A synergistic mindsets intervention protects adolescents from stress

Abstract

Social-evaluative stressors-experiences in which people feel they could be judged negatively-pose a major threat to adolescent mental health^1-3 and can cause young people to disengage from stressful pursuits, resulting in missed opportunities to acquire valuable skills. Here we show that replicable benefits for the stress responses of adolescents can be achieved with a short (around 30-min), scalable 'synergistic mindsets' intervention. This intervention, which is a self-administered online training module, synergistically targets both growth mindsets⁴ (the idea that intelligence can be developed) and stress-can-be-enhancing mindsets⁵ (the idea that one's physiological stress response can fuel optimal performance). In six double-blind, randomized, controlled experiments that were conducted with secondary and post-secondary students in the United States, the synergistic mindsets intervention improved stress-related cognitions (study 1, n = 2,717; study 2, n = 755), cardiovascular reactivity (study 3, n = 160; study 4, n = 200), daily cortisol levels (study 5, n = 118 students, n = 1,213 observations), psychological well-being (studies 4 and 5), academic success (study 5) and anxiety symptoms during the 2020 COVID-19 lockdowns (study 6, n = 341). Heterogeneity analyses (studies 3, 5 and 6) and a four-cell experiment (study 4) showed that the benefits of the intervention depended on addressing both mindsets-growth and stress-synergistically. Confidence in these conclusions comes from a conservative, Bayesian machine-learning statistical method for detecting heterogeneous effects⁶. Thus, our research has identified a treatment for adolescent stress that could, in principle, be scaled nationally at low cost.

Fig. 1. How young people’s social-evaluative stressors accumulate consequences for healthy development.

a,b, First, the individuals appraise both acute stressful events and their stress responses (a); and second, their mindset beliefs shape their appraisals and responses, which leads to differences in internalizing symptoms over time (b). This integrated model is rooted in established process models in affective science^,, recursive process models in psychology^, and mindset models^,,. a, Stressful events, such as a challenging exam or an argument with a friend, are appraised as either harmful and uncontrollable or more helpful and controllable, cultivating threat or challenge response tendencies, respectively. Then, the meaning of the stress response is appraised as either distressing and non-functional (harmful and uncontrollable) or as a resource that helps one address situational demands (helpful and controllable), which results in further threat- or challenge-type stress responses, respectively^,. Threat stimulates the hypothalamic–pituitary–adrenal (HPA) axis in the brain, the end-product of which is the catabolic adrenal hormone cortisol, in anticipation of damage or social defeat. Challenge is characterized by increased peripheral blood flow (hence the red depiction), and a faster return to homeostasis after stress offset. Threat, however, results in increased vascular resistance and less oxygenated blood flow to the periphery (hence the blue depiction) as HPA activation tempers sympathomedullary effects and produces a more prolonged stress response^,,. Threat leads to avoidance motivation and negative affect, whereas challenge elicits approach motivation and more positive affect relative to threat. SNS, sympathetic nervous system. b, Mindsets are situation-general beliefs about categories of events (for example, academic stressors) and responses (for example, feelings of worry) that shape appraisals at the event stage and next at the response stage^,,,. Individuals who respond with an optimized challenge-type stress response engage with and respond to future stressors more adaptively in a self-reinforcing, positive feedback cycle that results in better coping and performance.

Fig. 2. Procedures and results of studies 1 and 2.

a–d, Studies 1 and 2 (n = 2,534 and n = 790, respectively) showed that relative to the neutral control condition the synergistic mindsets intervention reduced negative appraisals of an immediate, hypothetical stressor (a,b), and an acute naturalistic stressor up to 3 weeks after the intervention (c,d). Participants were secondary school students (study 1) or undergraduates (study 2) attending public schools in the United States. Starbursts represent stressor onset. Dots correspond to the ATEs estimated with the Bayesian model. Thick lines represent the 10th to 90th percentiles; grey lines represent the 2.5th to 97.5th percentiles. The appraisals for each study were coded so that higher values corresponded to more negative appraisals, so negative treatment effects are consistent with a beneficial stress optimization effect. Average effect sizes appear in the text. Study 1, control n = 1,326; treatment n = 1,208. Study 2, control n = 403; treatment n = 387.

Fig. 3. In study 3, the synergistic mindsets intervention improved cardiovascular responses to the TSST.

a–c, Participants in study 3 (n  = 160) were undergraduate students in a laboratory experiment. a, Procedures for study 3. b,c, Coloured lines (b) and dots (c) correspond to the expected value of the outcome (b) or the ATE (c), estimated with the Bayesian model. The thick bands represent the 10th to 90th percentiles of the posterior distributions; grey lines represent the 2.5th to 97.5th percentiles. TPR (b) is measured in Dyn s × cm⁵, where. Time indicates the elapsed, cumulative physiological recording. Starbursts indicate TSST epochs that presented acute demands (that is, the stressful epochs). Baseline measurements were taken before the stress induction and random assignment to condition. Baseline scores were subtracted from all active epochs to compute reactivity scores for each minute. Preparation measurements were taken after intervention materials when participants planned their speech; speech delivery and mental mathematics measurements were taken during the speech and maths tasks, respectively; and finally, measurements were taken during a recovery period in which evaluative pressure (stress) was removed. The differences in TPR for the two groups were similar at baseline (see propensity score comparisons in the Supplementary Information). In c, ATEs and 10th to 90th percentiles are: preparation = −168 Dyn s × cm⁵ [−217, −121], speech = −223 [−274, −172], maths = −128 [−175, −80], recovery = −90 [−139, −41]. Control, n = 86; treatment, n = 74.

Fig. 4. In study 4, the synergistic mindsets intervention improved cardiovascular responses to the TSST, and this effect was larger than the effects of single-mindset interventions.

a–c, Participants in study 4 (n = 200) were undergraduate students in a laboratory experiment. a, Procedures for study 4. b, ATEs across outcomes. Dots correspond to the ATEs estimated with the Bayesian algorithm. Thick lines represent the 10th to 90th percentiles; grey lines represent the 2.5th to 97.5th percentiles. c, The entire posterior distributions of a difference between the treatment effects of the conditions (synergistic mindset versus single mindset) (that is, a test of the interaction effect hypothesis), estimated in the Bayesian model. Study 4 streamlined the TSST procedures to allow for more efficient data collection, so the maths epoch was removed. The pre-registration stated that the primary outcome would be TPR during the speech delivery epoch. All results were estimated with the multi-arm implementation of the BCF algorithm; cardiovascular outcomes (TPR, stroke volume, PEP) used targeted smoothing. Additional details for the study procedures are provided in the legend of Fig. 3. In a, starbursts represent stressor onset. Asterisks in b,c indicate a pre-registered outcome. Control, n = 44; growth only, n = 52; stress only, n = 65; synergistic, n = 39.

Fig. 5. In study 5, the synergistic mindsets intervention reduced negative self-regard.

a, Procedures for study 5. b,c, The synergistic mindsets intervention reduced negative self-regard (see Methods for the scoring of this measure) relative to controls overall and especially on intensely stressful days (b). The intervention also reduced (c) daily salivary cortisol levels overall relative to controls. Participants (n = 119 individuals; n ≤ 1,213 observations (total number of daily diary or cortisol observations for all participants)) were students from low-income families attending a public high school in the United States. Starbursts represent stressor measurements. Univariate marginal distribution plots are shown at the top in b,c. Thick coloured lines represent the 10th to 90th percentiles; grey lines represent the 2.5th to 97.5th percentiles. The vertical dashed line in b represents the cut-off point for high versus low daily stress intensity that was used to estimate subgroup CATEs. The unstandardized CATE for negative self-regard for high daily stress intensity was −0.48 scale points [−0.81, −0.14]; for low daily stress intensity days it was −0.23 scale points [−0.44, −0.02]. The ATEs for academic course credits and cortisol are presented in the text. Control, n = 58; treatment, n = 61.

Fig. 6. In study 6, the synergistic mindsets intervention reduced symptoms of general anxiety relative to controls.

Participants in study 6 (n = 351) were undergraduate students attending a public university in the United States. a, Procedures for study 6. b, The posterior distribution of treatment effects estimated by the Bayesian model, the red line is the CATE for each subgroup, the dark shading marks the interquartile range and the lighter shading marks the 10th to 90th percentiles. Although there was a small posterior probability of a null treatment effect among prior negative mindset participants, there was a higher probability of effects > 0.30 s.d. The prior mindset subgroups used to display treatment effects in b were generated by implementing a hands-off Bayesian decision-making algorithm that maximized the outcome differences among the mindset groups, without using information about magnitudes of treatment effects (see Supplementary Information). Control, n = 172; treatment, n = 179.

Extended Data Fig. 1. In study 3, prior mindsets moderated the treatment effect on TPR during stressful TSST epochs.

In (A), the expected value of TPR reactivity for each epoch and for each prior mindset group, by condition, in (B), an additive summary of the posterior distribution of treatment effects, by negative prior mindset levels, in (C) the conditional average treatment effects (CATEs) for each prior mindset subgroup for each epoch, and in (D) and the interaction between treatment and prior mindsets on TPR responses across TSST epochs. Note: TPR = total peripheral resistance (in dyne-sec x cm⁵). Dots correspond to the expected values (a), CATEs (b), and average of the posterior distribution of a difference in CATEs (C) estimated with the Bayesian algorithm. Thick lines represent the 10^th to 90^th %iles of the posterior distribution; grey lines represent the 2.5^th to 97.5^th %iles. ATE = average treatment effect. In (B), the red line corresponds to the expected partial treatment effect, which corresponds to the offset from the average treatment effect (ATE) at each level of the moderator, holding other potential moderators constant; the dark band is the 10^th to 90^th percentile of the posterior distribution and the light band is the 2.5^th to 97.5^th %iles. The prior mindset subgroups used to display treatment effects in (A), (C) and (D) were identified by implementing a hands-off Bayesian decision-making algorithm that maximized the differences among the mindset groups in terms of the outcome, without using information on the magnitudes of the treatment effects (see SI online). Control n = 86, Treatment n = 74.

Extended Data Fig. 2. In study 3, the synergistic mindsets intervention improved cardiovascular responses to the TSST.

Effects of the intervention on stroke volume (SV)—the amount of blood ejected from the heart during each beat, in ml—were tested because challenge (relative to threat) responses increase SV to facilitate actively addressing stressors^,,. Thus, we anticipated those experiencing challenge-type stress during the stressful TSST epochs should exhibit relatively higher stroke volumes as their bodies distribute oxygenated blood to optimize performance, whereas threatened individuals were expected to have lower stroke volumes during stressful epochs of the TSST as their bodies seek to concentrate blood in the core. SV values reported here are reactivity scores, which means that the average of the 5 min during the baseline epoch were subtracted from each. In (A) the darkest lines correspond to the expected value of the outcome, estimated in the Bayesian model. Dots correspond to the ATEs (B), expected values (C), CATEs (D), and average of the posterior distribution of a difference in CATEs (F) estimated with the Bayesian algorithm. ATE = average treatment effect. In (E), the red line corresponds to the expected partial treatment effect, which corresponds to the offset from the average treatment effect (ATE) at each level of the moderator, holding other potential moderators constant. In all panels, thick bands represent the 10^th to 90^th %iles of the posterior distribution; the lightest/grey lines represent the 2.5^th to 97.5^th %iles. In (B), ATE for Prep = 2.5 ml [1.5, 3.6], Speech = 4.0 ml [3.1, 5.0], Math = 2.9 [1.9, 4.0], Recovery = 0.9 [−.1, 2.0]. In (C), (D) and (F), the prior mindset subgroups used to display the different treatment effects were generated by implementing a hands-off Bayesian decision-making algorithm that maximized the differences among the mindset groups in terms of the outcome, without using information on the magnitudes of the treatment effects. Control n = 86, Treatment n = 74.

Extended Data Fig. 3. In study 3, the effect of the synergistic mindsets intervention on PEP reactivity in milliseconds across TSST epochs.

Pre-ejection period (PEP)—which assesses the contractile force of the heart by measuring the time from onset of ventricular depolarization to aortic valve opening—was examined to test for effects of the intervention on sympathetic nervous system (SNS) arousal. Challenge responses evoke more rapid onset of SNS arousal during a stressor and more rapid recovery to homeostasis after stress offset. Threat-type responses are associated with sustained vigilance for sources of harm and prolonged stress responses, thus threat is associated with slower recovery to baseline after stress offset^,. Whereas all participants should show PEP decreases (leading to a more rapid heart rate) relative to baseline during the stressful epochs^, (see Fig. 1), condition differences are expected to emerge during the recovery period, because controls should be slower to return to homeostasis relative to treated individuals. In (A) the darkest lines correspond to the expected value of the outcome, estimated in the Bayesian model. In (B), dots correspond to the ATEs. ATE = average treatment effect. In both panels, thick bands represent the 10^th to 90^th %iles of the posterior distribution; the lightest/grey lines represent the 2.5^th to 97.5^th %iles. In (B), a positive treatment effect of 2.13 ms [0.8, 3.4] was found during the recovery epoch, as expected. PEP values reported here are reactivity scores, which means that the average of the 5 min during the baseline epoch were subtracted from each. Control n = 86, Treatment n = 74.

Extended Data Fig. 4. In study 5, the synergistic mindsets intervention reduced daily negative self-regard relative to controls the most among people with negative prior mindsets, on their most highly stressful days.

Note: n = 119, n ≤ 1,213 observations. In (A), dots correspond to the average expected value of the outcome, and in (B) dots correspond to the CATEs, estimated by the Bayesian algorithm. CATE = Conditional Average Treatment Effect. In both panels, thick bands represent the 10^th to 90^th %iles of the posterior distribution; the lightest/grey lines represent the 2.5^th to 97.5^th %iles. The CATEs are: Low Daily Stress Intensity, Negative Prior Mindsets CATE = −.19 [−.48, .12], Positive Prior Mindsets CATE = −.23 [−.48, .022]; High Daily Stress Intensity, Negative Prior Mindsets CATE = −.57 [−1.11, −.12], Positive Prior Mindsets CATE = −.41 [−.75, −.07]. Hence, the CATE was 40% for negative prior mindsets participants on high-stress days relative to positive prior mindsets participants. The prior mindset subgroups used to display different treatment effects were generated by implementing a hands-off Bayesian decision-making algorithm that maximized the differences among the mindset groups in terms of the outcome, without using information on the magnitudes of the treatment effects. Control n = 58, Treatment n = 61.

Extended Data Fig. 5. In study 6, an additive summary of the posterior distribution of treatment effects shows greater reductions in anxiety in response to the treatment among those with negative prior mindsets, and this same result is supported when examining the CATEs for positive and negative prior mindsets.

Note: In (A), the red line corresponds to the expected partial treatment effect, which corresponds to the offset from the average treatment effect (ATE) at each level of the moderator, holding other potential moderators constant, estimated in the Bayesian algorithm. In (B) dots correspond to the CATEs, estimated by the Bayesian algorithm. CATE = Conditional Average Treatment Effect. In both panels, thick bands represent the 10^th to 90^th %iles of the posterior distribution; the lightest/grey lines represent the 2.5^th to 97.5^th %iles. Control n = 172, Treatment n = 179.