Brief digital narrative intervention for adolescent depression, anxiety, and insomnia during academic stress: a cluster randomized controlled trial

Abstract

Background

High-stakes examinations represent a significant but underrecognized threat to adolescent mental health, contributing to elevated symptoms of depression, anxiety, and insomnia. Despite the global scale of this problem, effective interventions during these critical periods remain scarce, largely due to implementation barriers in both clinical and educational settings. Scalable, low-resource solutions are urgently needed to address this mental health gap in adolescent care.

Methods

In a cluster randomized controlled trial (ChiCTR2200058881, N = 587), we examined the efficacy of the Guided Narrative Technique (GNT), a brief digital writing intervention, compared to a neutral writing group (NWG). Adolescents preparing for China’s College Entrance Examination within 100 days (M_age = 18.23, SD_age = 0.60) were assigned to GNT (n = 290) or NWG (n = 297) through class-level cluster randomization and completed three consecutive 20-minute daily sessions. The primary outcome was test anxiety, assessed across the intervention and follow-up period. Secondary outcomes were depression, general anxiety, and insomnia, assessed at baseline, post-intervention, and 15-day follow-up.

Results

For the primary outcome, GNT did not produce significantly greater reductions in overall test anxiety than NWG in the full sample. However, GNT was associated with greater reductions on the TAI worry subscale, representing the cognitive component of test anxiety (d = 0.18, 95% CI [0.01, 0.36]) in exploratory subgroup analyses among adolescents with elevated baseline test anxiety. For secondary outcomes, compared with NWG, GNT resulted in significantly greater reductions in depression (d = 0.35, 95% CI [0.16, 0.54]), general anxiety (d = 0.37, 95% CI [0.18, 0.56]), and insomnia (d = 0.23, 95% CI [0.04, 0.42]) during the intervention, with between-group differences also observed for depression and general anxiety at follow-up.

Conclusions

GNT did not significantly reduce overall test anxiety, but showed preliminary benefits for depression, general anxiety, insomnia, and the worry component among adolescents with high baseline anxiety, warranting further evaluation in adequately powered trials.

Trial registration

The study was registered as ChiCTR2200058881.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12916-026-05047-9.

Background

Academic pressure has become a global mental health crisis, with high-stakes examinations contributing to alarming rates of anxiety, depression, and sleep disorders in millions of adolescents annually [1, 2]. From China’s College Entrance Examination (CEE) to standardized tests worldwide, intense academic competition not only threatens students’ immediate mental health but also establishes pathways to chronic psychiatric morbidity and long-term health burden [3, 4]. Despite affecting up to 66% of students worldwide [5, 6], evidence-based interventions remain critically scarce during peak-stress exam periods when mental health support is most urgently needed yet systemically difficult to deliver.

China’s CEE exemplifies this crisis, affecting over 10 million students annually who face extreme psychological pressure with documented health consequences [7–9]. This prolonged stress exposure can lead to hypothalamic-pituitary-adrenal (HPA) axis dysregulation and allostatic load [10], creating cascading effects on immune function, cardiovascular health, and neurological development that extend far beyond the academic period [11]. Yet despite established long-term health consequences [5, 12], rigorous intervention research remains limited during this critical pre-CEE period, leaving millions of adolescents without accessible, evidence-based mental health protection.

While established treatments such as cognitive behavioral therapy (CBT) demonstrate efficacy for academic-related anxiety and depression [13], their resource requirements render them impractical during intensive exam periods. Expressive writing interventions show promise for reducing test anxiety and improving psychological outcomes [14, 15], but their emphasis on negative emotional expression has raised concerns about potential distress exacerbation in some students, and these approaches have not been specifically adapted for high-pressure academic contexts. This intervention gap represents a pressing challenge to protect adolescent mental health at a population scale, precisely when neurobiological vulnerability intersects with maximal environmental stress.

We developed the Guided Narrative Technique (GNT)—a brief digital mental health intervention specifically designed for implementation during critical pre-exam periods. GNT combines evidence-based emotional processing with cognitive reappraisal [16, 17], enabling comprehensive emotion regulation rather than mere stress ventilation. The intervention’s hybrid delivery model integrates structured classroom guidance with mobile-based writing tasks across three 20-minute sessions, designed for scalability within existing educational infrastructure while accommodating intensive study demands.

This cluster-randomized controlled trial addressed the mental health burden during pre-CEE preparation, a highly stressful and consequential period for millions of Chinese adolescents that remains understudied in school-based randomized intervention research. Testing GNT against neutral writing controls, we hypothesized it would significantly reduce greater test anxiety, depression, general anxiety, and insomnia than the control, with effects sustained at a 15-day follow-up. GNT would be particularly effective among students with elevated baseline test anxiety, demonstrating larger reductions in symptoms.

Methods

Study design

This two-arm, parallel cluster randomized controlled trial (RCT) was conducted at a public high school in Inner Mongolia Province, China. Ethical approval was obtained from the Ethics Committee of the School of Psychological and Cognitive Sciences, Peking University (ID: #2022-03-02). The study protocol is available in Additional file 1. Written consent was obtained from a designated school staff member on behalf of the institution, and informed consent was obtained from both the participants and their parents before the baseline assessment. Outcomes were measured at six time points: baseline, post-psychoeducation, two within-session assessments of test anxiety (conducted immediately after sessions 1 and 2), post-treatment (after the 3rd session), and a 15-day follow-up. All assessments and writing tasks were administered via Qualtrics. Data collection occurred between April and May 2022. This study was registered with the Chinese Clinical Trial Registry (ChiCTR2200058881) and reported in accordance with the CONSORT 2025 guideline, and the completed checklist is provided in Additional file 2.

All eligible participants completed a baseline assessment via Qualtrics and attended a 1-hour psychoeducational presentation delivered by two undergraduate psychology majors. The presentation introduced the benefits of structured writing for stress management and provided CEE preparation strategies. The three-day intervention was then implemented in classroom settings. Each day, participants watched an instructional video (pre-recorded by a graduate student in clinical psychology under the supervision of a principal investigator), then completed a 20-minute writing task on their cell phones via Qualtrics, followed by a brief assessment.

Participants

Participants were candidates preparing for the CEE, which was scheduled to take place within the next 100 days. Eligibility criteria were: (a) senior-year student aged 16–20 years; (b) no history of a mental disorder previously diagnosed by a licensed clinician in a medical setting (ascertained via student self-report and verified against school administrative records documenting hospital-based psychiatric diagnoses); (c) Chinese-speaking; (d) residing in the school boarding facilities to support consistent participation; and (e) owning a mobile phone to complete the writing intervention. No compensation was provided apart from access to the intervention.

Randomization and masking

Among the 16 initially eligible classes, head teachers from two classes declined student participation; therefore, 14 classes entered the randomization procedure. Following baseline assessment and the psychoeducational presentation, an independent researcher conducted class-level randomization using computer-generated random numbers (1:1 ratio), assigning classes to either the Guided Narrative Technique (GNT) or Neutral Writing Group (NWG), labelled as groups ‘A’ or ‘B’. The principal investigator and statisticians remained blinded to allocation until the final analysis. No significant between-class differences in age, gender, or baseline psychological outcomes were found prior to randomization.

Interventions

Guided Narrative Technique (GNT)

The GNT guided participants through three daily writing sessions with progressively structured emotional focus. On Day 1, participants wrote about their general thoughts and feelings regarding the upcoming mock examination, targeting emotional awareness and initial appraisals. On Day 2, participants explored negative emotions (e.g., worry, fear, self-blame), reflecting on their sources and impacts on daily life; this structured emotional disclosure prompted articulation and processing of negative affect. On Day 3, participants shifted to positive aspects, identifying positive emotions (e.g., persistence, optimism) and personal growth from exam preparation; this session targeted cognitive reappraisal through meaning-making and identification of personal strengths. Each session began with a brief standardized instructional video outlining the writing focus and procedures.

Neutral Writing Group (NWG)

The NWG participants wrote detailed, objective descriptions of their daily activities across three sessions. They were required to recall and document their day chronologically, including specific details such as wake-up time, meals, class activities, and break times. The instructions emphasized factual documentation without emotional content or personal opinions. Both groups wrote continuously for 20 min per session. All classrooms began the intervention on the same calendar date, and all assessments were administered simultaneously across conditions. Participants were assured that their writing content would remain confidential.

Measures

The primary outcome was test anxiety measured by the Test Anxiety Inventory (TAI), a 20-item self-report scale with two 8-item subscales: worry and emotionality [18]. Each item is rated on a 4-point Likert scale (1 = almost never to 4 = almost always). Total scores range from 20 to 80, with higher scores indicating greater anxiety (20–35: no/minimal; 36–50: mild; 51–65: moderate; 66–80: severe). The TAI includes two 8-item subscales: worry (cognitive concerns about performance) and emotionality (physical and emotional responses to testing). Although these subscales can be scored separately, prior work supports a dominant general factor, justifying use of the total score as an overall test anxiety index [19]. In this study, Cronbach’s α was 0.950 for the total score and 0.903 for each subscale.

The secondary outcomes were insomnia, assessed using the Insomnia Severity Index (ISI) [20], and emotional distress, assessed using the Revised Child Anxiety and Depression Scale (RCADS) [21]. RCADS includes 15 items for general anxiety and 10 for depression, rated on a 4-point Likert scale (0 = never, 3 = always). RCADS has demonstrated good validity and internal consistency in the Chinese population [22], with a reliability coefficient of 0.910 in this study (depression subscale = 0.841, anxiety subscale = 0.855). ISI is a 7-item self-report scale assessing the severity, consequences, and distress associated with insomnia. Each item is rated on a 0–4 scale, yielding a total score of 0–28, with higher scores indicating more severe insomnia. The ISI has shown good validity and reliability for evaluating treatment effectiveness in insomnia [23], with an internal consistency coefficient of 0.857 in this study.

Demographic information, including age and gender, was collected at baseline. Adverse events were documented through self-report at each assessment point.

Statistical analysis

All analyses were conducted using R 4.2.3, with a two-sided significance level of 0.05. Baseline group differences were assessed using independent t-tests and chi-square tests. Following an intent-to-treat approach [24], linear mixed models with two levels (repeated measurements nested within individuals) were employed using the lmerTest package. Due to the limited number of clusters (14 classrooms), a three-level model including classroom-level random effects could not be reliably estimated. We therefore used a two-level model, with repeated observations nested within individuals, as the primary analysis. To address potential classroom-level clustering, we conducted sensitivity analyses using cluster-robust standard errors. We also estimated design effects for each outcome using DEFF = 1 + (m − 1) × ICC, where m denotes the average classroom size. For outcomes with DEFF > 1.5, we conducted an additional conservative sensitivity check by applying a design-effect adjustment to the standard errors, and the substantive conclusions were unchanged.

Piecewise models with random slopes were selected as final models based on the lowest Akaike Information Criterion (AIC). Models adopted the maximal random-effects structure to ensure conservative and generalizable inference [25]. Age, gender, and baseline outcome scores were initially included as covariates. Age and gender were removed as they were not significant in any model; baseline outcome scores were significant in all models and were retained. Detailed information on model selection and final model specifications is provided in the Additional file 1: eMethods.

Treatment effectiveness was primarily evaluated using condition × time interactions. Between-group comparisons at post-treatment and follow-up, derived from the same longitudinal model, were examined as supplementary information. Interaction effect sizes were calculated by growth modeling analysis d (GMA d) [26], computed by multiplying the regression coefficient (b) by the time duration and dividing by the pooled baseline standard deviation. This metric corresponds to Cohen’s d for longitudinal data [27]. Between-group effect sizes (d) at post-treatment and follow-up were calculated by dividing model-estimated mean differences by the pooled baseline standard deviation across both conditions [28]. Exploratory subgroup analyses were conducted among participants with baseline TAI scores greater than 35, indicating mild or higher test anxiety.

The sample size was determined by a formal power analysis conducted prior to enrollment. Assuming a small-to-medium effect size (d = 0.25) based on previous expressive writing research, 576 participants (288 per group) were required to achieve 85% power at α = 0.05, after adjusting for 10% anticipated attrition and classroom-level clustering. The final enrolled sample (N = 587) exceeded this requirement and included all eligible consenting students. Full details of the sample size calculation are provided in Additional file 1: Study Protocol.

Results

Participant characteristics

In March 2022, 587 of 726 eligible high school seniors from an urban public school in Inner Mongolia Province enrolled in the study (see Fig. 1 for exclusions). Five days after baseline assessment, all participants attended the psychoeducation session, with 490 (83.5%; GNT = 242, NWG = 248) completing the post-session assessment. Participants attended an average of 2.33 sessions (SD = 1.06) in the GNT condition and 2.47 sessions (SD = 0.97) in the NWG condition. Overall, 409 participants (69.7%) completed all three sessions, and 475 (80.9%) attended at least two sessions. At the 15-day follow-up, 416 participants (70.9%; GNT = 201, NWG = 215) completed the assessment. No adverse events were reported in this study.

Fig. 1.

CONSORT Flow Chart. Note: GNT, Guided Narrative Technique. NWG, Neutral Writing Group.

Baseline characteristics are presented in Table 1. Participants (N = 587) had a mean age of 18.23 years (SD = 0.60), and 56.4% were female. At baseline, 59.3% of participants reported elevated test anxiety (TAI > 35). No significant between-group differences were observed on test anxiety, depression, anxiety, or insomnia at baseline.

Table 1.

Demographics and Psychological Characteristics of Participants in Baseline (N = 587)

Gender	GNT (n = 290) N (%) / Mean (SD)	NWG (n = 297) N (%) / Mean (SD)	Overall (N = 587) N (%) / Mean (SD)
Male	122 (42.07)	134 (45.12)	256 (43.61)
Female	168 (57.93)	163 (54.88)	331 (56.39)
Suffer from academic pressure
Yes (baseline TAI > 35)	169 (58.28)	179 (60.27)	348 (59.28)
No (baseline TAI ≤ 35)	121 (41.72)	118 (39.73)	239 (40.72)
Age	18.22 (0.61)	18.23 (0.60)	18.23 (0.60)
Test Anxiety
Total	39.37 (13.23)	39.94 (13.14)	39.66 (13.18)
Emotionality	16.19 (5.55)	16.48 (5.11)	16.34 (5.33)
Worry	15.30 (5.78)	15.48 (5.44)	15.39 (5.61)
Depressive and Anxiety Symptoms
Total	24.96 (11.69)	25.43 (11.39)	25.20 (11.53)
Depression	10.26 (4.98)	10.60 (5.01)	10.43 (4.99)
Anxiety	14.69 (7.49)	14.84 (7.14)	14.77 (7.31)
Insomnia	8.08 (4.92)	8.03 (4.98)	8.05 (4.95)

Note: TAI, Test Anxiety Inventory

Primary outcome

Following an intent-to-treat approach, all 587 participants were included in analyses. Descriptive statistics for all outcomes across assessment points are presented in Table 2, with corresponding sample sizes in Additional file 1: eTable 1. Full linear mixed model results are presented in Additional file 1: eTable 2.

Table 2.

Descriptive Statistics for Outcomes Across All Assessments [M (SD)]

		Baseline (T1)	Post-Psychoeducation (T2)	1st session (T3)	2nd session (T4)	Post-treatment (T5)	15-day follow-up (T6)
Test Anxiety (TAI)
Total	GNT	39.37 (13.23)	39.44 (12.22)	38.75 (12.69)	38.46 (12.31)	38.44 (12.11)	39.59 (12.87)
	NWG	39.94 (12.14)	40.85 (11.68)	39.16 (11.70)	38.87 (12.10)	39.14 (11.96)	40.3 (11.76)
Emotionality	GNT	16.19 (5.55)	16.25 (5.05)	16.19 (5.13)	16.25 (5.09)	16.11 (4.97)	16.42 (5.16)
	NWG	16.48 (5.11)	16.69 (4.56)	16.13 (4.72)	16.01 (4.67)	16.22 (4.63)	16.73 (4.65)
Worry	GNT	15.30 (5.78)	15.48 (5.32)	14.97 (5.36)	14.64 (5.18)	14.83 (5.07)	15.20 (5.54)
	NWG	15.48 (5.44)	16.00 (5.07)	15.31 (5.12)	15.09 (5.35)	15.19 (5.34)	15.67 (5.20)
Depression and Anxiety Symptoms (RCADS)
Total	GNT	24.96 (11.69)	22.64 (13.45)	–	–	20.25 (13.08)	20.72 (14.35)
	NWG	25.43 (11.39)	24.08 (12.67)	–	–	23.13 (12.37)	22.33 (13.75)
Depression	GNT	10.26 (4.98)	9.36 (5.53)	–	–	8.71 (5.65)	8.80 (6.00)
	NWG	10.60 (5.01)	10.30 (5.70)	–	–	9.96 (5.44)	9.67 (5.95)
Anxiety	GNT	14.69 (7.49)	13.28 (8.47)	–	–	11.54 (8.00)	11.92 (8.78)
	NWG	14.84 (7.14)	13.78 (7.60)	–	–	13.17 (7.53)	12.67 (8.27)
Insomnia (ISI)
	GNT	8.08 (4.92)	7.20 (5.08)	–	–	6.48 (4.73)	6.00 (5.09)
	NWG	8.03 (4.98)	7.41 (5.09)	–	–	7.43 (4.94)	6.87 (5.38)

Note: GNT, Guided Narrative Technique. NWG, Neutral Writing Group. TAI, Test Anxiety Inventory. RCADS, Revised Child Anxiety and Depression Scale. ISI, Insomnia Severity Index. Sample sizes for each assessment are detailed in Additional file 1: eTable 1

For test anxiety (TAI total score and worry and emotionality subscales), condition×time interactions were non-significant across both the intervention and follow-up phases (all ps > 0.10; Additional file 1:eTable 2), indicating that the groups did not differ in their trajectories of change. Temporal trends for test anxiety are illustrated in Fig. 2.

Fig. 2.

Estimated Mean Scores for Test Anxiety (TAI) at All Time Points (N = 587). Note: Data is from Linear Mixed Model Analysis. Panels represent (A) TAI total score, (B) TAI subscale of worry, and (C) TAI subscale of emotionality. TAI, Test Anxiety Inventory (scoring 20 to 80). The X-axis represents days from baseline (1: baseline, 6: post-psychoeducation, 7: process assessment after the 1st session, 8: process assessment after the 2nd session, 9: post-treatment, 24: 15-day follow-up). Background colors indicate different study periods: orange for the psychoeducation session, green for the 3-day writing intervention, and purple for the follow-up. Error bars represent standard error (SE) from the mixed-model analysis. ^* 0.01 < p < .05.

Secondary outcomes

Regarding depression and general anxiety, for the combined RCADS score, the condition×time interaction was significant during the intervention phase (b = 0.331, t₁₅₀₂ = 3.23, p = .001, d = 0.23), indicating greater reductions in the GNT group. At post-treatment, the GNT group had significantly lower scores than the NWG group (b = 2.619, t₁₄₆₂ = 3.99, p < .001, d = 0.39), and a smaller group difference was observed at 15-day follow-up (b = 1.520, t₁₈₁₄ = 1.97, p = .049, d = 0.10). Similar patterns were found for both subscales, with the GNT group showing superior outcomes in depression (condition×time: b = 0.124, t₁₅₀₀ = 2.72, p = .007, d = 0.14; post-treatment difference: b = 1.052, t₁₄₆₉ = 3.61, p < .001, d = 0.35) and anxiety (condition×time: b = 0.206, t₁₅₀₄ = 3.19, p = .001, d = 0.16; post-treatment difference: b = 1.570, t₁₄₈₁ = 3.83, p < .001, d = 0.37). The condition × time interaction during the follow-up phase was non-significant, indicating no differential change between groups during this period. Temporal changes are presented in Fig. 3 and detailed results of dimensions are in Additional file 1: eResults.

Fig. 3.

Estimated Depression, Anxiety (RCADS), and Insomnia (ISI) Over Time (N = 587). Note: Data is from Linear Mixed Model Analysis. Panels represent (A) RCADS total score, (B) RCADS subscale of depression, (C) RCADS subscale of anxiety, and (D) insomnia. RCADS, Revised Child Anxiety and Depression Scale (scoring 0 to 75). ISI, Insomnia Severity Index (ranging from 0 to 28). The X-axis represents days from baseline (1: baseline, 6: post-psychoeducation, 9: post-treatment, 24: 15-day follow-up). Background colors indicate different study periods: orange for the psychoeducation session, green for the 3-day writing intervention, and purple for the follow-up period. Error bars represent standard error (SE) from the mixed-model analysis. ^* p < .05, ^*** p < .001.

For insomnia (Fig. 3D), the condition×time interaction during the intervention phase was significant (b = 0.089, t₁₄₉₆ = 2.11, p = .035, d = 0.14), indicating greater reductions in the GNT group. At post-treatment, the GNT group reported significantly lower ISI scores than the NWG group (b = 0.633, t₁₄₇₄ = 2.35, p = .019, d = 0.23), though both groups decreased in the period (GNT: b = − 0.171, t₁₅₀₀ = − 5.64, p < .001, d = − 0.29; NWG: b = − 0.082, t₁₄₉₃ = − 2.75, p = .006, d = − 0.14). The condition × time interaction during follow-up was non-significant, indicating that the groups did not differ in their rate of change during this phase. At the 15-day follow-up, the between-group difference was small and not statistically significant (b = 0.54, t₁₈₂₄ = 1.71, p = .087, d = 0.20).

Exploratory analysis: subsample with elevated test anxiety

Among participants with elevated baseline test anxiety (TAI > 35; n = 348), a significant condition×time interaction emerged for the worry subscale during the intervention phase (b = 0.105, t₁₂₁₂ = 2.10, p = .036, d = 0.18), indicating greater reductions in the GNT group. Between-group comparisons revealed that the GNT group had significantly lower worry scores than the NWG group at post-treatment (b = 0.699, t₅₆₀ = 2.18, p = .029, d = 0.25) and 15-day follow-up (b = 1.067, t₁₃₄₀ = 2.43, p = .015, d = 0.38), though both groups showed significant reductions during the intervention(GNT: b = − 0.217, t₁₂₁₀ = − 6.10, p < .001, d = − 0.35; NWG: b = − 0.113, t₁₂₁₄ = − 3.26, p = .001, d = − 0.19).

No significant interactions were observed for TAI total score or emotionality during the intervention or follow-up phases (all ps > 0.10). However, at the 15-day follow-up, the GNT group showed significantly lower total test anxiety (b = 2.124, t₁₃₀₈ = 2.19, p = .029, d = 0.34), and emotionality (b = 0.851, t₁₃₇₈ = 2.04, p = .041, d = 0.31) compared with NWG. These temporal patterns are illustrated in Fig. 4, with full model results in Additional file 1: eTable 3.

Fig. 4.

Estimated Mean Score of Test Anxiety Inventory Over Time in Subgroup (n = 348). Note: Data is from the Linear Mixed Model Analysis. Panels represent (A) TAI total score, (B) TAI subscale of worry, and (C) TAI subscale of emotionality. TAI, Test Anxiety Inventory (scoring 20–80). The X-axis represents days from baseline (1: baseline, 6: post-psychoeducation, 7: process assessment after the 1st session, 8: process assessment after the 2nd session, 9: post-treatment, 24: 15-day follow-up). Background colors indicate different study periods: orange for the psychoeducation session, green for the 3-day writing intervention, and purple for the follow-up period. Error bars represent standard error (SE) from the mixed-model analysis. ^* 0.01 < p < .05.

Discussion

This cluster-randomized controlled trial examined whether GNT and neutral writing exercises differentially reduced academic stress-related symptoms among adolescents. During the intervention period, GNT demonstrated superior effectiveness relative to NWG in reducing overall emotional distress (RCADS total score) and insomnia (ISI total score) with an interaction GMA d = 0.23 and 0.14, respectively. During the follow-up period approaching the examination, the between-group differences were maintained but did not further diverge. For test anxiety, however, both conditions showed similar reductions, with no significant between-group differences in change trajectories.

This pattern suggests that GNT may function primarily as an early stress-buffering intervention, conferring protection against escalating distress rather than producing cumulative symptom reductions during the highest-pressure phase. The intervention appeared to provide short-term reductions in emotional distress and insomnia at a critical pre-exam stage, without additional divergence once academic demands intensified. This finding highlights the time-sensitive nature of brief interventions delivered before major examinations and suggests that repeated or booster sessions closer to the examination date may be necessary to sustain or amplify their relevance during peak stress periods.

Delivered digitally through structured writing exercises, GNT targets a critical yet often overlooked period, the final months before a high-stakes, consequential examination, when psychological support is most needed but logistically challenging to deliver. Rather than functioning as a long-term preventive program [9], GNT serves as a timely, pre-exam strategy, designed to help students process academic stress, regulate emotions, and develop adaptive coping before distress escalates. The observed effects likely reflect short-term self-regulatory benefits that help prevent anxiety escalation rather than large absolute symptom reductions. Its low-cost, scalable digital format aligns with recommendations for universal school-based mental health interventions in low- and middle-income countries [29], offering a practical, resource-efficient solution for broader implementation.

The effect sizes observed in this study are consistent with comparable interventions. The differential improvement in emotional distress favoring GNT (GMA d = 0.23) is similar to those reported for universal school-based programs (Hedges’ g = 0.18–0.23) [30, 31] and numerically similar to, though slightly higher than, the mean effect size for traditional expressive writing on adolescent emotional distress (d = 0.11, 95% CI [0.02, 0.19]) [14]. For insomnia, the between-group effect at post-treatment (d = 0.23) falls within the confidence interval of a 6-week app-based cognitive behavioral intervention for adolescent insomnia (d = 0.41, 95% CI [0.10, 0.72]) [32], although this comparison should be interpreted cautiously given the wide confidence interval and differences in population and intervention intensity. These comparisons are intended as contextual benchmarking rather than claims of direct equivalence given differences in study design, populations, and outcome measures. Notably, GNT’s practical significance is enhanced by timing: the intervention occurred during the final weeks before the College Entrance Examination when academic pressure was already high and continuing to intensify. A 0.23 SD advantage during this critical period, when anxiety typically escalates, suggests the intervention provided meaningful stress buffering precisely when students needed it most.

Beyond traditional expressive writing, GNT integrates cognitive reappraisal with emotional disclosure, enhancing adaptive emotion regulation and cultural responsiveness [9, 33]. By guiding students to acknowledge distress while reframing academic challenges, GNT may facilitate shifts in attentional processing and self-regulation, reducing excessive worry and rumination that drive anxiety and depression [34]. This structured approach mitigates the risk of distress amplification observed in solely disclosure-based writing [35] while aligning with collectivist norms of emotional expression [33]. These features may explain GNT’s superior outcomes compared to single-dimensional regulatory interventions [9, 14, 15], and underscore the importance of designing mental health interventions that accommodate different emotion regulation styles rather than assuming a one-size-fits-all approach [36].

Although GNT was designed to target pre-exam distress, its effects on depression and anxiety symptomatology should be interpreted as short-term emotional buffering during a period of intensifying academic stress rather than durable mental health benefits. Whether these effects have any bearing on longer-term outcomes, including the well-documented link between chronic academic stress and risk for clinical anxiety and depression [3, 5], remains an open question requiring longitudinal designs with extended follow-up and clinical outcome assessments.

The observed improvements in insomnia are consistent with theoretical models describing the reciprocal association between stress and sleep dysregulation [37], although the present study did not directly assess the underlying mechanisms. Possible explanations include reduced perceived stress, enhanced emotion regulation through gratitude practice, or improved coping strategies. However, distinguishing among these pathways would require mediation analyses beyond the scope of the present study. These findings add to evidence highlighting bidirectional links between sleep and emotional functioning in academic contexts [38] and suggest that brief school-based interventions targeting pre-exam stress warrant further investigation as a means of supporting adolescent sleep health [39].

The sustained post-treatment benefits among highly test-anxious students support the value of targeted interventions for vulnerable subgroups. Precision mental health models advocate for identifying “what works for whom,” and our findings suggest that brief screening for academic stress could optimize intervention delivery [40]. Future studies could employ computational text analysis to examine whether specific narrative features (e.g., use of insight words, causal reasoning, emotional tone shifts) are associated with symptom improvement.

From a global mental health perspective, GNT addresses the critical gap between high mental health needs and limited resources [29]. Its meaningful effects, achieved through a low-cost digital format, challenge the notion that effective interventions require extensive clinical infrastructure [30, 31] or academic disruption [41]. Future research should evaluate its cost-effectiveness, analyze its long-term mental health benefits, examine durability beyond 15 days, identify key mechanisms, and optimize implementation across diverse educational settings. Cultural adaptations should refine the balance between emotional expression and cognitive reframing to align with local norms [6, 7, 33].

Several limitations should be considered. First, the study was conducted in a single high school, which may limit generalizability despite the large sample and does not fully exclude contamination between classes via peer discussion. Second, due to privacy and administrative restrictions on CEE performance data, we could not obtain CEE performance data and therefore could not evaluate whether psychological improvements translated into academic gains. Third, the 15-day follow-up, restricted by the approaching examination, limits conclusions about long-term effects. After this assessment, students entered the final 40-day preparation period, during which additional data collection would have imposed undue burden and was not ethically feasible. While meta-analyses suggest expressive writing benefits may persist or grow over time [14], future research should examine whether GNT’s effects are sustained beyond the pre-exam period. Forth, the current design does not estimate effects relative to a no-treatment condition, which should be addressed in future studies (e.g., via a waitlist control). Fifth, the study was not powered to examine potential moderators such as gender. Finally, we did not assess broader attrition-related covariates (e.g., family background), and the absence of such variables limits our ability to verify the MAR assumption underlying our mixed-model analyses. If this assumption is violated, the estimated treatment effects may be biased.

Conclusions

This study did not demonstrate effectiveness of GNT on its primary outcome of overall test anxiety. However, GNT was associated with short-term improvements in depression, general anxiety, and insomnia among adolescents facing high-stakes academic stress, with modest but meaningful effect sizes, and exploratory analyses suggested benefits on the cognitive (worry) component among those with elevated baseline test anxiety. Given their secondary and exploratory nature, these findings are encouraging but preliminary. Future research should confirm and extend them across broader populations and diverse educational settings.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

GNT

Guided narrative technique

NWG

Neutral writing group

CEE

College entrance examination

TAI

Test Anxiety Inventory

RCADS

Revised Child Anxiety and Depression Scale

ISI

Insomnia Severity Index

Author contributions

All authors read and approved the final manuscript. YL: conceptualisation, study design, project administration, intervention implementation, data analysis, funding acquisition, writing—original draft, and writing—review & editing; JF: conceptualisation, study design, project administration, intervention implementation, data analysis, and writing—review & editing; YZ: conceptualisation, study design, funding acquisition, investigation, methodology, data analysis, project administration, supervision, validation, writing—original draft, and writing—review & editing.

Funding

This study was supported by grants from the National Natural Science Foundation of China (32371139, 32000776) and the Beijing Natural Science Foundation (QY24067). The foundations had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Data availability

Authors wishing to access the data should contact the corresponding author and provide a statistical analysis plan addressing a new research question. The study trial management group will discuss all requests.

Declarations

Ethics approval and consent to participate

Ethical approval was obtained from the Ethics Committee of the School of Psychological and Cognitive Sciences, Peking University (ID: #2022-03-02). Written consent was obtained from a designated school staff member on behalf of the institution, and informed consent was obtained from both the participants and their parents before the baseline assessment.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.