Effects of app delivered self hypnosis on stress management

Abstract

Stress and stress-related chronic illness are increasing worldwide while mental health care access remains limited. Recent neurophysiological advances support the effectiveness and safety of hypnosis for stress management. In this retrospective observational study, we studied app-delivered hypnosis in 84,395 users across 282,893 stress reduction sessions. Users rated pre- and post-session stress on a 10-point Likert Scale. Data analysis utilized Linear Mixed Effects (LME) models to accommodate repeated measures and missing data. Effects of session type, user hypnotizability, age, sex, and membership were assessed. Pre-to-post stress reduction occurred consistently in each of the first 10 sessions (Cohen’s d values ranging from −0.71 to −0.78), demonstrating significant improvement in stress management. Across the first 10 sessions, greater stress reduction was observed with interactive and regular-length sessions, higher hypnotizability, older age groups, and paying members. Findings provide evidence that disseminable digital formulations of hypnosis contribute meaningfully to stress reduction.

Introduction

Since the COVID-19 pandemic, increasing numbers of people are reporting high stress levels and stress-related mental and physical illness^1,2. Although psychosocial stress is ubiquitous to human experience, and the body’s stress response holds critical adaptive significance, chronic unmanaged stress has been implicated in mental health disorders, cardiovascular disease, metabolic syndrome, cancer, sleep impairment, dementia, autoimmunity, and other chronic illnesses^3,4, likely via dysregulation of the hypothalamic-pituitary-adrenal (HPA) axis, with implications for immune dysfunction, chronic inflammation, and cellular injury^4,5.

The relationship between stress and health highlights the need for proven stress management tools to help people manage life’s inevitable stressors. In clinical settings, first-line treatment of anxiety and stress-related mental health conditions (e.g., adjustment disorder, generalized anxiety disorder, post-traumatic stress disorder, and major depressive disorder) has emphasized pharmacotherapy and psychotherapy⁶. These effective interventions are indicated for patients with anxiety disorders and generally take time to yield discernible symptom improvement. Complementary interventions are increasingly appreciated and utilized, including among people who may not necessarily meet formal diagnostic criteria for anxiety disorders, with exercise, mindfulness meditation, and controlled breathwork yielding moderate to large effect sizes in studies^7,8. However, global population surveys show that among those in need of mental health care, only about 1 in 4 (27.6%) receive it, highlighting concerns about stigma, inaccessibility, cost, and insufficient numbers of trained clinicians⁹.

Stress management interventions are thus increasingly available in accessible digital formats¹⁰. The nascent literature on mental health digital applications (apps) appears promising¹⁰, though mixed¹¹, with concerns for small sample sizes and few randomized controlled trials. Thomas Insel, former director of the National Institute of Mental Health, has described the next steps in digital mental health care as “the development of engagement, integration, quality improvement, and better outcomes.”¹² A subset of such apps offer hypnosis, a long-studied intervention with a favorable risk/benefit ratio¹³. A systematic review of hypnosis apps concluded that while most apps target relaxation and stress, less than 3% are evidence-based¹⁴. One prospective randomized clinical trial (RCT) investigating a hypnosis app in 72 patients in a craniofacial pain clinic found significant improvements in anxiety and pain¹⁵. A recent RCT of an online hypnosis intervention in pregnant women showed lower stress and improved childbirth-related experiences¹⁶. Additional recent studies of app-delivered hypnosis for irritable bowel syndrome, menopausal hot flashes, and nicotine dependency offer encouraging evidence^17–19.

Hypnosis is a state of enhanced focal attention (absorption) with reduced peripheral awareness (dissociation) and increased cognitive flexibility (suggestibility)²⁰. Despite hypnosis being the oldest Western conception of psychotherapy, modern advances in brain imaging are only recently uncovering its complex neurophysiology¹³. Hypnosis-based treatments have been found to be effective for an array of conditions, including pain, nicotine dependency, insomnia, trauma-related symptoms, phobias, depression, and gastrointestinal symptoms^21–23. Many studies show hypnosis to be effective for clinical anxiety, and for managing stress by modulating stress reactivity^24,25. Particular benefits of hypnosis for these targets include its rapid effects and adaptability to a range of settings, enabling people to feel less anxious during surgical interventions and their immediate aftermath, in patients hospitalized in intensive care^21,26, and in reducing post-surgical opioid use²⁷.

Several factors may influence the effects of hypnosis. The first is hypnotizability, a natural tendency towards hypnotic states, which is positively associated with responsiveness to hypnotic interventions²⁰. Hypnotizability has been found to be a stable trait through adulthood²³, though potentially modifiable²⁸. Multiple validated scales, including the hypnotic induction profile (HIP), are useful for evaluating hypnotizability in clinical settings²⁹. The second is the mode of delivery of the hypnotic intervention. In view of hypnotizability as a natural tendency or trait, self-directed hypnosis does not differ from clinician-guided hypnosis and is potentially more effective for highly hypnotizable individuals³⁰. The third is repeated use of hypnosis. Research has generally shown improved outcomes with practice, such that hypnosis may be viewed akin to a skill²³.

The existing body of literature supporting the flexibility, safety, and rapid effects of hypnosis suggests the viability of accessible app-based outlets for self-directed hypnosis, or self-hypnosis¹³. While rapid socioeconomic digitalization has been linked to mental health burden³¹, such technological advances have also allowed for the expansion of mental health care on broad scales. In this retrospective observational study, we investigated the effects of app-delivered self-hypnosis on stress management in a very large sample of 84,395 users. We hypothesized a priori a significant reduction in self-rated stress from pre-to-post intervention. We predicted that better stress management would be associated with full-length (vs. brief) sessions, interactive (vs. non-interactive) sessions, and higher hypnotizability. On an exploratory basis, we examined the effects of sex, age, and membership status (pre-existing enrollment as paying members of the app) on stress reduction.

Results

Stress reduction at each session

At baseline users reported an overall moderate level of stress (Table 1). Analysis of pre-to-post change in stress level at each session indicated consistent within-session stress reduction throughout the first 10 sessions (Table 1). According to our primary LME model, the first 10 hypnosis sessions showed effect sizes ranging from Cohen’s d values of −.071 (session 4) to −0.78 (session 1). The mean stress change ranged from −1.623 (95%CI = −1.643, −1.603) to −1.447 (95%CI = −1.501, −1.392) within the first 10 sessions. While the greatest mean stress reduction was observed in session 1, and the smallest reduction was noted in session 6, the stress reduction was relatively consistent across the sessions, demonstrated by the mean change scores and the effect sizes (Table 1 and Fig. 1). More than half of the users only completed one session; however, both those who did only one session (Mean = −1.57 (95% CI = −1.60, −1.54), Effect Size = −0.76) and those who did more than one session (Mean = −1.69 (95% CI = −1.72, −1.66), Effect Size = −0.80) demonstrated significant stress reductions at session 1.

Table 1.

Estimated mean stress reduction and effect sizes of the first 10 hypnosis sessions in the whole sample

Session	N LME	Pre score mean (SD)	Change score mean	Std Err	95% LB	95% UB	Effect Size
1	77513	5.59 (2.09)	−1.623	0.01	−1.643	−1.603	−0.78
2	34153	5.64 (2.09)	−1.538	0.015	−1.566	−1.509	−0.74
3	19677	5.69 (2.07)	−1.496	0.019	−1.532	−1.459	−0.72
4	13219	5.71 (2.06)	−1.462	0.022	−1.505	−1.419	−0.71
5	9805	5.72 (2.05)	−1.464	0.025	−1.513	−1.415	−0.71
6	7749	5.75 (2.04)	−1.447	0.028	−1.501	−1.392	−0.71
7	6290	5.78 (2.04)	−1.485	0.03	−1.545	−1.425	−0.73
8	5272	5.73 (2.02)	−1.452	0.033	−1.518	−1.387	−0.72
9	4449	5.78 (2.01)	−1.511	0.035	−1.581	−1.442	−0.75
10	3886	5.84 (1.97)	−1.466	0.037	−1.539	−1.393	−0.74

Fig. 1. Estimated mean stress reduction and sample sizes of the first 10 hypnosis sessions in the whole sample.

The estimated mean change for each session represents the slope of an LME model, with stress score as the dependent variable and time as the independent variable (pre-score = 1, post-score = 2). The sample size for each session reflects the number of users whose stress scores contributed to the model for that session.

Longitudinal stress reduction across sessions

While stress reduction was observed consistently within sessions, our analysis did not detect meaningful longitudinal changes of this effect across sessions, suggesting repeat use produced consistent but not increasing stress reduction over time across the first 10 sessions and subsequent sessions. The LME results estimated a slope of 0.035 (95%CI = 0.030, 0.040) across the first 10 sessions and a slope of 0.0016 (95%CI = 0.0014, 0.0019) across all sessions.

Similarly, we found no meaningful evidence of change in baseline stress (pre-intervention scores over repeat sessions) across the first 10 sessions (β = 0.034, [95%CI = 0.030, 0.038]) or across all sessions (β = −0.0003, [95%CI = −0.0005, −0.0001]).

User participation based on first session response

We found no clinically meaningful association between first session pre-to-post change in stress scores and number of sessions (Spearman rho = −0.048 (95% CI = −0.058, −0.037) (Table 2), suggesting users were not more likely to participate in repeat sessions based on first session response. We also found that users with higher first session pre-intervention (baseline) stress scores did not participate in a meaningfully different number of subsequent sessions (Spearman rho = 0.011, 95% CI = 0.004, 0.018). Similarly, the association between the number of repeat sessions and first session post-intervention stress scores, although statistically significant, was not clinically meaningful (Spearman rho = −0.022, 95% CI = −0.033, −0.012).

Table 2.

Spearman rank order correlation between number of subsequent hypnosis sessions with session 1 pre, post, and change stress scores

	Pre-score (Session 1)	Post-score (Session 1)	Change score (Session 1)
Spearman rank order correlation with number of subsequent sessions (ρ)	0.011	−0.022	−0.048
95% confidence interval	0.004, 0.018	−0.033, −0.012	−0.058, −0.037
Number of Session 1 Users1	74,835	36,083	33,405

¹This represents the number of session 1 users who had the data.

Sensitivity analysis by completeness of the data

The effect sizes from the LME models in the entire sample ranged from −0.82 to −0.94 when only subjects with complete data for both pre- and post-scores were included in the models. This suggests that the results reported in Fig. 1 and Table 1 may have been underestimated due to data missingness, which is partly attributable to treating pre-scores of 10 and post-scores of 1 as unusable or missing.

Sensitivity analysis by software update

The LME model effect sizes ranged from −0.66 to −0.72, reflecting the conservative approach before the software update, where pre-intervention scores of 10 and post-intervention scores of 1 were reset to missing (Table 3). The LME model effect sizes ranged from −0.76 to −0.93 after the software update (Table 3). Thus, the sensitivity analysis indicates the overall estimated changes above (i.e., based on the combined sample; Fig. 1 and Table 1) may have been underestimated (Fig. 2).

Table 3.

Estimated mean stress reduction and effect sizes after software update

Software update	Session	N LME	Pre score mean (SD)	Change score mean	Std Err	95% LB	95% UB	Effect aize
Before	1	58118	5.47 (2.06)	−1.462	0.012	−1.486	−1.438	−0.71
Before	2	24541	5.57 (2.07)	−1.395	0.018	−1.431	−1.359	−0.67
Before	3	13850	5.64 (2.05)	−1.368	0.024	−1.414	−1.321	−0.67
Before	4	9297	5.65 (2.03)	−1.338	0.028	−1.393	−1.284	−0.66
Before	5	6991	5.68 (2.02)	−1.359	0.031	−1.421	−1.298	−0.67
Before	6	5564	5.71 (2.00)	−1.339	0.034	−1.406	−1.272	−0.67
Before	7	4562	5.74 (2.01)	−1.392	0.037	−1.465	−1.32	−0.69
Before	8	3819	5.71 (1.99)	−1.375	0.041	−1.456	−1.294	−0.69
Before	9	3264	5.75 (1.98)	−1.419	0.043	−1.505	−1.334	−0.72
Before	10	2896	5.8 (1.92)	−1.384	0.046	−1.473	−1.294	−0.72
After	1	19395	5.94 (2.13)	−1.983	0.017	−2.017	−1.949	−0.93
After	2	9612	5.82 (2.12)	−1.788	0.024	−1.835	−1.742	−0.84
After	3	5827	5.79 (2.11)	−1.693	0.029	−1.751	−1.636	−0.8
After	4	3922	5.83 (2.13)	−1.653	0.034	−1.721	−1.586	−0.78
After	5	2814	5.82 (2.11)	−1.643	0.041	−1.724	−1.562	−0.78
After	6	2185	5.85 (2.14)	−1.643	0.048	−1.736	−1.549	−0.77
After	7	1728	5.89 (2.14)	−1.662	0.054	−1.767	−1.556	−0.78
After	8	1453	5.77 (2.10)	−1.585	0.057	−1.696	−1.474	−0.76
After	9	1185	5.83 (2.08)	−1.695	0.061	−1.815	−1.576	−0.81
After	10	990	5.96 (2.10)	−1.615	0.064	−1.741	−1.489	−0.77

Fig. 2. Estimated mean stress reduction and sample sizes of the first 10 hypnosis sessions before and after the software update at sessions 1–10.

Red bars and blue bars indicate estimated mean changes before and after the software update, respectively. The estimated mean change for each session represents the slope of an LME model, with stress score as the dependent variable and time as the independent variable (pre-score = 1, post-score = 2). The sample size for each session reflects the number of users whose stress scores contributed to the model for that session.

Differences in stress reduction based on subgroups

Analysis of the effects of HIP scores, our measure of hypnotizability, on stress reduction revealed a positive correlation. In session 1, the mean stress reduction was greater for users with high HIP scores (7–10) (β = −0.51 [95%CI = −0.59, −0.44]) and medium HIP scores (4–6) (β = −0.24 [95%CI = −0.32, −0.16]) compared to those with low HIP scores (0–3) (Fig. 3). Similarly, users with high HIP scores had greater stress reduction than those with medium HIP scores (β = −0.28 [95%CI = −0.34, −0.21]).

Fig. 3. Mean stress reduction in session 1 by baseline hypnotizability.

The estimated mean change represents the slope of an LME model for each HIP category, with stress score as the dependent variable and time, HIP category, and their interactions as independent variables (time = 1 for pre-score, time = 2 for post-score). Users with HIP scores of 0–3, 4–6, and 7–10 were categorized as “Low”, “Medium,” and “High” hypnotizability, respectively. N represents the number of users belonging to each HIP category at session 1.

HIP scores had a small but consistently significant effect size on stress change score at each of the first 10 sessions, ranging from −0.15 (95% CI = −0.18, −0.12) to −0.11 (95% CI −0.14, −0.07) (Table 4). Thus, higher hypnotizability was associated with greater stress reduction within these sessions.

Table 4.

Correlations between HIP score and stress reduction in the first 10 sessions

Session	N	Spearman rho (ρ)	Lower CI	Upper CI
1	15,933	−0.14 ***	−0.16	−0.13
2	8614	−0.13 ***	−0.15	−0.11
3	5694	−0.13 ***	−0.16	−0.11
4	4224	−0.11 ***	−0.14	−0.08
5	3358	−0.15 ***	−0.18	−0.12
6	2688	−0.15 ***	−0.18	−0.11
7	2240	−0.11***	−0.15	−0.07
8	1948	−0.14 ***	−0.18	−0.09
9	1665	−0.13 ***	−0.18	−0.08
10	1478	−0.11 ***	−0.16	−0.05

Small effect sizes (−0.15 < ρ < −0.11) were observed in sessions 1–10. N represents the number of users with both pre-to-post stress score and baseline hypnotizability score at each session (i.e., those with missing data for either pre-to-post score or HIP score were excluded from this analysis). The symbol ‘***’ indicates statistical significance at the 0.05 alpha level.

There was a small but statistically significant correlation between hypnotizability and participation in repeat sessions, rho=0.14.

Analysis of session types (interactive vs. non-interactive, regular vs. brief) showed that the standard/interactive sessions were associated with a larger reduction in stress scores compared to the brief/non-interactive sessions. The standard/interactive session type is closer to a typical clinician-administered intervention. In session 1, the mean change stress score observed in the standard/interactive condition was 1.8 times greater than that in the brief/non-interactive condition (M = −2.02, SD = 1.89 versus M = −1.13, SD = 1.79, respectively). Superiority in stress reduction in the standard/interactive types compared to the brief/non-interactive types was sustained in subsequent sessions. A LME model showed that the main effects of interactive and standard-length sessions were positively associated with larger stress reduction (interactive sessions: β = −0.48 [95%CI = −0.55, −0.40]; standard sessions: β = −0.27 [95%CI = −0.35, −0.20] in session 1), but their interaction effect was not significant (β = 0.12 [95%CI = −0.03, 0.27]). Similar trends were observed in the first 10 sessions.

We compared the change in stress reduction between those who completed ten sessions of stress reduction within four weeks and those who took longer. All effect sizes were small (.10 or less), so we concluded that more immediate use of the exercise was not associated with better stress control.

In session 1, the pre-to-post stress reduction was smaller for each increment in the ordinal age category (β = 0.09 [95%CI = 0.05, 0.13]), which indicates that the stress reduction was greater for the youngest age group (i.e., under 25) compared to the older groups. Across the first 10 sessions, the pre-to-post stress reduction was greater for each increment in the ordinal age category (β = −0.02 [95%CI = −0.03, −0.01], indicating stress reduction was greater in older age groups across the first 10 sessions. Similar patterns were observed when we used categorical age groups (55–64 [second oldest] vs. under 25 [youngest]: β = −0.09 [95%CI = −0.17, −0.02]; 65 and older [oldest] vs. under 25 [youngest]: β = −0.12 [95%CI = −0.20, −0.03]). These findings suggest that younger users were more likely to respond well to the initial session, while older users, who completed more repeat sessions, benefited more from multiple uses compared to younger repeat users.

The pre-to-post stress reduction was greater among female than male users in session 1 (β = 0.12 [95%CI = 0.05, 0.20]). The pre-to-post stress reduction was slightly greater for male users than female users across the remaining first 10 sessions, but the difference was not clinically meaningful (β = −0.02 [95%CI = −0.043, −0.003]) (Fig. 4).

Fig. 4. Mean stress reduction across the first 10 sessions by sex.

The estimated mean changes for each sex were derived from an LME model, with pre-post change in stress score as the dependent variable and session, sex, and their interaction as independent variables.

The pre-to-post intervention stress reduction was greater for members than non-members in session 1 (β = −0.21 [95%CI = −0.25, −0.17]) and across the first 10 sessions (β = −0.04 [95%CI = −0.06, −0.02]). Across all sessions, no meaningful difference was observed (β = −0.004 [95%CI = −0.007, −0.001]) (Fig. 5).

Fig. 5. Mean stress reduction across the first 10 sessions by membership status.

The estimated mean changes were derived from an LME model, with pre-post change in stress score as the dependent variable and session, membership status, and their interaction as independent variables.

Discussion

As more people worldwide acknowledge stress and stress-related illness as key challenges to their mental and physical well-being, widely disseminable digital interventions unconstrained by clinician availability, cost, and insurance coverage considerations emerge as a useful modern strategy. However, the proliferation of mental health apps thus far has outpaced confirmation of their effectiveness. To our knowledge, this is the first study to demonstrate the effects of hypnosis delivered remotely through a digital application for stress reduction at scale, in a very large sample of people. For all users of the Reveri app (84,395 subjects), there were only ten users who complained of worsening symptoms or other problems, all minor. Our study thus provides encouraging evidence for an accessible, safe, and inexpensive formulation of a well-studied and rapid-acting intervention for stress.

Our observations may be viewed in light of several neurobiological findings. Evidence from neuroimaging studies suggests that hypnosis involves activity changes in the central executive network (CEN), salience network (SN), and default mode network (DMN)^32,33. These changes likely represent responding to suggestion (CEN), altering attention (SN), and disengaging from self-referential thinking (DMN)^20,33, which cumulatively may contribute to changes in stress perception underlying the stress reduction observed in the current study.

We observed modest yet consistent relationships between hypnotizability and benefit from hypnosis sessions. While directionally consistent with a considerable amount of research supporting a link between hypnotizability and treatment outcome³⁴, our observation suggests that, at a large scale, one does not need to be highly hypnotizable to benefit from hypnosis. Nevertheless, the consistent relationship across sessions suggests that moderate and high hypnotizability are a reliable factor in responding to hypnosis-based interventions. A key limitation of this finding is the experimental nature of the hypnotizability assessment modality; while traditional clinical evaluations are done in-person, we opted to use a remote app-based version of the Hypnotic Induction Profile (HIP) to remain consistent with the scalability of the approach. Further studies are needed to evaluate the comparability of remote app-based HIP with in-person administration.

While users were able to reduce stress at each use, longitudinal changes with repeated use were not observed in our sample. These observations contrast with existing literature on hypnosis, which shows repeated use to facilitate resolution of a problem (analogous here to lower pre-intervention stress scores over time) and allow for more benefit at each use with practice (analogous to significant change from pre-to-post)²⁴. However, methodological factors may help account for these findings. In clinical settings, patients are often prescribed daily practice of hypnosis for a given problem, sometimes multiple times per day in a structured manner, while repeated use in our study does not necessarily indicate daily or routine practice. Thus, investigation of potential cumulative and practice effects may be more feasible with controlled, prospective study designs, as opposed to the real-world, retrospective design that enabled our usefully large sample size. Nevertheless, the emerging picture here may be of stress reduction akin to as-needed benzodiazepine use, which can help rapidly lower anxiety when it occurs but may not necessarily prevent anxiety, and in contrast to selective serotonin reuptake inhibitors (SSRIs), which require daily administration to reach therapeutic effects of diminished baseline anxiety. While this was a self-selected sample rather than one with diagnosed anxiety disorders, the side effect profile of benzodiazepines and other sedative-hypnotics (e.g., dependency and cardiorespiratory effects) highlights the long-recognized need for safe and effective anxiolytic interventions³⁵.

We acknowledge various study limitations. First, as data were obtained from a commercial mobile application, under the terms of the user agreement, demographic data were limited and in some cases unavailable for analysis. Second, the study did not have a control condition, which limited causal inference capability in the whole sample and in subgroups (i.e., the different session types). Third, users were not necessarily receiving the intervention in a uniform or structured manner, a design choice that has the potential to introduce confounding variables, though the very large sample size helps mitigate this problem. Fourth, our measure of stress was a simple 10-point Likert scale rather than a multi-item validated questionnaire (e.g., PSS, PANAS, SSSQ)^36–38. This was done to elicit involvement and responsiveness in a broad population of people with varying levels of distress accessing a public-use app, and to measure immediate stress and changes in stress, rather than stress over the preceding week or month as assessed by other questionnaires. Fifth, because of our chosen study design, we cannot define the clinical vs. non-clinical composition of the study population, which does limit interpretation of subgroup effects. However, our focus was on stress and stress reduction in the general population, encompassing a range of experiences from worry about stressors to anxious responses to stressors³⁹, and future studies could focus on app-delivered hypnosis for clinical vs. non-clinical samples. Future RCTs could be useful to better characterize the potential for effectiveness of app-delivered self-hypnosis by addressing these limitations. Ultimately, our investigation provides support for the utility and safety of digital delivery of a rapidly effective stress reduction intervention at-scale.

Methods

Intervention

We observed a very large group of people’s use of a mobile app called Reveri (www.Reveri.com), which guides users through self-hypnosis targeting a range of problems, including stress, pain, poor sleep, and attention difficulties. We focused on users of the “Manage Stress and Anxiety” session module, an audio-guided hypnotic intervention involving suggestion of sensorimotor changes, guided visualization, and cognitive restructuring. This study followed a Strengthening Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cross-sectional studies⁴⁰.

Users could select session version (interactive or non-interactive) and length (standard or brief). In the interactive version, users could speak aloud or tap on their mobile screen to respond to audio prompts (e.g., “Is your arm floating up in the air?), which would trigger follow-up instructions (e.g., “Imagine a balloon tied to your wrist and let your arm float up.”). In the non-interactive version, users listened to the recorded hypnotic intervention without responding to prompts. Standard length sessions were 11–13 minutes (interactive) or 10 minutes (non-interactive), and brief sessions were 4–5 minutes (interactive) or 4 minutes (non-interactive). Brief sessions maintained the essential components of the intervention with less preparatory explanation.

To measure baseline hypnotizability, users were guided through a remotely administered Hypnotic Induction Profile (HIP)³⁴, which employs a scoring system ranging from 0 to 10 (the Induction Score, IND), where scores higher than 6 indicate moderate to high hypnotizability, and scores lower than 4 indicate low hypnotizability. The HIP is a standardized assessment of trait hypnotizability and requires approximately 5–10 minutes to administer^41,42. Independent evaluations have shown that the HIP is moderately positively correlated with laboratory-based measures of hypnotizability⁴³. Higher HIP scores have been shown to correspond with increased functional MRI (fMRI) connectivity between the left dorsolateral prefrontal cortex (dlPFC) and the dorsal anterior cingulate cortex (dACC)⁴⁴. The remote HIP, which is equivalent to the traditional HIP with minor logistical modifications⁴⁵, assesses sensorimotor changes associated with the key elements of hypnosis (absorption, dissociation, suggestibility) through a series of brief instructions and questions (e.g., “Look up towards the top of your head as you close your eyelids slowly”; “When you stroke the middle finger of your left hand, you will develop movement sensations in that finger”; “Do your left arm and hand feel as though they are not as much a part of your body as your right arm and hand?”; “When your arm went up were you aware of a relative difference in your sense of control between the hypnotized and non-hypnotized arm?”; “Did you experience a sensation you could describe as a lightness, buoyancy or floating?”).

Users were prompted to rate their current stress level on a 10-point Likert Stress Scale immediately before and after sessions, with 1 indicating the lowest stress and 10 indicating the highest stress. This simple scale was utilized to maximize engagement and data collection, and based on evidence of comparability to clinical measures⁴⁶. Such a scale demonstrated sensitivity to differential reduction in stress/anxiety in an RCT during catheterization surgery comparing intravenous opioids to supportive nursing care and hypnosis²¹.

Data collection

Data were collected from all 91,170 users, who logged a total of 290,696 stress reduction sessions between November 2021 and January 2025. Of these users, 6775 had missing data on their stress scores throughout the sessions, resulting in an analytic dataset consisting of N = 84,395 users and 282,893 sessions.

Users were asked to report sex and age group with enrollment only after April 1, 2024 and June 7, 2024, respectively. The number of users in sex and age groups is described in Table 5. We examined only the first 10 sessions for age and sex effects due to the lack of data on participants who repeated more than 10 sessions.

Table 5.

Demographics of users

	Sample size (N)	Percentage (%)
Sex
Female	8536	48.3%
Male	9022	51.0%
Non-binary/non-conforming	115	0.7%
Total	17,673	100%
Age group
<25 years old	617	5.5%
25–34 years old	2588	23.2%
35–44 years old	3742	33.5%
45–54 years old	2296	20.6%
55–64 years old	1338	12.0%
≥65 years old	584	5.2%
Total	11,165	100%
Membership
Member	34,218	40.5%
Non-member	50,177	59.5%
Total	84,395	100%
Hypnotizability
Low (0–3)	7,617	21.6%
Medium (4–6)	9,635	27.4%
High (7–10)	17,956	51.0%
Total	35,208	100%

Outcome variable

Our outcome throughout the analyses was the stress scores assessed at both pre-intervention and post-intervention, or the change between the immediate post- and pre-intervention stress scores (i.e., change scores), depending on the statistical methodologies used.

Independent variables

HIP scores ranged from 0 to 10. Session version was binary-coded as 1 for interactive and 0 for non-interactive. Session length was binary-coded as 1 for regular and 0 for brief. Age was collected in categories: under 25, 25–34, 35–44, 45–54, 55–64, and 65 or older; categories were re-coded as ordinal values (0, 1, 2, 3, 4, 5, respectively). We refer to this re-coded age as the ordinal age category. Sex initially had 3 categories: male, female, and non-binary/non-conforming. Due to extremely small counts of the lattermost category, we included only male and female in the analysis for sex effects. Membership was binary-coded as 1 for paying members and 0 for non-members.

Handling of software update

As mentioned, our analytic data included 84,395 users, who logged a total of 282,893 stress reduction sessions between November 2021 and January 2025. A total of 62,607 users who installed the app before a software update (December 22, 2023 for IOS and January 15, 2024 for Android) completed 211,261 sessions, and 21,788 users who installed the application after the software update completed 71,632 sessions. Accordingly, before the software update, 24,790 users (41.8% of users) completed multiple hypnosis sessions; after the software update, 15,928 users (54.7% of users) completed multiple hypnosis sessions. Prior to the software update, pre- and post-intervention stress scores not reported by users within a fixed time interval were assigned by the software default values of 10 for pre-scores and 1 for post-scores. The software update eliminated the time limit and default score assignment. Thus, we stratified the data into two separate cases: “Before the software update”, where we treated pre-scores of 10 and post-scores of 1 as unusable or missing data to avoid biasing results, and “After the software update”, representing sessions that only included new users after the update, with values of 10 and 1 preserved. The assumption for the “Before the software update” was conservative for those who truly scored 10 or 1, as their true scores were also treated as missing data. We first analyzed the whole sample, combining the “Before” and “After” software update groups. As a sensitivity analysis, we analyzed the two cases separately to assess for different trends (See Sensitivity analysis by software update).

Handling of missing data

Our method of handling missing data, including values assigned as missing due to the software issue, was with the Linear Mixed-Effects (LME) model using restricted maximum likelihood estimation. We selected the LME as our main statistical analysis due to its robustness in handling missing data and capability to model both fixed and random effects, thus accounting for the variability both within and between users over time⁴⁷. We used a random intercept model assuming a linear change. In LME, it is assumed that the missing data mechanism is random conditional on observed information in the overall sample. This assumption was reasonable since missingness occurred when users failed to respond within the time limit, which is unrelated to stress levels.

Primary analysis

For our primary aim, we estimated the mean change in the primary outcome from pre- to post-intervention based on mixed effects modeling using the first 10 hypnosis sessions. We reported 95% confidence intervals and Cohen’s d values, which represent effect sizes.

Intervention effects across sessions

As exploratory aims, to examine whether the degree of first-session stress reduction predicted the likelihood of repeat use, we calculated the Spearman Rank-Order correlations between available session 1 change scores and the number of sessions. We also tested the relationships of session 1 pre-score (baseline stress) and session 1 post-score with repeat use using Spearman Rank-Order correlations. To test whether repeated use was associated with greater or smaller changes in stress in response to the intervention, we compared the slopes based on mixed effects models across sessions.

Potential moderators of intervention effects

We tested hypnotizability-intervention interaction for session 1 through LME modeling, as well as Spearman Rank-Order correlation analysis between HIP score and stress reduction over the first 10 sessions. In addition, we tested the relationship between interactive vs. non-interactive and standard vs. brief length versions using the LME. To examine whether age, sex, or paying membership influenced stress reduction over sessions, we fitted LME models that included interactions with these variables across sessions.

Author contributions

N.T. and C.S. co-drafted the initial manuscript, led figure and table development, and assisted with study design, data interpretation, and revisions. E.N. contributed to data analysis, manuscript writing, and editing. B.J. and B.K. independently oversaw data analysis and manuscript preparation. N.Z. contributed to background research, study design, and editing. B.N. facilitated project coordination and ethical compliance and contributed to editing. C.K., A.J., and A.F. assisted with background search and editing. J.M. independently provided clinical insight, oversaw data analysis, and directed manuscript preparation. D.S. conceptualized the study, supervised the project, contributed to manuscript revision, and provided domain expertise.

Data availability

The deidentified participant data used in this manuscript is not publicly available at this time pursuant to a Data Access Agreement between Stanford and Reveri Health, Inc. Reveri Health retains ownership of the data and exclusive rights to its distribution. For further information on data access requests, interested readers may contact Reveri Health, Inc. (dan@reveri.com) directly.This study involved a retrospective analysis of de-identified data collected by Reveri Health, Inc., with user consent provided during enrollment consistent with Reveri’s Privacy Policy. Data collection was performed by Reveri; Stanford University researchers conducted the analysis without access to identifying patient health information (PHI). In his roles as Co-Founder and Scientific Advisor at Reveri Health, Dr. Spiegel only accessed personal health information in response to rare requests from users seeking mental health support. The Stanford IRB reviewed the study (Protocol #76729) and determined it did not meet the definition of human subjects research under 45 CFR 46.102 or 21 CFR 50.3; therefore, no further review was required.

Code availability

The statistical code used to analyze the data in this study was written in R (version 4.4.1). All results were produced using the variables and parameters specified in the Methods section, including the linear mixed effects models.

Competing interests

D.S. is Co-Founder and Scientific Advisor at Reveri Health, Inc., the vendor that collected the data used in this study. The study was reviewed by the Stanford University Conflict of Interest Office. In accordance with their recommendations, senior faculty members with no financial or professional relationships with Reveri Health, Inc. (B.J. and J.M.) independently oversaw the data analysis and manuscript preparation. A.F. served as a consultant for Reveri Health Inc. during the period in which data for this study were collected, but had no role in the data collection process.

Supplementary information

The online version contains supplementary material available at 10.1038/s41746-025-02182-0.