’ stress reactivity and recovery at age 14 and substance use by age 14,cannabis growing equipment substance use by age 16, and initiation of substance use between ages 14 and 16.

Because participants provided multiple cortisol samples and emotion ratings throughout the protocol, we utilized a multilevel framework with observations nested within participants . Specifically, saliva samples and emotion ratings were variables at Level 1 and substance use was measured at Level 2. Multilevel models allow for missing data at Level 1, such that participants could be missing data on a sampling occurrence and still be included in analyses. Models included 905 total observations for cortisol, and 1,299 total observations for emotion. Number of cortisol samples and emotion ratings did not vary by gender, grade, poverty status reported substance use at age 14, and baseline levels of each emotion, all ps > .05. Multilevel models also allow for the number of observations and the specific timing of the collection of each saliva sample to vary across participants, so that the cortisol response to stress can be accurately modeled. This framework leveraged all data and enabled both stress reactivity and recovery to be modeled simultaneously. Participants reported substance use at ages 14 and 16, which enabled testing of whether stress reactivity and recovery at age 14 were related to substance use at age 14, substance use at age 16, and substance use initiation over two years among non-users. Substance use was collected at the level of the participant and was therefore included as a predictor of stress reactivity and recovery , and models tested whether differences in the magnitude of stress reactivity and recovery at age 14 related to whether adolescents had ever used each substance by ages 14 and 16. It is important to note that we consistently model cortisol and emotion at age 14 as the outcome, even though differences in the stress response are thought to be a risk factor for substance use at age 16. This approach is necessary statistically, as other approaches are unable to simultaneously model stress reactivity and recovery with this number of time points. This modeling also allows for piece wise modeling.

There are a total of four samples for cortisol and five reports of emotion, both of which allow for piece wise assessment. Although three time points are generally needed to predict a linear trend, this modeling of all time points allows for HPA axis recovery to be computed using only two time points and for emotion reactivity to be computed using two time points. Alternative approaches include creating another index to test as a predictor of substance use, but these indices generally involve exaggerated error terms or violate statistical assumptions by assuming no error for each value . Conceptually this model is appropriate because, just as a correlation reflects a bidirectional association, this model tests the association between substance use and differences in stress reactivity and recovery, irrespective of which is the predictor versus outcome. A similar approach has been used in previous papers . Adolescents’ substance use was dummy-coded . Separate models predicted cortisol, anger, sadness, and happiness as a function of adolescents’ substance use. Prior research has highlighted that multilevel models are generally robust to violations of assumptions, including having skewed outcome variables . Piece wise modeling was used so that reactivity and recovery could be modeled simultaneously within the same model, and reactivity and recovery were estimated separately by calculating separate time terms at Level 1 . Reactivity was calculated as the number of minutes before the sample’s peak level, and all subsequent values were coded as 0. Recovery was calculated as the numbers of minutes following peak level, and all prior values were coded 0.The study was designed to test whether HPA axis reactivity and recovery and emotional reactivity and recovery were related to use of different substances. Given that substances have different effects and results may not carryover across substances, analyses of HPA axis and emotional reactivity and recovery for each substance were treated as a separate family of analyses. Three related measures of emotion were administered. Because the emotion items showed a high factor loading using exploratory factor analysis both at baseline and across assessments , we completed separate analyses of each emotion and incorporated a correction for the degree to which emotion items were related to one another . The high inter-relatedness of items suggests that analyses are largely non-independent, which resulted in a critical p-value of .046. Piece wise multilevel models were used to examine whether substance use was related to differences in cortisol reactivity and recovery simultaneously. Time was centered at the second cortisol sample, 30 min post-task onset, because salivary cortisol tends to peak 20-30 min following stress onset. Separate time terms were calculated for reactivity and recovery .

Reactivity time was coded as the number of minutes prior to the 30 min. sample and was coded as 0 for samples following 30 min. post-task onset,cannabis plant growing and recovery time was coded as the number of minutes following the 30 min. sample and as 0 for samples before 30 min. Post task onset. Coefficients for reactivity time and recovery time represent the change in cortisol per minute. Models tested interactions between substance use and time variables as predictors of cortisol to determine whether the magnitude of cortisol reactivity and recovery differed between adolescents who had versus had never used each substance by age 14. Models were repeated with substance use at age 16, and interactions were probed to examine cortisol reactivity and recovery at age 14 for adolescents who had versus had never used each substance by age 16. These models did not suggest that cortisol reactivity or recovery was related to use of alcohol, marijuana, cigarettes, or vaping nicotine use by ages 14 or 16 . There was also consistently no main effect of poverty status on cortisol across all models, ps > .05. Again, reactivity and recovery were modeled simultaneously within the same model using all five reports of emotion, and reactivity and recovery were estimated by calculating separate time terms . As expected, participants reported feeling the most extreme levels of emotion during the TSST using the retrospective report. This report was coded as how participants felt at the midpoint of the TSST, which lasted about 15 min. in total, and each assessment was coded with respect to the number of minutes before or after the middle of the TSST. Therefore, the baseline report was 7.5 min. before the middle of the TSST, and the reports following the TSST were 7.5, 22.5, and 52.5 min. following the middle of the TSST. For the reactivity time variable, baseline emotion was coded as -7.5 and all subsequent time points were coded as 0, so that this coefficient would only measure changes in emotion between prior to the TSST and during the TSST. Ideally, three or more time points would be used to estimate a linear trajectory. However, within this experimental context, it was not feasible to include another assessment of emotion that would improve estimation of emotion reactivity. For the recovery time variable, the ratings of emotion at baseline and during the TSST were coded as 0, and the subsequent three time points were coded with respect to the number of minutes following the middle of the TSST . Coefficients for the time variables represent the rate of change in emotion per minute. First, models tested whether emotion reactivity to and recovery from the TSST at age 14 were related to whether adolescents had ever used substances by age 14 .

Although difficulties with stress regulation are related to more frequent substance use among users , less is known regarding whether psychobiological responses to stress relate to substance use and precede initiation of use in adolescence. Therefore, the present study investigated whether dampened HPA axis and emotion responses to stress were related to substance use in a sample of Mexican-origin adolescents who had experienced high levels of adversity. Findings suggested that differences in HPA axis and emotion responses to social-evaluative stress relate to—and in some cases temporally precede—substance use among these adolescents, although associations varied by poverty status and sex. Dampened cortisol reactivity was related to use of alcohol by age 14 and vaping nicotine by age 16 among youth above the poverty line, although there was no evidence that cortisol reactivity related to initiation of use of substances between ages 14 and 16. In turn, dampened emotion responses to stress were related to substance use primarily in female adolescents. Among female adolescents, blunted anger reactivity was related to marijuana use by age 14, and blunted sadness reactivity and recovery were related to use of alcohol by age 16 and use of marijuana by ages 14 and 16. Blunted happiness reactivity was related to use of alcohol by age 16, regardless of sex, and to the emergence of use of marijuana and cigarettes between ages 14 and 16 among female adolescents who had not used these substances by age 14. Differences in associations between stress reactivity and recovery and substance use by poverty status and sex may be due to differences in adolescents’ access to substances or differences in motivation for substance use.Dampened cortisol reactivity was related to use of alcohol by age 14 and vaping of nicotine by age 16 for youth above, but not below, the poverty line. These findings align with prior work suggesting that blunted cortisol responses to stress relate to riskier substance use four years later among adolescents . Differences in stress physiology have been related to greater substance use among users , as well as greater risk for substance use initiation among youth . Inability to elicit a cortisol response from a stressor may suggest inflexibility of the HPA axis, such that people are unable to mobilize biological resources in the context of stress. Dampened cortisol reactivity may be indicative of difficulties with regulating stress, as strategies for emotion regulation have been linked with psychobiological responses to stress . For instance, prior studies have found that adolescents and adults with poorer emotion regulation show blunted cortisol reactivity to stress, often characterized by consistently high levels of cortisol . Furthermore, moderate cortisol responses to stress can promote executive function including emotion processing and behavioral inhibition during stress . Stress responses may be particularly tied to emotion regulation during adolescence, when youth are particularly sensitive to social threat and are still developing strategies for emotion regulation . Although associations with alcohol use at age 14 were cross-sectional, most evidence regarding the effect of substance use on HPA axis function has been observed among heavy users , and we do not have heavy use in this sample given participants’ age. Therefore, a potentially more probable pathway is that differences in the stress response may confer risk for substance use. Interestingly, youth above the poverty line who used alcohol by age 14 had higher levels of baseline salivary cortisol compared to youth who had never used alcohol by age 14.