Effectiveness of physical therapeutic scoliosis exercise (PSSE) intervention for adolescent idiopathic scoliosis: a systematic review and meta-analysis

Abstract

Objective

To systematically evaluate the effects of physiotherapeutic scoliosis-specific exercise (PSSE) on Cobb angle, angle of trunk rotation (ATR) and quality of life (QoL) in patients with adolescent idiopathic scoliosis (AIS), and to explore the modifying role of exercise period, thereby providing an evidence-based foundation for the clinical optimisation of exercise prescriptions.

Methods

Following PRISMA guidelines, the EMBASE, PubMed, Cochrane Library, and SinoMed databases were searched (up to 15 June 2025) without language restriction; a total of 18 RCTs (involving 1,014 patients with AIS, mean age 13.98 ± 1.81 years) were included. Risk of bias was assessed using the Cochrane RoB 2 (11 low-risk, 7 medium-risk). Standardised mean differences (SMDs) and 95% confidence intervals (CIs) were calculated using a random-effects model, and the frequency of exercise was quantified by subgroup analyses (low: ≤2 times/week; high: ≥3 times/week).

Results

The combined effect of PSSE on Cobb angle in patients with AIS had an SMD of − 1.00, 95% confidence area [− 1.47, − 0.53], P < 0.001; the combined SMD of ATR was − 1.15, 95% CI [− 1.94, − 0.36], P < 0.001; the combined SMD of QoL was 0.63, 95% CI [0.15, 1.11], P < 0.01. Subgroup analyses showed that high-frequency exercise practice (≥ 3 times/week) was more effective.

Conclusion

PSSE effectively improves Cobb angle, ATR, and QoL in patients with AIS. High-frequency exercise forms are recommended; however, the actual application needs to be individualised based on patient characteristics. Spreading RCTs are recommended to further define the optimal exercise dose and identify key influencing factors.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12891-025-09218-2.

Introduction

Scoliosis is a three-dimensional deviation of the spinal axis that manifests as abnormal coronal curves, fixed vertebral rotations, and changes in sagittal physiological curvature [1]. Adolescent idiopathic scoliosis (AIS) is the most common type of scoliosis [2], with a prevalence of 0.47–5.2% [3]. It is referred to as “idiopathic” due to its unknown aetiology and is characterised by a scoliosis with a Cobb angle greater than 10°, occurring in adolescents aged 13 to 18 years, with a predominance in young females. In untreated adolescents with idiopathic scoliosis, the condition may continue to progress after skeletal maturity, potentially resulting in back pain, pulmonary dysfunction, and even self-image issues [4–7], all of which significantly affect quality of life (QoL) [8].

Currently, treatment methods for AIS mainly include surgical and non-surgical approaches. Surgical treatment, although highly effective, is associated with high trauma, high risk, difficulty, and postoperative complications. Common non-surgical treatments include standard care (e.g., observation or use of braces) and exercise therapy. According to recommendations from the Scoliosis Research Society (SRS), non-surgical treatment can be used when the scoliosis angle is less than 45°. For patients with moderate scoliosis and a Risser sign of 0–2, bracing is usually required [9]. Several studies have shown that physical activity (PA) is positively associated with AIS [10, 11]. In addition, other studies have shown that exercise therapy has a positive effect on both physiological and psychological aspects of patients with AIS [12, 13]. PSSE is a specific exercise programme tailored to the location, degree, and clinical characteristics of a patient’s scoliosis [14]. It is defined as “a curve-specific exercise programme individually adapted to the patient’s clinical characteristics” [15]. The Society of Scoliosis Orthopaedics and Rehabilitation Treatment (SOSORT) recommends PSSE as the first step in the treatment of mild AIS [14]. Its standard features include three-dimensional self-correction, training in activities of daily living, and stabilisation of corrected posture [16]. PSSE was endorsed by the International SOSORT in 2011 and is considered one of the most important treatments for AIS [17]. In recent years, studies have shown that PSSE therapy has long-term effects in reducing the Cobb angle in patients with AIS [18]. Clinical evaluation and imaging are the primary screening methods for scoliosis, with Cobb angle [19] and ATR [20] being important indicators for assessing the effectiveness of interventions in patients with scoliosis. The present meta-analysis also employed the QoL score as an assessment of AIS, which not only affects the physical appearance of patients but also leads to psychological problems such as low self-esteem and anxiety due to the change in appearance. The QoL score allows for a comprehensive evaluation of the impact on the patient’s life across multiple dimensions, including physiological and psychological aspects. The SRS-22 scale, known for its high degree of reliability and cultural relevance, is widely used in evaluating the outcomes of non-surgical treatment of AIS. Its usage rate has reached 40.5%; therefore, the SRS-22 was selected as the QoL assessment tool in this study.

In recent years, treatment strategies for AIS have been optimised through technological advancements, and conservative treatments have been used far more often than surgical ones [21]. As a result, physiotherapy and bracing have become the main interventions. During the course of our study, we found that Schroth therapy (Schroth), as a representative form of physiotherapeutic scoliosis-specific exercise (PSSE) for scoliosis, is based on the classical principles of physiotherapy [22]. The Schroth method was founded in 1920 by Katharina Schroth and has since been continuously optimised, with approximately 3,000 patients with scoliosis treated per year [16]. Its effectiveness has been validated in several meta-analyses [23]. This method not only improves spinal deformity but also strengthens the pulmonary function and respiratory muscle strength of the patient. Furthermore, Schroth-based PSSE combined with a 3D-printed brace has proven more effective in improving the primary curvature Cobb angle, trunk rotation angle, coronal imbalance, patient satisfaction, and quality of life compared with the use of a 3D-printed brace alone [24]. Although several reviews and meta-analyses have been published on AIS, Dimitrijević et al. (2022) [1] Specifically evaluated the effectiveness of Schroth therapy on all three core scoliosis indices, providing high-level evidence for supporting non-surgical treatment of AIS. Schroth therapy has been presented as a potential alternative or complement to brace therapy, particularly effective in improving quality of life and mild deformity, aligning with the conservative treatment goals of the SOSORT clinical guidelines. However, the review did not include a funnel plot or Egger’s test, which may have led to the omission of studies with negative outcomes. In 2024, Zhenghui Wang et al. [25] proposed a network meta-analysis incorporating Schroth, SEAS, active self-correction (active self-correction), side-shift training (side-shift), FITS and DoboMed, addressing the limitation that traditional meta-analyses do not allow for multidirectional comparisons. However, the reported optimal improvement in Cobb angle with SEAS was based on only five studies, making the clinical significance uncertain. For the following reasons, there remains a need to review and summarise the most recent literature: First, to update high-quality evidence: this study integrates new literature published in the last three years, significantly expanding the inclusion size of high-quality randomised controlled trials (RCTs) and enhancing the timeliness and reliability of the conclusions. Second, to expand the scope of the study, most existing reviews focus primarily on Schroth therapy, and there are insufficient systematic assessments of other PSSEs. This paper broadens the focus on all PSSE interventions for meta-analysis. Third, cross-language literature search: the search strategy did not set a language restriction, and a total of English, Chinese, and Korean literature was included to minimise publication bias. Fourth, to evaluate exercise dose-effect analysis: exercise frequency was innovatively used as a subgroup variable to quantitatively assess its effects of different frequency regimens on the three core outcome indicators (Cobb angle, trunk rotation angle, and quality of life) of adolescent idiopathic scoliosis (AIS).

In summary, the aim of this study was to systematically evaluate the effects of PSSE on Cobb angle, ATR, and QoL in patients with AIS through meta-analysis. It also investigates the effects of different exercise doses on these outcomes, with the aim of providing clinicians with an evidence-based recommendation for the “optimal exercise frequency” (OEF).

Methodology

Study design

This research protocol has been registered with the PROSPERO platform (registration number: CRD42024607029 at https://www.crd.york.ac.uk/PROSPERO/. The information described herein is consistent with the original registration. During study implementation, we strictly followed the Preferred Reporting Items for Systematic Evaluation and Meta-Analyses (PRISMA) guidelines [24] and referred to PRISMA recommendations specific to the fields of exercise, rehabilitation, sports medicine, and sports science for study conduct and reporting.

Study inclusion and exclusion criteria

This study adhered to the PICOS principles recommended by the PRISMA guidelines [26], with clearly defined inclusion and exclusion criteria. The included intervention studies were randomised controlled trials (RCTs) that examined the effects of physiotherapeutic scoliosis-specific exercise (PSSE) on Cobb angle, ATR, and QoL in patients with adolescent idiopathic scoliosis (AIS).

Population (P): Adolescent patients aged 10–18 years with a diagnosis of idiopathic scoliosis and a lateral curvature of the spine (Cobb angle) greater than 10° were included. Cases of non-idiopathic scoliosis were excluded.

Intervention (I): The intervention was PSSE. The main genres of PSSE were Schroth, Lyon, SEAS, Dobomed, BSPTS, SideShift, FITS, and FED. Twenty-two studies were included that employed PSSE therapeutic interventions, as well as combined therapies of PSSE with exercises such as Pilates, traditional exercises, and ball and sling. Studies that did not use PSSE bases were excluded.

Control (C): The control groups included in this study were treated with non-surgical physiotherapy, including the non-intervention group (watchful waiting), usual care, and other treatments (e.g., bracing). Specifically, the control groups included standard care, exercises combining balls and slings, spinal orthoses made by 3D printing technology, manipulative orthopaedic massage therapy, standard therapeutic training, brace therapy, and observational studies without any intervention.

Outcome (O): The primary outcome measures were Cobb angle, ATR, and QoL. Cobb angle was measured using radiographs, ATR was measured with a Scoliometer during the Adam’s Forward Flexion Trial, and ATR was measured by photogrammetry, and QoL was evaluated using the Society for the Study of Scoliosis-22 (SRS-22) questionnaire, SRS-22r questionnaire, and SRS-23 assessment. Currently, the main scales used for the assessment of quality of life in patients with AIS include the SRS-22 scale and the EuroQol 5-dimension (EQ-5D) questionnaire. The 3-level version of the EQ-5D is one of the most widely used tools for measuring QoL [27], which assesses the health status of an individual through the five dimensions of mobility, self-care, activities of daily living, pain/discomfort, and anxiety/depression. Assessing personal health status, each dimension contains three options that reflect the respondent’s ability to manage each dimension [28]. The SRS-22 scale, on the other hand, includes five dimensions: function/mobility, pain, self-image/appearance, mental health, and treatment satisfaction, comprising a total of 22 entries [29]. If at least one outcome indicator met the inclusion criteria, the study was included in the review.

Study design (S): Randomised controlled trials (RCTs) were included in this study.

The following exclusion criteria were applied: (1) reviews, conference reports, commentaries, and prospective studies; (2) repetitively published studies; (3) studies with incomplete raw data, studies that did not assess any outcome metrics, or studies from which outcome data could not be extracted; (4) studies that did not set up a control group; (5) studies of non-physical interventions; (6) studies involving surgical treatments; and (7) studies from which full text could not be obtained.

Search strategy

Five databases—EMBASE, PubMed, Cochrane Library, and SinoMed—were searched, covering the period from the inception of each database to 15 June 2025. The search strategy used a combination of keywords, including “adolescent idiopathic scoliosis”, “PSSE”, “physical therapeutic scoliosis”, “scoliosis”, “Schroth”, “Lyon”, “SEAS”, “Dobomed”, “BSPTS”, “SideShift”, “FITS”, “FED”, and related terms. Specific search strategies for each database are provided in the Supplementary Information (SI). Database searches were conducted independently by two reviewers (JQQ and ZBY). To avoid missing eligible studies, we also screened the reference lists of the included literature and relevant reviews.

Study selection and data extraction

Two researchers (JQQ and ZBY) independently extracted the data, and a third researcher (WH) resolved any disagreements. To effectively manage the included literature, EndNote software (version X9) was used, and duplicate studies were excluded.

Two researchers (JQQ and ZBY) independently screened the titles and abstracts of the literature. For studies with unclear relevance, the full text was reviewed. Any disagreements during the screening process were discussed and resolved in a meeting chaired by the third author (WH). In parallel, excluded studies were documented along with reasons for their exclusion.

In the included literature, the researchers extracted several key data points, including the first author’s name, year of publication, mean age of the participants, pre-treatment Cobb angle, ATR, Risser sign, type of scoliosis, sample size, details of the intervention, duration of the intervention, and outcome measures. Data extraction strictly followed predefined study grouping criteria.

To obtain information that might be missing from the literature, the researchers actively contacted the relevant authors via email. However, for literature that was unable to obtain key information or lacked important data, the research team decided to exclude them from subsequent in-depth analyses to ensure the accuracy and reliability of the study.

Methodological quality assessment

The quality assessment and risk of bias of this study were assessed by two independent authors (JQQ and ZBY). In cases of disagreement, a third author (WH) was consulted. The assessment was conducted using the Cochrane Risk of Bias 2 (RoB2) tool for randomised trials with RevMan 5.4 software. The RoB2 tool systematically assessed the risk of bias for each study from the following five domains: risk of bias in the randomisation process, risk of bias against interventions, risk of bias against missing data, risk of bias in the measurement of outcomes, and risk of bias in the reporting of outcomes. Each domain made risk judgments for articles based on predefined questions, and the risk of bias was graded as “low risk”, “some concern”, and “high risk”. The final risk rating for each study was based on the combined assessment of the five areas.

The results of the evaluation were categorised as “low”, “high”, or “some problems”. A study was judged to be “low” if all items were rated as low risk, “high” if any item was high risk, and “some problems” otherwise. Studies were required to meet rigour requirements for all core assessment criteria, including, random sequence generation: verifiable methods such as computer randomisation or random number tables were used; allocation concealment: mechanisms such as central randomisation or use of sealed envelopes to ensure that the grouping could not be predicted by the researcher; blinding: subjects, researchers and outcome assessors were blinded and blinding was not compromised; and data completeness: data for key endpoints were missing at ≤ 10% and intention-to-treat was used; selective reporting: prespecified outcome indicators were fully reported; and other biases: no major design flaws (e.g., uncorrected baseline imbalances). A study was considered to be at “low risk” if all of the above criteria were met. Studies with serious flaws in any of the core categories, e.g., randomisation was pseudo-randomised by date of birth, medical record number, etc.; lack of allocation concealment or only verbal assignment; blinding was not performed and outcomes were susceptible to subjective influences; data were missing in >20% of cases or inappropriate analyses were used; key outcomes were concealed and not reported; there were conflicts of interest or serious design errors. The presence of any of these issues was considered “high risk” [30].

Statistical analyses

All statistical analyses were conducted using STATA version 15.0 software. Mean changes in Cobb angle, ATR, and QoL between the intervention and control groups before and after the trial were primarily analysed to evaluate the effectiveness of the intervention in improving AIS outcomes. The standardised mean difference (SMD) was used as the effect size in this meta-analysis. Data were based on the mean changes before and after the experiment in the control and intervention groups, and the standard deviation values were standardised to the post-test data. A random effects model was used to estimate the overall effect size, as a high degree of quantitative variation in the I² statistic between studies [31]. A random effects model should be used due to I² >50%. Meta-analysis was performed in STATA 15 using the “metan” package to calculate SMD and generate forest plots; sensitivity analysis using the “metaninf” package; publication bias analysis using the “metafunnel” package to generate funnel plots; and Egger’s test using the “metabias” package. Subgroup analyses based on exercise dose were conducted using the “by()” statement in the “metan” package. This study hypothesised that heterogeneity may be related to exercise frequency. Exercise regimens conducted ≤ 2 times/week were categorised as a low-frequency group (group 1), while those conducted ≥ 3 times/week were categorised as high-frequency (group 2) [32].

Results

Study selection and characteristics

A total of 559 duplicate studies were immediately excluded from the 1,486 articles retrieved from the five databases. Of the remaining 927 studies, 767 irrelevant studies were excluded after screening titles and abstracts for relevance. Following full-text review of the remaining 160 studies, 38 were excluded for not meeting the inclusion criteria Ultimately, 22 studies were included in this systematic review and meta-analysis. Among these, 18 articles met the final inclusion criteria for the meta-analysis after quality assessment using the Cochrane Risk of Bias 2 (RoB2) for randomised trials. The study selection process is illustrated in Fig. 1.

Fig. 1.

PRISMA 2020 flowchart depicting records identified, screened, and included after the final selection criteria for the current meta-analysis

Table 1 presents the characteristics of the studies included in the meta-analysis.

Table 1.

Characteristics of the study

study	age		N		Type	Subject Type				Program Type		Intensity	Outcome
	E	C	E	C		Risser Sign	Cobb Angle	ATR	QoL	E	C
Gichul Kim2016 [33]	15.6 ± 1.1	15.3 ± 0.8	12	12	NA	NA	SEG (23.63 ± 1.5)	NA	NA	Schroth exercise	Pilates	60 m,3t/1w,12w	Cobb angle
							PEG (24.0 ± 2.6)
Hikmet Kocaman2021 [34]	14.07 ± 2.37	14.21 + 2.19	14	14	LenKe Ⅰ,RT = 3, LT = 5, RTLL = 6 VS.LenKe Ⅰ, RT = 3, RT = 5, RTLL = 6	1.64 + 1.34	SG Cobb-T (17.64 ± 4.01), Cobb-L (15.80 ± 3.42)	SG ATR-T (15.17 ± 4.02), ATR-L (4.29 ± 2.73)	SG (3.49 ± 0.13)	Schroth exercise	traditional exercises	90 m,3t/w,10w	Cobb angle, ATR, QoL (SRS-22)
							CG Cobb-T (17.29 ± 3.45), Cobb-L (15.17 ± 4.02)	CG ATR-T (8.43 ± 2.50), ATR-L (4.43 ± 2.38)	CG (3.48 ± 0.24)
Jung-Hyun Kim2015 [62]	15.04 ± 2.42	15.72 ± 2.06	15	15	NA	NA	Schroth (18.01 ± 4.29) Ball & sling (14.98 ± 3.66)	NA	NA	Schroth exercise	Ball and sling exercise	90 m,3t/w,8w	Cobb angle
LIANG Junhao2023 [35]	14.25 ± 2.15	14.89 ± 2.02	55	55	NA	NA	CG(24.41 ± 1.47) EG(24.41 ± 1.47)	CG(5.25 ± 1.23) EG(5.32 ± 1.20)	NA	Combined application of Schroth orthopaedic gymnastics on the basis of the control group	3D Printed Spinal Orthoses Combined with Conventional Rehabilitation	90 m,3t/w,6 m	Cobb angle, ATR
LU Yuelun2022 [36]	13.8 ± 2.1	13.7 ± 3.9	20	20	NA	NA	EG (14°±4°) CG (15°±5°)	EG (5°±2°) CG (5°±2°)	NA	Schroth therapy combined with orthopaedic massage	Chinese osteopathy	10 m,1t/1w,6 m	Cobb angle, ATR
Pil-Neo HwangBo2016 [63]	18.14 ± 1.60	18.88 ± 1.55	8	8	NA	NA	SEG (22.07 ± 6.81) PEG (21.20 ± 3.95)	Not available	NA	Schroth exercise	Pilates exercise	3t/1w,12w	Cobb angle
R.A. MOHAMED2021 [37]	14.50 ± 1.20	14.90 ± 1.40	17	17	NA	NA	SG (20.42 ± 2.57) PG (20.21 ± 2.80)	SG (8.05 ± 0.65) PG (8.29 ± 0.68)	NA	Schroth exercise	proprioceptive neuromuscular facilitation	60 m,3t/1w,6 m	Cobb angle, ATR
Sanja Schreiber2015 [64]	13.5	13.3	25	25	NA	SG (1.76 (1.10 to 2.45))	Not available	NA	SG (4.25 (4.09 to 4.40)) CGl (4.14 (3.96 to 4.31))	Schroth exercises added to standard care	Standard care	5t/1 h,2w, 1 h/1w + 30–45 m/1d, 6 m	QoL (SRS-22)
						CG (1.44 (0.77 to 2.11))
Sanja Schreiber2016 [38]	13.5	13.3	25	25	NA	Schroth (1.76 (1.10 to 2.45))	Schroth + standard of care (29.1 ± 8.9) Standard of care (27.9 ± 8.8)	NA	NA	Schroth exercises added to standard care	Standard care	5t/1 h,2w, 1 h/1w + 30–45 m/1d, 6 m	Cobb angle
						Control (1.44 (0.77 to 2.11))
Woo-Jin Lee2017 [65]	11.00 ± 1.03	10.30 ± 1.15	14	14	NA	NA	SSE (14.26 ± 1.09) TSE (13.61 ± 1.88)	NA	NA	side shift exercise	trunk stabilization exercise	3t/1w,8w,24t	Cobb angle
Akyurek, Elcin2022 [47]	13.86 ± 1.86 (11–17)	13.86 ± 1.86 (11–17)	14	14	NA	PG (2.50 ± 1.65) CG (2.71 ± 2.05)	PG (9.47 ± 8.02) CG(22.14 ± 6.3)	PG (5.53 ± 3.09) CG(6.00 ± 3.53)	NA	PSSE group	Control group	45–60 min, 2/wk, 8w	ATR, JR Error, POTSI, ATSI, dWRVAS
Alves de Araújo, M. E.2012 [39]	21.5 ± 3.5(18–25)	21.5 ± 3.5(18–25)	20	11	NA	NA	EG (7.6°±3.5) CG (7.1°±2.8)	NA	NA	Pilates	No therapeutic intervention	2t/1w,3 m 60 min	Cobb angle
Dufvenberg, M.2025 [40]	12.6 ± 1.4	12.6 ± 1.5	45	45	NA	NA	SSE (31.1 ± 4.1) PA (31.2 ± 4.4)	SSE (10.8 ± 3.3) PA (11.3 ± 3.1)	NA	Scoliosis specific exercise (SSE)	Scoliosis specific exercise (SSE)	30 + 30/d,+1t/1 h,3 m	Cobb angle, ATR
Gao, C.2019 [66]	12.22 ± 1.3	12.13 ± 1.2	23	22	NA	NA	OE (29.13 ± 4.32) OI (28.64 ± 3.91)	NA	NA	Scoliosis specific exercise (SSE) + orthotic	orthotic	10–15 m/1t	Cobb angle
Kamel, M. I.2022 [41]	12.91 ± 1.40	13.00 ± 1.71	30	30	NA	NA	CS + KT (19.29 ± 2.32) CS (18.46 ± 1.98)	NA	NA	Kinesio Taping + Core Stabilization	Core Stabilization	60 m,3t/1w,8w	Cobb angle
Kisa, E. P.2024 [42]	11.76 ± 2.26	11.40 ± 2.73	25	25	NA	3D exercise(2.56 ± 1.38) conventional exercise༈2.38 ± 1.20༉	3D exercise(16.96 ± 3.12) conventional exercise༈16.48 ± 5.73༉	3D exercise(6.82 ± 1.48) conventional exercise༈6.56 ± 1.93༉	NA	3D exercise	conventional exercise	50–70 m,3t/1w,24w	Cobb angle, ATR
Kuru, Tugba2016 [48]	12.9 ± 1.4	12.8 ± 1.2	15	15	NA	NA	Supervised exercise (33.4 ± 8.9) Control group (30.3 ± 6.6)	Supervised exercise (11.9 ± 5.2) Control group (8.4 ± 2.96)	SRS-23 Supervised exercise (0.16 ± 0.18) Control group (− 0.04 ± 0.13)	​Schroth Clinical Exercise	Traditional Exercises	60–90 min, 3t/w, 6w + 18wTraditional Exercises	ATR, QoL
Monticone, M.2014 [43]	12.5 ± 1.1	12.4 ± 1.1	55	55	NA	NA	EG (19.3 ± 3.9) CG (19.2 ± 2.5)	EG (7.1 ± 1.4) CG (6.9 ± 1.3)	NA	Active self-correction + task-oriented exercises + cognitive education	Conventional spinal exercises	60 m, 1t/1w, 42 m(clinic) +30 m, 2t/1w, 42 m༈home༉	Cobb angle, ATR
Shen, X.2023 [44]	13.75 ± 1.02	13.47 ± 1.01	30	29	NA	intervention group(2.67 ± 0.76)Control group༈2.93 ± 1.03༉	intervention group(19.90 ± 2.35)Control group༈20.86 ± 1.57༉	intervention group(7.73 ± 0.91)Control group༈8.07 ± 0.89༉	SRS-22 intervention group(4.04 ± 0.32)Control group༈4.06 ± 0.3༉	Balance training combined with Schroth therapy	Schroth’s 3-dimensional exercises	90 min, 3t/w, 6w	Cobb angle, ATR, QoL
Sun, X. L.2023 [45]	12.6 ± 2.2	13.6 ± 1.9	10	8	Rigo A	0—4	EG (5.5 ± 5.6) CG (18.9 ± 4.2)	EG (​6.1 ± 6.07​) CG (​8.35 ± 6.89​)	NA	PSSE + Apparatus Assisted Release Therapy	appliance-assisted relaxation therapy	24w, >3 h/w	Cobb angle, ATR, vital capacity
Yagci, G.2019 [46]	14.2 ± 1.2	14.0 ± 1.3	15	15	NA	2—3	CS (30.0 ± 9.3) SEAS (27.6 ± 8.0)	CS (10.7 ± 5.4) SEAS (9.6 ± 4.6)	CS (4.0 ± 0.5) CEAS (4.1 ± 0.4)	scientific exercise approach to scoliosis	core stabilization exercise	16w, 20–40 min, 7/wk	Cobb angle, ATR, QoL (SRS), WRVAS, POTSI
Zapata, K. A.2023 [67]	11.6 ± 1.1	12.5 ± 1.4	34	19	EG (5 thoracic, 6 thoracolumbar, 14 lumbar) CG (10 thoracic, 8 thoracolumbar, 4 lumba)	EG (1.4 ± 1.4) CG(1.7 ± 1.6)	EG (16.2 ± 2.5) CG (17.1 ± 2.9)	NA	NA	Schroth-based method	standard-of-care observation	24w ≥ 8 h face-to-face guidance AND home exercise 15 min/day x 5 days/week (75 min/week)	Brace prescription, Curve magnitude

CG Control group, EG Experiment group, SG Schroth group, PG PNF group, NA Not available

A total of 1,014 patients were included in this meta-analysis. The main intervention in the included studies was PSSE, with a blank control or non-surgical physical intervention in the control group. The duration of intervention ranged from 2 to 24 weeks. All studies used improvement as the primary outcome indicator.

Quality evaluation

A total of 15 studies, involving 750 patients with AIS, assessed changes in Cobb angle before and after treatment [33–46]. In addition, 11 studies enrolling 609 patients with AIS assessed changes in ATR before and after treatment [34–37, 40, 42–44, 46–48]. Another 5 studies enrolling 197 patients with AIS assessed patients’ QoL using the Scoliosis Research Society-22 (SRS-22) scale and the SRS-22r scale, and SRS-23 [34, 44, 46, 48, 49].

The quality assessment showed that 12 RCT articles included in the study were rated as low risk, 6 as medium risk, and 4 as high risk. Studies identified as high-risk articles were excluded, resulting in the inclusion of 18 RCT studies in this meta-analysis. Figures 2 and 3 present the results of the risk of bias assessment for the included studies.

Fig. 2.

Risk of bias map for each study; red: high risk; yellow: potential risk; green: low risk

Fig. 3.

Risk of bias map

Meta-analysis of PSSE on AIS

Cobb angle

Figure 4 illustrates 15 studies involving a total of 750 participants. Each study provided effect values (Effect) and corresponding 95% confidence intervals (95% CI) and the weight (%) assigned based on the random-effects model.

Fig. 4.

Forest plot of Cobb angle change values (using random effects model)

Effect of PSSE on Cobb angle: 15 studies evaluated the effect of PSSE on Cobb angle. The pooled results showed a significant effect of PSSE on Cobb angle compared with control (SMD = − 1.00, 95% CI [− 1.47, − 0.53], P < 0.001, I² = 88.1%).

The 95% confidence intervals (CIs) for the combined effect values did not include zero, indicating statistical significance of the overall results. These analyses suggest that PSSE had a significant positive effect on outcome indicators. However, the high degree of heterogeneity warrants further investigation to identify the potential sources of heterogeneity.

To assess the robustness of the findings, sensitivity analyses were conducted using the “metaninf” function. The results showed that, when studies were sequentially excluded, the combined effect sizes (SMDs) remained generally stable, with none of the 95% confidence intervals crossing the zero line, suggesting that no single study had a decisive influence on the overall results (see Fig. 5). Therefore, the results of the meta-analysis findings possessed good stability and credibility.

Fig. 5.

Cobb angle sensitivity analysis (using random effects model)

The publication bias of the included studies was assessed using Egger’s regression test and funnel plot, as shown in Fig. 6. The results showed that in the scatter distribution, the dot distribution of this paper was roughly symmetrical, and there was no obvious publication bias. The slope of the regression line was gentle and did not show significant skewness, and there was no systematic overestimation and underestimation effect in the small-sample studies. 95% CI confidence interval bands covered the zero point, and the Egger’s test was statistically insignificant (p > 0.05). The small number of discrete points in the upper right corner and the sparse points in the lower left corner indicate that the effect sizes of high-precision studies (large samples) are close to the combined effect, and the results of high-quality studies are stable; and the low-precision studies (small samples) are dispersed, but there is no centralised bias, and there is no “drawer problem”.

Fig. 6.

Cobb angle funnel plot

To further validate publication bias, Egger’s linear regression was used to test for small-study effects as shown in Fig. 7. The results showed that the scatter points were symmetrically distributed around the zero point (SND = 0), with no obvious skewed pattern; the regression line (slope = − 0.86) was close to the level, and the 95% CI covered the zero point; the Egger’s test statistically confirmed that there was no significant bias (p = 0.890). Based on these, the risk of publication in this study was considered low.

Fig. 7.

Cobb angle publication bias test (Egger’s linear regression)

To understand the effect of exercise frequency on Cobb angle, ATR, and QoL, subgroup analyses were first conducted for Cobb angle. To control the study variables, we grouped studies based on different intervention frequencies, single exercise durations, and total intervention periods. The grouping criteria were as follows: intervention frequency of low frequency ≤ 2 times, medium frequency 3–4 times, and high frequency ≥ 5 times [50, 51]; single exercise duration was grouped according to short < 30 min, medium 30–60 min, and long ≥ 60 min [52]. Total intervention period was grouped according to short-term (≤ 12 weeks), medium-term (12–24 weeks), and long-term (≥ 24 weeks) [53]. Following grouping, only studies with a single exercise session duration of ≥ 60 min and an intervention period of ≤ 12 weeks were retained. In the process of grouping, the study of LU Yuelun (2022) [36] was excluded due to substantial differences in frequency of exercise, duration of a single exercise session, and total period of intervention, making it unsuitable for direct comparison with the other studies. Subgroup analyses of Cobb’s Corner are presented in Fig. 8.

Fig. 8.

Subgroup analysis of Cobb angle

Subgroups were divided into a low-frequency subgroup (≤ 2 times/week), containing 2 studies (Alves de Araujo, 2012 [39], Dufvenberg, 2025 [40]). High-frequency subgroup (≥ 3 times/week): contains 8 studies (e.g., Kim 2016 [33], Kocaman 2021 [34], etc.). Firstly, looking at the heterogeneity, low-frequency group: I² = 65.1%, p = 0.090. high-frequency group: I² = 84.1%, P < 0.001. Overall heterogeneity: I² = 84.0%, p < 0.001, suggesting that there was a significant difference between studies. Turning to the differences between subgroups, test for heterogeneity between groups: p = 0.326 (not significant), suggesting that the frequency subgroups (≤ 2/w vs. ≥3/w) did not significantly affect the effect size. Finally, look at the total effect value: overall combined effect: −0.78 [− 1.27, − 0.30], which is significantly effective and supports the overall effectiveness of the intervention.

The intervention was significantly effective in the high-frequency group (≥ 3 times/week) (combined effect of − 0.87), whereas there was no significant effect in the low-frequency group. Subgroup analyses failed to capture the effect of intervention frequency on intervention effectiveness (p = 0.326 between groups), but the high heterogeneity of the high-frequency group itself (I² = 84.1%) suggests that other confounders (e.g., intensity of the intervention, population characteristics) may have influenced the results.

ATR

Figure 9 presents the results of forest plotting from 11 studies (609 participants in total). Effect of PSSE on ATR. There were 11 studies assessing the effect of PSSE on ATR. The pooled results of the 11 trials showed a significant effect of PSSE on ATR compared with controls (SMD = 1.15, 95% CI [− 1.94, − 0.36], p < 0.001, I² = 94.7%).

Fig. 9.

Forest plot of changing values of ATR (using random effects model)

As shown in Fig. 10, sensitivity analysis for ATR revealed fluctuations ranged from SMD_min = − 1.94 to SMD_max = − 0.36. Using extreme study-case scenarios, when Mohamed (2021) was excluded, the combined effect was SMD = − 0.36 (95% CI: −0.08, − 0.64); when Dufvenberg (2025) was excluded, the combined effect increased to SMD = − 1.94 (95% CI: −2.19, − 1.69). These findings suggest that the overall combined effect was reliable and not unduly influenced by a single study.

Fig. 10.

ATR sensitivity analysis (using random effects model)

As shown in Fig. 11, the distribution of scatter points in the funnel plot showed asymmetry, which was reflected in the blank lower right quadrant and the absence of small-sample negative studies; the regression line was skewed upward to the right (slope > 0), which was consistent with the statistical test; the 95% confidence interval band did not cover the null point (SND = 0); and the visual and statistical results collectively supported that there was a risk of publication bias.

Fig. 11.

ATR funnel diagram

Publication bias of included studies was assessed by Egger’s regression test funnel plot as in Fig. 12. To assess publication bias for the effect of ATR improvement, statistical tests showed intercept (bias) = − 8.271 (95% CI: −15.371, − 1.172, t = − 2.64, p = 0.027); slope (slope) = 1.772 (95% CI: −0.333 to 3.877, p = 0.089).

Fig. 12.

ATR publication bias test (Egger’s linear regression)

The same grouping criteria were used for ATR to perform subgroup analyses, as shown in Fig. 13.

Fig. 13.

ATR subgroup analysis

The subgroups were divided into a low-frequency subgroup (≤ 2 times/week), containing 2 studies, high-frequency subgroup (≥ 3 times/week), containing 5 studies. Firstly, in terms of heterogeneity, the low frequency group: I² = 71.5%, p = 0.061; high frequency group: I² = 96.0%, p < 0.001; and overall heterogeneity: I² = 94.2%, p < 0.001, the reliability of the combined results is doubtful. Heterogeneity between groups, p = 0.233; no statistically significant difference in frequency subgroups could explain the difference in effects.

The overall effective combined effect of the intervention was − 1.06 (95% CI: −2.02 to − 0.09) but statistically insignificant in subgroup analyses (p = 0.233); there was very high heterogeneity in the high-frequency group, and an extreme negative effect (− 4.96) coexisted with a significant positive effect (+ 0.56).

QoL

Figure 14 presents the findings from five studies, encompassing a total of 197 participants, which evaluated the effect of PSSE on QoL. These studies assessed QoL using validated scales, including the SRS-22, SRS-22r, and SRS-23 scales. The pooled results from the five trials showed a significant effect of PSSE on QoL when compared with the control group (SMD = 0.63 (95% CI: 0.15 to 1.11, p < 0.01).

Fig. 14.

Forest plot of QoL change values (using random effects model)

The leave-one-out method was used to assess the stability of the combined effect sizes (Fig. 15). Sequential exclusion of results revealed that the largest fluctuations occurred upon excluding the high-impact study by Kocaman (2021), with the effect size decreasing to 0.15 (95% CI: −0.02 to 0.32), which still maintained a positive trend. The smallest fluctuations occurred when excluding the neutral-effect study by Yagci (2019), with the effect size increasing to 0.89 (95% CI: 0.41 to 1.37). Based on this, the point estimates remained above 0 (range: 0.15–0.89) for all contexts, and the study results were directionally robust.

Fig. 15.

Sensitivity analysis of QoL change values (using random effects model)

In this study, Egger’s regression analysis was used to test for small-study effects. As shown in Fig. 16, the funnel plot was constructed with effect size (θ) on the horizontal axis and precision on the vertical axis. The 95% confidence intervals for the Egger regression line and its intercept (dashed line) were superimposed on the plot. The scatter distribution showed an overall symmetric trend with no significant visual bias. The statistical results of the Egger regression (Fig. 17) were further analysed quantitatively: slope = 0.040 (95% CI: −4.83 to 4.91, t = 0.03, p = 0.981), intercept = 1.752 (95% CI: −12.91 to 16.42, t = 0.38, p = 0.729). The p-value for testing the null hypothesis of “no small study effect” was 0.729, which was greater than 0.05, indicating that no significant publication bias was detected in the current evidence.

Fig. 16.

QoL funnel diagram

Fig. 17.

QoL publication bias test

QoL was also sub-grouped using the same criteria, and the results of the specific subgroup analyses are presented in Fig. 18. Only the high-frequency group (≥ 3 times/week) met the eligibility criteria, comprising four studies, with no low-frequency group comparisons. Of these, Kocaman (2021) [34] (SMD = 1.72) was much higher compared to the average effect size of the other studies (mean SMD = 0.66). Other factors may have contributed to the high effect sizes, e.g., population differences, dose of intervention, etc. Due to the lack of low-frequency controls, it was not possible to determine the minimum threshold of the onset of action and the association of the frequency of intervention with the effect of the intervention. It was only possible to confirm that the intervention was significantly beneficial for HF users and should be interpreted with caution in practical clinical situations.

Fig. 18.

QoL subgroup analysis

Discussion

The aim of this systematic meta-analysis study was to empirically evaluate the intervention effects of PSSE on the three core outcome indicators in AIS (Cobb angle, ATR, and QoL), and innovatively explore the modulation effect from the perspective of exercise dose. This meta-analysis pooled data from 18 studies, representing 810 participants.

Although the aetiology of spinal curvature has not yet been clearly defined, current consensus suggests it may be categorised into soft tissue abnormalities and bone-related problems. Soft tissue problems primarily involve persistent bilateral muscle imbalance, which may lead to central nervous system adaptation variability and ultimately to scoliosis. Bone-related problems involve structural abnormalities in the skeletal spine and dysregulation in bone metabolism, which influence the initiation and progression of scoliosis. Early diagnosis and timely intervention are, therefore, of critical importance in managing AIS [15]. In this study, in order to investigate the effect of PSSE on the treatment of AIS, Cobb angle, ATR, and QoL were selected to quantify the treatment effect. Cobb angle and ATR are well-established indicators for assessing scoliosis [54], while QoL is used as an indicator of patient satisfaction with the treatment.

By analysing the data from the included literature, we observed that PSSE had a positive effect on all three indicators of Cobb angle, ATR, and QoL. Among various PSSE methods, Schroth therapy remains the most widely used by patients and physicians due to its structured characteristics [55].

The present study pooled the effect of PSSE on Cobb angle and yielded an SMD of approximately − 1.00, 95% CI (− 1.47 to − 0.53), indicating a statistically significant effect relative to the control group. However, the actual improvement in angle tends to be on the low side due to the SMD as a standardised effect size. For example, Ceballos-Laita et al. [56] reported that using Schroth only resulted in a mean decrease of 3.18° in Cobb angle, which falls below the commonly accepted clinical threshold of 5° [57]. Thus, although the intervention shows statistical validity and can serve as a clinical reference, its clinical significance needs to be interpreted with caution. Considering the high heterogeneity in this study (I² = 88.1%), we conducted a subgroup analysis to explore whether exercise dose was a key variable influencing the effect of the intervention. However, frequency stratification did not significantly reduce the heterogeneity between the groups (p < 0.001). In fact, heterogeneity remained higher in the high-frequency group (I² = 84.1%), possibly due to confounding variables such as intervention type. For instance, Monticone (2014) [43] combined PSSE with cognitive therapy, whereas Schreiber (2016) [58] implemented PSSE alone. Conversely, heterogeneity was lowest in the low-frequency group, suggesting that the effects of the different interventions converged when the frequency of exercise was ≥ 3 times per week and that a frequency of ≥ 3 times per week was the threshold to produce a significant efficacy. This result is also consistent with the findings of Dimitrijević [59, 60] et al. In terms of ATR, PSSE also demonstrated a positive intervention effect. Subgroup analyses showed relatively low inter-study heterogeneity in the low-frequency group (I² = 71.5%), but a high degree of heterogeneity remained in the high-frequency group. However, the difference in effect sizes between groups was not statistically significant (p = 0.233), suggesting that exercise dose may not be an independent determinant of heterogeneity. In terms of QoL, PSSE showed a positive intervention effect. However, the QoL group was only included in the high-frequency intervention study, and future RCTs of the low-frequency group are needed to be added to support the dose-response relationship. The present study explored the relationship between the frequency of the intervention and the intervention effect through a forest plot. The results showed that high-frequency interventions (≥ 3 times/week) were overall significantly effective (combined effect 0.77, 95% CI: 0.27–1.27). However, some contradictory points remain. Firstly, both Cobb’s Corner and ATR data showed that the effect was stronger in the high-frequency group than in the low-frequency group (− 0.87, − 0.41), and the difference between the groups was not statistically significant (p = 0.326), indicating that there are other factors besides frequency affecting the effect of the intervention and that the frequency of the intervention is a non-independent factor. Additionally, some studies demonstrated extreme effect sizes, such as R.A. MOHAMED (2021), with an effect size of − 4.96. LIANG (2023) and Shen, X. (2023) with (+ 0.56), which also suggests that the intervention group is not as effective as the control group, or there is a potential harm. Overall, while HF protocols are generally effective, for specific populations, such as the LIANG study, the treatment effect may be reversed, and patient-specific individualised treatment approaches are required. In future studies, further research of heterogeneity is warranted, including parsing heterogeneity by patient age as well as severity, validating the minimum effective dose in the low-frequency group, and reviewing the quality of the raw data on the extreme effect sizes, to inform clinical practice.

Compared with other meta-analyses, the innovation of this paper is that most of the existing literature uses Cobb angle as the only or main indicator of intervention effect, neglecting the comprehensive assessment of functional status and quality of life of AIS patients. TThis study involves the simultaneous incorporation of three types of outcome variables: the Cobb angle, the angle of trunk rotation (ATR), and quality of life (QoL), to comprehensively analyse the intervention effect of PSSE from multiple dimensions, such as morphology correction, trunk symmetry, and subjective health perception. This multidimensional analysis not only enriches the evaluation system of PSSE efficacy but also helps to promote the concept of AIS treatment from “image improvement” to a “function-oriented” comprehensive rehabilitation model. In exercise intervention research, the frequency of intervention is considered to be a key parameter in determining the intensity and efficacy of treatment, but there has been a lack of systematic discussion in the field of AIS for a long time. In this paper, a subgroup analysis model was established based on the dosage factor to deeply analyse the differences in the intervention effects of PSSE under different exercise dosage structures and to preliminarily reveal the non-linear law that “moderate intensity” may bring better therapeutic efficacy.

Limitations of this study include the fact that during the literature search, it was found that most of the studies focused on Schroth therapy, and the number of studies on other therapies was low. Therefore, this study was unable to assess the extent of improvement in Cobb angle, ATR, and QoL with other exercise intervention modalities. Second, there were differences in the inclusion criteria of the studies. For example, patients had different values of initial Cobb angle, and the Cobb angle in the intervention group ranged from 14° to 33.26°, which is a wide range of data and makes it difficult to quantify the effect of the intervention. In addition, there was a lack of a valid basis for improvement in severe patients with a Cobb angle greater than 45°. According to the recommendations of most PSSE studies, exercise intensity should be 60–90 min, 3–5 times per week for 10–24 weeks [61]. However, data from the literature included in this study showed that the shortest duration of exercise was 14 days and the longest was 26 weeks, which is clearly not in line with the recommended duration of exercise. In addition, there were differences in the number of people set in the intervention and control groups, and some of the control groups were not blank controls. All of these issues may have some impact on the results of this study.

Future research should focus on innovation and standardisation of study design. A substantial proportion of the heterogeneity observed in current meta-analyses is attributable to methodological differences among the original studies. To address this, several recommendations can be proposed. First, the promotion of individual patient data (IPD) meta-analysis is essential. IPD approaches enable the integration of individual data from original trials, thereby allowing for precise adjustment of confounding variables (e.g., Risser staging, Lenke staging) and accurate analysis of the dose-effect relationship and the source of heterogeneity. Second, there is a pressing need to establish a core outcome set for PSSE trials. In collaboration with the International Society of Scoliosis and Spondylolisthesis (SOSORT), a consensus-based standard set of outcomes indicators should be established. In parallel, reporting criteria for exercise dose. Third, strengthening the grey literature search mechanism. This includes the systematic inclusion of clinical trial registries, platforms, and conference abstracts to reduce the risk of publication bias of ATR and other indicators. Finally, a precise exploration of population segmentation. Heterogeneity of populations included in studies weakens clinical applicability.

Conclusion

This study systematically evaluated the effectiveness of physiotherapeutic scoliosis exercises (PSSE) as an intervention for patients with adolescent idiopathic scoliosis (AIS). The results showed that PSSE was statistically significant in improving Cobb angle, angle of trunk rotation (ATR), and quality of life (QoL). Subgroup analyses suggested that higher exercise frequencies may enhance the intervention effect, although this dose-response relationship remains multifactorial. Sensitivity analyses with publication bias assessment showed robust results with some risk of bias in ATR. Although some effects did not reach the threshold of clinical significance, PSSE is overall a safe and effective conservative treatment modality with potential clinical applications. Future research should prioritise large-sample, multicentre randomised controlled trials to optimise intervention parameters and strengthen the evidence base for individualised intervention strategies.

Abbreviations

PSSE

Physiotherapeutic Scoliosis-Specific Exercises

AIS

adolescents idiopathic scoliosis

ATR

Angle of trunk rotation

QoL

quality of life

SRS-22

Scoliosis Research Society-22

Rob2

Randomized Trials Risk of Bias Tool

Authors’ contributions

The first author, Jia Qiqi, contributed the most to the data collection and analysis as well as the writing of the manuscript.Second author: Boyuan Zhang: assisted in data analysis and writing of the manuscript, second to the first author.Corresponding author Wang Hong Zheng Weina: provided comments on the revision of the paper and guidance on the data analysis.

Funding

This study was supported by the “Theory and Practice of National Sports Health Promotion Service“(6002950).

Data availability

The datasets generated and/or analysed during the current study are available in the MEGA repository,https://mega.nz/file/R2hlBDwY#HOUKpQWAY8zjsqVTbfg1ZH1yXEDkadKIM6kuTVkJnkQ.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not Applicable.

Competing interests

The authors declare no competing interests.