The Standalone Effect of Safety Behavior Manipulations: A Systematic Review and Meta-Analysis

Safety behaviors (SBs) are overt actions and mental strategies used to prevent or reduce anxiety, avoid a potential threat, or prevent a negative outcome (Salkovskis, 1991). They are a key maintaining factor in cognitive-behavioral conceptualizations of anxiety, obsessive-compulsive, and trauma-related disorders (Ehlers & Clark, 2000; McManus, Sacadura & Clark, 2008) and have been associated with a host of adverse clinical symptoms. For example, SBs are associated with the development and worsening of anxiety (Goodson & Haeffel, 2018; Deacon and Maak, 2008; Olantunji et al, 2011), increased perceptions of threat (Engelhard et al., 2015; Lovibond et al, 2008), maintenance of catastrophic beliefs about anxiety (Salkovskis et al. 1999), decreased confidence (Abbasi, Yazkhasti, & Abedi, 2023), decreased perceptions of control (Milosevic & Radomsky, 2013), increased negative perceptions of others (Alden & Bieling, 1998; McManus et al., 2008), and decreased quality of life (Kirk et al, 2019).

SBs also play a role in the treatment of anxiety-related disorders. While findings have not been uniform, fading SBs (i.e., systematically having individuals reduce and eliminate SBs) during treatment for anxiety enhances outcomes (Wells et al., 1999; McManus et al., 2008; Goodson & Haeffel 2018), while SB use during treatment results in smaller clinical gains (Helbig-Lang et al 2014). Additionally, the presence of SBs at post-treatment has been associated with greater risk of relapse (Beesdo-Baum et al, 2012; Patel et al., 2023; Zech et al., 2023). SB interventions have also been used in anxiety prevention studies with varying levels of effectiveness (Gorman, Goodson, & Haffel, 2023, Arai et al., 2023).

Initially, the treatment literature on SBs primarily investigated their role within the context of established cognitive and behavioral treatments for anxiety disorders (Salkovskis, 1999; Wells et al., 1995; Lovibond et al, 2009). However, over the past decade, a growing body of research has investigated the impact of stand-alone SB treatment interventions as well as non-treatment experimental manipulations (Patel & Cougle, 2024a, Summers & Cougle, 2018; Wilver et al., 2020). A number of studies, employing a wide-range of methodologies, have investigated the efficacy of targeting SBs (Gorman et al., 2023; Olatunji et al., 2011; Patel & Cougle, 2024b, Raines et al, 2023; Schmidt et al., 2012). However, little is known about the general efficacy of these standalone SB interventions, as no systematic review or meta-analysis has been conducted which focuses on their outcomes.

Past Systematic Reviews and Meta-Analyses on Safety Behaviors

We located four published meta-analyses or systematic reviews examining SBs in the context of anxiety (Helbig-Lang & Petermann, 2010; Blakey and Abramowitz, 2016; Mulders et al., 2016, and Goetz et al., 2016). In their review, Helbig-Lang and Petermann (2010) proposed a two-axis classification system for SBs (cognitive vs. behavioral and preventative vs. restorative) and recommended identifying and abandoning SBs during anxiety-related treatments. Meulders and colleagues (2016) conducted a meta-analysis investigating exposure interventions with instructions to either decrease SBs or increase SBs during exposure. Results suggested little impact of the effects of SB-related instructions, with small effects for studies that included decreasing SBs (standardized mean difference = 0.31, [−0.04, 0.66]) and small-to-negative effects for studies that included increasing SBs (standardized mean difference = −.13, [−0.37, 0.11]). Goetz et al (2016) carried out a systematic review of the literature focused on preventative and restorative SBs. The authors found little evidence for any detrimental effects associated with restorative SB use. With respect to preventative SB use, the authors found mixed results, with approximately half of included studies showing negative treatment effects, and the other half showing no impact. Finally, Blakey and Abramowitz (2016) reviewed SBs from an inhibitory learning perspective. The authors highlighted several pathways through which SBs may impede fear reduction, including three pathways specific to inhibitory learning: 1) SBs do not allow for full expectancy violation; 2) SBs diminish generalization of safety learning; and 3) SBs prevent the development of distress tolerance.

None of the systematic reviews or meta-analyses to date have investigated the effects of stand-alone interventions or experimental manipulations targeting SBs. This is an important omission, as there has been a proliferation of such studies over the past decade. However, the efficacy of these approaches remains unclear. A meta-analytic review may be able to shed light on differential effectiveness across different types of SB intervention studies by examining important moderators associated with better (or worse) corresponding clinical symptoms. For example, several studies have sought to examine the effect of reducing SB within a non-treatment experimental context, where no treatment rationale is given for changing these behaviors (Patel & Cougle, 2024; Summers & Cougle, 2018; Wilver et al., 2020). These studies specifically sought to examine the causal role of these behaviors in psychopathology (Cougle, 2025). In contrast, other studies provided an explicit rationale as to why this approach may lead to changes in symptoms within a treatment context (i.e., treatment seeking sample, clinically oriented psychoeducation, and change in SB as a therapeutic tool). It is possible that the difference between a treatment and non-treatment experimental context may moderate effects on SB and clinical symptoms. Further, it is unknown whether therapist-guided SB interventions are associated with better symptom improvements compared to interventions that do not require a therapist. This is an important question given the relative lack of therapists trained in evidence-based treatments for anxiety disorders (Frank et al., 2020).

In the present study, we conducted a systematic review and meta-analysis on the effect of directly targeting SBs. We evaluated the within-group and between-group efficacy of targeting SBs. We evaluated studies that both examined an SB manipulation relative to a control condition as well as single group trials. However, we limited our meta-analyses to between-group effects, given the issues of reliably calculating within-group effects for meta-analyses (Cuijpers et al., 2017). We examined the between-group effects of decreasing SBs on (1) SBs and (2) primary clinical symptoms as well as the between-group effects of increasing SBs on (3) SBs and (4) clinical symptoms. Additionally, we tested moderators associated with clinical symptoms. We also explored whether study type (experiment vs. treatment) was a moderator of observed effects on SBs and symptoms. Further, we hypothesized that effect size of the SB manipulation on SB use would be a moderator for effects on clinical symptoms, such that higher effect size on SB use would be associated with greater effect size on clinical symptoms.

Methods

Search Strategy

The initial search was performed in June 2025 on APA PsycINFO, Academic Search Complete, and PubMed for articles published up to June 2025. We updated our search in September 2025 prior to conducting analyses. The following search string was used to conduct the search: (“Safety behav*” OR “Safety seeking behav*”) AND (therapy OR treatment OR trial OR intervention OR experiment* OR effect* OR test) AND (reduction OR change OR increase OR decrease OR fading OR role) AND (panic OR phobia OR social phobia OR agoraphobia OR obsessive-compulsive disorder OR OCD OR generalized anxiety disorder OR GAD OR posttraumatic stress disorder OR PTSD OR health anxiety OR social anxiety OR appearance OR mood OR depression). This string allowed us to identify articles that explicitly observed SBs in a therapeutic or experimental context in relation to psychopathology. This systematic review was conducted according to the preferred reporting elements for systematic reviews and meta-analyses (PRISMA; Page et al., 2021) and meta-analyses of psychotherapy studies (MAP-24; Flückiger et al., 2018). Additionally, a protocol of procedures was developed and registered on PROSPERO (ID: CRD420251063442). See Supplementary Table 1 for concordance with PRISMA checklist for meta-analyses.

All search results were compiled in Zotero and imported into the web-based platform Covidence for deduplication and screening. This review was focused on identifying treatment or experimental studies that directly targeted SBs (whether increase or decrease) and measured the effect on symptoms of psychopathology. Further, they included pre- and post-treatment measure of these symptoms. The goal of the search was to identify how these procedures effect SBs and their subsequent impact on symptoms of psychopathology. The initial search yielded 428 articles after duplicates were removed (n = 271; Figure 1).

Figure 1.

Study Selection Flow Chart

Screening procedures were conducted using Covidence systematic review software. Titles and abstracts for these 428 studies were screened for relevance twice independently by two co-authors (JG and TP). Conflicts were resolved by the two screeners meeting and discussing the articles until consensus was reached. Title and abstract screening resulted in the exclusion of 372 irrelevant studies. For the remaining 58 studies, full texts were retrieved, and screening of full text articles was conducted to make inclusion decisions. All full text articles were screened for relevance twice independently by two of the co-authors (JG and TP). Conflicts were resolved by a discussion of the articles until consensus was reached by the two reviewers. Full text review resulted in the exclusion of 32 studies. A total of 26 studies were included.

Eligibility Criteria

Abstract Review

For studies to be included at this phase, articles were required to: 1) be peer-reviewed and published in scientific journals; 2) provide quantitative results; 3) be written in English; 4) include participants (or a subset of participants separately reported) aged 18 years or older; 5) describe a treatment or experimental study; and 6) measure SBs. Articles were excluded if they violated a criterion for inclusion or if they represented a case study, case series, or review article. If no abstract was provided, articles were included for review in the subsequent phase.

Full-Text Review

For the full-text phase, articles were required to: 1) have an English full-text available; 2) be a peer-reviewed manuscript; 3) be an empirical article with quantitative results; 4) only include adult participants or separate adult results from child/adolescent; 5) be a treatment or experimental study; 6) directly target SBs (rather than a study that examined a treatment that included components of SBs reductions such as exposure and response prevention); 7) directly target SB outside of the context of exposure (e.g., Meulders et al., 2016; Riccardi et al., 2017; Schmidt et al., 2012); and 8) report effect sizes or information required to calculate effect sizes for between-group differences. Importantly, we do include within-group effect sizes in our systematic review, but we did not examine these in our meta-analyses, given issues regarding effect and standard error calculation in within-group effect sizes resulting in bias in meta-analyses (Cuijpers et al., 2017).

Data Extraction

Two authors (JG and TP) independently extracted the following data for each study: country, sample size, experimental vs. treatment, increase or decrease SB manipulation, control condition, duration of manipulation, SB measure and timeframe, symptom measure and timeframe, effect size and descriptive statistics for both SB and symptoms, sex, race, age, inclusion criteria, research focus, and whether a clinician was involved in procedures. In the absence of effect sizes, the available data were used to calculate Cohen’s d effect sizes.

Synthesis of Results for Systematic Review

All studies included in the review were systematically reviewed to narratively describe sample descriptives, the focus of the studies, details on the experimental or treatment procedures, measures used to assess symptoms, within-group effect sizes, and between-effect sizes. Effect sizes were narratively described using Cohen’s d and the following conventions: .20 = small, .50 = medium, and .80 = large. Though the primary focus of our meta-analysis was the between-group effects, we chose to summarize within-group effects in our review for several reasons. First, it allowed for the inclusion of additional studies to more robustly describe the state of the literature. Second, the inclusion of these effects contextualizes heterogeneity in between-group effects, which we tested in our moderator analyses. Third, these effects help to inform how future studies should account for potential placebo or expectancy effects (e.g., large within-group effects with small to no between group effects). Finally, they provide descriptive information on clinically significant change that can be used for planning future studies.

Quality Assessment

Risk of bias for RCTs was assessed using the Cochrane Risk of Bias tool version 2 (Sterne et al., 2019). Ratings assessed the following domains: randomization, deviation from intended intervention, missing data, measurement of symptoms, and selective reporting. For non-randomized controlled trials, we used the Cochrane Risk Of Bias In Non-randomised Studies - of Interventions (ROBINS-I). This tool assessed the latter four criterion from the previous tool in addition to controlling for confounding factors, intervention classification, and selection of participants.

Meta Analysis Data Analytic Plan

All analyses were completed in R version 4.4.2 using the meta and dmetar packages (Balduzzi et al., 2019; Harrer et al., 2019; R Core Team, 2024; Viechtbauer, 2010). Effect sizes used in the analyses were Cohen’s d, calculated using standardized mean differences (SMDs). Using these effect sizes, meta-analyses were conducted for the between-group effect sizes at post-treatment. Heterogeneity was measured using I², Q, and tau statistics. I² describes the variation across study results not due to chance as a percentage with 25% considered low, 50% moderate, and 75% high heterogeneity (Higgins et al., 2003). Q describes the weighted sum of the squared difference between the effects of a study and the pooled effect across other studies in a meta-analysis (Cochran, 1954). Finally, tau represents the standard deviation of the true effect size.

To explore the study aims, we planned to conduct four meta-analyses. First, we conducted a between-group meta-analysis of the effect of decreasing SBs on SB frequency at post to examine the pooled effect of such a manipulation. Second, we conducted a between-group meta-analysis of the effect of decreasing SBs on the primary symptoms measured in studies. Third, we conducted a between-group meta-analysis of the effect of increasing SB on SB frequency at post. Finally, we conducted a between-group meta-analysis of the effect of increasing SB on the primary symptoms measured in studies. We separated our analyses by increase or decrease studies given these effects would be opposite signs of each other, and they represent two distinct empirical questions to test (i.e., does increasing SB increase psychopathology?; Does decreasing them lower psychopathology?).

Meta-regressions were conducted using restricted maximum likelihood to test potential moderators of effect size. Moderator analyses are presented with intercepts (B0; i.e., the mean effect size after controlling for the moderator), the moderator effect (B), and the R² analogue (i.e., proportion of the between-study variance (tau²) that can be explained by the moderator). Possible moderators included year of publication, sample size, study type (treatment vs. non-treatment), duration of manipulation, manipulation type (daily reminder vs. single session vs. multiple session), sex, race/ethnicity, age, focus of research (e.g., social anxiety, PTSD, appearance anxiety), and whether a clinician was involved in facilitating procedures (e.g., a therapist).

To account for possible publication bias (i.e., tendency to publish significant results but not nonsignificant ones) funnel plots were generated. Funnel plots model the relationship between the sample size of a study using standard error and effect sizes. To interpret these plots, the trim and fill method and Egger’s (1997) regression tests were used. The trim and fill method imputes studies that counteract any effect sizes that significantly deviate from the mean; while Egger’ (1997) regression test uses studies’ standard errors to predict their standardized effects allowing for the observation of bias as a regression intercept with an associated p-value. Data and analytic code can be requested by contacting the corresponding author.

Results

Systematic Review

Characteristics of Studies

A full summary of the 26 included studies can be found in Table 1. The studies included in this study were conducted in four different countries: the United States (n = 20), the United Kingdom (n = 3), the Netherlands (n = 1), and Japan (n = 2). Fifteen of the studies were non-treatment experimental studies and 11 were treatment trials. Nineteen of the studies were RCTs and seven were within-group trials. With regard to the clinical symptoms of focus, eight of the studies examined social anxiety, seven examined appearance concerns, three examined eating disorder symptoms, three examined health anxiety, two examined anxiety (broadly measured), two examined obsessive-compulsive symptoms, and one study examined posttraumatic stress disorder. With regards to SBs, 10 of the studies targeted appearance-related SBs (e.g., mirror checking, social comparison, and body checking), eight targeted social SBs (e.g., social avoidance, eye contact), three examined health-related SBs (e.g., searching symptoms online), two examined general anxiety-related SBs, two examined obsessive-compulsive SBs related to checking, and one study examined posttraumatic stress SBs (e.g., avoidance of trauma-related stimuli). A total of 2,048 individuals participated in the studies. The average age was 24.90 years. A total of 26.63% of the participants identified as male, and 67.30% identified as white. Fourteen of the studies recruited participants from the community, and 12 studies recruited participants from a university setting. Thirteen of the studies included completely remote procedures, 10 were conducted at universities, and three were conducted in academic medical centers/hospital settings. All 11 of the treatment trials included symptom-related rationale for reducing SBs, nine experimental studies used deception and did not inform participants why SBs were being changed, and the remaining six studies did not clearly report whether participants were aware of the purpose of changing SBs.

Table 1.

Summary of studies included in the review (N = 26)

Citation	Total N, Age (Mean) and Gender (% female)	Study Type and Focus	Study Design and Treatment conditions	Measure of SB and Clinical Symptom	Within-group Effect Sizes	Between-group Effect Sizes
Arai, Ishikawa, et al, 2022	N = 59 M = 19.82 % NR	Treatment, decrease social anxiety SBs	Treatment RCT comparing Single session Safety Aid Elimination Intervention to a single session health lecture for social anxiety.	SB = Subtle Avoidance Frequency Examination Social anxiety = Social Anxiety Interaction Scale	SB d = 0.17 Symptom d = 0.67	SB d = 0.28 Symptom d = 0.55
Arai, Seki, et al, 2022	N = 6 M = 28.3 % 0	Treatment, decrease social anxiety SBs	Within-group study examining the effect of false safety behavior elimination therapy on social anxiety	SB = Subtle Avoidance Frequency Examination Social anxiety = Liebowitz Social Anxiety Scale	SB d = 0.26 Symptom d = 0.83
Bailey et al., 2017	N = 50 M = 35.8 % 100	Experiment, increase appearance-related SB	Experiment comparing days when instructed to body check and days when instructed not to body check	SB = Measuring wrist Symptom = Eating Disorder-15	Symptom d = 0.21
Bedford & Schmidt, 2023	N = 38 M = 20.84 % 88.3	Treatment, decrease posttraumatic SBs	Treatment RCT comparing Safety Behavior Elimination for Traumatic Stress to a physical health behavior control	SB = Posttraumatic Safety Behavior Questionnaire Symptom = Posttraumatic Stress Disorder Checklist for DSM-5	SB d = 0.31 Symptom d = 1.37	SB d = 0.31 Symptom d = 0.36
Cougle et al., 2020	N = 94 M = 32.99 % 85.1	Treatment, decrease social anxiety SBs	Treatment RCT comparing a text-message based safety behavior fading intervention to present-centered control	SB = Subtle Avoidance Frequency Examination Symptom = Social Phobia Inventory	SB d = 1.27 Symptom d = 1.66	SB d = 0.43 Symptom d = 0.41
Deacon & Mack, 2008	N = 56 M = 20.0 % 53.6	Experiment, increase contamination-related SBs	Experiment examining effect of safety behavior manipulation on contamination fears	SB = Safety behavior checklist Symptom = Padua Inventory Contamination Fear Subscale	SB d = 3.32 Symptom d = 0.61
Gorman et al., 2023	N = 130 M = 19.0 % 69.5	Treatment, decrease anxiety-related SBs	Treatment RCT comparing a SB reduction intervention to an academic skills control	SB = Safety Behavior Assessment Form Symptom = Beck Anxiety Inventory	SB d = 0.37 Symptom d = 0.13	SB d = 0.30 Symptom d = 0.04
Gray et al., 2019	N = 96 M = 20.8 % 59.4	Experiment, increase anxiety-related SBs	Experiment comparing individuals when using anxiety-related SBs to baseline, and comparing with others who are not using SBs	SB = Avoidance or impression management Symptom = Mood Thermometer	Symptom (Avoid/Not Avoid) d = 0.69 Symptom (Impression management/no impression management) d = 0.23	Symptom (Avoidance/No avoidance) d = 0.73 Symptom (Impression management/no impression management) d = 0.76
Jakes et al., 2025	N = 106 M = 35.1 % 73.6	Treatment, decrease health anxiety SBs	Treatment RCT comparing an SB reduction intervention for health anxiety to a present-centered control	SB = Safety behavior checklist Symptom = Whitely Index	SB d = 0.36 Symptom d = 0.30	SB d = 0.17 Symptom d = 0.15
Laker & Waller, 2022	N = 39 M = 20.47 % 100	Experiment, increase body comparisons	Experiment examining effect of increasing body comparisons on body satisfaction	SB = Comparison of self survey Symptom = Body Satisfaction Scale	SB d = NR Symptom d = 0.24
Olatunji et al., 2011	N = 60 M = 19.33 % 75.67	Experiment, increase health-related SBs	Experiment comparing a group instructed to increase health behaviors to a health behavior monitoring control group	SB = Safety Behavior Checklist Symptom = Short Health Anxiety Inventory	SB d = 2.82 Symptom d = 0.23	SB d = 3.05 Symptom d = 0.76
Olatunji, 2015	N = 60 M = 19.33 % 75.67	Experiment, increase health-related SBs	Experiment comparing a group instructed to increase health behaviors to a health behavior monitoring control group	SB = Health behavior checklist Symptom = Disgust Propensity	SB d = 3.29 Symptom d = 0.32	SB d = 3.57 Symptom d = 0.30
Patel & Cougle 2024a	N = 94 M = 19.44 % 100	Experiment, decrease appearance-related SBs	Experiment comparing an appearance-related SB fading condition to a self-monitoring control	SB = Appearance Behavior Checklist Symptom = Social Appearance Anxiety Scale	SB d = 0.95 Symptom d = 1.02	SB d = 0.88 Symptom d = 0.65
Patel & Cougle 2024b	N = 203 M = 36.06 % 100	Treatment, decrease appearance-related SBs	Treatment RCT comparing an appearance-related SB fading condition to an unhealthy behavior fading control	SB = Appearance Behavior Checklist Symptom = Social Appearance Anxiety Scale	SB d = 0.99 Symptom d = 1.46	SB d = 0.30 Symptom d = 0.34
Patel & Cougle 2025a	N = 39 M = 27.34 % 100	Treatment, decrease appearance-related SB	Treatment open trial investigating the effects of a 4-week appearance-related SB fading condition.	SB = Appearance Behavior Checklist Appearance concerns = Social Appearance Anxiety Scale	SB d = 1.07 Symptom d = 1.20
Patel & Cougle 2025b	N = 40 M = 24.98 % 0.00	Treatment, decrease appearance-related SB	Treatment open trial investigating the effects of a 4-week appearance-related SB fading condition.	SB = Appearance Behavior Checklist Appearance concerns = Social Appearance Anxiety Scale	SB d = 1.95 Symptom d = 1.78
Patel et al., 2024a	N = 132 M = 18.77 % 81.6	Experiment, decrease social anxiety SB	Experiment comparing 2-week social anxiety SB fading condition to a no instructions control.	SB = Subtle Avoidance Frequency Examination Loneliness = UCLA Loneliness Scale	SB d = 0.99 Symptom d = 0.50	SB d = 0.59 Symptom d = 0.20
Patel et al., 2024b	N = 94 M = 19.44 % 0.00	Experiment, decrease appearance-related SB	Experiment comparing 4-week appearance-related SB fading protocol to appearance-related SB self-monitoring control.	SB = Appearance Behavior Checklist Distress Tolerance = Distress Tolerance Inventory	SB d = 0.96 Symptom d = 0.25	SB d = 0.89 Symptom d = 0.54
Plasencia et al., 2016	N = 72 M = 32.73 % 48.6	Experiment, decrease social anxiety SB	Experiment comparing 1-hour single-session SB-free social interaction to a no instructions control.	SB = Social Behaviors Questionnaire Authenticity = Self-Experience Questionnaire	SB d = 0.79 Symptom d = 0.83	SB d = 0.49 Symptom d = 0.50
Raines et al., 2023	N = 22 M = 48.60 % 37.0	Treatment, decrease anxiety-related SB	Treatment open trial investigating the effects of an 8-session weekly group SB reduction treatment.	SB = Safety Behavior Assessment Form Anxiety = Overall Anxiety Severity and Impairment Scale	SB d = 1.50 Symptom d = 0.85
Stentz & Cougle, 2022	N = 115 M = 19.06 % 77.5	Experiment, decrease social anxiety SB	Experiment RCT comparing 2-week social anxiety SB reduction to no instructions control.	SB = Subtle Avoidance Frequency Examination Social anxiety = Social Phobia Inventory	SB d = 0.90 Symptom d = 1.37	SB d = 0.69 Symptom d = 1.28
Stentz et al., 2022	N = 84 M = 19.27 % 100	Experiment, decrease appearance-related SB	Experiment RCT comparing 2-week appearance related SB reduction to no instructions control.	SB = Appearance Behavior Checklist Bulimic symptoms = Eating Disorder Inventory - bulimic subscale	SB d = 2.48 Symptom d = 0.44	SB d = 2.49 Symptom d = 0.51
Summers & Cougle, 2018	N = 68 M = 19.30 % 100	Experiment, increase or decrease appearance-related SB	Experiment RCT comparing 1-week appearance related SB reduction to SB increase instructions control.	SB = Appearance Behavior Checklist BDD symptoms = Yale-Brown Obsessive Compulsive Scale Modified for BDD-Self-report	SB d = 1.64 Symptom d = 0.78	SB d = 1.06 Symptom d = 0.15
van Uijen & Toffolo, 2015	N = 90 M = 22.36 % 81.1	Experiment, decrease checking SB	Experiment RCT comparing 1-week checking SB increase to a no-instructions control.	SB = past-day checking checklist OCD symptoms = Checking Cognitions Scale	SB d = 2.91 Symptom d = 0.28	SB d = 3.47 Symptom d = 0.75
Wilver et al., 2020	N = 84 M = NR % 100	Experiment, decrease appearance-related SB	Experiment RCT comparing 2-week daily SB fading to no-instructions control.	SB = Appearance Behavior Checklist BDD symptoms = Yale-Brown Obsessive Compulsive Scale Modified for BDD-Self-report	SB d = 2.48 Symptom d = 1.79	SB d = 2.49 Symptom d = 1.07
Zech et al., 2025	N = 201 M = 28.96 % 89.6	Treatment, decrease social anxiety SB	Treatment RCT comparing four-week daily SB fading to daily unhealthy behavior fading control.	SB = Subtle Avoidance Frequency Examination Social anxiety = Social Phobia Inventory	SB d = 1.33 Symptom d = 1.64	SB d = 0. 36 Symptom d = 0.26

Note: NR = not reported; NA = not applicable RCT = Randomized Controlled Trial; SB = safety behavior; BDD = body dysmorphic disorder; d = Cohen’s d effect size; If effect sizes were not reported, Cohen’s d were calculated.

With regard to SB manipulations, 13 studies involved sending daily reminders via a digital intervention to reduce SBs, seven involved multi-session manipulations with a clinician (e.g., five 50-minute therapy sessions), five involved a single-session treatment or manipulation procedures with a clinician, and one involved a group therapy setting lasting for eight sessions. The duration of SB manipulations varied from 10 minutes to eight weeks. Specifically, 10 studies manipulated SBs over the course of one month, four over two weeks, three over one week, two over three weeks, and one for 10 minutes, one hour, two hours, two days, three days, 5 weeks, and 8 weeks, respectively. With regard to control conditions, seven were either waitlist (in treatment studies) or no instruction controls (in non-treatment experiments); five were self-monitoring conditions occurring within non-treatment contexts, where individuals were asked to monitor their SB use daily but not change their SB use; two of the studies used a control condition asking participants to reduce unhealthy behaviors for one month; two involved a single-session lecture on health habits; two studies used a present-centered control; and one involved increasing studying habits. Interestingly, only five studies involved a control condition that included contact with a clinician.

Within-Group Effects

Nineteen studies examined the within-group effect of decreasing SBs. Notably, two studies examined targeting SB in two separate samples resulting in a total of 21 potential effect sizes. Of these, 20 (95.2%) reported an effect size on reducing SBs. Across these studies, effect sizes ranges from 0.17 to 2.48. The mean effect size was d = 1.08 (95% CI [0.78, 1.38]). With regard to clinical symptoms, all 21 effect sizes were reported ranging from 0.13 to 2.06 with a mean effect size of d = 0.99 (95% CI [0.78, 1.20]). Importantly, of these 21 effect sizes, 12 came from treatment studies and nine came from non-treatment experimental studies. For treatment studies, the mean effect was d = 0.88 (95% CI [0.51, 1.25]) for SBs and d = 1.18 (95% CI [0.83, 1.53]) for primary symptoms. For non-treatment studies, the mean effect was d = 1.36 (95% CI [0.84, 1.88]) for SBs and d = 0.82 (95% CI [0.43, 1.21]) for primary symptoms.

With regard to SB increase studies, eight experimental studies examined within-group effects. Of these, one study reported on two separate samples with a total of nine effect sizes. Of these, the within-group effect on SB was only reported in five studies. The effect sizes ranged from 1.72 to 3.32 with a mean of d = 2.81 (95% CI [2.00, 3.62]). With regard to symptoms, all nine studies reported effect sizes that ranged from 0.21 to 0.69. The mean effect size was d = 0.35 (95% CI [0.22, 0.48]).

Risk of Bias

A full summary of risk of bias can be found in Supplemental Table 2 and 3. Briefly, 13 RCT studies were deemed to have some concerns for risk of bias, and six studies were determined to have low bias. With regard to nonrandomized studies, all seven studies were determined to have moderate risk of bias due to moderate risk of uncontrolled confounds.

Meta-Analyses

Pooled Effect of Decreasing Safety Behaviors on SB Frequency

First, we sought to examine the pooled effect of manipulations on reducing SB relative to a control condition. A total of 16 (N = 1,624) studies that examined SB reduction relative to a control, were initially planned to be included, but we found that two studies (Stentz et al., 2022; and Wilver et al., 2020) were drawn from the same sample examining different clinical symptoms. Thus, we only included the original effect on SB in our meta-analysis resulting in 16 total studies in our initial analysis. Summary of the effects of the 16 studies can be found in Supplemental Figure 1. We found that SB reduction manipulations/treatments outperformed control conditions, showing a significant medium effect in reducing SB use (mean Cohen’s d = 0.75, 95% CI [0.35, 1.15], t = 4.02, p =.001). The effect was significantly heterogenous (Q (14) = 105.36, p < .001). The standard deviation of the true effect was 0.66, 95% CI [0.46, 1.13], and I² indicated high heterogeneity with 86.7%, 95% CI [79.7%, 91.3%], of the variance in the observed effects reflecting variance in the true effect.

Prior to examining moderators, we first examined the potential for any outlier effects (Viechtbauer & Cheung, 2010) influencing the pooled effect, given high heterogeneity. Importantly, we found one outlier effect (Wilver et al., 2020). A summary of this analysis can be found in Figure 2. We again found that SB reduction manipulations/treatments outperformed control conditions, showing a significant medium effect (mean Cohen’s d = 0.52, 95% CI [0.36, 0.68], t = 7.05, p <.001). With the removal of this effect, the pooled effect was no longer significantly heterogenous (Q (13) = 21.00, p =.073). The standard deviation of the true effect was 0.17, 95% CI [0.00, 0.39], and I² indicated low heterogeneity with 38.1%, 95% CI [0.00%, 67.2%] of the variance in the observed effects reflecting variance in the true effect. Given the low heterogeneity, we did not investigate all possible moderators. Rather, we only examined whether study type (treatment vs. non-treatment) moderated the effect, consistent with our hypotheses. Of note, we found that study type was a significant moderator (B0 = .76, CI [.64, .88], p <.001; B = −0.45, CI [−0.62, −0.29], p < .001, R² analogue = 1.00), suggesting that the pooled effect for non-treatment experiments is .76, and the effect for treatments is .31.

Figure 2.

Forest plot of Cohen’s d effect size observing between-group effects of safety behavior reduction once removing outlier effect. Higher value indicate lower safety behavior use relative to control.

The funnel plot for this analysis showed slight indication of publication bias (Supplemental Figure 2). Eger’s regression test indicated a nonsignificant potential for funnel plot asymmetry (B0 = 2.68, CI [−0.64, 6.00], t(12) = 1.58, p = .139). Trim and fill analysis showed evidence of two missing studies with effects below the mean effect. Incorporating these studies resulted in an imputed mean effect of d = 0.46, CI [0.28, 0.63] suggesting that the effect in the present analysis (i.e., 0.52) is overestimated and the true effect may be slightly lower than what was found. Taken together, publication bias and small-study effects cannot be ruled out in the present findings.

Pooled Effect of Decreasing Safety Behaviors on Clinical Symptoms

Next, we examined the pooled effect of reducing SB on symptoms (k = 16). Summary of the effects of the 16 studies can be found in Figure 3. We found that there were no outlier effects. We found that SB reduction manipulations/treatments outperformed control conditions, showing a significant small-to-medium effect (mean Cohen’s d = 0.43, 95% CI [0.27, 0.59], t = 5.64, p <.001). The pooled effect was significantly heterogenous (Q (15) = 27.69, p =.024). The standard deviation of the true effect was 0.19, 95% CI [0.02, 0.44], and I² indicated low heterogeneity with 45.8%, 95% CI [2.8%, 69.8%], of the variance in the observed effects reflecting variance in the true effect. In observing moderators, no variable moderated the pooled effect. However, given these effects sizes were in response to a manipulation of SB, we examined SB effect size as a moderator of effects. We found that between-group SB effect size was a significant moderator (B0 = .24, CI [.04, .44], p =.023; B = 0.23, CI [0.03, 0.43], p = .029, R² analogue = 0.85), suggesting that for every 1-point increase in SB effect size, the effect size for clinical symptoms increased by 0.23.

Figure 3.

Forest plot of Cohen’s d effect size observing between-group effects of safety behavior reduction on primary symptoms. Higher value indicate lower clinical symptoms relative to control.

The funnel plot for this analysis showed indication of publication bias (Supplemental Figure 3). Eger’s regression test indicated a significant potential for funnel plot asymmetry (B0 = 4.38, CI [1.40, 7.35], t(14) = 2.88, p = .012). Trim and fill analysis showed evidence of four missing studies with effects below the mean effect. Incorporating these studies resulted in an imputed mean effect of d = 0.31, CI [0.12, 0.51], suggesting that the effect in the present analysis (i.e., .43) is overestimated and the true effect may be slightly lower than what was found. Taken together, publication bias and small-study effects cannot be ruled out in the present findings.

Pooled Effect of Increasing Safety Behaviors

Next, we examined the effect of increasing SBs relative to a control condition. Five studies were included in these analyses; however one study (Gray et al., 2019) did not assess SBs using pre-post procedures, and another study had two control conditions (van Uijen & Toffolo, 2015). This resulted in five effects (N = 368) in the present analysis. Summary of the effects of the five studies can be found in Figure 4. We found that SB increase manipulations outperformed control conditions, showing a significant large effect (mean Cohen’s d = 2.96, 95% CI [2.14, 3.78], t = 10.00, p <.001). The effect was significantly heterogenous (Q (4) = 14.97, p = .005). The standard deviation of the true effect was 0.57, 95% CI [0.19, 1.82], and I² indicated high heterogeneity with 73.3%, 95% CI [29.4%, 75.8%], of the variance in the observed effects reflecting variance in the true effect. Given the low number of effects, we did not conduct any moderator analyses due to being underpowered to perform these analyses. The funnel plot for this analysis showed indication of publication bias (Supplemental Figure 5), but we were unable to conduct Eger’s regression test or trim and fill analyses. Taken together, this analysis should be interpreted with caution as publication bias and small study effects likely biased the estimate.

Figure 4.

Forest plot of Cohen’s d effect size observing between-group effects of safety behavior increase. Higher value indicate higher safety behavior use relative to control.

Pooled Effect of Increasing Safety Behaviors on Clinical Symptoms

Next, we examined the effect of increasing SBs on symptoms relative to a control condition. Five studies with seven effect sizes were included in this analysis. Summary of the effects of the five studies can be found in Figure 5. We found that SB increase manipulations outperformed control conditions, showing a significant moderate effect (mean Cohen’s d = 65, 95% CI [0.46, 0.83], t = 8.54, p <.001). The effect was not significantly heterogenous (Q (6) = 3.30, p = .770). The standard deviation of the true effect was 0, 95% CI [0.00, 0.34], and I² indicated no heterogeneity with 0%, 95% CI [0.0%, 70.8%], of the variance in the observed effects reflecting variance in the true effect. Given the low number of effects and no heterogeneity, we did not conduct any moderator analyses. The funnel plot for this analysis did not indicate the possibility of publication bias (Supplemental Figure 6), but we were unable to conduct Eger’s regression test or trim and fill analyses due to a low number of studies. Taken together, this meta-analysis should be interpreted with caution. While some metrics indicate a lack of heterogeneity, publication bias may have biased the effects.

Figure 5.

Forest plot of Cohen’s d effect size observing between-group effects of safety behavior increase on primary. Higher value indicate higher clinical symptoms relative to control.

Discussion

The aim of the present systematic review and meta-analysis was to examine the general efficacy of stand-alone SB manipulations (i.e., treatments or non-treatment experiments) on SBs and symptoms of psychopathology. To accomplish this goal, we reviewed RCTs and single group trials that targeted SBs and measured the effect of this manipulation on clinical symptoms before and after the manipulation. A total of 26 relevant studies met our inclusion criteria for the systematic review, 19 of which were included in our meta-analyses.

Concerning within-group effects, our findings suggest that SB reduction protocols can lead to reductions in SB frequency, both in treatment and non-treatment contexts, and that these reductions are mirrored by reductions in clinical symptoms. We found large within-group effect sizes for both SB frequency (d = 1.08) and clinical symptoms (d = 1.03), which were comparable between treatment and non-treatment experimental studies. Moreover, the mean impact of protocols calling for increased SB use was even larger (d = 2.81), though these increases in SB frequency were matched by relatively small increases in clinical symptoms (d = 0.35). Taken together, these findings provide preliminary evidence that standalone SB reduction/elimination may have a salutary impact on clinical symptoms, and their increase leads to worse symptomatology. It remains unclear why clinical symptoms were less responsive to SB increase studies versus SB reduction/elimination studies. One possible explanation could be due to the fact that these SB increase studies were examined in healthy college samples rather than symptom-elevated or clinical samples that were included in the SB decrease studies. Therefore, individuals in decrease studies may have had more room for clinical improvement, whereas individuals in increase studies quickly reached a peak level of symptoms that could be contributed by SB use alone within a short period of time.

Regarding between group-effects, we found that the effect of reducing SBs on these behaviors was moderate in size (d = 0.52), suggesting that relative to control, these manipulations were successful in reducing the targeted mechanisms. Though there was not a significant amount of heterogeneity around this effect size, we did find that study type moderated the little heterogeneity around the effect size. Specifically, experimental manipulation saw larger effect sizes (d = 0.76) compared to treatment trials (d = 0.31). One possible reason for these findings could be that many of the treatment trials were examined in community-based treatment-seeking samples, whereas the experimental non-treatment studies were mostly among college students. It is possible that demand effects may have influenced the experimental studies. It is also possible that the control groups that were used in the non-treatment studies were weaker compared to the treatment trials. For example, several studies (Plasencia et al., 2016; Stentz & Cougle, 2022; Wilver et al., 2020) used no-instruction control groups in the non-treatment trials; whereas many of the treatment studies used an active control condition that appeared to indirectly lead to reductions in SBs (Patel et al., 2024b; Zech et al., 2025).

We also found that SB reduction manipulations led to greater reductions in clinical symptoms compared to control groups across 16 studies, resulting in a small-to-medium effect size (d = 0.43). These findings may suggest that the standalone effect of targeting SBs can lead to significant improvement in symptoms, but this improvement is slightly smaller than what might be seen in multi-component treatment approaches that target multiple mechanisms. For example, a recent meta-analysis of CBT for anxiety disorders found a medium between-group effect size (g = 0.51) compared to active controls for anxiety disorders (Hofmann et al., 2025). As hypothesized, we did find that the between-group effects on SB frequency moderated the effects on clinical symptoms, such that a one-point increase in the effect size of SB reduction corresponded to a 0.23 increase in the between-group effect on symptoms.

Taken together, the current findings allow for cautious optimism in the state of the standalone SB treatment literature. With the advent of digital interventions that often target a single mechanism, the present findings are consistent with the notion that targeting SBs via such an approach (i.e., SB fading; Cougle et al., 2020) could be fruitful in delivering efficacious treatments to more people who need them. Future research should continue to examine the efficacy of this approach in different conditions as well as examine how standalone SB interventions could be improved to maximize efficacy.

One thing future research will need to consider is the potential role expectancy effects may play within SB interventions. There is strong evidence demonstrating that expectancy is strongly related to symptom change within psychological treatment studies (Dear, 2025; Southworth & Kirsch, 1988). Eleven of the included treatment trials included explicit rationale for reducing SBs and how this would be associated with symptom reduction. It is possible that the effects seen in these studies may be in part due to expectancy effects. Further, the difference in effect size seen between decrease and increase studies may be explained by these expectancy effects. Future research will want to consider the role of expectancy effects and design studies that mask or modify these expectations to better examine the mechanistic role SB play in psychopathology.

We also evaluated five studies that have examined the effects of increasing SBs. All included studies occurred in experimental non-treatment contexts seeking to examine the causal relationships of these behaviors on symptoms. We found that the between-group effect of these manipulations were large (d = 2.96), but this was not surprising, as the control conditions within these trials consisted of no instructions or benign manipulations (e.g., increase studying behaviors; Summer & Cougle, 2018). We found that the pooled between-group effect size of these manipulations on symptoms were medium in size (d = 0.65), providing further support that increasing these behaviors lead to greater psychopathology. While SBs have long been considered to be maintenance factors of psychopathology (Wellz et al., 1995), the present study aligns with this theory. However, these findings should be interpreted with caution as there were a relatively low number of studies in both analyses, preventing the ability to examine heterogeneity and publication bias.

It is important to highlight several limitations with the present review and meta-analyses. First, while there was no significant evidence of publication bias on the between-group effect of SB reduction on SBs, there was evidence to suggest that publication bias may have impacted the effect size for clinical symptoms. Future research will benefit from preregistration and subsequent publication of findings, even if they are null. Second, we were unable to examine publication bias in the SB increase studies. Third, there was evidence that there is some risk of bias in most of the studies included in the present review. Fourth, we were unable to conduct meta-analyses on within-group effect sizes due to issues of reliably calculating standard errors and confidence intervals (Cuijpers et al., 2017). Future studies, whether they are RCTs or single-group trials, would benefit with including the correlation of measures at both baseline and post. Fifth, most of the participants included in this review and meta-analysis were young women. Future studies should continue to examine the efficacy of reducing SBs in more diverse samples. Sixth, there was heterogeneity in both the type of SBs and clinical symptoms that were included within the studies in our meta-analyses. This fact potentially biases our meta-estimates. While the literature on standalone manipulations of SBs is still nascent, future meta-analyses will be necessary to evaluate the state of this literature as it continues to grow. It is possible that with new and more representative studies within various subfields (e.g., social anxiety vs. appearance concerns) greater heterogeneity will arise, signaling the importance of future meta-analyses to characterize effect sizes. Additionally, it is important to note that despite the overall review including 26 studies, our meta-analyses only included subsets of these studies. Thus, the present meta-analyses were limited by a relatively low number of studies and participants. While our goal was to summarize the current state of the literature and to provide insight into the effects of SB manipulations, our findings should be interpreted with caution given the early stage of this literature and the preliminary nature of our meta-analyses. As this literature continues to develop, future meta-analyses will benefit from the increase in available studies and participants. Finally, there was one study that was not included in analyses due to being a significant outlier (Wilver et al., 2020). Wilver and colleagues (2020) found a much larger between-group effect on SBs than was seen by other studies, but this may have been due to their use of a no-instruction control group coupled with a strong within-group effect for the SB reduction condition.

Clinical Implications and Future Directions

Despite the limitations, the present study yields several key implications for future research and practice. The present findings lend support to the growing literature indicating the efficacy of directly targeting SBs as a means to reduce clinical symptoms of anxiety-related psychopathology. Interestingly, a majority of the studies did not involve a clinician to facilitate procedures. Of the 11 treatment studies included in this review, only four included clinicians. With the advent of digital interventions, future studies should continue to examine the efficacy of standalone SB reduction as a means to improve symptoms of anxiety or fear-related disorders. Further, these interventions should be evaluated in additional conditions. While general anxiety, social anxiety, and appearance anxiety are well represented in the literature included in this study, more work is needed to examine and understand the efficacy of SB reduction for other conditions, such as obsessive-compulsive and posttraumatic stress disorder.

Future research should continue to examine SB manipulation in experimental (non-treatment) contexts. Our findings suggest that these methods may lead to greater SB reduction (and increase). The approach of using deception to facilitate SB manipulation compared to an active control could prove an excellent means of testing the causal role of SBs in psychopathology. For example, it has recently been argued that experimental procedures that occur outside of the lab may prove to be especially useful in theory testing (Cougle, 2025). In fact, several of the experimental studies included in this review follow similar procedures as described by Cougle (2025), where individuals are instructed to reduce SB without any treatment rationale and compared to a control group such as self-monitoring.

Future research should also examine how to best facilitate SB reduction within a treatment context, given our moderation finding (i.e., increase in SB reduction leads to increase in symptom reduction). While many of the studies included in this review used a protocol consisting of asking individuals to reduce SB with daily reminders, this approach could be further improved as a standalone approach. The inclusion of summary tracking of daily behaviors could be helpful to monitor progress for the participant. Further, more comprehensive psychoeducation could prove to be helpful in facilitating adherence and motivation to reduce SBs, especially given the finding that experimental studies were better at reducing SBs compared to treatments. Many of the non-clinician treatment trials consisted of a brief one paragraph psychoeducation followed by instructions to reduce SB. This approach could be improved with interactive components and psychoeducational videos that can be revisited.

Conclusion

The present systematic review and meta-analysis aimed to describe and quantify the effects of standalone SB manipulations. Our findings lend support for the role SBs play in anxiety-related psychopathology and suggest that SB reduction may be an efficacious way to reduce symptoms of psychopathology. Importantly, our review highlights the need for more studies in examining this promising approach. While the number of studies examining this approach have been growing in the past decade, more research is needed to better understand the efficacy of targeting SBs across different conditions. Future research should continue to explore both non-treatment experimental and treatment methods to robustly understand the efficacy of targeting SBs and their role in maintaining psychopathology.