Effects of Graded Exposure on Functional Improvement, Quality of Life, and Mental Health in Chronic Musculoskeletal Pain: A Systematic Review and Meta-Analysis

Abstract

Background

Chronic musculoskeletal pain (CMP) is a prevalent and disabling condition associated with impaired physical function and reduced quality of life. Graded exposure therapy (GEP), a cognitive-behavioral intervention targeting pain-related fear and avoidance, has been proposed as a non-pharmacological approach for CMP management.

Objective

To evaluate the effects of GEP compared with non-GEP interventions in individuals with CMP.

Methods

We searched electronic databases from inception to February 2025 for randomized controlled trials (RCTs) comparing GEP and non-GEP interventions in individuals with CMP. Primary outcomes included pain intensity and pain-related disability. Secondary outcomes included pain-related fear and cognition, health-related quality of life, psychological outcomes, sleep-related outcomes, behavioral adaptation, self-efficacy, and physical functioning. The certainty of evidence was assessed using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach.

Results

Eight RCTs involving 554 participants were included. Compared with non-GEP interventions, GEP was associated with improvements in pain-related disability and pain-related fear and cognition, including pain catastrophizing (Pain Catastrophizing Scale [PCS]: mean difference [MD] = −4.45, 95% confidence interval [CI] −7.20 to −1.69) and kinesiophobia (Tampa Scale of Kinesiophobia [TSK]: MD = −4.57, 95% CI −7.11 to −2.03). Disability outcomes have showen beneficial effects across several instruments, although many estimates are based on individual studies. Effects on pain intensity were inconsistent, with significant reductions observed for Neuropathic Pain Scale (NPS) (MD = −2.66, 95% CI −4.09 to −1.23) and Fibromyalgia Impact Questionnaire Pain subscale (FIQ-Pain) (MD = −2.51, 95% CI −3.48 to −1.54). The certainty of evidence ranged from very low to moderate, with most outcomes being downgraded owing to the risk of bias and imprecision.

Conclusion

GEP may improve pain-related disability and pain-related fear and cognition in individuals with CMP. However, the limited certainty of evidence and variability across populations, interventions, and outcome measures warrant cautious interpretation. Further high-quality trials with standardized outcomes and longer-term follow-up are needed.

Background

Chronic musculoskeletal pain (CMP) refers to pain involving musculoskeletal structures that persists for more than three months, represents a substantial global health challenge with profound effects on individuals and healthcare systems.^1,2 Musculoskeletal pain conditions, including low back pain and neck pain, contribute substantially to disability worldwide and are associated with reduced physical function, impaired quality of life, and substantial healthcare burden.^3,4 According to the Global Burden of Disease Study 2021, low back pain, one of the most prevalent musculoskeletal pain conditions, affected 619 million people globally in 2020, with projections indicating 843 million prevalent cases by 2050.³ Beyond its clinical burden, chronic pain also imposes considerable economic costs, with the annual economic burden in the United States estimated at $560–635 billion due to medical expenditures and productivity losses.¹

CMP is not only a physical condition but also a biopsychosocial phenomenon, involving biological, psychological, and social factors that influence pain, disability, and treatment outcomes.^1,2 Psychological factors, including depression, anxiety, and pain-related fear, are commonly reported among individuals with CMP and may contribute to functional limitations and poorer treatment outcomes.⁵ This complex nature has shifted treatment approaches from purely biomedical models toward integrated, multidisciplinary interventions, with psychological therapies increasingly being recognized as important components of CMP management.^6,7

Among psychological and behavioral interventions, graded exposure therapy (GEP) has emerged as a promising approach for managing CMP. GEP targets the fear-avoidance cycle, in which pain-related fear leads to avoidance of movements or activities, contributing to disability and functional limitations.^8,9 It involves a structured and hierarchical exposure process to feared movements or activities, aiming to reduce catastrophic beliefs and fear responses through corrective learning.¹⁰ Unlike graded activity, which primarily focuses on progressive increases in activity levels according to predefined goals, GEP specifically targets pain-related fear and avoidance behaviors through exposure-based procedures.¹¹ Although GEP shares behavioral principles with cognitive behavioral therapy (CBT), CBT represents a broader psychological intervention framework that incorporates multiple cognitive and behavioral strategies, whereas GEP primarily emphasizes exposure to feared activities.^12,13

Despite the growing interest and theoretical support, the comparative effectiveness of GEP relative to other behavioral interventions remains uncertain. Previous randomized controlled trials (RCTs) have compared GEP with active and inactive interventions, including graded activity, CBT, and usual care, but findings have been inconsistent. For example, de Jong et al reported greater improvements with GEP than with graded activity in patients with chronic low back pain and high pain-related fear; however, these differences were not consistently maintained at long-term follow-up.¹⁴ Other trials have compared GEP with interventions such as CBT or usual care, but the findings have not established whether GEP provides additional benefits over these approaches. Psychological interventions such as CBT may overlap with GEP in addressing pain-related beliefs and behaviors, but CBT incorporates broader cognitive and behavioral strategies rather than systematically targeting fear avoidance through exposure. Trost et al also highlighted variability in treatment responses and challenges in optimizing exposure-based interventions.¹⁵ Recent reviews have reported substantial heterogeneity in study protocols, patient populations, interventions, and outcome measures, limiting conclusions regarding the comparative effectiveness of GEP across different intervention types.^12,13 To our knowledge, no meta-analysis has specifically evaluated the effectiveness of GEP compared with non-GEP interventions in individuals with chronic musculoskeletal pain, including both adults and adolescents. Previous reviews have examined exposure-based therapies broadly¹¹ but did not specifically distinguish GEP from other interventions, such as graded activity, CBT, and inactive comparators, or systematically explore potential sources of heterogeneity. Therefore, whether GEP provides additional benefits beyond other interventions remains unclear and warrants further investigation.

This systematic review and meta-analysis aimed to evaluate the effectiveness of GEP compared to non-GEP interventions in individuals with CMP. We synthesized evidence from RCTs and assessed the effects on pain intensity, disability, pain-related fear, and other patient-reported outcomes. Potential sources of heterogeneity, including pain condition, participant age, mode of delivery, and type of comparator, were also explored.

Methods

Study Design and Registration

This systematic review and meta-analysis was conducted and reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement.¹⁶ The review protocol was registered retrospectively in PROSPERO (CRD420261469624).

Search Strategy

Four electronic databases, including PubMed, the Cochrane Library, Embase, and PsycINFO, were systematically searched from inception to 26 February 2025. The search strategy combined MeSH terms and free-text terms based on the following concepts: “chronic musculoskeletal pain”, “graded exposure”, “GEP”, “randomized controlled trial”, and “randomized”. The search strategy was adapted to the syntactic requirements of each database. Endnote X9 was used for literature management and screening. The detailed search strategy is provided in Supplementary Method 1.

The search strategy was adapted to the syntactic requirements of each database. Additionally, we hand-searched reference lists of included studies and relevant systematic reviews to identify additional eligible studies.

Study Selection

Two reviewers (Qiong Zhang and Yapeng Wu) independently screened titles, abstracts, and full-text articles according to predefined eligibility criteria. Disagreements were resolved through discussion or consultation with a third reviewer (Chao Zhang). The study selection process followed a two-stage approach: an initial screening of titles and abstracts, followed by a full-text review of potentially eligible studies.

Inclusion and Exclusion Criteria

Inclusion Criteria

Studies were included if they met the following eligibility criteria: (1) Participants: individuals with CMP or chronic pain conditions with substantial musculoskeletal involvement, with a pain duration of ≥3 months. (2) Interventions: Studies that explicitly applied GEP principles or graded exposure protocols, including graded exposure combined with standardized treatments such as physical therapy and pain education. GEP was defined as a systematic behavioral intervention with the following essential components: (a) identification of feared activities or movements; (b) hierarchical exposure beginning with activities eliciting lower levels of fear; and (c) progression based on reductions in fear or anxiety responses. (3) Comparisons: Studies comparing GEP with non-GEP interventions were included. Non-GEP comparators were categorized into two groups: (a) conventional control groups, including standard care, cognitive behavioral therapy (CBT), graded activity (GA), and other exercise-based interventions; and (b) no-intervention control groups, including waitlist controls or no-treatment conditions. (4) Outcomes: Studies reporting the outcomes of interest, included pain intensity, pain-related disability, pain-related fear, quality of life, sleep quality, pain-related anxiety and depression, pain-related distress, pain self-efficacy, coping strategies, and daily activity levels. Outcomes were assessed at short-term follow-up (≤6 months).¹⁷ (5) Study design: Randomized controlled trials published in peer-reviewed journals.

Exclusion Criteria

Studies were excluded if they met any of the following eligibility criteria: (1) cancer-related pain, including pain caused by tumor progression, cancer treatment (chemotherapy or radiotherapy), or cancer-related procedures; (2) pain secondary to acute injuries, surgery, or diseases of non-musculoskeletal organs and tissues; and (3) Studies were excluded when the independent effect of GEP could not be determined.

Outcome Measures

Primary Outcomes

The primary outcomes were pain intensity and pain-related disability. Detailed definitions of outcome measures and assessment procedures are provided in Supplementary Method 2.

(1) Pain intensity was assessed using validated instruments, including generic pain rating scales, such as the Visual Analog Scale (VAS),¹⁸ and Numerical Rating Scale (NRS);¹⁹ multidimensional pain assessment tools, including the Brief Pain Inventory (BPI),²⁰ McGill Pain Questionnaire (MPQ),²¹ and McGill Pain Questionnaire–Short Form (SF-MPQ);²² and condition-specific instruments, including the Fibromyalgia Impact Questionnaire Pain subscale (FIQ-Pain),²³ Neuropathic Pain Scale (NPS)²⁴ and the pain intensity item of the Örebro Musculoskeletal Pain Questionnaire (ÖMPQ).²⁵

(2) Pain-related disability was assessed using validated instruments, including generic disability measures, such as the Pain Disability Index (PDI),²⁶ World Health Organization Disability Assessment Schedule 2.0, 12-item version (WHODAS 2.0–12),²⁷ and Functional Disability Inventory (FDI);²⁸ condition-specific disability questionnaires, including the Oswestry Disability Index(ODI)²⁹ and Quebec Back Pain Disability Scale (QBPDS);³⁰ and domain-specific functional assessments, including the Radboud Skills Questionnaire (RASQ)³¹ and Walking Ability Questionnaire (WAQ).^32,33

Secondary Outcomes

Secondary outcomes encompassed several domains, including pain-related fear and cognition, health-related quality of life, sleep-related outcomes, psychological outcomes, behavioral adaptation and self-efficacy, and physical functioning and activity levels.

(1) Pain-related fear and cognition were assessed using validated instruments covering five related domains based on the fear-avoidance model: (a) pain catastrophizing: Pain Catastrophizing Scale (PCS)³⁴ and Pain Catastrophizing Scale-Child Version (PCS-C);³⁵ (b) kinesiophobia: Tampa Scale of Kinesiophobia (TSK);³⁶ (c) fear-avoidance beliefs: Fear-Avoidance Beliefs Questionnaire (FABQ),³⁷ FABQ-Physical Activity subscale (FABQ-PA),³⁷ and Psychological Inflexibility in Pain Scale (PIPS);³⁸ (d) fear of pain: Fear of Pain Questionnaire-Child Version (FOPQ-C);³⁹ and (e) perceived harmfulness: Photograph Series of Daily Activities (PHODA)⁴⁰ and Photograph Series of Daily Activities-Youth Edition (PHODA-YE).⁴¹

(2) Health-related quality of life and sleep-related outcomes were assessed using validated instruments, including the 36-Item Short Form Health Survey (SF-36)⁴² and Brunnsviken Brief Quality of Life Scale (BBQ)⁴³ for health-related quality of lifeand the Insomnia Severity Index (ISI)⁴⁴ for insomnia severity.

(3) Psychological outcomes were assessed using validated instruments targeting anxiety and depression. Pain-related anxiety was assessed using the Pain Anxiety Symptoms Scale (PASS-20),⁴⁵ whereas generalized anxiety symptoms were assessed using the 7-item Generalized Anxiety Disorder Scale (GAD-7).⁴⁶ Depressive symptoms were assessed using the Patient Health Questionnaire-9 (PHQ-9).⁴⁷ The Hospital Anxiety and Depression Scale (HADS)⁴⁸ was used to assess symptoms of both anxiety and depression.

(4) Behavioral adaptation and self-efficacy were assessed using validated instruments measuring self-efficacy and coping resources, including the Pain Self-Efficacy Questionnaire (PSEQ)⁴⁹ and Pain Processing and Coping Questionnaire (FESV-BW).⁵⁰

(5) Physical functioning and activity levels were assessed using complementary objective and subjective measures, including the Tri-axial Research Tracker (RT3)⁵¹ accelerometer for quantitative assessment of daily physical activity levels and the Activities of Daily Living Scale (ADL scale)⁵² for evaluation of self-reported functional performance in routine activities.

Data Extraction

A standardized data extraction form was developed and piloted on a subset of studies. Two reviewers independently extracted the data. Disagreements were resolved through discussion. If consensus could not be reached, a third reviewer was planned to adjudicate. Data extraction was performed using a standardized form capturing the following variables: study characteristics (first author, publication year, country of origin, study design); participant demographics (sample size, mean age, sex distribution, pain condition, pain duration); intervention characteristics (graded exposure protocol, treatment duration, session frequency, and concurrent interventions); comparator details; outcome measurement instruments and assessment time points; and outcome data (means, standard deviations, effect sizes, and confidence intervals [CI]). For multi-arm trials, comparisons between the GEP and each eligible non-GEP comparator were extracted separately when different comparator types were reported. When multiple assessment time points were available, the longest available follow-up assessment was selected for meta-analysis.

Risk of Bias Assessment

Two reviewers independently assessed the risk of bias for each included study using the Cochrane Risk of Bias version 2.0 tool (RoB 2.0).⁵³ Five domains were evaluated: bias arising from the randomization process, bias due to deviations from intended interventions, bias due to missing outcome data, bias in measurement of the outcome, and bias in selection of the reported result. Each domain was rated as “low risk of bias”, “some concerns”, or “high risk of bias”. Disagreements were resolved through discussion or consultation with a third reviewer. Risk of bias assessments were presented using traffic light plots and summary figures.

Statistical Analysis and Certainty Assessment

For continuous outcomes, mean differences (MDs) with 95% confidence intervals (CIs) were calculated when the same measurement instrument was used across studies. Effect sizes were interpreted according to Cohen’s criteria: small (0.2), moderate (0.5), and large (0.8) effects. Statistical heterogeneity was assessed using the I² statistic. I² values of 25%, 50%, and 75% represented low, moderate, and high heterogeneity, respectively. A random-effects model was used for all analyses to account for expected heterogeneity between studies. Subgroup analyses were performed based on clinically relevant factors, including type of pain condition, age group, mode of delivery, and type of control group. Publication bias was assessed using funnel plots when ≥10 studies were available for a specific outcome; however, no outcome met this criterion, and funnel plots were therefore not performed.⁵⁴ Statistical analyses were performed using Review Manager 5.4 (RevMan 5.4, The Cochrane Collaboration, Copenhagen, Denmark).

Certainty of Evidence

The certainty of evidence for the primary outcomes (pain intensity and pain-related disability) was assessed using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach. Evidence from RCTs was initially rated as high certainty and was downgraded based on risk of bias, inconsistency, indirectness, imprecision, and publication bias when applicable. The certainty of evidence was categorized as high, moderate, low, or very low.⁵⁵ Clinical relevance was evaluated by comparing observed treatment effects with available minimal important difference (MID) thresholds. MID values were obtained from published literature, including recommendations for interpreting clinical importance in chronic pain trials.^56,57 Sensitivity analyses were performed when sufficient studies were available to assess the robustness of pooled estimates. Leave-one-out analyses were conducted by sequentially removing each study and recalculating the pooled estimate to evaluate the stability of the findings.

Results

Study Selection

A systematic search identified 6,728 records from the four databases. After removal of 280 duplicates, 6,448 records remained for screening. Among these, 67 records were identified as ineligible by automated screening tools and were removed after manual confirmation, leaving 6,381 records for title and abstract screening. After screening the titles and abstracts, 65 full-text articles were selected for eligibility. Ultimately, 8 studies^58–65 met the inclusion criteria and were included in this systematic review and meta-analysis. The study selection process is illustrated in Figure 1.

Figure 1.

PRISMA flow diagram.

Study Characteristics

Eight studies,^58–65 comprising 554 participants from six countries (Canada, the Netherlands, Sweden, Germany, Spain and the USA), were included in the final synthesis. Sweden contributed the largest number of studies (n = 3). The sample sizes ranged from 34⁶¹ to 140.⁵⁹ Overall, 56.7% of participants were female, and the mean ages of participants ranged from 14.1 to 53.5 years. One study included adolescent participants,⁶² whereas seven studies focused on adult populations.^{58–61,63–65} Four studies investigated participants with low back pain or spinal pain,^58,60,61,63 while four studies examined other CMP conditions.^59,62,64,65 GEP was delivered face-to-face in seven studies,^58,60–65 whereas one study⁵⁹ delivered GEP through an internet-based platform. Among the included RCTs, seven comparisons involved active control conditions,^58,60–65 whereas three comparisons involved inactive controls, including waitlists and no-treatment controls.^59,63,64 The number of treatment sessions ranged from 6 to 17. The characteristics of all included studies are summarized in Table 1.

Table 1.

Characteristics of Included Studies

Studies	Country	Sample Size	Participants	Age (Years)	Gender	Interventions	Outcomes
Leeuw et al (2008).60	The Netherlands	85	CLBP, RDQ > 3, TSK>33	45.32±9.45	44Male/41Female	GEP: A 16-session intervention (twice weekly) involving systematic exposure. GE: A 26-session program (twice weekly) using graded activity.	QBPDS, PHODA, PCS, RT3, SF-MPQ
Linton et al (2008).61	Sweden	34	Spinal pain, TSK>35	GEP:46±9.9; TAU: 49 ± 7.3	16Male/18Female	GEP: A psychologist-led intervention comprising 13–15 sessions graded exposure combined with standard care TAU: Participants received routine care (physician-guided pain management and physical therapy) for 14 weeks.	TSK, PCS, PHODA, QBPDS, ADL Scale, ÖMPQ
Woods et al (2008).63	Canada	44	CLBP, 18–65yrs, TSK≥38	46.45±11.9	15Male/29Female	GEP: Eight 45-minute sessions over 4 weeks (twice weekly) systematic graded exposure. GE: Eight 45-minute sessions over 4 weeks (twice weekly) of graded activity. WLC: No intervention.	PDI, HADS, SF-MPQ, PSEQ, TSK-17, FABQ, PASS-20, PCS
Hollander et al (2016).65	The Netherlands	46	CRPS-I patients, PHODA≥34	44.85±11.23	9Male/37Female	GEP: Systematic graded exposure over 17 weeks, 17 hours, 1-h sessions. TAU: Standardized physical therapy treatment,17 weeks, 17 hours, 30-min sessions.	RASQ, WAQ, NPS, PCS, PHODA, SF-36
Glombiewskiet al (2018).58	Germany	88	CLBP, TSK≥35	GEP-long: 52.7±9.4;GEP-short: 51.8±9.2;CBT:53.5±9.0	40Male/48Female	GEP-long:15 weekly 50-minute sessions, including education and 10 exposure sessions graded exposure treatment. GEP-short: 10 weekly 50-minute sessions with 5 exposure sessions. CBT:15 weekly 50-minute sessions CBT.	QBPDS, PDI, NRS, PIPS, PASS-20, FESV_BW, HADS
Hedman-Lagerlöf et al (2018).59	Sweden	140	Fibromyalgia	50.3±10.9	3Male/137Female	GEP: A 10-week, 8-module online program focused on systematic exposure. WLC: participants received no active treatment during the 10-week study period.	WHODAS 2.0–12, BBQ, PHQ-9, GAD-7, ISI, PRS, PIPS, FIQ
Mateos et al (2019).64	Spain	49	Women with Chronic pelvic pain,18–65yrs,TSK>33	GEP: 42.26±9.57;MT:40.67±11.70;WLC: 42.40±6.15	49 Female	GEP: Combined manual therapy (twice weekly) with graded exposure therapy (once weekly) for 6 weeks. MT: Received manual therapy twice weekly for 6 weeks. WLC: Received only an informational booklet about chronic pelvic pain.	FABQ-PA, BPI, ODI
Simons, L.E. et al (2024).62	USA	68	CMP,8–17yrs, FOPQ-C≥35, FDI≥13	GEP:14.1±2.92;TAU:14.2±2.24	12Male/55Female/1Nonbinary	GEP:12 sessions over 6–8 weeks of systematic graded exposure. TAU: 12 sessions over 6–8 weeks combining CBT therapy for comprehensive pain management.	FOPQ-C, PHODA-YE, FDI, VAS, PCS-C

Abbreviations: ADL Scale, Activities of Daily Living Scale; BBQ, Brunnsviken Brief Quality of Life Scale; BPI, Brief Pain Inventory; CBT, Cognitive Behavioral Therapy; CLBP, Chronic low back pain; CRPS-I, Complex regional pain syndrome type I; FABQ, Fear-Avoidance Beliefs Questionnaire; FABQ-PA, Fear-Avoidance Beliefs Questionnaire–Physical Activity subscale; FDI, Functional Disability Inventory; FESV_BW, Pain Processing and Coping Questionnaire; FIQ, Fibromyalgia Impact Questionnaire; FOPQ-C, Fear of Pain Questionnaire–Child Version; GAD-7, 7-item Generalized Anxiety Disorder scale; GE, Graded Exercise; GEP, Graded Exposure Therapy; HADS, Hospital Anxiety and Depression Scale; ISI, Insomnia Severity Index; MPQ, McGill Pain Questionnaire; MT, Manual Therapy; NPS, Neuropathic Pain Scale; NRS, Numerical Rating Scale; ODI, Oswestry Disability Index; ÖMPQ, Örebro Musculoskeletal Pain Questionnaire; PASS-20, Pain Anxiety Symptoms Scale–20; PCS, Pain Catastrophizing Scale; PCS-C, Pain Catastrophizing Scale–Child Version; PDI, Pain Disability Index; PHODA, Photograph Series of Daily Activities; PHODA-YE, Photograph Series of Daily Activities–Youth English; PHQ-9, Patient Health Questionnaire-9; PIPS, Psychological Inflexibility in Pain Scale; PRS, Pain Reactivity Scale; PSEQ, Pain Self-Efficacy Questionnaire; QBPDS, Quebec Back Pain Disability Scale; RASQ, Radboud Skills Questionnaire; RDQ, Roland Disability Questionnaire; RT3, Tri-axial Research Tracker; SF-36, 36-Item Short Form Health Survey; SF-MPQ, Short-Form McGill Pain Questionnaire; TAU, Treatment As Usual; TSK, Tampa Scale of Kinesiophobia; VAS, Visual Analog Scale; WAQ, Walking Ability Questionnaire; WHODAS 2.0–12, 12-item World Health Organization Disability Assessment Schedule 2.0; WLC, Waiting List Control.

Risk of Bias

Risk of bias was assessed using the RoB 2.0 tool, and the results are presented in Supplementary Table 1 and Supplementary Figure 1. Overall, six studies (75%)^{58–60,62,64,65} were rated as having some concerns and two studies (25%)^61,63 were classified as high risk of bias. Across the five domains, the randomization process showed relatively low risk of bias, with six studies^{58–60,62,64,65} rated as low risk of bias, one study⁶³ having some concerns, and one study⁶¹ rated as high risk of bias. For deviations from intended interventions, five studies^{58–60,64,65} demonstrated low risk of bias, one study⁶³ had some concerns, and two studies^61,62 were rated as high risk of bias. Regarding missing outcome data, five studies^{58–60,64,65} were rated as low risk of bias, one study⁶² raised some concerns, and two studies^61,63 were considered high risk of bias. The measurement of outcome domain showed the greatest risk of bias, with only two studies^60,65 rated as low risk of bias, two studies^58,64 having some concerns, and four studies^59,61–63 rated as high risk of bias. Finally, for selection of reported results, four studies^58,59,62,64 were rated as low risk of bias, two studies^60,65 raised some concerns, and two studies^61,63 were rated as high risk of bias.

Results of Meta-Analysis

Pain Intensity

Overall Assessment

Pain intensity was assessed using eight different measurement instruments across eight studies,^58–65 including 554 participants. Due to the diversity of pain assessment scales employed, separate meta-analyses were conducted for each measurement instrument. Tests for subgroup differences indicated statistically significant differences between measurement instruments (χ² = 25.72, df = 7, P = 0.0006; I² = 72.8%) (Figure 2).

Figure 2.

Forest plot of comparison graded exposure to non-graded exposure for pain intensity.

Abbreviations: BPI, Brief Pain Inventory; CI, confidence interval; FIQ-Pain, Fibromyalgia Impact Questionnaire-Pain; MPQ, McGill Pain Questionnaire; NA, not applicable; NPS, Numeric Pain Scale; NRS, numerical rating scale; ÖMPQ, Örebro Musculoskeletal Pain Questionnaire; SD, standard deviation; SF-MPQ, Short-Form McGill Pain Questionnaire; VAS, visual analogue scale.

Compared with non-GEP groups, statistically significant reductions in pain intensity were observed for NPS (MD = −2.66, 95% CI: −4.09 to −1.23, P = 0.0003; I² = Not applicable (NA), P = NA; low certainty evidence) and FIQ-Pain (MD = −2.51, 95% CI: −3.48 to −1.54, P < 0.00001; I² = NA, P = NA; moderate certainty evidence). No statistically significant differences were observed for VAS (MD = −0.52, 95% CI: −1.74 to 0.70, P = 0.40; I² = NA, P = NA; low certainty evidence), NRS (MD = 0.24, 95% CI: −0.67 to 1.16, P = 0.60; I² = 0%, P = 0.89; very low certainty evidence), BPI (MD = −1.75, 95% CI: −3.63 to 0.13, P = 0.07; I² = 79%, P = 0.03; low certainty evidence), MPQ (MD = 0.70, 95% CI: −9.52 to 10.92, P = 0.89; I² = NA, P = NA; low certainty evidence), SF-MPQ (MD = −3.07, 95% CI: −7.08 to 0.94, P = 0.13; I² = 0%, P = 0.79; very low certainty evidence), or ÖMPQ (MD = 0.00, 95% CI: −1.40 to 1.40, P = 1.00; I² = NA, P = NA; very low certainty evidence) (Figure 2; Supplementary Table 2).

Subgroup Analysis

Pain Type

Tests for subgroup differences among measurement instruments showed no statistically significant differences within the LBP subgroup (χ² = 2.51, df = 3, P = 0.47) or the other CMP subgroup (χ² = 7.55, df = 3, P = 0.06). In the LBP subgroup, no significant differences between GEP and non-GEP groups were observed in VAS (MD = −0.70, 95% CI: −2.17 to 0.77, P = 0.35; I² = 48%, P = 0.17), NRS (MD = 0.24, 95% CI: −0.67 to 1.16, P = 0.60; I² = 0%, P = 0.89), MPQ (MD = 0.70, 95% CI: −9.52 to 10.92, P = 0.89; I² = NA, P = NA), and MPQ-SF (MD = −3.07, 95% CI: −7.08 to 0.94, P = 0.13; I² = 0%, P = 0.79).In the other CMP conditions, GEP was associated with greater reductions in pain intensity than non-GEP for NPS (MD = −2.66, 95% CI: −4.09 to −1.23, P = 0.0003; I² = NA, P = NA) and FIQ-PAIN (MD = −2.51, 95% CI: −3.48 to −1.54, P<0.00001; I² = NA, P = NA). No significant differences were observed in VAS (MD = −0.52, 95% CI: −1.74 to 0.70, P = 0.40; I² = NA, P = NA) or BPI (MD = −1.75, 95% CI: −3.63 to 0.13, P = 0.07; I² = 79%, P = 0.03) (Table 2).

Table 2.

Subgroup Analysis of Pain Intensity

Subgroup			N	Sample	MD, 95% CI	P for MD	I2	P for I2	Chi2	df	P for Subgroup Differences
Type of pain	LBP	NRS	2	40/54	0.24[−0.67, 1.16]	0.60	0%	0.89	2.51	3	0.47
		MPQ	1	38/35	0.70[−9.52, 10.92]	0.89	NA	NA
		SF-MPQ	2	30/29	−3.07[−7.08, 0.94]	0.13	0%	0.79
		ÖMPQ	1	13/21	0.00[−1.40, 1.40]	1.00	NA	NA
	Others	NPS	1	18/17	−2.66[−4.09, −1.23]	0.0003	NA	NA	7.55	3	0.06
		FIQ	1	70/70	−2.51[−3.48, −1.54]	<0.00001	NA	NA
		VAS	1	33/35	−0.52[−1.74, 0.70]	0.40	NA	NA
		BPI	2	32/33	−1.75[−3.63, 0.13]	0.07	79%	0.03
Age group	Adults	NRS	2	40/54	0.24[−0.67, 1.16]	0.60	0%	0.89	24.69	6	0.0004
		BPI	2	32/33	−1.75[−3.63, 0.13]	0.07	79%	0.03
		MPQ	1	38/35	0.70[−9.52, 10.92]	0.89	NA	NA
		SF-MPQ	2	30/29	−3.07[−7.08, 0.94]	0.13	0%	0.79
		NPS	1	18/17	−2.66[−4.09, −1.23]	0.0003	NA	NA
		FIQ	1	70/70	−2.51[−3.48, −1.54]	<0.00001	NA	NA
		ÖMPQ	1	13/21	0.00[−1.40, 1.40]	1.00	NA	NA
	Adolescents	VAS	1	33/35	−0.52[−1.74, 0.70]	0.40	NA	NA	NA	NA	NA
Mode of intervention delivery	Face-to-face model	VAS	1	33/35	−0.52[−1.74, 0.70]	0.40	NA	NA	14.93	6	0.02
		NRS	2	40/54	0.24[−0.67, 1.16]	0.60	0%	0.89
		BPI	2	32/33	−1.75[−3.63, 0.13]	0.07	79%	0.03
		MPQ	1	38/35	0.70[−9.52, 10.92]	0.89	NA	NA
		SF-MPQ	2	30/29	−3.07[−7.08, 0.94]	0.13	0%	0.79
		NPS	1	18/17	−2.66[−4.09, −1.23]	0.0003	NA	NA
		ÖMPQ	1	13/21	0.00[−1.40, 1.40]	1.00	NA	NA
	Online model	FIQ	1	70/70	−2.51[−3.48, −1.54]	<0.00001	NA	NA	NA	NA	NA
Type of control group	Conventional control	VAS	1	33/35	−0.52[−1.74, 0.70]	0.40	NA	NA	13.31	6	0.04
		BPI	1	16/16	−0.82[−1.94, 0.30]	0.15	NA	NA
		MPQ	1	38/35	0.70[−9.52, 10.92]	0.89	NA	NA
		SF-MPQ	1	15/13	−3.60[−9.17, 1.97]	0.20	NA	NA
		NRS	2	40/54	0.24[−0.67, 1.16]	0.60	0%	0.89
		NPS	1	18/17	−2.66[−4.09, −1.23]	0.0003	NA	NA
		ÖMPQ	1	13/21	0.00[−1.40, 1.40]	1.00	NA	NA
	No-intervention group	FIQ	1	70/70	−2.51[−3.48, −1.54]	<0.00001	NA	NA	2.43	2	0.30
		SF-MPQ	1	15/16	−2.50[−8.29, 3.29]	0.40	NA	NA
		BPI	1	16/17	−2.74[−4.06, −1.42]	<0.0001	NA	NA

Abbreviations: BPI, Brief Pain Inventory; CI, confidence interval; FIQ-Pain, Fibromyalgia Impact Questionnaire pain subscale; LBP, low back pain; MD, mean difference; MPQ, McGill Pain Questionnaire; NA, not available; NPS, Neuropathic Pain Scale; NRS, Numerical Rating Scale; SF-MPQ, Short-Form McGill Pain Questionnaire; VAS, Visual Analog Scale.

Age Group

Tests for subgroup differences among measurement instruments were significant in adults (χ² = 24.69, df = 6, P = 0.0004), whereas subgroup differences could not be assessed in adolescents because only one measurement instrument was available. In adult participants, GEP was associated with greater reductions in pain intensity than non-GEP for NPS (MD = −2.66, 95% CI: −4.09 to −1.23, P = 0.0003; I² = NA, P = NA) and FIQ-pain (MD = −2.51, 95% CI: −3.48 to −1.54, P<0.00001; I² = NA, P = NA).No significant differences were observed for BPI (MD = −1.75, 95% CI: −3.63 to 0.13, P = 0.07; I² = 79%, P = 0.03), VAS (MD = −0.70, 95% CI: −2.17 to 0.77, P = 0.35; I² = 48%, P = 0.17), NRS (MD = 0.24, 95% CI: −0.67 to 1.16, P = 0.60; I² = 0%, P = 0.89), MPQ (MD = 0.70, 95% CI: −9.52 to 10.92, P = 0.89; I² = NA, P = NA), or MPQ-SF (MD = −3.07, 95% CI: −7.08 to 0.94, P = 0.13; I² = 0%, P = 0.79). In adolescents, no significant difference was observed for VAS (MD = −0.52, 95% CI: −1.74 to 0.70, P = 0.40; I² = NA, P = NA). Other pain intensity instruments were not available in adolescent participants (Table 2).

Mode of Intervention Delivery

Tests for subgroup differences among measurement instruments were significant within the face-to-face intervention subgroup (χ² = 14.93, df = 6, P = 0.02). In the face-to-face subgroup, GEP was associated with greater reductions in pain intensity than non-GEP for NPS (MD = −2.66, 95% CI: −4.09 to −1.23, P = 0.0003; I² = NA, P = NA). No significant differences were observed for BPI (MD = −1.75, 95% CI: −3.63 to 0.13, P = 0.07; I² = 79%, P = 0.03), VAS (MD = −0.60, 95% CI: −1.39 to 0.20, P = 0.14; I² = 0%, P = 0.38), NRS (MD = 0.24, 95% CI: −0.67 to 1.16, P = 0.60; I² = 0%, P = 0.89), MPQ (MD = 0.70, 95% CI: −9.52 to 10.92, P = 0.89; I² = NA, P = NA), or MPQ-SF (MD = −3.07, 95% CI: −7.08 to 0.94, P = 0.13; I² = 0%, P = 0.79). One study that included an online intervention subgroup and reported a statistically significant reduction in FIQ-Pain with GEP compared with non-GEP (MD = −2.51, 95% CI: −3.48 to −1.54, P<0.00001; I² = NA, P = NA). Other pain intensity instruments were not available for participants receiving online interventions (Table 2).

Type of Control Group

Tests for subgroup differences among measurement instruments were significant within the conventional control subgroup (χ² = 13.31, df = 6, P = 0.04), whereas no significant differences were observed within the no-intervention control subgroup (χ² = 2.43, df = 2, P = 0.30). Compared with conventional control groups, GEP demonstrated greater reductions in pain intensity measured by NPS (MD = −2.66, 95% CI: −4.09 to −1.23, P = 0.0003; I² = NA, P = NA). No significant differences were observed for VAS (MD = −0.60, 95% CI: −1.39 to 0.20, P = 0.14; I² = 0%, P = 0.38), BPI (MD = −0.82, 95% CI: −1.94 to 0.30, P = 0.15; I² = NA, P = NA), MPQ (MD = 0.70, 95% CI: −9.52 to 10.92, P = 0.89; I² = NA, P = NA), MPQ-SF (MD = −3.60, 95% CI: −9.17 to 1.97, P = 0.20; I² = NA, P = NA), or NRS (MD = 0.24, 95% CI: −0.67 to 1.16, P = 0.60; I² = 0%, P = 0.89). Compared with no-interventiongroups, GEP showed significant improvements in BPI (MD = −2.74, 95% CI: −4.06 to −1.42, P<0.0001; I² = NA, P = NA) and FIQ-PAIN (MD = −2.51, 95% CI: −3.48 to −1.54, P<0.00001; I² = NA, P = NA).No significant difference was observed for MPQ-SF (MD = −2.50, 95% CI: −8.29 to 3.29, P = 0.40; I² = NA, P = NA) (Table 2).

Pain Disability

Overall Assessment

Pain-related disability was assessed in eight studies (n = 554)^58–65 using seven measurement instruments. Analysis by measurement instrument revealed substantial heterogeneity between different scales (χ^{2 =} 46.84, df = 6, P<0.00001, I² = 87.2%) in Figure 3.

Figure 3.

Forest plot of comparison graded exposure to non-graded exposure for pain-related disability.

Abbreviations: CI, confidence interval; FDI, Functional Disability Inventory; NA, not applicable; ODI, Oswestry Disability Index; PDI = Pain Disability Index; QBPDS, Quebec Back Pain Disability Scale; RASQ, Roland Activity Scale Questionnaire; SD, standard deviation; WAQ, Walking Ability Questionnaire; WHODAS 2.0–12, World Health Organization Disability Assessment Schedule 2.0–12-item version.

Compared with non-GEP groups, significant reductions in pain-related disability were observed in the GEP groups when assessed using ODI (MD = −9.99, 95% CI: −13.58 to −6.40, P<0.00001; I² = 95%, P<0.0001; low certainty evidence), WHODAS 2.0–12(MD = −16.19, 95% CI: −22.10 to −10.28, P<0.00001; I² = NA, P = NA; moderate certainty evidence), RASQ (MD = −1.21, 95% CI: −2.01 to −0.41, P = 0.003; I² = NA, P = NA; low certainty evidence), and WAQ (MD = −3.79, 95% CI: −6.85 to −0.73, P = 0.02; I² = NA, P = NA; low certainty evidence). No statistically significant effects were observed for PDI (MD = −1.35, 95% CI: −5.07 to 2.38, P = 0.48; I² = 50%, P = 0.11; very low certainty evidence), QBPDS (MD = −1.48, 95% CI: −6.24 to 3.28, P = 0.54; I² = 0%, P = 0.70; low certainty evidence) or FDI (MD = 0.20, 95% CI: −5.87 to 6.27, P = 0.95; I² = NA, P = NA; low certainty evidence) (Figure 3; Supplementary Table 2).

Subgroup Analysis

Type of Pain

Tests for subgroup differences among measurement instruments showed a statistically significant difference within the other CMP subgroup (χ² = 46.71, df = 4, P < 0.00001), whereas no significant difference was observed within the LBP subgroup (χ² = 0, df = 1, P = 0.96). In the LBP subgroup, no significant differences between GEP and non-GEP groups were observed for PDI (MD = −1.35, 95% CI: −5.07 to 2.38, P = 0.48; I² = 50%, P = 0.11) or QBPDS (MD = −1.48, 95% CI: −6.24 to 3.28, P = 0.54; I² = 0%, P = 0.70). In participants with other CMP conditions, GEP was associated with greater reductions in disability scores than non-GEP for WHODAS 2.0–12 (MD = −16.19, 95% CI: −22.10 to −10.28, P<0.00001; I² = NA, P = NA), RASQ (MD = −1.21, 95% CI: −2.01 to −0.41, P = 0.003; I² = NA, P = NA), WAQ (MD = −3.79, 95% CI: −6.85 to −0.73, P = 0.02; I² = NA, P = NA), and ODI (MD = −9.99, 95% CI: −13.58 to −6.40, P<0.00001; I² = 95%, P<0.0001). No significant difference was observed for FDI (MD = 0.20, 95% CI: −5.87 to 6.27, P = 0.95; I² = NA, P = NA) (Table 3).

Table 3.

Subgroup Analysis of Pain Disability

	Subgroup		N	Sample	MD,95% CI	P for MD	I2	P for I2	Chi2	df	P for Subgroup Differences
Type of pain	LBP	PDI	4	70/83	−1.35[−5.07, 2.38]	0.48	50%	0.11	0	1	0.96
		QBPDS	4	91/110	−1.48[−6.24, 3.28]	0.54	0%	0.70
	Others	WHODAS 2.0-12	1	70/70	−16.19[−22.10, −10.28]	<0.00001	NA	NA	46.71	4	<0.00001
		RASQ	1	13/11	−1.21[−2.01, - 0.41]	0.003	NA	NA
		WAQ	1	6/6	−3.79[−6.85, −0.73]	0.02	NA	NA
		FDI	1	33/35	0.20[−5.87, 6.27]	0.95	NA	NA
		ODI	2	32/33	−9.99[−13.58, −6.40]	<0.00001	95%	<0.0001
Age group	Adults	PDI	4	70/83	−1.35[−5.07, 2.38]	0.48	50%	0.11	46.37	5	<0.00001
		ODI	2	32/33	−9.99[−13.58, −6.40]	<0.00001	95%	<0.0001
		QBPDS	4	91/110	−1.48[−6.24, 3.28]	0.54	0%	0.70
		WHODAS 2.0-12	1	70/70	−16.19[−22.10, −10.28]	<0.00001	NA	NA
		RASQ	1	13/11	−1.21[−2.01, −0.41]	0.003	NA	NA
		WAQ	1	6/6	−3.79[−6.85, −0.73]	0.02	NA	NA
	Adolescents	FDI	1	33/35	0.20[−5.87, 6.27]	0.95	NA	NA	NA	NA	NA
Mode of intervention delivery	Face-to-face model	PDI	4	70/83	−1.35[−5.07, 2.38]	0.48	50%	0.11	24.12	5	0.0002
		ODI	2	32/33	−9.99[−13.58, −6.40]	<0.00001	95%	<0.0001
		QBPDS	4	91/110	−1.48[−6.24, 3.28]	0.54	0%	0.70
		RASQ	1	13/11	−1.21[−2.01, −0.41]	0.003	NA	NA
		WAQ	1	6/6	−3.79[−6.85, −0.73]	0.02	NA	NA
		FDI	1	33/35	0.20[−5.87, 6.27]	0.95	NA	NA
	Online model	WHODAS 2.0-12	1	70/70	−16.19[−22.10, −10.28]	<0.00001	NA	NA	NA	NA	NA
Type of control group	No-intervention group	PDI	1	15/16	−6.60[−15.08, 1.88]	0.13	NA	NA	7.48	2	0.02
		WHODAS 2.0-12	1	70/70	−16.19[−22.10, −10.28]	<0.00001	NA	NA
		ODI	1	16/17	−21.37[−27.70, −15.04]	<0.00001	NA	NA
	Conventional control	ODI	1	16/16	−4.59[−8.95, −0.23]	0.04	NA	NA	5.27	5	0.38
		QBPDS	4	91/110	−1.48[−6.24, 3.28]	0.54	0%	0.7
		RASQ	1	13/11	−1.21[−2.01, −0.41]	0.003	NA	NA
		WAQ	1	6/6	−3.79[−6.85, −0.73]	0.02	NA	NA
		FDI	1	33/35	0.20[−5.87, 6.27]	0.95	NA	NA
		PDI	3	55/67	−0.09[−4.24, 4.06]	0.97	52%	0.12

Abbreviations: CI, confidence interval; FDI, Functional Disability Inventory; LBP, low back pain; MD, mean difference; NA, not available; ODI, Oswestry Disability Index; ÖMPQ, Örebro Musculoskeletal Pain Questionnaire; PDI, Pain Disability Index; QBPDS, Quebec Back Pain Disability Scale; RASQ, Radboud Skills Questionnaire; WAQ, Walking Ability Questionnaire; WHODAS 2.0–12, 12-item World Health Organization Disability Assessment Schedule 2.0.

Age Group

Tests for subgroup differences among measurement instruments were significant in adults (χ² = 46.37, df = 5, P < 0.00001). In adults, GEP was associated with greater reductions in disability scores than non-GEP for ODI (MD = −9.99, 95% CI: −13.58 to −6.40, P<0.00001; I² = 95%, P<0.0001), WHODAS 2.0–12 (MD = −16.19, 95% CI: −22.10 to −10.28, P<0.00001; I² = NA, P = NA), RASQ (MD = −1.21, 95% CI: −2.01 to −0.41, P = 0.003; I² = NA, P = NA), and WAQ (MD = −3.79, 95% CI: −6.85 to −0.73, P = 0.02; I² = NA, P = NA). No significant differences were observed for PDI (MD = −1.35, 95% CI: −5.07 to 2.38, P = 0.48; I² = 50%, P = 0.11) or QBPDS (MD = −1.48, 95% CI: −6.24 to 3.28, P = 0.54; I² = 0%, P = 0.70). Only one study was available in adolescents, which showed no significant difference between GEP and non-GEP groups for FDI (MD = 0.20, 95% CI: −5.87 to 6.27, P = 0.95; I² = NA, P = NA) (Table 3).

Mode of Intervention Delivery

Tests for subgroup differences among measurement instruments were significant within the face-to-face delivery subgroup (χ² = 24.12, df = 5, P = 0.0002). Participants receiving face-to-face GEP demonstrated greater reductions in disability scores than those receiving face-to-face non-GEP for ODI (MD = −9.99, 95% CI: −13.58 to −6.40, P<0.00001; I² = 95%, P <0.0001), RASQ (MD = −1.21, 95% CI: −2.01 to −0.41, P = 0.003; I² = NA, P = NA), and WAQ (MD = −3.79, 95% CI: −6.85 to −0.73, P = 0.02; I² = NA, P = NA). No significant differences were observed for PDI (MD = −1.35, 95% CI: −5.07 to 2.38, P = 0.48; I² = 50%, P = 0.11), QBPDS (MD = −1.48, 95% CI: −6.24 to 3.28, P = 0.54; I² = 0%, P = 0.70), or FDI (MD = 0.20, 95% CI: −5.87 to 6.27, P = 0.95; I² = NA, P = NA). One study was available for the online delivery subgroup and reported a significant reduction in WHODAS 2.0–12 scores with GEP compared with non-GEP (MD = −16.19, 95% CI: −22.10 to −10.28, P<0.00001; I² = NA, P = NA). Other disability instruments were only available for face-to-face delivery studies (Table 3).

Type of Control Group

Tests for subgroup differences among measurement instruments showed a significant difference within the no-intervention control subgroup (χ² = 7.48, df = 2, P = 0.02), whereas no significant difference was observed within the conventional control subgroup (χ² = 5.27, df = 5, P = 0.38). Compared with non-GEP, GEP was associated with significantly greater reductions in disability scores on the WHODAS 2.0–12 (MD = −16.19, 95% CI: −22.10 to −10.28, P<0.00001; I² = NA, P = NA) and ODI (MD = −21.37, 95% CI: −27.70 to −15.04, P<0.00001; I² = NA, P = NA). No significant difference was observed on PDI (MD = −6.60, 95% CI: −15.08 to 1.88, P = 0.13; I² = NA, P = NA).

Compared with conventional control groups, GEP demonstrated greater reductions in disability scores for ODI (MD = −4.59, 95% CI: −8.95 to −0.23, P = 0.04; I² = NA, P = NA), RASQ (MD = −1.21, 95% CI: −2.01 to −0.41, P = 0.003; I² = NA, P = NA), and WAQ (MD = −3.79, 95% CI: −6.85 to −0.73, P = 0.02; I² = NA, P = NA). No significant differences were observed for QBPDS (MD = −1.48, 95% CI: −6.24 to 3.28, P = 0.54; I² = 0%, P = 0.7), FDI (MD = 0.20, 95% CI: −5.87 to 6.27, P = 0.95; I² = NA, P = NA), or PDI (MD = −0.09, 95% CI: −4.24 to 4.06, P = 0.97; I² = 52%, P = 0.12) in Table 3.

Pain-Related Fear and Cognition

Pain-related fear and cognition were assessed across five domains: pain catastrophizing, kinesiophobia, fear-avoidance beliefs, fear of pain, and perceived harmfulness.

Pain Catastrophizing

Pain catastrophizing was assessed using two measurement instruments across five studies (n = 254).^60–63,65 Analysis by measurement instrument revealed no significant heterogeneity between scales (χ² = 0.53, df = 1, P = 0.47, I² = 0%) in Supplementary Figure 2.

When analyzed by specific measurement instruments, GEP was associated with greater reductions in pain catastrophizing measured by PCS across four studies (MD = −4.45, 95% CI: −7.20 to −1.69, P = 0.002; I² = 42%, P = 0.14). No statistically significant difference was observed for PCS-C in one adolescent study (MD = −1.96, 95% CI: −8.08 to 4.16, P = 0.53; I² = NA, P = NA) in one study each (Supplementary Figure 2).

Subgroup Analysis: Type of Pain

Tests for subgroup differences among measurement instruments showed no statistically significant difference within the other CMP subgroup (χ² = 0.58, df = 1, P = 0.45). In LBP participants, GEP was associated with greater reductions in pain catastrophizing measured by PCS compared with non-GEP (MD = −4.22, 95% CI: −7.34 to −1.09, P = 0.008; I² = 56%, P = 0.08). In participants with other CMP conditions, no statistically significant difference was observed for PCS (MD = −5.24, 95% CI: −11.06 to 0.58, P = 0.08; I² = NA, P = NA) or PCS-C (MD = −1.96, 95% CI: −8.08 to 4.16, P = 0.53; I² = NA, P = NA) (Supplementary Table 3).

Subgroup Analysis: Age Group

No within-subgroup comparison among measurement instruments was performed because only one measurement instrument was available within each age subgroup. Among adults, GEP was associated with greater reductions in pain catastrophizing measured by PCS compared with non-GEP (MD = −4.45, 95% CI: −7.20 to −1.69, P = 0.002; I² = 42%, P = 0.14). In adolescents, no significant difference was observed for PCS-C (MD = −1.96, 95% CI: −8.08 to 4.16, P = 0.53; I² = NA, P = NA) (Supplementary Table 3).

Subgroup Analysis: Mode of Intervention Delivery

Tests for subgroup differences among measurement instruments showed no statistically significant difference within the face-to-face delivery subgroup (χ² = 0.53, df = 1, P = 0.47) Participants receiving face-to-face GEP showed greater reductions in pain catastrophizing measured by PCS than those receiving face-to-face non-GEP (MD = −4.45, 95% CI: −7.20 to −1.69, P = 0.002; I² = 42%, P = 0.14). No statistically significant difference was observed for PCS-C (MD = −1.96, 95% CI: −8.08 to 4.16, P = 0.53; I² = NA, P = NA) (Supplementary Table 3).

Subgroup Analysis: Type of Control Group

In subgroup analysis by type of control group, no statistically significant difference in treatment effects among measurement instruments was observed within the conventional control subgroup (χ² = 0.15, df = 1, P = 0.70). Compared with no-intervention control groups, a significant reduction in pain catastrophizing measured by PCS was observed in the GEP group (MD = −11.00, 95% CI: −18.17 to −3.83, P = 0.003; I² = NA, P = NA). Compared with conventional control groups, a significant reduction in pain catastrophizing measured by PCS was observed in the GEP group (MD = −3.31, 95% CI: −6.29 to −0.33, P = 0.03; I² = 4%, P = 0.37), whereas no significant difference was observed when assessed using PCS-C (MD = −1.96, 95% CI: −8.08 to 4.16, P = 0.53; I² = NA, P = NA). Only one study was available in the no-intervention subgroup, precluding a comparison among measurement instruments (Supplementary Table 3).

Kinesiophobia

In kinesiophobia assessment, two studies (n = 78)^61,63 reported a significant reduction in TSK scores in the GEP group compared with the non-GEP group (MD = −4.57, 95% CI: −7.11 to −2.03, P = 0.0004; I² = 0%, P = 0.80) (Supplementary Figure 3).

Subgroup Analysis: Type of Pain, Age Group and Mode of Intervention Delivery

In subgroup analyses by type of pain (LBP only), age group (adults only), and mode of intervention delivery (face-to-face mode only), the results were consistent across all three analyses. The GEP group showed a significant reduction in TSK scores compared with the non-GEP group (MD = −4.57, 95% CI: −7.11 to −2.03, P = 0.0004; I² = 0%, P = 0.80), based on three studies (Supplementary Table 4).

Subgroup Analysis: Type of Control Group

In subgroup analysis by type of control group, GEP was associated with a significant reduction in TSK scores compared with conventional control groups (MD = −5.06, 95% CI: −8.30 to −1.82, P = 0.002; I² = 0%, P = 0.64), whereas no significant differences was observed compared with no-intervention control groups (MD = −3.80, 95% CI: −7.89 to 0.29, P = 0.07; I² = NA, P = NA) (Supplementary Table 4).

Fear-Avoidance Beliefs

In the assessment of fear-avoidance beliefs, four measurement instruments were evaluated across four studies (n = 391).^58,59,63,64 Analysis by measurement instrument showed no heterogeneity between scales (χ² = 0.46, df = 2, P = 0.79, I² = 0%). Significant reductions were observed in FABQ (MD = −13.03, 95% CI: −21.90 to −4.15, P = 0.004; I² = 22%, P = 0.26), PIPS (MD = −9.79, 95% CI: −13.47 to −6.12, P<0.00001; I² = 93%, P<0.00001), and FABQ-PA (MD = −10.66, 95% CI: −13.77 to −7.54, P<0.00001; I² = 72%, P = 0.06) (Supplementary Figure 4).

Subgroup Analysis: Type of Pain

Tests for subgroup differences among measurement instruments showed statistically significant differences within the LBP subgroup (χ² = 5.89, df = 1, P = 0.02) and the other pain conditions subgroup (χ² = 8.44, df = 1, P = 0.004). In the LBP subgroup, GEP was associated with a significant reduction in fear-avoidance beliefs measured by FABQ total score compared with non-GEP (MD = −13.03, 95% CI: −21.90 to −4.13, P = 0.004; I² = 22%, P = 0.26), whereas no significant difference was observed for PIPS (MD = −0.33, 95% CI: −5.47 to 4.80, P = 0.90; I² = 0%, P = 0.74). In other pain conditions, GEP was associated with greater reductions in fear-avoidance beliefs measured by PIPS (MD = −19.72, 95% CI: −24.98 to −9.13, P<0.00001; I² = NA, P = NA) and FABQ-PA (MD = −10.66, 95% CI: −13.77 to −7.54, P<0.00001; I² = 72%, P = 0.06) (Supplementary Table 5).

Subgroup Analysis: Age Group

Tests for subgroup differences among measurement instruments showed no statistically significant differences within the adult subgroup (χ² = 0.46, df = 2, P = 0.79). In the adult subgroup, GEP was associated with significant reductions in fear-avoidance beliefs measured by FABQ total score (MD = −13.03, 95% CI: −21.90 to −4.13, P = 0.004; I² = 22%, P = 0.26), FABQ-PA (MD = −10.66, 95% CI: −13.77 to −7.54, P < 0.00001; I² = 72%, P = 0.06), and PIPS (MD = −9.79, 95% CI: −13.47 to −6.12, P < 0.00001; I² = 93%, P < 0.00001) (Supplementary Table 5).

Subgroup Analysis: Mode of Intervention Delivery

Tests for subgroup differences among measurement instruments showed statistically significant differences within the face-to-face delivery subgroup (χ² = 12.53, df = 2, P = 0.002). In the face-to-face delivery subgroup, GEP was associated with significant reductions in fear-avoidance beliefs measured by FABQ total score (MD = −13.03, 95% CI: −21.90 to −4.13, P = 0.004; I² = 22%, P = 0.26) and FABQ-PA (MD = −10.66, 95% CI: −13.77 to −7.54, P < 0.00001; I² = 72%, P = 0.06), whereas no significant difference was observed for PIPS (MD = −0.33, 95% CI: −5.47 to 4.80, P = 0.90; I² = 0%, P = 0.74). Only one study was available in the online intervention subgroup, which reported a significant reduction by PIPS in the GEP group compared with the non-GEP group (MD = −19.72, 95% CI: −24.98 to −9.13, P < 0.00001; I² = NA, P = NA) in Supplementary Table 5.

Subgroup Analysis: Type of Control Group

Tests for subgroup differences among measurement instruments showed no statistically significant differences within either the no-intervention subgroup (χ² = 0.57, df = 1, P = 0.45) or the conventional control subgroup (χ² = 2.82, df = 2, P = 0.24). Compared with the no-intervention subgroup, GEP was associated with a significant reduction in FABQ-PA scores (MD = −13.21, 95% CI: −17.29 to −9.13, P < 0.00001; I² = NA, P = NA), whereas no significant difference was observed for FABQ (MD = −8.30, 95% CI: −20.36 to 3.76, P = 0.18; I² = NA, P = NA). Compared with conventional control groups, GEP was associated with significant reductions in FABQ (MD = −18.60, 95% CI: −31.70 to −5.50, P = 0.005; I² = NA, P = NA), FABQ-PA (MD = −7.07, 95% CI: −11.90 to −2.24, P = 0.004; I² = NA, P = NA), and PIPS (MD = −9.79, 95% CI: −13.47 to −6.12, P < 0.00001; I² = 93%, P < 0.00001) (Supplementary Table 5).

Fear of Pain

In the assessment of fear of pain, one study (n = 68)⁶² used the Fear of Pain Questionnaire for Children (FOPQ-C). No significant differences were observed between GEP and non-GEP groups in FOPQ-C pain (MD = −0.98, 95% CI: −6.69 to 4.73, P = 0.74) or FOPQ-C activity scores (MD = 1.84, 95% CI: −3.04 to 6.72, P = 0.46) (Supplementary Figure 5). No significant differences were observed between the effects measured by the FOPQ-C pain and activity subscales (χ² = 0.54, df = 1, P = 0.46).

Subgroup analyses by type of pain, age group, mode of intervention delivery, and type of control group were based on the same single study and, therefore, additional comparative analyses could not be performed (Supplementary Table 6).

Perceived Harmfulness

In the assessment of perceived harmfulness, three studies (n = 166)^60,62,65 used two versions of the PHODA to compare GEP and non-GEP groups. Analysis by measurement instrument showed substantial heterogeneity across scales (χ² = 41.34, df = 2, P<0.00001, I² = 95.2%). The GEP group showed significantly lower perceived harmfulness scores on the adult version (PHODA) (MD = −27.49, 95% CI: −35.75 to −19.22, P<0.00001; I² = 91%, P = 0.001), whereas no significant differences were observed on the youth version (PHODA-YE) for either the pain (MD = −0.22, 95% CI: −1.30 to 0.86, P = 0.69; I² = NA, P = NA) or worry subscales (MD = −0.30, 95% CI: −1.41 to 0.81, P = 0.60; I² = NA, P = NA) (Supplementary Figure 6).

Subgroup Analysis: Type of Pain

In subgroup analysis by type of pain, a statistically significant difference in treatment effects among measurement instruments was observed within the other pain conditions subgroup (χ² = 45.19, df = 2, P<0.00001). Only one measurement instrument was available in the LBP subgroup; therefore, differences between the measurement instruments could not be assessed. In the LBP subgroup, the GEP group showed significantly lower perceived harmfulness scores than the non-GEP group on PHODA (MD = −15.28, 95% CI: −26.29 to −4.27, P = 0.007; I² = NA, P = NA). In the other pain conditions subgroup, GEP was associated with lower perceived harmfulness scores on PHODA (MD = −43.23, 95% CI: −55.74 to −30.72, P<0.00001; I² = NA, P = NA), whereas no significant differences were observed on PHODA-YE pain (MD = −0.22, 95% CI: −1.30 to 0.86, P = 0.69; I² = NA, P = NA) or PHODA-YE worry (MD = −0.30, 95% CI: −1.41 to 0.81, P = 0.60; I² = NA, P = NA) (Supplementary Table 7).

Subgroup Analysis: Age Group

In subgroup analysis by age group, no statistically significant difference in treatment effects among measurement instruments was observed within the adolescent subgroup (χ² = 0.01, df = 1, P = 0.92). For adults, the GEP group showed significantly lower PHODA scores than the non-GEP group (MD = −27.49, 95% CI: −35.75 to −19.22, P<0.00001; I² = 91%, P = 0.001). For adolescents, no significant differences were observed for PHODA-YE pain scores (MD = −0.22, 95% CI: −1.30 to 0.86, P = 0.69; I² = NA, P = NA) or PHODA-YE worry scores (MD = −0.3, 95% CI: −1.41 to 0.81, P = 0.60; I² = NA, P = NA) (Supplementary Table 7).

Subgroup Analysis: Mode of Intervention Delivery and Type of Control Group

In subgroup analyses by mode of intervention delivery and type of control group, all studies used face-to-face delivery and conventional control conditions; therefore, the findings were identical across both subgroup analyses. A statistically significant difference in treatment effects was observed among the measurement instruments within both the face-to-face subgroup (χ² = 41.34, df = 2, P<0.00001) and the conventional control subgroup (χ² = 41.34, df = 2, P<0.00001). The GEP group showed significantly lower PHODA scores than the non-GEP group (MD = −27.49, 95% CI: −35.75 to −19.22, P<0.00001; I² = 91%, P = 0.001). No significant differences were observed for PHODA-YE pain scores (MD = −0.22, 95% CI: −1.30 to 0.86, P = 0.69; I² = NA, P = NA) or PHODA-YE worry scores (MD = −0.30, 95% CI: −1.41 to 0.81, P = 0.60; I² = NA, P = NA) (Supplementary Table 7).

Quality of Life

Two studies (n = 175)^59,65 assessed quality of life using the SF-36 and BBQ. No statistically significant difference in treatment effects among measurement scales was observed (χ² = 0.45, df = 2, P = 0.80, I² = 0%). The GEP group showed higher SF-36 physical component summary (PCS) scores (MD = 18.24, 95% CI: 2.95 to 33.53, P = 0.02; I² = NA, P = NA) and higher BBQ scores (MD = 16.97, 95% CI: 9.22 to 24.72, P<0.0001; I² = NA, P = NA), whereas no significant difference was observed for SF-36 mental component summary (MCS) scores (MD = 12.03, 95% CI: −1.82 to 25.88, P = 0.09; I² = NA, P = NA) (Supplementary Figure 7).

Subgroup Analysis: Type of Pain

In subgroup analysis by type of pain, only participants with other pain conditions were included. No statistically significant difference in treatment effects among measurement instruments was observed within this subgroup (χ² = 0.45, df = 2, P = 0.80). The GEP group showed higher SF-36-PCS scores than the non-GEP group (MD = 18.24, 95% CI: 2.95 to 33.53, P = 0.02; I² = NA, P = NA) and higher BBQ scores (MD = 16.97, 95% CI: 9.22 to 24.72, P<0.0001; I² = NA, P = NA). However, no significant difference was observed in SF-36-MCS (MD = 12.03, 95% CI: −1.82 to 25.88, P = 0.09; I² = NA, P = NA) (Supplementary Table 8).

Subgroup Analysis: Age Group

In subgroup analysis by age group, only adults were included. No statistically significant difference in treatment effects among measurement instruments was observed within the adult subgroup (χ² = 0.45, df = 2, P = 0.80). The GEP group showed higher SF-36-PCS scores than the non-GEP group (MD = 18.24, 95% CI: 2.95 to 33.53, P = 0.02; I² = NA, P = NA) and higher BBQ scores (MD = 16.97, 95% CI: 9.22 to 24.72, P<0.0001; I² = NA, P = NA). No significant difference was observed in SF-36-MCS (MD = 12.03, 95% CI: −1.82 to 25.88, P = 0.09; I² = NA, P = NA) (Supplementary Table 8).

Subgroup Analysis: Mode of Intervention Delivery

In subgroup analysis by mode of intervention delivery, no statistically significant difference in treatment effects among measurement instruments was observed within the face-to-face subgroup (χ² = 0.35, df = 1, P = 0.56). For face-to-face delivery, the GEP group showed higher SF-36-PCS scores than the non-GEP group (MD = 18.24, 95% CI: 2.95 to 33.53, P = 0.02; I² = NA, P = NA),whereas no significant difference was observed in SF-36-MCS scores (MD = 12.03, 95% CI: −1.82 to 25.88, P = 0.09; I² = NA, P = NA). Only one study was available in the online subgroup, which assessed quality of life using BBQ; therefore, differences among measurement instruments could not be assessed. For online delivery, the GEP group showed higher BBQ scores than the non-GEP group (MD = 16.97, 95% CI: 9.22 to 24.72, P < 0.0001; I² = NA, P = NA) in Supplementary Table 8.

Subgroup Analysis: Type of Control Group

In subgroup analysis by control group type, no statistically significant difference in treatment effects was observed among the measurement instruments within the conventional control subgroup (χ² = 0.35, df = 1, P = 0.56). Compared with conventional controls, the GEP group showed significantly better physical quality of life assessed by SF-36-PCS (MD = 18.24, 95% CI: 2.95 to 33.53, P = 0.02; I² = NA, P = NA). No significant difference was observed for SF-36-MCS (MD = 12.03, 95% CI: −1.82 to 25.88, P = 0.09; I² = NA, P = NA). Only one study was available in the no-intervention subgroup, which assessed quality of life using the BBQ. Therefore, comparisons among measurement instruments could not be performed. Compared with no-intervention controls, the GEP group demonstrated significantly higher quality of life scores on the BBQ (MD = 16.97, 95% CI: 9.22 to 24.72, P<0.0001; I² = NA, P = NA) (Supplementary Table 8).

Sleep-Related Outcomes

Sleep-related outcomes was assessed in only one study (n = 140).⁵⁹ The GEP group showed a significant reduction in insomnia severity measured by the ISI (MD = −2.96, 95% CI: −5.18 to −0.74, P = 0.009) (Supplementary Figure 8).

In subgroup analyses of sleep-related outcomes, only one study was available; therefore, comparisons between subgroups and tests for differences among measurement instruments could not be performed. The effect estimate for ISI remained consistent across all predefined subgroup categories, including type of pain (other pain conditions), age group (adults), intervention delivery mode (online), and control group type (no-intervention control), with the GEP group showing lower insomnia severity than the control group (MD = −2.96, 95% CI: −5.18 to −0.74, P = 0.009; I² = NA, P = NA) (Supplementary Table 9).

Pain-Related Anxiety

Pain-related anxiety was assessed in three studies comparing GEP and non-GEP groups using two measurement instruments.^58,59,63 Tests for subgroup differences among measurement instruments showed no statistically significant difference (χ² = 0.16, df = 1, P = 0.69, I² = 0%) (Supplementary Figure 9).

When assessed using the GAD-7, the GEP group showed significantly lower anxiety scores than the non-GEP group (MD = −3.37, 95% CI: −5.04 to −1.70, P<0.0001; I² = NA, P = NA). However, no significant difference was observed between groups when assessed using PASS-20 (MD = −2.30, 95% CI: −7.23 to 2.63, P = 0.36; I² = 69%, P = 0.02) (Supplementary Figure 9).

Subgroup Analysis

Type of Pain

In subgroup analysis by type of pain, no significant difference was observed between the GEP and non-GEP groups in LBP studies assessed using PASS-20 (MD = −2.30, 95% CI: −7.23 to 2.63, P = 0.36; I² = 69%, P = 0.02). In studies involving other pain conditions, the GEP group showed significantly lower anxiety scores than the non-GEP group when assessed using GAD-7 (MD = −3.37, 95% CI: −5.04 to −1.70, P<0.0001; I² = NA, P = NA) (Supplementary Table 10).

Age Group

In subgroup analysis by age group, only adults were included. Tests for subgroup differences among measurement instruments showed no statistically significant difference within the adult subgroup (χ² = 0.16, df = 1, P = 0.69). The findings were consistent with the overall analysis, with no significant difference observed for PASS-20 (MD = −2.30, 95% CI: −7.23 to 2.63, P = 0.36; I² = 69%, P = 0.02), whereas a significant reduction was observed for GAD-7 (MD = −3.37, 95% CI: −5.04 to −1.70, P<0.0001; I² = NA, P = NA) (Supplementary Table 10).

Mode of Intervention Delivery

In subgroup analysis by mode of intervention delivery, no significant difference was observed between the GEP and non-GEP groups in face-to-face interventions assessed using PASS-20 (MD = −2.30, 95% CI: −7.23 to 2.63, P = 0.36; I² = 69%, P = 0.02). In online interventions, the GEP group showed significantly lower anxiety scores than the non-GEP group when assessed using GAD-7 (MD = −3.37, 95% CI: −5.04 to −1.70, P<0.0001; I² = NA, P = NA) (Supplementary Table 10).

Type of Control Group

In subgroup analysis by type of control group, tests for subgroup differences among measurement instruments showed no statistically significant difference within the no-intervention subgroup (χ² = 2.31, df = 1, P = 0.13). Compared with no-intervention controls, the GEP group showed significantly lower anxiety scores on PASS-20 (MD = −13.10, 95% CI: −25.55 to −0.65, P = 0.04; I² = NA, P = NA) and GAD-7 (MD = −3.37, 95% CI: −5.04 to −1.70, P<0.0001; I² = NA, P = NA). In the conventional control subgroup, comparisons among measurement instruments were not performed. No significant difference was observed for PASS-20 when comparing GEP with conventional control groups (MD = −0.29, 95% CI: −5.66 to 5.08, P = 0.92; I² = 69%, P = 0.04) Supplementary Table 10.

Pain-Related Depression

Pain-related depression was assessed in three studies^58,59,63 using three measurement instruments. Tests for subgroup differences among measurement instruments showed a statistically significant difference in treatment effects (χ² = 11.62, df = 1, P = 0.0007). When analyzed by individual instruments, the GEP group showed significantly lower depressive symptom scores than the non-GEP group when assessed using PHQ-9 (MD = −3.45, 95% CI: −5.17 to −1.73, P<0.0001; I² = NA, P = NA). However, no significant difference was observed for HADS (MD = 0.45, 95% CI: −0.98 to 1.88, P = 0.54; I² = 78%, P = 0.003) (Supplementary Figure 10).

Subgroup Analysis

Type of Pain

In subgroup analysis by type of pain, HADS was used in the LBP subgroup, whereas PHQ-9 was used in participants with other pain conditions. As measurement instruments differed concurrently with pain type, the effects of measurement instrument and pain condition could not be separately distinguished. No significant difference was observed between the effects measured using HADS and PHQ-9 across pain-type subgroups (χ² = 2.61, df = 1, P = 0.11). No significant difference was observed between the GEP and non-GEP groups in LBP participants assessed using HADS (MD = 0.45, 95% CI: −0.98 to 1.88, P = 0.54; I² = 78%, P = 0.003). In participants with other pain conditions, the GEP group showed significantly lower depressive symptom scores than the non-GEP group when assessed using PHQ-9 (MD = −3.45, 95% CI: −5.17 to −1.73, P<0.0001; I² = NA, P = NA) (Supplementary Table 11).

Age Group

In subgroup analysis by age group, only adults were included. Tests for subgroup differences among measurement instruments showed a statistically significant difference in treatment effects within the adult subgroup (χ² = 16.29, df = 2, P = 0.0003). The GEP group showed significantly lower depressive symptom scores than the non-GEP group when assessed using PHQ-9 (MD = −3.45, 95% CI: −5.17 to −1.73, P<0.0001; I² = NA, P = NA). However, no significant differences was observed for HADS (MD = 0.45, 95% CI: −0.98 to 1.88, P = 0.54; I² = 78%, P = 0.003) (Supplementary Table 11).

Mode of Intervention Delivery

In subgroup analysis by mode of intervention delivery, HADS was used in the face-to-face subgroup, whereas PHQ-9 was used in the online subgroup. Therefore, the effects of measurement instrument could not be separated from those of intervention delivery mode. No significant difference was observed between the effects measured using HADS and PHQ-9 across intervention delivery subgroups (χ² = 2.61, df = 1, P = 0.11). No significant difference was observed between the GEP and non-GEP groups in face-to-face delivery assessed using HADS (MD = 0.45, 95% CI: −0.98 to 1.88, P = 0.54; I² = 78%, P = 0.003). In online delivery, the GEP group showed significantly lower depressive symptom scores than the non-GEP group when assessed using PHQ-9 (MD = −3.45, 95% CI: −5.17 to −1.73, P<0.0001; I² = NA, P = NA) (Supplementary Table 11).

Type of Control Group

In subgroup analysis by type of control group, no significant difference was observed between the effects measured using HADS and PHQ-9 within the no-intervention subgroup (χ² = 0.35, df = 1, P = 0.56). Compared with no-intervention controls, the GEP group showed significantly lower depressive symptom scores on PHQ-9 (MD = −3.45, 95% CI: −5.17 to −1.73, P<0.0001; I² = NA, P = NA), whereas no significant difference was observed for HADS (MD = −5.10, 95% CI: −10.31 to 0.11, P = 0.06; I² = NA, P = NA). In the conventional control subgroup, no significant difference was observed for HADS (MD = 0.90, 95% CI: −0.59 to 2.39, P = 0.24; I² = 78%, P = 0.01) (Supplementary Table 11).

Pain-Related Distress

Pain-related distress was assessed using the PRS in one study (n = 140).⁵⁹ The GEP group showed significantly lower pain-related distress scores than the non-GEP group (MD = −7.53, 95% CI: −10.27 to −4.79, P<0.00001; I² = NA, P = NA) (Supplementary Figure 11).

In subgroup analyses, only one study was available; therefore, meaningful comparisons between subgroups could not be performed. The effect estimate for PRS remained consistent across all predefined subgroup categories, including type of pain (other pain conditions), age group (adults), intervention delivery mode (online), and control group type (no-intervention control), with the GEP group showing lower pain-related distress scores than the control group (MD = −7.53, 95% CI: −10.27 to −4.79, P<0.00001; I² = NA, P = NA) (Supplementary Table 12).

Pain Self-Efficacy

Pain self-efficacy was assessed in one study,⁶³ which included two control conditions. No significant differences were observed between the GEP and non-GEP groups in either the no-intervention group (MD = −2.70, 95% CI: −9.71 to 4.31, P = 0.45; I² = NA, P = NA) or the conventional control group (MD = −5.20, 95% CI: −14.90 to 4.50, P = 0.29; I² = NA, P = NA) (Supplementary Figure 12). Subgroup analyses by type of pain (LBP only), age group (adults only), and mode of intervention delivery (face-to-face only) included only a single subgroup and therefore did not permit between-subgroup comparisons (Supplementary Table 13).

Coping Strategies

In pain coping strategies assessment, one study⁵⁸ showed a significant reduction in FESV-BW scores in the GEP group compared with the non-GEP group (MD = −12.92, 95% CI: −20.22 to −5.61, P = 0.0005; I² = 0%, P = 0.52) (Supplementary Figure 13; Supplementary Table 14).

Subgroup analyses by type of pain, age group, mode of intervention delivery, and type of control group were not applicable because only one study was available. The included study involved LBP participants, adults, face-to-face intervention delivery, and a conventional control group; therefore, subgroup interaction tests could not be performed.

Daily Activity Level

In daily activity level assessment, two studies^60,61 were included, employing different measurement approaches. Analysis by measurement instrument showed no heterogeneity between the different measurement methods (χ^{2 =} 0.11, df = 1,P = 0.74, I² = 0%).

When analyzed by measurement method, one study⁶⁰ using the RT3 accelerometer (objective measurement of activity counts) found no significant difference (MD = 8579.00, 95% CI: −41,577.28 to 58735.28, P = 0.74; I² = NA, P = NA). Another study using the ADL scale (subjective patient-reported outcome)⁶¹ also showed no significant difference (MD = 1.40, 95% CI: −6.66 to 9.46, P = 0.73; I² = NA, P = NA) (Supplementary Figure 14; Supplementary Table 15).

In subgroup analyses by type of pain, age group, mode of intervention delivery, and type of control group, all included studies involved LBP participants, adults, face-to-face intervention delivery, and conventional control groups, respectively. Therefore, subgroup interaction tests were not performed beyond the measurement-method comparison. The subgroup interaction test was not significant (χ² = 0.11, df = 1, P = 0.74) (Supplementary Table 15).

Sensitivity Analyses

Leave-one-out sensitivity analyses were conducted for PCS and QBPDS, which included sufficient numbers of studies. The direction of effects remained consistent after the removal of individual studies. However, exclusion of Woods 2008⁶³ attenuated the pooled effect estimate for PCS and resulted in loss of statistical significance, suggesting that this study contributed to the precision of the pooled estimate. Detailed results are presented in Supplementary Table 16.

Discussion

This meta-analysis synthesized RCTs comparing the effectiveness of GEP with non-GEP interventions in patients with CMP. The pooled evidence indicated that GEP resulted in greater improvements than non-GEP approaches in disability and pain-related fear and cognition, whereas effects on pain intensity varied across assessment instruments, clinical populations, and comparator interventions. Studies on broader psychosocial outcomes, including anxiety, depression, and quality of life, showed mixed or limited results. Overall, these findings suggest that GEP may not consistently reduce pain intensity compared to non-GEP strategies, but its benefits appear more pronounced for improving functional outcomes and reducing activity-related fear, indicating a predominantly function-oriented therapeutic profile.

Physical and Functional Outcomes: Pain Intensity and Pain-Related Disability

Our meta-analysis revealed that the effects of GEP varied across outcome domains and measurement instruments. Regarding pain intensity, the findings were inconsistent across different measures. Significant improvements were observed with NPS and FIQ-Pain, whereas no significant differences were detected with other measures, including VAS, NRS, BPI, and MPQ. This variability may reflect differences in the dimensions of pain assessed by these instruments, as well as variations in patient populations and clinical characteristics across studies. Because GEP primarily targets pain-related fear and avoidance, its effects on pain intensity may depend on the extent to which these mechanisms contribute to individual pain experiences. Given the heterogeneity of measurement instruments and the limited number of studies contributing to individual outcomes, these findings should be interpreted cautiously. For pain-related disability, GEP showed beneficial effects on several measures, such as the ODI, but not consistently across all instruments, suggesting that its effects may vary according to the functional domains assessed.

Notably, in patients with chronic LBP (CLBP), GEP did not demonstrate statistically significant benefits for either pain or disability, whereas significant effects were observed in patients with fibromyalgia and complex regional pain syndrome. This difference may partly reflect the nature of the comparator interventions used in CLBP trials. Several CLBP studies have included active comparators, such as graded activity and cognitive-behavioral therapy, which may already incorporate components of progressive activity and behavioral modification. Direct comparative trials have also reported similar outcomes between graded exposure protocols and these active treatments in patients with CLBP.^58,60 Therefore, when comparator interventions already address fear-avoidance and promote activity engagement, the additional benefit of formalized graded exposure may be attenuated.

The dissociation between pain and disability outcomes, with improvements in function despite inconsistent effects on pain intensity, is consistent with the fear-avoidance model, in which catastrophic interpretations of pain may contribute to avoidance behaviors and persistent disability.^66–68 GEP may influence this pathway by addressing fear-related beliefs through graded exposure and facilitating engagement in previously avoided activities.⁶⁹ Thus, improvements in physical functioning may occur without a substantial reductions in pain intensity, suggesting that functional recovery may not necessarily depend on pain reduction alone. However, evidence regarding different delivery modalities remains limited. Most of the included studies evaluated face-to-face GEP, whereas only one study examined online delivery; therefore, conclusions regarding the effectiveness of online GEP remain uncertain.

These findings challenge the traditional symptom-focused treatment paradigms and highlight the importance of functional restoration as a therapeutic goal. GEP may be particularly relevant when functional limitations are driven by fear-avoidance mechanisms because systematic exposure may facilitate the resumption of activities despite persistent pain. However, in CLBP, where graded activity may already incorporate elements of graded exposure,^60,66 the additional benefit of formalized GEP may be more limited.

Pain-Related Fear and Cognition: Catastrophizing, Kinesiophobia, Fear-Avoidance Beliefs, Fear of Pain, and Perceived Harmfulness of Activities

GEP significantly reduced pain catastrophizing, kinesiophobia, fear-avoidance beliefs, and perceived harmfulness compared to non-GEP interventions, indicating beneficial effects across several activity-related cognitive-emotional domains. However, these effects appeared to be domain-specific, with improvements observed mainly in activity-related fear cognitions rather than general fear of pain. The greatest effects were observed in studies involving populations with fibromyalgia and CRPS populations.

These findings are consistent with predictions from the fear-avoidance model.^66–68 GEP operates through systematic behavioral experiments in which patients engage in feared movements or activities, allowing them to test and revise catastrophic expectations regarding activity-related harm.^4,10 Through repeated exposure, expectancy violations may occur when feared activities do not result in anticipated catastrophic outcomes, potentially modifying cognitive appraisals and emotional responses through inhibitory learning mechanisms.^4,70 Importantly, this process primarily targets beliefs about activity consequences rather than fear of pain sensations, which may explain the observed specificity of effects. This interpretation is consistent with previous research showing that exposure-based interventions may produce greater improvements in movement-related fear than in broader psychological distress.^62,71 The relatively stronger effects observed in fibromyalgia and CRPS may reflect the relevance of fear-avoidance mechanisms in conditions where protective avoidance behaviors contribute substantially to functional limitations.^66–68 These cognitive-emotional improvements may also contribute to functional gains observed in disability outcomes, as reductions in catastrophizing and kinesiophobia may facilitate engagement in previously avoided activities and help disrupt the cycle of avoidance that perpetuates disability.^69,72,73

From a clinical perspective, GEP may be particularly relevant for patients with prominent activity-related fear and catastrophizing, and treatment selection should be guided by the assessment of these mechanisms. For broader concerns, such as work-related fears or fear of pain sensations, GEP may require combination with interventions targeting these specific cognitive-emotional processes. GEP may be most applicable to conditions such as fibromyalgia and CRPS, in which fear-avoidance mechanisms appear to play a prominent role.

Psychological Outcomes: Anxiety, Depression, and Pain-Related Distress

GEP demonstrated variable effects on broader psychological outcomes, with inconsistent findings across measures of anxiety and depression. Significant improvements were observed in some instruments, whereas no significant differences were detected in others, suggesting that the effects of GEP may depend on the specific psychological constructs assessed. These findings may reflect the mechanistic focus of GEP, which primarily uses behavioral experiments to expose patients to feared movements and modify activity-related threat beliefs.^10,74 Such processes may have more direct effects on fear-related cognitions than on broader affective symptoms associated with anxiety and depression.⁷⁵ Previous research suggests that integrated approaches combining cognitive restructuring, behavioral activation, and other psychological strategies may be beneficial for addressing comorbid anxiety and depressive symptoms in chronic pain populations.^76,77 Clinically, for patients with substantial mood disturbances beyond activity-related fears, GEP may be complemented by evidence-based interventions specifically targeting anxiety and depression.

Quality of Life and Sleep-Related Outcomes

GEP showed variable effects across quality-of-life domains and limited evidence regarding sleep-related outcomes. Improvements were observed in physical health-related quality of life measured by SF-36-PCS, as well as overall quality of life assessed by the BBQ, whereas mental health-related quality of life measured by SF-36-MCS showed a non-significant trend. This pattern may reflect the primary therapeutic focus of GEP, as systematic exposure to feared movements may facilitate activity engagement and reduce activity-related disability,^78,79 potentially improving physical functioning without comprehensively affecting broader aspects of psychological well-being.⁷⁵ Evidence regarding sleep-related outcomes remains preliminary, as only one study included these data and has demonstrated modest improvements. GEP may potentially influence sleep indirectly through improvements in pain-related distress and daytime activity;^80–82 however, sleep disturbances are multifactorial, and targeted interventions such as cognitive-behavioral therapy for insomnia may be required for patients with persistent sleep problems. Clinically, these findings suggest that GEP may be particularly relevant for improving physical functioning, whereas patients with substantial mental health or sleep-related difficulties may benefit from integrated multimodal approaches.

Evidence Certainty and Clinical Interpretation

Despite improvements observed across several outcomes, the certainty of evidence was generally limited, with most outcomes rated as having low or very low certainty. Downgrading was primarily attributable to the risk of bias, reflecting the methodological limitations common in behavioral intervention trials, and imprecision due to small sample sizes and wide CIs. Clinical relevance was further considered by comparing observed effects with available MID thresholds. Although several outcomes demonstrated statistically significant differences, the probability of achieving clinically meaningful improvements varied across outcome measures and was influenced by the certainty of the underlying evidence. Sensitivity analyses were performed only for outcomes with sufficient numbers of studies. Most leave-one-out analyses showed consistent effect estimates; however, the PCS effect was attenuated and lost statistical significance after removal of one study, indicating that this finding should be interpreted with caution. Overall, current evidence suggests that GEP may provide benefits across several outcome domains; however, the limited certainty, small evidence base, and variable robustness of findings require cautious interpretation.

Strengths

This systematic review and meta-analysis provided a comprehensive quantitative synthesis of evidence comparing GEP and non-GEP approaches in individuals with CMP. We conducted systematic searches across multiple databases to identify relevant studies and minimize the potential risk of publication bias. In addition to pain intensity and disability, this review examined a broad range of clinically relevant outcomes, including pain-related fear and cognition, psychological outcomes, quality of life, sleep-related outcomes, and behavioral measures, to provide a more comprehensive assessment of the effects of GEP. We further explored the potential sources of heterogeneity through pre-specified subgroup analyses based on pain condition type, age group, intervention delivery mode, and type of control group. These findings may provide clinically relevant insights into the potential role of GEP and may help inform future research directions, guideline development, and evidence-based clinical decision-making for CMP management.

Limitations

Despite the strengths of this meta-analysis, several limitations should be considered. First, the included studies varied in terms of patient populations, pain conditions, intervention characteristics, and outcome measures. Additional variability in factors such as baseline fear levels, therapist characteristics, and treatment adherence may also have influenced treatment effects, although these factors have not been consistently reported across studies. Such clinical and methodological differences may have contributed to the heterogeneity, and residual variability may remain despite the use of pre-specified subgroup analyses. Some subgroup analyses were based on a limited number of studies or participants, with certain control categories represented by only one study; therefore, these findings should be considered exploratory and interpreted cautiously.

Second, most included studies evaluated face-to-face GEP, whereas evidence regarding online delivery was limited. Future trials should directly compare different delivery modalities to determine whether treatment differ according to intervention format. In addition, several studies had concerns regarding risk of bias in outcome measurement, largely reflecting the subjective nature of patient-reported outcomes and the practical challenges of blinding participants, therapists, and outcome assessors in behavioral interventions.

Finally, although the review was registered in PROSPERO, the registration was completed retrospectively after the review had commenced. Therefore, the potential benefits of prospective registration in reducing selective reporting and protocol deviations may have been limited.

Future Research Directions

Future research should address several important gaps in the evidence base for GEP in CMP. First, larger, well-designed trials are needed for underrepresented pain conditions, including neuropathic pain and headache disorders, to determine whether the effects observed in current populations are generalizable across different pain conditions. Second, a systematic investigation of exposure parameters, including intensity, duration, frequency, and hierarchy design, is needed to better understand the active components of GEP and optimize intervention protocols. Comparative studies should also explore whether the combination of GEP with other evidence-based interventions targeting psychological or pain-related processes provides additional benefits.

Digital technologies may offer new opportunities to enhance GEP delivery and patient engagement. Virtual reality and augmented reality may provide immersive exposure environments,^83–85 whereas mobile health applications and wearable devices may facilitate real-time monitoring and personalized feedback.⁸⁶ AI-based approaches may potentially support treatment personalization by predicting treatment response; however, their clinical utility requires rigorous empirical evaluation. Finally, longer-term follow-up studies (≥12 months) are needed because current evidence is largely limited to short-term outcomes. These studies should evaluate the durability of the treatment effects, relapse prevention, and sustained functional improvements. Such evidence will help clarify the role of GEP within comprehensive chronic pain management and inform future evidence-based clinical guidelines.

Conclusions

This meta-analysis suggests that GEP may improve functional disability and fear-related cognitive outcomes in patients with CMP, including reductions in pain catastrophizing, kinesiophobia, fear-avoidance beliefs, and perceived harmfulness. However, the certainty of the evidence remains limited because of concerns related to the risk of bias, imprecision, and heterogeneity across interventions and outcome measures. The Effects of GEP on pain intensity and broader psychological or sleep-related outcomes were inconsistent. Subgroup analyses suggested that more consistent effects were observed in fibromyalgia and complex regional pain syndrome, whereas evidence in chronic low back pain remained less consistent. GEP may be particularly relevant for patients with prominent fear-avoidance mechanisms; however, further high-quality RCTs with standardized outcome assessments, larger sample sizes, and longer-term follow-up are needed to clarify its effectiveness and clinical applicability.

Funding Statement

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Systematic Review Registration

This systematic review was registered in PROSPERO (CRD420261469624).

Abbreviations

ADL Scale, Activities of Daily Living Scale; BBQ, Brunnsviken Brief Quality of Life Scale; BPI, Brief Pain Inventory; CBT, Cognitive Behavioral Therapy; CI, Confidence Interval; FABQ, Fear-Avoidance Beliefs Questionnaire; FABQ-PA, Fear-Avoidance Beliefs Questionnaire–Physical Activity subscale; FDI, Functional Disability Inventory; FESV_BW, Pain Processing and Coping Questionnaire; FIQ-Pain, Fibromyalgia Impact Questionnaire Pain subscale; FOPQ-C, Fear of Pain Questionnaire-child version; GAD-7, 7-item Generalized Anxiety Disorder scale; GE, Graded Exercise; GEP, Graded Exposure Therapy; GRADE, Grading of Recommendations Assessment, Development and Evaluation; HADS, Hospital Anxiety and Depression Scale; ISI, Insomnia Severity Index; MD, Mean difference; MID, Minimal important difference; MPQ, McGill Pain Questionnaire; MT, Manual Therapy; NA, Not Available; NPS, Neuropathic Pain Scale; NRS, Numerical Rating Scale; ODI, Oswestry Disability Index; ÖMPQ, Örebro Musculoskeletal Pain Questionnaire; PASS-20, Pain Anxiety Symptoms Scale; PCS, Pain Catastrophizing Scale; PCS-C, Pain Catastrophizing Scale–Child Version; PDI, Pain Disability Index; PHODA, Photograph Series of Daily Activities; PHODA-YE, Photograph Series of Daily Activities-Youth English; PHQ-9, Patient Health Questionnaire-9; PIPS, Psychological Inflexibility in Pain Scale; PNE, Pain Neuroscience Education; PRS, Pain Reactivity Scale; PSEQ, Pain Self-Efficacy Questionnaire; QBPDS, Quebec Back Pain Disability Scale; RASQ, Radboud Skills Questionnaire; RCT, Randomized controlled trial; RoB 2.0, Risk of Bias tool version 2.0; RT3, Tri-axial Research Tracker; SF-MPQ, Short-Form McGill Pain Questionnaire; SF-36, 36-Item Short Form Health Survey; TAU, Treat As Usual; TSK, Tampa Scale of Kinesiophobia; VAS, Visual Analog Scale; WAQ, Walking Ability Questionnaire; WHODAS 2.0-12, 12-item World Health Organization Disability Assessment Schedule 2.0; WLC, Waiting List Control.

Data Sharing Statement

All data generated details are available from the corresponding author upon request.

Human Ethics and Consent to Participate Declarations

The ethics statement is not applicable.

Author Contributions

All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

The authors report no conflicts of interest in this work.