Population-Based RCT of a Digital Cognitive-Behavioral Guided Self-Help Intervention for Anxiety, Depression, and Eating Disorders in College Students

Abstract

Scalable approaches like digital cognitive-behavioral therapy guided self-help (D-CBTgsh) may help close the treatment gap for college students with mental disorders. In a randomized clinical trial (ClinicalTrials.gov: NCT04162847) across 26 U.S. colleges, populations were offered a mental health screen (39,194 assessed). Students with clinical levels or high-risk for anxiety, depression, and/or eating disorders (N=6,205) were randomized to screening+D-CBTgsh or screening+referral-to-college-provided-care. Screening+D-CBTgsh reduced prevalence of any mental disorder (primary outcome) at 6 weeks (OR=0.80, 95% CI=0.70–0.91), 6 months (OR=0.77, 95% CI=0.68–0.88), and 2 years (OR=0.82, 95% CI=0.72–0.93). Services uptake was greater in screening+D-CBTgsh (74.4%) vs screening+referral (30.2%) at 6 months (OR=6.72, 95% CI=6.01–7.52) and 2 years (OR=1.83, 95% CI=1.64–2.04), including for minoritized groups. Screening+D-CBTgsh (vs. screening+referral-to-college-provided-care) also improved dimensional outcomes of generalized anxiety, social anxiety, depression, eating disorder symptoms, and mental health functioning. Findings supported transdiagnostic prevention and intervention benefits of screening+DCBTgsh and its viability as a scalable, population-based approach.

Introduction

Mental disorders and demand for services in college students are increasing worldwide. Epidemiological data from the U.S. suggests that three quarters of lifetime mental disorders emerge during the typical college age range (18–24 years old).¹ In college students, anxiety, depression, and eating disorders (EDs) are among the most common, with up to 40–60% of college students globally experiencing them.^2–5 Left untreated, mental health symptoms become greater in severity, frequency, and duration over time, which can have lasting impacts on educational attainment, employment, and social relationships.⁶ Thus, access to timely interventions for these concerns is crucial.

Although many campus counseling centers have increased capacity for service provision in recent years, demand has risen to exceed capacity^7,8—a situation only worsened by the COVID-19 pandemic.⁹ Of U.S. students with mental health problems, only 20–34% receive treatment,^8,10 with this gap more pronounced for students of color and those with financial hardships.^10,11 Increasing the number of mental health providers on college campuses is costly. Furthermore, this neither addresses the problem of students waiting to seek services until they are at a crisis level (if they seek treatment at all) nor does it alleviate campus leaders’ concerns about demand continuing to increase. These issues highlight the critical need for a new, more efficient and proactive model for early identification and provision of services to populations of college students with anxiety, depression, and ED symptoms.

Digital technologies offer great potential for improving access to mental health services on college campuses. Such technologies have been successfully used to offer screens and provide intervention for anxiety, depression, and EDs across a variety of individuals, including U.S. college students,¹² among whom smartphone ownership is nearly ubiquitous (97% of 18–29 year-olds own a smartphone).¹³ Yet, a key challenge remains that, there has not been a systematic, efficient way to identify students within the population with or at high-risk for mental health problems and connect them to timely, tailored care implemented broadly on college campuses. Preliminary work such as the Screening and Treatment for Anxiety and Depression (STAND) program¹⁴ explored college student screening and triage models for clinical levels of general anxiety or depression in open trials. Nonetheless, evidence from large-scale randomized studies—particularly those addressing those at risk for or at clinical levels of five mental health disorders (generalized anxiety disorder (GAD), social anxiety disorder (SAD), panic disorder (PD), depression, and EDs) relevant to college students—remains lacking.

Related to the overarching challenge of needing a streamlined way to identify students with mental health needs and connect them with tailored care, there have recently been calls for more efficient approaches to the dissemination of psychological interventions at the population level—as one key part of employing a more strategic approach to address population mental health and ensure the mental health of all individuals.^15,16 A population-level approach would ideally begin with making the entire population aware of the availability of mental health screening. Such screening would also offer feedback on symptoms and refer students to potential mental health support. Those for whom mental health support was appropriate could be provided with immediate access to low-cost digital prevention and intervention tools. It would also be beneficial if such tools addressed multiple sequentially and concurrently comorbid mental health problems, regardless of whether they were present at baseline.¹⁶ This approach would offset in-person clinical demands on college campuses, improve outcomes at the population level,¹⁷ and ultimately represent a paradigm shift in how mental health was addressed on college campuses.

The current study was designed to answer the call to deliver efficient and effective population-based access to screening plus mental health services. The primary goal was to examine the effectiveness of digital cognitive-behavioral therapy guided self-help (D-CBTgsh), following guideline-based mental health screening with feedback,^18,19 for both prevention of the development of new disorders and treatment of anxiety, depression, and EDs among students at U.S. colleges and universities. Full populations of students on participating U.S. college and university campuses were invited to complete screens and receive tailored feedback on their symptoms. Those identified as at risk for and with clinical levels of GAD, SAD, PD, depression, and/or EDs, based on self-report, and not currently in treatment were offered the opportunity to participate in the next phase of the study (i.e., the randomized clinical trial). Students who consented were randomized to screening+D-CBTgsh or screening+referral-to-college-provided-care on their respective college campus (control). We hypothesized that screening+D-CBTgsh would be associated with lower prevalence of having any mental disorder compared to screening+referral to college provided care. That is, we hypothesized that screening+D-CBTgsh would be superior to screening+referral in the prevention and treatment of GAD, SAD, PD, depression, and EDs, even if these disorders were not present at baseline. We also hypothesized that uptake of services would be greater in the screening+D-CBTgsh group compared to screening+referral to college provided care, including for students from minority demographic groups (e.g., non-White, Hispanic/Latino, non-heterosexual, first-generation, with financial hardship). In addition to examining disorder prevalence, we also compared the randomized conditions on changes in dimensional disorder-specific symptoms as well as on changes in quality of life and functioning. We hypothesized that improvements on these dimensions would be superior in screening+D-CBTgsh compared to screening+referral to college provided care. Furthermore, we examined if any of these outcomes were moderated by baseline clinical status (i.e., having self-reported clinical levels of symptoms vs. at high-risk for development of them).

Results

Descriptive Statistics

We sent email invitations to full populations of students at 26 participating U.S. colleges and universities to complete a survey about student health and well-being. There were 39,194 students who answered the screen. Of those, 19,241 had a clinical level of or were at high-risk for self-reported anxiety, depression, and/or EDs. Of 7,080 who met all inclusion criteria and consented to participate, N=6,205 completed the baseline survey and were randomized, with 3,103 randomized to screening+D-CBTgsh and 3,102 randomized to screening+referral-to-college-provided-care. Among randomized participants, 4,538 met criteria for at least one clinical disorder (2,264 in screening+D-CBTgsh and 2,274 in screening+referral-to-college-provided-care), and 1,667 were at high-risk for at least one disorder without meeting criteria for a clinical disorder (839 and 828, respectively), based on self-report. We use the terms “clinical” and “high-risk” subgroups to refer to participants who screened positive for a clinical-level disorder or at high-risk at baseline, respectively. Overall, 72.6% of consented participants completed at least one follow-up assessment (screening+D-CBTgsh: 2,120 of 3,103 participants [68.3%]; screening+referral-to-college-provided-care: 2,387 of 3,102 participants [77.0%]). It is also important to mention that 99.67% of those who completed the screen owned a smartphone and thus, this was not a criterion that excluded many participants. See Figure 1 for the flow diagram and Table 1 for sample characteristics.

Fig.1 |. Participant flow diagram.

This figure presents the CONSORT flow of participants from screening through the 2-year follow-up. A total of 39,194 students were assessed for eligibility; 6,205 were randomized to screening + D-CBTgsh (n = 3,103) or screening + referral to college-provided care (n = 3,102). Follow-up completion at 6 weeks, 6 months, and 2 years is shown for each group. All randomized participants were included in analyses. D-CBTgsh= digital cognitive-behavioral therapy guided self-help.

Table 1.

Baseline Characteristics

	n (%)
Characteristics	Intervention (n=3,103)	Control (n=3,102)
Age, mean (SD), y	20.2 (4.0)	20.2 (4.0)
Sex assigned at birth (Male)	821 (26.5%)	820 (26.4%)
Sexual orientation
Heterosexual	2,120 (68.3%)	2,108 (68.0%)
Gay/Lesbian	109 (3.5%)	102 (3.3%)
Bisexual	480 (15.5%)	458 (14.8%)
Other	394 (12.7%)	434 (14.0%)
Race
White	2,134 (68.8%)	2,144 (69.1%)
Asian	473 (15.2%)	466 (15.0%)
Black or African American	231 (7.4%)	234 (7.5%)
Native Hawaiian or Pacific Islander	11 (0.4%)	10 (0.3%)
American Indian or Alaskan Native	35 (1.1%)	42 (1.4%)
Multiracial	219 (7.1%)	206 (6.6%)
Hispanic ethnicity	519 (16.7%)	527 (17.0%)
Year in school
Undergraduate year 1	1,320 (42.5%)	1,313 (42.3%)
Undergraduate year 2	902 (29.1%)	922 (29.7%)
Undergraduate year 3	449 (14.5%)	460 (14.8%)
Undergraduate year 4	351 (11.3%)	337 (10.9%)
Undergraduate year 5	81 (2.6%)	70 (2.3%)
Financial hardship
Not very hard	1,116 (36.0%)	1,131 (36.5%)
Somewhat hard	1,036 (33.4%)	1,028 (33.1%)
Hard	585 (18.9%)	590 (19.0%)
Very hard	366 (11.8%)	353 (11.4%)
Parent education
Bachelor’s degree or higher	2,026 (65.3%)	2,022 (65.2%)
Less than bachelor’s degree	1,077 (34.7%)	1,080 (34.8%)
School sector
Public	2,687 (86.6%)	2,700 (87.0%)
Private	416 (13.4%)	402 (13.0%)
Diagnosis (based on self-report)
Depression	1,836 (59.2%)	1,829 (59.0%)
Generalized anxiety disorder	1,323 (42.6%)	1,276 (41.1%)
Social anxiety disorder	721 (23.2%)	697 (22.5%)
Panic disorder	418 (13.5%)	432 (13.9%)
Eating disorder	577 (18.6%)	574 (18.5%)
Clinical Status
Clinical	2,264 (73.0%)	2,274 (73.3%)
High-risk	839 (27.0%)	828 (26.7%)

Note. Financial hardship was assessed using item, “Financial difficulty during the past month was assessed using an item: “In the past month during college, how hard has it been for you to pay for expenses like tuition, textbooks, course materials, food, housing, and medical care?”; Diagnosis (based on self-report) refers to number of students who met clinical level for each disorder. Clinical status was determined at baseline and categorized as high-risk (elevated symptoms without meeting criteria for any clinical disorder) or Clinical (meeting criteria for at least one clinical disorder).

Primary Outcome

There was significantly lower prevalence of having any mental disorder in the screening+D-CBTgsh condition vs the screening+referral-to-college-provided-care condition at the 6-week, 6-month, and 2-year follow-ups (Table 2). Compared to screening+referral-to-college-provided-care, those randomized to screening+D-CBTgsh had 4.3% lower prevalence of having any mental disorder at 6 weeks (odds ratio [OR] = 0.80, 95% confidence interval [CI] = 0.70, 0.91), 4.9% lower prevalence at 6 months (OR = 0.77, 95% CI = 0.68, 0.88), and 3.8% lower prevalence at 2 years (OR = 0.82, 95% CI = 0.72, 0.93). See also Figure 2.

Table 2.

Number of People with at Least One Clinical Mental Disorder at Each Timepoint

Time	Intervention (n=3,103) n (%)	Control (n=3,102) n (%)	p-value	z-value	Odds ratio	95% CI
Baseline	2,264 (73.0%)	2,274 (73.3%)	.719	−0.36	0.98	0.85–1.12
6-week	1,764 (56.8%)	1,896 (61.1%)	<.001	−3.51	0.80	0.71–0.91
6-month	1,743 (56.2%)	1,894 (61.1%)	<.001	−4.00	0.77	0.68–0.88
2-year	1,723 (55.5%)	1,841 (59.3%)	.002	−3.11	0.82	0.72–0.93

Note. Multilevel logistic regression coefficients were tested using Wald z-tests, which do not have associated degrees of freedom. Odds ratios (ORs) and 95% confidence intervals (CIs) were obtained by exponentiating the estimated log-odds coefficients. An OR greater than 1 indicates higher odds of having at least one clinical mental disorder in the intervention group compared to the control group, and an OR less than 1 indicates lower odds. Tests were two-tailed.

Figure 2. Prevalence of mental disorders and uptake of mental health services by condition.

(A) Percentage of participants meeting criteria for at least one of five mental disorders at each time point by treatment condition. (B) Six-month service uptake by condition. Uptake was defined as any access to the mobile platform in the intervention condition, and, in the control condition, reporting ≥1 counseling/therapy visit or taking a new (i.e., not taken at baseline) antidepressant or anti-anxiety medication at the 6-month follow-up. D-CBTgsh= digital cognitive-behavioral therapy guided self-help. D-CBTgsh=digital cognitive-behavioral therapy guided self-help.

Clinical status (i.e., clinical vs. high-risk) moderated the prevalence differences between conditions at 6 weeks (OR = 0.71, 95% CI = 0.54, 0.92, p = .01, z = −2.56), but not at 6-month (OR = 0.83, 95% CI, 0.64–1.09; p = .17, z = −1.36) or 2-year follow-up (OR = 1.16, 95% CI = 0.89.51, p = .26, z = 1.13). Specifically, at 6 weeks there was significantly lower caseness among clinical participants assigned to screening+D-CBTgsh compared to screening+referral-to-college-provided-care (OR = 0.74, 95% CI = 0.64, 0.85, p < 0.01, z = −4.33). However, there was no significant difference between the two conditions among high-risk participants at 6 weeks (OR = 1.04, 95% CI = 0.83, 1.30, p = .74, z = 0.33). Thus, whereas only the clinical participants showed lower caseness from screening+D-CBTgsh compared to screening+referral-to-college-provided-care at 6 weeks, the full sample showed greater benefit at 6 months and 2 years.

Secondary Outcomes

Uptake

See Figure 2 and Table 3 for outcomes on uptake of services across conditions. The majority of those randomized to screening+D-CBTgsh (74.4%) began the program, whereas only 30.2% of those randomized to screening+referral-to-college-provided-care reported any therapy or new medication (antidepressants or anti-anxiety) at 6-month follow-up. Odds of mental health services uptake were nearly 7 times greater for intervention vs control participants (OR = 6.72, 95% CI = 6.01–7.52), and baseline clinical status did not moderate this effect (OR = 0.92, 95% CI = 0.71–1.19, p = .53, z = −0.62). Notably, compared to screening+referral-to-college-provided-care, those randomized to screening+D-CBTgsh had significantly higher likelihood of uptake of mental health services across all subgroups including White and non-White, non-Hispanic/Latino and Hispanic/Latino, heterosexual and non-heterosexual, non-first generation and first generation, and low and high financial hardship. Between-group differences in uptake of services remained statistically significant even when considering any therapy or new medication (antidepressants or anti-anxiety) reported by control participants over the entire 2-years they were in the study (compared to the 6 months those in D-CBTgsh had digital platform access). Specifically, 61.4% of control participants reported receiving any mental health services during the 2 years they were in the trial. Odds were nearly 2 times greater for the 6-month uptake within D-CBTgsh (74.4%; OR = 1.83, 95% CI = 1.64, 2.04). There was no significant interaction between study condition and clinical status in these models (OR = 1.04, 95% CI = 0.82, 1.32, p = .74, z = 0.34). Moreover, the intervention continued to show significantly higher likelihood of services uptake across all subgroups compared to those in the control group, including White and non-White, non-Hispanic/Latino and Hispanic/Latino, heterosexual and non-heterosexual, non-first generation and first generation, and low financial hardship and high financial hardship.

Table 3.

Comparison of Uptake by Condition and by Minority/Majority Subgroups

Variable	Intervention	Control	p-value	z-value	Odds ratio	95% CI
	n/total n (%)	n/total n (%)
Overall Uptake	2,308/3,103 (74.4%)	938/3,102 (30.2%)	<.001	33.50	6.72	6.01–7.52
Race
White	1606/2134 (75.3%)	653/2144 (30.5%)	<.001	28.18	6.96	6.08–7.97
Non-White	702/969 (72.4%)	285/958 (29.7%)	<.001	18.11	6.21	5.09–7.57
Hispanic Ethnicity
Non-Hispanic	1921/2584 (74.3%)	786/2575 (30.5%)	<.001	30.33	6.63	5.87–7.50
Hispanic	387/519 (74.6%)	152/527 (28.8%)	<.001	14.12	7.35	5.57–9.70
Sexual orientation
Heterosexual	1514/2120 (71.4%)	583/2108 (27.7%)	<.001	27.41	6.56	5.73–7.50
Non-Heterosexual	794/983 (80.8%)	355/994 (35.7%)	<.001	19.29	7.58	6.17–9.31
First generation
Not-first generation	1513/2026 (74.7%)	606/2022 (30.0%)	<.001	27.35	6.96	6.05–7.99
First generation	795/1077 (73.8%)	332/1080 (30.7%)	<.001	19.32	6.35	5.27–7.66
Financial hardship
Low hardship	1621/2152 (75.3%)	638/2159 (29.6%)	<.001	28.82	7.34	6.41–8.40
High hardship	687/951 (72.2%)	300/943 (31.8%)	<.001	17.07	5.58	4.58–6.79

Note. For the intervention group, uptake refers to the percentage of participants who accessed the mobile platform during the six months they had access to it. For the control condition uptake refers to endorsing having had at least 1 session of counseling or psychotherapy (in the past 6 months) or by reporting taking a new (i.e., were not taking these medications at baseline) antidepressant or anti-anxiety medication in the six months following the referral to services. Multilevel logistic regression coefficients were tested using Wald z-tests, which do not have associated degrees of freedom. Tests were two-tailed. No adjustments were made for multiple comparisons. Odds ratios (ORs) and 95% confidence intervals (CIs) were obtained by exponentiating the estimated log-odds coefficients. OR>1 indicates higher odds of having uptakes for the intervention group compared to the control group.

Dimensional Outcomes

GAD:

Although both the screening+D-CBTgsh and screening+referral-to-college-provided-care groups showed significant within group reductions from baseline in GAD symptoms at all time points (Supplementary Table 1), only from baseline to 6-weeks were there significantly greater reductions in GAD symptoms in response to screening+D-CBTgsh (vs. screening+referral-to-college-provided-care; difference in mean change (DMC) = −0.35, 95% CI, −0.58, −0.12; Table 4). At the same time, baseline clinical status moderated the association between study condition and reduction of GAD symptoms at 6 weeks (β = −0.65, 95% CI = −1.16, −0.14, t(18600) = −2.49, p = .01) and 6 months (β = −0.56, 95% CI = −1.07, −0.05, t(18600) = −2.16, p = .03; Supplementary Table 2). Simple slope analyses found that reductions were significantly greater in participants with clinical levels of baseline symptoms receiving screening+D-CBTgsh (vs. screening+referral-to-college-provided-care) at 6 weeks (β = −0.53, 95% CI = −0.80, −0.26, t(13610) = −3.81, p = .001, d = −0.14), but was not significantly different at 6 months (β = −0.19, 95% CI = −0.46, −0.08, t(13610) = −1.36, p = .17, d = −0.05). Among high-risk participants, GAD symptom changes at 6 weeks (β = 0.12, 95% CI = −0.28, 0.52, t(4995) = 0.59, p = .55, d = −0.03 and 6 months (β = 0.37, 95% CI = −0.03, 0.77, t(4995) = 1.84, p = .07, d = −0.10) did not significantly differ between study conditions. Thus, only clinical status participants with GAD changed more in screening+D-CBTgsh than control and only at 6 weeks. At other time points differences between conditions and clinical status as a moderator were not significant.

Table 4.

Descriptive Statistics and Time (Baseline-to-Follow-Up) × Condition Interaction Statistics for Dimensional Outcomes Analysis

	M (SD)
Variable	Intervention (n=3,103)	Control (n=3,102)	β (95% CI)	SE	t 18610	d	p
GADQ-IV
Baseline	6.56 (3.98)	6.60 (3.94)
6-week	5.34 (3.84)	5.73 (3.92)	−0.35 (−0.58, −0.12)	0.12	−2.94	−0.09	.003
6-month	5.43 (4.02)	5.50 (4.02)	−0.03 (−0.26, 0.20)	0.12	−0.25	−0.01	.803
2-year	5.37 (3.96)	5.49 (3.93)	−0.08 (−0.31, 0.15)	0.12	−0.65	−0.02	.513
SPDQ
Baseline	12.44 (6.05)	12.38 (6.06)
6-week	11.63 (6.16)	12.06 (6.33)	−0.49 (−0.84, −0.14)	0.18	−2.74	−0.08	.006
6-month	11.20 (6.40)	11.53 (6.41)	−0.39 (−0.74, −0.04)	0.18	−2.16	−0.06	.031
2-year	11.17 (6.30)	11.40 (6.03)	−0.29 (−0.64, 0.06)	0.18	−1.63	−0.05	.103
PDSR
Baseline	4.66 (6.96)	4.72 (6.96)
6-week	2.69 (5.75)	2.84 (5.80)	−0.09 (−0.49, 0.31)	0.20	−0.43	−0.02	.668
6-month	3.60 (6.41)	3.62 (6.44)	0.04 (−0.36, 0.44)	0.20	0.20	0.01	.845
2-year	3.75 (6.60)	3.82 (6.58)	−0.01 (−0.41, 0.39)	0.20	−0.04	−0.00	.965
PHQ-9
Baseline	11.65 (5.91)	11.68 (5.89)
6-week	8.32 (5.52)	9.43 (5.76)	−1.08 (−1.43, −0.73)	0.18	−6.09	−0.19	<.001
6-month	8.61 (5.95)	9.29 (6.13)	−0.66 (−1.01, −0.31)	0.18	−3.72	−0.11	<.001
2-year	8.29 (5.79)	8.73 (5.74)	−0.41 (−0.76. −0.06)	0.18	−2.32	−0.07	.020
EDE-Q-Global
Baseline	1.81 (1.40)	1.83 (1.39)
6-week	1.55 (1.24)	1.74 (1.34)	−0.16 (−0.23, −0.09)	0.03	−4.41	−0.12	<.001
6-month	1.48 (1.22)	1.69 (1.33)	−0.19 (−0.26, −0.12)	0.04	−5.31	−0.15	<.001
2-year	1.62 (1.29)	1.67 (1.29)	−0.02 (−0.09, 0.05)	0.04	−0.62	−0.02	.533
SF-12-MCS
Baseline	34.11 (10.73)	33.89 (10.69)
6-month	36.41 (11.49)	35.95 (11.63)	0.24 (−0.44, 0.92)	0.35	0.69	0.02	.489
2-year	37.85 (11.33)	36.93 (11.40)	0.71 (0.03, 1.39)	0.35	2.04	0.06	.042
SF-12- PCS
Baseline	53.65 (7.63)	53.85 (7.30)
6-month	53.21 (8.55)	53.43 (8.31)	−0.01 (−0.51, 0.48)	0.25	−0.04	0.00	.967
2-year	53.15 (7.84)	53.42 (7.87)	−0.06 (−0.56, 0.43)	0.25	−0.24	−0.01	.807

Note. Statistics are from three-level linear mixed-effects models (time nested within participants, participants nested within schools) with random intercepts for schools and participants. B coefficients represent the Time × Condition interaction effect at each follow-up compared to baseline. Tests were two-tailed. No adjustments were made for multiple comparisons. PHQ-9 = Patient Health Questionnaire-9; GADQ-IV = Generalized Anxiety Disorder Questionnaire-IV; SPDQ = Social Phobia Diagnostic Questionnaire; PDSR = Panic Disorder Self-Report; EDE-Q-Global = Eating Disorder Examination-Questionnaire-Global score; SF-12- MCS/PCS= 12-Item Short-Form Health Survey-Mental Component Summary Score/Physical Component Summary Score; SF-12 MCS and PCS were not assessed at 6-weeks; Higher scores indicate greater symptom severity for all measures except the SF-12 PCS/MCS, for which higher scores indicate better physical and mental health functioning; Degrees of freedom (df) were 18,610 for all outcomes except SF-12 PCS/MCS (df =12,410); d = Cohen’s d effect size

SAD:

Although there were significant reductions from baseline within both the screening+D-CBTgsh and screening+referral-to-college-provided-care groups in SAD symptoms at all time points (Supplementary Table 1), these symptoms changed significantly more from screening+D-CBTgsh (vs. control) at 6 weeks (DMC: −0.49, 95% CI, −0.84, −0.14) and 6 months (DMC = −0.39, 95% CI, −0.74, −0.04) but not at 2 years (Table 4). Also, clinical status did not moderate the association between study condition and reduction in SAD symptoms at any time point (Supplementary Table 2). Thus, in all participants SAD symptoms reduced more from screening+D-CBTgsh (vs. control) at 6 weeks and 6 months but not 2 years.

PD:

Although there were significant reductions from baseline within both the screening+D-CBTgsh and screening+referral-to-college-provided-care groups in PD symptoms at all time points (Supplementary Table 1), there were no significant differences between conditions on changes in PD symptoms at any time point (Table 4). Furthermore, clinical status did not moderate these effects at any timepoint (Supplementary Table 2). Thus, within the full sample, there were no significant differences between the effects of the treatment vs. control group on PD symptoms at any timepoint.

Depression:

There were significant reductions from baseline within both the screening+D-CBTgsh and screening+referral-to-college-provided-care groups in depression symptoms at all time points (Supplementary Table 1), and these symptoms changed significantly more from screening+D-CBTgsh (vs. control) at 6 weeks (DMC = −1.08, 95% CI = −1.43, −0.73), 6 months (DMC = −0.66, 95% CI = −1.01, −0.31) and 2 years (DMC = −0.41, 95% CI = −0.76, −0.06; Table 4). Baseline clinical status moderated this effect only at 6 weeks (β = −0.76, 95% CI = −1.53, −0.001, t(18600) = −1.76, p = .049; Supplementary Table 2). At this timepoint, simple slope analyses found that reductions were significantly greater in clinical status participants receiving screening+D-CBTgsh (vs. screening+referral-to-college-provided-care) for depression symptoms (β = −1.30, 95% CI = −1.72, −0.89, t(13610) = −6.18, p < .001, d = −0.23). However, within high-risk participants there was no significant difference on change in depression symptoms between screening+D-CBTgsh and control at 6 weeks (β = −0.54, 95% CI = −1.10, 0.02, t(4995) = −1.88, p = .06, d = −0.11). Thus, although only clinical status participants benefitted more from screening+D-CBTgsh (vs. screening+referral-to-college-provided-care) at 6 weeks, at 6 months and 2 years the full sample showed greater reduction in depression symptoms in screening+D-CBTgsh.

ED:

There were significant reductions from baseline within both the screening+D-CBTgsh and screening+referral-to-college-provided-care groups in ED symptoms at all time points (Supplementary Table 1), and these symptoms changed significantly more from screening+D-CBTgsh (vs. control) at 6 weeks (DMC: −0.16, 95% CI, −0.23, −0.09) and at 6 months (DMC = −0.19, 95% CI = −0.26, −0.12) but not at 2 years (Table 4). However, at 6 months, the effect was moderated by clinical status (β = −0.22, 95% CI = −0.38, −0.06, t(18600) = −2.69, p = .01; Supplementary Table 2). In particular, simple slope analyses found that reductions were significantly greater in clinical status participants receiving screening+D-CBTgsh (vs. screening+referral-to-college-provided-care; β = −0.25, 95% CI = −0.34, −0.17, t(13610) = −5.73, p < .001, d = −0.19). However, among participants with high-risk, changes at 6 months in ED symptoms (β = −0.03, 95% CI = −1.15, 0.08, t(4995) = −0.58, p = .56, d = −0.03) did not significantly differ between study conditions. Thus, although ED symptoms changed significantly more from screening+D-CBTgsh (vs. control) in the full sample at 6 weeks, only the clinical participants showed significantly greater reduction at 6 months.

SF-12 mental health:

Although there were significant increases in mental health-related quality of life from baseline within both the screening+D-CBTgsh and screening+referral-to-college-provided-care groups at all time points (Supplementary Table 1), these symptoms changed significantly more from screening+D-CBTgsh (vs. control) at 2 years (DMC = 0.71, 95% CI = 0.03, 1.39) but not at 6 months (Table 4). These effects were not moderated by baseline clinical status (Supplementary Table 2). Thus, the whole sample improved more from screening+D-CBTgsh (vs. screening+referral-to-college-provided-care) at 2 years in mental health-related quality of life, but not 6 months.

SF-12 physical health:

Physical quality of life appeared to deteriorate from baseline across all follow-up points (Table 1), though these effects were quite small, and there were no significant differences between the screening+D-CBTgsh and screening+referral-to-college-provided-care groups at any time point (Table 4). In addition, baseline clinical status did not moderate the effects of screening+D-CBTgsh vs. control (Supplementary Table 2). Thus, regardless of treatment condition, and baseline clinical status, participants’ physical quality of life declined, with very small effect sizes.

D-CBTgsh Engagement

On average, the 2,308 participants who accessed the D-CBTgsh platform visited it 11.7 times (SD=25.0; median=4; IQR=1–13; range=1–776). The total average time participants spent on the D-CBTgsh platform was 98.8 minutes (SD=266.3; median=18.0; IQR=2.7–95.1; range=0–381.2). Participants accessed an average of 2.4 unique intervention modules (SD=3.0; median=1; IQR=0–3; range=0–18) and received an average of 14.9 messages from their coach (SD=6.3; median=14; IQR=13–18; range=0–34).

Discussion

In this pragmatic effectiveness randomized clinical trial, conducted under real-world conditions, we found that offering screening+D-CBTgsh for those with self-reported clinical levels or at risk for GAD, SAD, PD, depression, and/or EDs was associated with lower prevalence of having any mental disorder (i.e., self-reported clinical levels of any of the 5 disorders) compared to screening+referral-to-college-provided-care in a sample of 6,205 college students recruited via population-based screening at 26 colleges and universities. These differences emerged early in the intervention period (6 weeks), remained at post-intervention (6 months), and were sustained through long-term follow-up (2 years). In particular, compared to the control group, those randomized to screening+D-CBTgsh had 4.3% lower prevalence of having any mental disorder at 6 weeks (132 fewer people), 4.9% lower prevalence at 6 months (151 fewer people), and 3.8% lower prevalence at 2 years (118 fewer people). These differences in prevalence have major positive public health implications, particularly when considered at even broader population levels (e.g., all U.S. college students).¹⁶ Thus, from the perspective of prevalence reduction, screening+D-CBTgsh was successful in managing some of the most common mental health problems college students face through combined targeted prevention and intervention (i.e., successfully preventing the onset of new disorders and successfully treating established disorders). Importantly, participants in this study were recruited via mass online invitations to screening within 26 college and university campuses; our sample was therefore not the typical “treatment-seeking” sample often studied in randomized clinical trials of digital or other types of interventions for mental health problems. As such, results provide support for the effectiveness of D-CBTgsh even when disseminated more broadly—i.e., through mass, population-based screening invitations and to those not specifically seeking treatment.

Notably, our primary outcome criteria simultaneously examined both decrease in self-reported clinical levels of disorders in the clinical group as well as prevention in both the high-risk and clinical group of developing any of the five disorders we measured not present at baseline. It was crucial to measure these outcomes simultaneously because the five disorders in this study (GAD, SAD, PD depression, EDs) can be both concurrently and sequentially present.^20–35 That is, individuals can have clinical levels of one disorder (e.g., GAD) at baseline and at a later time point show remission from that disorder but simultaneously develop another disorder that was not present at baseline (e.g., SAD, PD, depression, and/or ED).^20–35 Similarly, subthreshold levels of one disorder can also predict the development of another disorder later.^25,26,32,36 Thus, our main outcome measure was unique in its ability to simultaneously show both decrease in clinical levels of disorders as well as prevention of onset of new disorders and represents a high bar for outcome.

Interestingly, the significant interaction between baseline clinical status and study condition at 6-week follow-up suggested that screening+D-CBTgsh produced faster reductions in disorder prevalence for students with baseline clinical disorders. However, students who were at high-risk for mental disorders at baseline did not show significantly different short-term trajectories across conditions. At the same time, the absence of moderation of clinical status at later follow-ups (6 months and 2 years) suggested that the intervention ultimately supported both prevention and treatment at the population level.

With regard to uptake of mental health services, screening+D-CBTgsh was far superior to screening+referral-to-college-provided-care following population-based mental health screening and provision of personalized feedback; this difference was not moderated by clinical status. In particular, 74.4% of students offered D-CBTgsh began it, whereas only 30.2% of students offered a referral to college-provided care reported accessing any mental health services at any point over the 6-month follow-up (i.e., the period that the intervention group had access to D-CBTgsh). This uptake rate compared favorably with other digital mental health interventions, with a recent systematic review finding that among studies conducted across the globe reporting uptake (only 39%), the average uptake was 79% (range: 32%–100%), though definitions varied widely.³⁷ Notably, students in the control group were referred to services available on their college campuses for mental health treatment, including student counseling centers and student health centers, where most services are free for students. As such, one could argue college students have substantially better access to mental health care than most U.S. adults, who rarely have access to fully-covered or free mental health treatment, including care that is located in such close physical proximity to where they live and/or work or attend classes. Moreover, the difference in mental health services uptake between the intervention and control groups was still significant even when comparing services uptake across two years in the control group to the 6-month period of the intervention group. Importantly, uptake was also stronger in screening+D-CBTgsh vs screening+referral-to-college-provided-care across race, ethnicity, sexual orientation, first generation status, and financial hardship status. Furthermore, students from minority groups, including those who identified as non-White, Hispanic/Latino, gay/lesbian/bisexual/other, having high financial hardship, and first-generation in the control group were significantly less likely to receive any treatment at any point in the control condition than in the intervention group. These findings offer support for the notion that not only can digital interventions improve access to mental health care in general but that they can increase access for those from vulnerable groups, highlighting the role that digital mental health can play in promoting mental health equity.³⁸

Of note, this study was conducted during the COVID-19 pandemic period, with the vast majority of participants recruited after the pandemic began. It is possible that the pandemic may have influenced both the availability of usual care and the demand for and interest in a digital tool to support mental health. However, most college campuses transitioned to offering counseling services remotely during the pandemic. For example, the Center for Collegiate Mental Health reported that 99% of 570 college counseling centers surveyed between August and October 2020 reported they were able to transition to telehealth services.³⁹ This suggests that, in general, services were likely available through the counseling center to those in the control group. However, it is also important to note that we assessed uptake of any session of therapy regardless of whether it was obtained through the counseling center or elsewhere.³⁹ Moreover, our uptake differences were largely consistent with a trial of an ED-specific digital intervention (vs. referral to college-provided care) that was conducted prior to the pandemic.⁴⁰ It is also important to note that we included in our definition of uptake use of any new medications in case these were easier to access. Thus, our outcome measure was not entirely focused on counseling center availability. At the same time, even if the pandemic resulted in decreased access to traditional services for some individuals, we view our results as an important reflection of the digital intervention’s relevance in contexts where access to traditional services is constrained—whether due to a public health crisis like the COVID-19 pandemic or other common barriers to care.

In addition to the reductions in prevalence of disorders demonstrated, the dimensional results provided important evidence of the intervention’s broader impacts on population mental health. Across generalized anxiety, social anxiety, depression, and ED, overall, students randomized to screening+D-CBTgsh showed consistently greater symptom reductions than those referred to usual care, with effects emerging as early as 6 weeks and persisting for social anxiety and ED symptoms through 6 months, and, for depression through 2 years. Panic symptoms were the only dimensional symptom domain for which we did not detect significant between-group differences at any timepoint; this may reflect the absence of facilitated interoceptive exposure from the intervention, a core component of effective PD treatment. At 2 years we also found greater effects of screening+D-CBTgsh (vs. control) on mental health functioning. Notably, both the intervention and control group led to significant improvement on all dimensional measures across all timepoints (including panic disorder), except physical health functioning. This was true even when there was not differential change between the two conditions. At the same time, physical health functioning demonstrated decline; however, these effects were extremely small. Taken together, these findings demonstrated that benefits of screening+D-CBTgsh extended beyond reducing disorder prevalence to also reducing disorder specific symptom levels across the population. However, any observed improvements were more specific to mental health rather than reflecting broader shifts in global physical functioning as assessed by the physical component of the SF-12.

These dimensional effects aligned with the prevalence findings and help to clarify how the intervention produced population-level reductions in disorder rates. Moderation analyses further demonstrated that the intervention often had its strongest effects on dimensional symptoms amongst students who met clinical criteria at baseline. For GAD and depression symptoms at 6 weeks, and ED symptoms at 6 months, students with clinical levels of symptoms at baseline in the intervention condition experienced larger improvements than those with clinical disorders in the control condition. In contrast, students identified as being at high-risk for mental disorders at baseline did not show significant between-condition differences in changes in GAD and depression symptoms at 6 weeks or ED symptoms at 6 months. This pattern may demonstrate that screening+D-CBTgsh provided the most meaningful symptom reduction for students screening positive for clinical disorders, while still offering smaller yet beneficial changes for students at elevated risk. Overall, findings reinforce the value of a transdiagnostic digital intervention that simultaneously supports prevention and treatment.

Strengths and Limitations

A primary strength of this study was the use of population-wide emails alerting all students to the opportunity to access guideline-based mental health screening for identifying mental health problems. All students who completed the screen, met study inclusion criteria, and those consented were randomized. Students in both study conditions were offered mental health screening, provided tailored feedback, and provided information on how to access a mental health service. Thus, this trial generated data on the benefit of D-CBTgsh compared to referral to college-provided care among screen-detected populations for both prevention and intervention of multiple disorders.⁴¹ Another strength of our study was that we screened for and aimed to prevent and treat five disorders simultaneously (GAD, SAD, PD, depression, and EDs) using a transdiagnostic intervention approach. Furthermore, our primary outcome measure required participants to meet a very high bar of not having clinical levels of any disorders at each timepoint.

Additional strengths included the very large number of participants recruited from across the U.S., representativeness of the population, broad inclusion criteria, and long-term follow-up. Moreover, we examined differences in uptake between screening+D-CBTgsh and screening+referral to college provided care. Few studies, let alone population-based access to screening studies, have examined degree of uptake of scalable mental health treatments among those randomized across treatment conditions.³⁷ This is important because uptake or starting the treatment is a needed first, critical step to symptom improvement. At the same time, it is important to note that the large sample size in this study allowed for power to detect even small effects. Notably, and particularly in the field of public health, small effects scaled across an entire population can be highly relevant and impactful⁴²—for example, when considering further scaling of our D-CBTgsh intervention to even larger populations of college students.

Regarding limitations, overall engagement with the intervention was relatively low, with users accessing 2.4 unique modules of the program on average (of 6–8 that were available for the user’s primary problem, with additional modules made available to address comorbid problems). Engagement in D-CBTgsh is a known challenge and is often brief.⁴³ However, recent studies have shown that even very brief engagement with digital or in-person interventions can still demonstrate efficacy.⁴⁴ Indeed, even with this level of engagement compared to the control group, screening+D-CBTgsh was associated with attaining a lower prevalence of having any of five mental disorders (GAD, SAD, PD, depression, and EDs) at 6 weeks, 6 months, and 2 years. Future work should address the issue of improving engagement with D-CBTgsh health apps deployed at the population level as well as investigate patterns of optimal engagement and develop and adopt common definitions of engagement.⁴⁵ Indeed, engagement comparisons across digital mental health interventions are difficult due to substantial variability in how engagement is defined and reported. In addition, results suggest the superiority of screening+D-CBTgsh vs screening+referral-to-college-provided-care (i.e., the standard practice at the end of mental health screens—referring to available resources) both for decreasing population prevalence of mental health disorders and for increasing access to care—key population health-related outcomes. Future research may directly compare the effectiveness of D-CBTgsh to other control conditions (e.g., in-person CBT) as well as expand this transdiagnostic intervention approach to additional prevalent conditions not addressed here, such as alcohol use disorder, substance use disorder, obsessive-compulsive disorder, and post-traumatic stress disorder, to provide even more comprehensive coverage of mental health needs on college campuses. Future studies should also examine the prediction of outcome from baseline variables using machine learning to develop a validated predictive algorithm⁴⁶ Such an algorithm could be used to evaluate stratified models of care in college settings wherein students predicted to derive sustained benefit from D-CBTgsh are allocated to this service, and those unlikely to benefit from D-CBTgsh are directed to more intensive intervention (e.g., psychotherapy). This could leverage machine learning to support population-level precision mental health care and optimize service delivery across campuses.

Conclusions

Overall, these findings support the transdiagnostic prevention and intervention benefits of screening+D-CBTgsh, which reduced the population prevalence of mental disorders, lowered symptom severity across multiple domains, and produced the largest early improvements among students with clinical-level presentations. The intervention also substantially increased access to care—including for students from minoritized groups—and demonstrated markedly higher uptake than referral to college-provided services, underscoring its viability as a scalable, population-based approach to improving college mental health.

Methods

This pragmatic effectiveness randomized controlled superiority trial complied with all relevant ethical regulations. All procedures were approved by the Institutional Review Board (IRB) of Washington University. All participating universities reviewed the approval, consent forms, and study protocols and deferred to Washington University’s IRB (201901073). Informed consent was obtained from all participants. Those who consented and completed the screen were given a US$5 gift card and entered into a raffle for 1 of 40 US$100 gift cards. Completion of assessments was compensated via gift cards (baseline and 6-weeks: US$5 each; 6-month: US$10; 2-year: US$20) and entry into raffles (1 of 5 US$50 raffles for completion of a given survey at each timepoint and 1 of 10 US$100 raffles for completion of all four study surveys. California schools did not participate in raffles due to state laws. Usage of the intervention itself was not compensated. This report follows the Consolidated Standards of Reporting Trials (CONSORT) reporting guidelines. All analyses reported in the current paper were preregistered on October 7, 2019 (NCT04162847) with no deviations. Also, see Fitzsimmons-Craft et al.⁴⁷ for the 2021 protocol paper. Although there are no deviations from the trial registry, this paper includes minor deviations from the published protocol as described within the methods.

Participants and Procedure

Participants were recruited from October 2019 to November 2021, with data collection completed by October 2023. We recruited participants using population-based invitations to screening implemented within 26 U.S. colleges and universities that varied in geographical location and environment (urban/rural, private/public, small/large); see Supplementary Table 3 for additional details on the participating colleges/universities. At 17 schools, email invitations were sent to all enrolled students to complete a screening, whereas at 9 schools, only first- and second-year students were invited to complete a screening. We initially focused on first- and second-year students to maximize the likelihood that participants would remain enrolled and in college for the duration of follow-up assessments. However, as the study progressed and recruitment opportunities varied across campuses, some schools were amenable to and opted to expand invitations to additional class years. We also implemented follow-up procedures intended to maximize retention and data collection even post-graduation, including the collection of alternative (non-institutional) email addresses to facilitate contact with the study participants. Email invitations informed students that their college was “conducting a campus-wide research study that involved a survey about student health and well-being.” Students were also informed that following the survey, they would receive “personalized mental health feedback and information about resources.” Finally, they were told that their responses could make them “eligible for a second phase of the study, which would involve randomization to one of two study conditions designed to support student mental health as well as 3 follow-up surveys to assess how they were doing, for which they would be compensated.” No other information was provided at the screening invitation phase. Students were only informed of the two possible study conditions (i.e., access to a coached mobile program or a referral to treatment on their campus) once they were determined to be eligible for and invited to participate in the RCT. Participants under 18 years of age and non-undergraduates were ineligible for the screen and, therefore, the study (see Figure 1 for CONSORT flow chart).

Eligibility criteria for enrollment in the randomized clinical trial included: being at risk for or screening positive for clinical levels based on self-report, for the 3 anxiety disorders (i.e., generalized anxiety disorder [GAD], social anxiety disorder [SAD], and/or panic disorder [PD]), depression, and/or ED (except anorexia nervosa [who were excluded and provided referral information]; operational definitions for all psychopathologies are provided in the Measures section), age ≥18, no psychotherapy in the past month, owning a smartphone, and an undergraduate student. Of note, students who endorsed suicidal ideation (on the Patient Health Questionnaire-9⁴⁸ item 9) received additional, immediate follow-up (including contact information for the National Suicide Prevention Hotline) but were not excluded. Individuals who met inclusion criteria were invited to participate in the next phase of the study (i.e., the randomized clinical trial), and those who consented to participate and completed baseline assessments were subsequently randomized through the survey platform (i.e., Qualtrics) within each college/university to one of the two study conditions. Given the nature of the interventions, participants were not blinded to study condition. Those randomized to screening+D-CBTgsh received personalized feedback and a link to enroll in the program, and those randomized to screening+referral-to-college-provided-care (control) received personalized feedback and a recommendation to seek services using available campus-specific mental health resources. For additional details, see Supplementary Table 3 for resources provided to students randomized to the control group on each campus and Supplementary Note 1 for screening, safety monitoring, and referral procedures.

Study Conditions

Intervention condition: Screening+D-CBTgsh

Following invitations deployed to the entire population of participating colleges, completion of mental health screens, and completion of baseline measures, participants randomized to the screening+D-CBTgsh condition received personalized feedback and a link to enroll in the D-CBTgsh program. The D-CBTgsh platform, hosted by SilverCloud Health, included three evidence-based guided cognitive-behavioral therapy (CBT) programs, one for each mental health problem (i.e., anxiety, depression, and EDs).^12,40,49,50 Six to eight 20-minute-long modules were offered for each problem. Content addressed key treatment targets early on (e.g., decreasing avoidance for anxiety disorders). Participants with multiple mental health problems either at a clinical level or at high-risk selected their initial treatment focus and were subsequently offered key modules to address comorbidities. Some students were also offered modules outside of the core ones offered for the 5 disorders based on coach’s clinical judgment and participant feedback. These modules were consistent with CBT and available within SilverCloud. Optional supplemental modules included the following topics: challenging times, grief, relaxation, sleep difficulties, anger management, communication/relationship, and self-esteem. A total of 12.5% of students in the intervention group received one or more of the optional modules. Thus, the program was personalized for users’ unique preferences and needs (Supplementary Figure 1). If participants with clinical problems at baseline did not demonstrate improvement in any problems reported after six weeks, they were provided with a referral to usual care available on their respective college campus while still encouraged to continue with the program. Improvement was defined as any reduction in symptom scores on the weekly symptom checklist(s) offered within the digital program corresponding to conditions for which participants met the clinical cutoff at baseline (i.e., anxiety, depression, and/or ED). Access to the platform was provided for six months from the time of randomization.

All participants in the screening+D-CBTgsh condition had access to a trained coach. Coaches, who were recruited by the study team, had a bachelor’s degree or higher and were enrolled in mental health training programs at the master’s, PhD, or PsyD level. Coaches were trained by the study team in CBT, motivational interviewing, risk management, and the mobile platform environment, and attended weekly supervision with a licensed psychologist. Qualifications of the coaches and training procedures were consistent with those used by SilverCloud Health. Additional details on coach characteristics, training, and supervision are available in Fitzsimmons-Craft et al.⁵¹ Overall, 75 coaches provided support to participants in the intervention. Of coaches completing an exit survey, mean age was 31 years (SD = 7.67, range = 23–53), 81% were women, 57% were White, and 60% were master’s students.⁵¹ Coaches had variable prior mental health intervention experience; many had no prior formal CBT training. Coaches participated on a volunteer basis, with some completing the role as part of clinical practicum requirements and accruing clinical hours. Coaches’ roles involved supporting and enhancing user motivation; monitoring progress; facilitating goal setting and ensuring accountability; providing feedback on technique usage and encouraging practice; answering questions; and monitoring and addressing clinical risks. Additional information regarding coaching procedures and fidelity monitoring are provided in Supplementary Note 2, Supplementary Figure 2, and Supplementary Figure 3.

Control condition: Screening+referral-to-college-provided-care

Following invitations deployed to the entire population of participating colleges, completion of mental health screens, and completion of baseline measures, participants randomized to the screening+college-provided care condition received personalized feedback and a recommendation to seek care. They were given information and contact details on resources available at their respective health or counseling center (see Supplementary Table 3).

Measures

Please see Supplementary Table 4 for specific criteria for clinical vs. high-risk status. Assessments included brief, validated mental health questionnaires and were taken from the whole sample. GAD, SAD, and PD were assessed with the Generalized Anxiety Disorder Questionnaire-IV⁵² (GAD-Q-IV; range, 0–13), Social Phobia Diagnostic Questionnaire⁵³ (SPDQ; 0–27), and Panic Disorder Self-Report⁵⁴ (PDSR; 0–23), respectively. Participants were considered to be at clinical levels if they met self-reported DSM-5⁵⁵ diagnostic criteria on the GAD-Q-IV, SPDQ, and PDSR. Participants were considered high-risk if they scored below clinical levels but greater than or equal to 5.7 on the GAD-Q-IV, 7.4 on the SPDQ, or 8.8 on the PDSR. Depression was assessed using the Patient Health Questionnaire-9 (PHQ-9),⁴⁸ with scores of 10 or higher indicating clinical levels and scores between 5 and 9 indicating high-risk status. EDs were assessed using the Stanford-Washington University EDs Screen (SWED)⁵⁶ and Eating Disorders Examination Questionnaire (EDE-Q).⁵⁷ Participants were deemed to be at clinical levels based on the SWED if they had 6 or more episodes of binge eating, vomiting, and/or diuretic/laxative use in the past three months. They were considered high-risk if they had a weight concern score of 47 or higher and at least one purging episode (vomiting or diuretic/laxative use) in the past three months. Additionally, quality of life was assessed using the 12-Item Short-Form Health Survey (SF-12)⁵⁸, which yields the physical (PCS) and mental (MCS) component summary scores, indicating the level of physical and mental functioning, respectively. Each subscale ranges from 0–100, with higher scores indicating better functioning. The GAD-Q-IV, SPDQ, PDSR, PHQ-9, and EDE-Q were administered at baseline, 6 weeks, 6 months, and 2 years. The SF-12 was administered at baseline, 6 months and 2 years.

Similar to Fitzsimmons-Craft et al.,⁴⁰ uptake of services in the intervention condition was defined by any use of D-CBTgsh (based on objective use data) during the 6 months they had access. In the control condition it was defined by students endorsing having had at least 1 session of counseling or psychotherapy or by reporting taking a new (i.e., were not taking these medications at baseline) antidepressant or anti-anxiety medication during the same 6-month follow-up assessment or cumulatively during the 2 years they were part of the study. For the control condition, these service-use items were assessed via self-report at the 6-month and 2-year follow-ups, reflecting use in the prior 6 and 18 months, respectively and assessed cumulatively to determine 2-year uptake for the control group.

Statistical Analysis

Power analysis

It is important to note that before we officially ended participant recruitment for this study, we reran the Monte Carlo power simulations with new missingness specifications and a lower than anticipated sample size. The missingness specifications below deviate from the published protocol⁴⁷ as it reflected an earlier version of the power analysis that was incorrect. Prior to discontinuing recruitment due to the challenges posed by COVID, we conducted revised power analysis incorporating a higher-than-anticipated rate of missingness and submitted these updated simulations to NIH to obtain approval for terminating participant recruitment. To ensure adequate power, a power analysis based on 1,000 Monte Carlo simulations that mirrored study conditions was conducted to determine the sample and cell size required to achieve at least 80% power and 0.05 level two-sided tests to detect minimally meaningful differences. We assumed a 0.005 intra-school correlation, with approximately equal numbers of students per condition per campus, with two-year outcome as the primary endpoint. We utilized a standardized effect size of 0.25, a random effect variance of 0.2, a residual error variance of 1, a critical Z value of 1.96 for a two-tailed significant test with an alpha of .05, and a conservative attrition rate of 50%. With a sample size of 6,254 and with the rate of post-treatment and follow-up assessment completion (approximately 50%), using Monte Carlo simulations, we had at least 80% power to detect a small (5%) difference in rates of clinical disorder status between the treatment and control groups. Power simulations assumed prevalence rates consistent with those specified in the protocol paper:⁴⁷ For high-risk participants, we assumed 24% vs. 18% prevalence of clinical disorders at two years, and for the clinical participants, 50% vs. 37.5%. For service uptake, we assumed 26% vs. 20% among high-risk participants and 60% vs. 50% among clinical participants. Power analyses did not adjust for multiplicity as the primary research question of interest was to assess differences in clinical status between the treatment and control groups, and all other analyses were secondary and exploratory. Our initial plan made a sharp distinction between our hypothesis-driven, confirmatory analysis of our primary outcome versus the additional secondary and exploratory analyses we planned to conduct. Monte Carlo power simulations were conducted in R statistical software (version 4.3.2).

Handling Missing Data

We deviated from our protocol paper⁴⁷ in using multiple imputation instead of full-information maximum likelihood estimation (FIML) to handle dropout and address complex missing-data patterns. FIML was deemed faulty for the present data structure because it would not allow incorporation of the full set of auxiliary variables related to dropout and outcomes, thereby significantly increasing risk of biased parameter estimates. Details regarding the rationale for and implementation of the imputation procedure are provided in Supplementary Note 3. Overall, our imputation strategy consisted of three steps: 1) specification of the imputation model, 2) estimation of the parameters, and 3) drawing imputed values from the predictive distribution. For our imputation model, we used the Markov Chain Monte Carlo (MCMC) method and the predictive mean matching algorithm, with a maximum of 100 iterations, to generate 10 imputed datasets to increase confidence and minimize the standard error of the parameter estimates.^59,60 All analyses employed the mice package (version 3.16.0) in R (version 4.3.2).⁶⁰ To pool parameter estimates across multiply imputed datasets into a single estimate and estimate its variance, we used Rubin’s rules under the MAR assumption.⁶¹ Here, the variance of our pooled parameter estimate was calculated by combining the within-imputation variance and between-imputation variance. When such pooling is conducted under the MAR assumption, the pooled estimates are unbiased.^62,63

2.4.2. Analysis of primary and secondary outcomes

Since our goal was to examine the transdiagnostic prevention and/or treatment of all anxiety, depression, and EDs (i.e., all comorbid disorders), our primary outcome examined group differences in the number of participants with no mental disorders (i.e., no clinical levels of anxiety [GAD, SAD, and PD], depression, and/or EDs, based on self-report) at each assessment. Caseness, defined as meeting or not meeting the validated self-report clinical cutoff for any of the 5 disorders at each timepoint was our most important outcome measure because it simultaneously captured both remission of any of the 5 disorders that were at clinical levels at baseline (intervention) as well as the high likelihood that participants could develop a disorder not present^20–35 or not at high-risk at baseline (prevention).^25,26,32,36 For each disorder, presence (‘yes’) was defined as scoring at or above the clinical cutoff on the corresponding self-report measure, and absence (‘no’) was defined as scoring below the clinical cutoff. Analyses were conducted using a multilevel logistic regression model, accounting for the nested data structure, with repeated assessments (baseline, 6-weeks, 6-months, 2-years; Level 1) nested within participants (Level 2) and participants nested within schools (Level 3). The model included a random intercept that was allowed to vary across schools and participants (participants nested within schools). Fixed effects included time, study condition (intervention vs control), and their interaction. Pairwise comparisons from the model were used to evaluate differences between study conditions at each time point. Furthermore, to examine moderation by baseline clinical status (high risk vs. clinical), we estimated models including time (6-weeks, 6-months, 2-years) × study condition × clinical status interactions. Baseline was excluded from the model because the number of positive cases at baseline was perfectly collinear with clinical status (the moderator). Any 3-way interaction was probed within each time point using pairwise contrasts of estimated marginal means of the study condition × clinical status interaction. This approach deviated from the original protocol⁴⁷, which proposed both combining and analyzing separately the high-risk and clinical groups. Instead, using clinical status as a moderator examined the same question, but reduced the number of statistical tests, thereby diminishing bias due to Type II error, and loss of power associated with separate subgroup analyses.

We used multilevel logistic regression to compare the secondary outcome of uptake between the screening+D-CBTgsh and screening+referral-to-college-provided-care groups, accounting for the nested data structure (participants nested within schools). Study condition was included as a fixed effect, and a random intercept was specified for schools. The main uptake analysis examined uptake at the 6-month follow-up. As an alternative, we also examined 2-year uptake, where uptake was coded as use of D-CBTgsh during the 6-month access period for the intervention group and as any uptake reported at either the 6-month or 2-year follow-up for the control group. Furthermore, we examined moderation by clinical status of uptake between screening+D-CBTgsh and screening+referral-to-college-provided-care groups within high-risk versus clinical subgroups and within majority and minority groups based on race (White and non-White), sexual orientation (heterosexual and non-heterosexual), Hispanic/Latino ethnicity (non-Hispanic/Latino and Hispanic/Latino), first-generation status (first-generation [individuals whose highest level of parental education was less than a bachelor’s degree] and non-first-generation), and financial hardship (low [individuals who endorsed not very hard or somewhat hard financial hardship in the past month] and high [individuals who endorsed hard or very hard financial hardship in the past month]).

For our secondary dimensional outcomes, we used multilevel models that accounted for the three-level structure of the data (time points within participants, participants within schools). The dependent variable was the observed score at each time point (baseline, 6 weeks, 6 months, 2 years). Time was modeled as a categorical variable so that follow-up scores were compared directly to baseline. Separate sets of models were run examining different secondary outcome measures, including the total GADQ-IV, SPDQ, PDSR, PHQ-9, EDE-Q-Global, and SF-12 scores. Each model included a random intercept that was allowed to vary across schools and participants (participants nested within schools), with time, study condition, and their interaction as fixed effects. Simple slope analyses were conducted to examine symptom changes within study condition at each time point. Moderation by baseline clinical status was tested for each dimensional outcome. When significant, follow-up analyses compared differences in dimensional outcomes between study conditions within each clinical status subgroup. Between-group effect sizes (d = (model-estimated between-group difference in change) / (baseline SD of outcome)) and within-group effect sizes were calculated (d = b/√(residual variance)) to ease interpretation,⁶⁴ where d values of 0.2, 0.5, and 0.8 denoted small, moderate, and large effect sizes, respectively. All statistical tests were two-tailed, and results were considered significant at p < .05. Analyses employed the lme4 package (version 1.1.35.3) for mixed-effects models and emmeans package (version 1.10.2) for pairwise contrasts in R statistical software (version 4.4.1).

Data Availability Statement:

Deidentified individual participant data that underlie the results reported in this article (text, tables, figures, and appendices) are available in the National Institute of Mental Health (NIMH) Data Archive (https://nda.nih.gov) and can be accessed through the Non-Disclosure Agreement (NDA) controlled-access mechanism. The shared repository includes, the data dictionary, and all variables necessary to reproduce the reported analyses. Researchers who provide a methodologically sound proposal will be considered for data access. The study protocol is published.⁴⁷ To gain access through the NIMH Data Archive, requestors must: (1) have an institutional affiliation with a Federal-Wide Assurance from the U.S. Department of Health and Human Services; (2) complete and submit a Data Use Certification with institutional official signature; (3) receive approval from the National Institutes of Health (NIH) Data Access Committee. Data requestors will need to sign a data access agreement.⁶⁵

Code Availability Statement:

The analysis code used in this study has been deposited in and available on the NDA website described above.⁶⁵