The efficacy and safety of pilates exercise in patients with knee osteoarthritis: a systematic review with meta-analysis of randomized controlled trials

Abstract

Background

Evidence on the effects of Pilates exercise in patients with knee osteoarthritis (KOA) remains limited. This meta-analysis aimed to evaluate its efficacy and safety.

Methods

We searched PubMed, Web of Science, EMBASE, Cochrane Library, and Google Scholar up to May 21, 2024, for randomized controlled trials (RCTs). Outcomes included Visual Analogue Scale (VAS), Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC), knee range of motion (ROM), and adverse events. Risk of bias and evidence certainty were assessed; subgroup, sensitivity, and publication bias analyses were performed.

Results

Eight RCTs involving 322 participants were included. Risk of bias ranged from low to high, mainly due to lack of blinding. Compared to blank controls, Pilates significantly reduced WOMAC scores (SMD = −1.70; 95% CI: −3.14 to −0.25). One study comparing Pilates to health education reported significant reductions in VAS (MD = −1.74; 95% CI: −2.51 to −0.97) and WOMAC (SMD = −1.42; 95% CI: −2.39 to −0.45), but no significant ROM improvement. Compared to other exercises, Pilates showed no significant effects on VAS, WOMAC, or ROM. Funnel plots showed asymmetry for VAS and WOMAC; Egger’s test indicated possible publication bias (p = 0.0039 and 0.0154, respectively). No adverse events were reported, but active monitoring was limited. GRADE rated evidence as “very low” for VAS and WOMAC, and “low” for ROM.

Conclusion

Pilates may relieve pain and improve physical function, but has limited effects on ROM. Due to insufficient adverse event reporting, safety remains unclear. Given the low study quality, high heterogeneity, and possible bias, results should be interpreted cautiously. Future studies should use standardized protocols, report long-term adherence, and assess mental health benefits.

1. Introduction

Knee Osteoarthritis (KOA) is a prevalent, disabling, progressive degenerative joint disease, affecting over 30% of individuals aged 65 and above [1]^, with a global number of patients exceeding 250 million [2]. With the accelerated aging of the population, KOA has become a leading cause of disability and diminished quality of life among older adults worldwide, imposing significant societal and economic burdens [3,4]. The latest guidelines for KOA from the American Academy of Orthopaedic Surgeons (AAOS) [5], the American College of Rheumatology (ACR) [6], and the Osteoarthritis Research Society International (OARSI) [7] all strongly recommend exercise as a first-line treatment for KOA patients.

Exercise can alleviate joint pain, improve cardiovascular function, enhance body coordination, elevate quality of life, and have beneficial effects on mental health. According to the WHO’s ‘Global Recommendations on Physical Activity for Health,’ individuals aged 65 and above are advised to partake in either 150 min of moderate-intensity or 75 min of vigorous-intensity aerobic activity each week, accompanied by muscle-strengthening activities on two or more days [8]. However, previous systematic reviews have found that 87% of KOA patients fail to meet such recommendations, largely due to barriers such as pain, low self-efficacy, and lack of structured supervision [9].

Pilates exercise originated in the 1920s, blending practical movement styles from dance, gymnastics, and yoga with philosophical concepts. It adheres to six fundamental principles: concentration, control, centering, precision, flow of movement, and breath [10]. Pilates primarily involves isometric contractions of core muscles, emphasizing the importance of breath and proprioception. It focuses on activating specific muscles in a controlled manner, following a functional sequence, and is classified within the realm of mind-body exercise [11,12]. Previous research findings indicate that Pilates can improve lower limb strength, flexibility, functional independence, reduce fall risk, and enhance overall well-being in middle-aged and older adults [13–15]. It is worth noting that certain features of Pilates may offer unique benefits for patients with KOA. The primary goal of Pilates is to enhance core muscle stability and strength, thereby improving pelvic and trunk stability [16]. This improved proximal control may contribute to favourable changes in lower limb biomechanics and lead to a more balanced distribution of stress across the knee joint [17]. Breath control, as one of the core principles of Pilates, has been found to improve core stability and trunk control. A study comparing abdominal curl exercises performed with and without Pilates breathing techniques found that abdominal muscle activation was greater during Pilates-style breathing [18]. Enhanced core muscle activation may contribute to improved lower limb joint stability, strength ratios, and motor unit recruitment [19]. In addition, proper breathing techniques may help regulate autonomic nervous system activity, promote relaxation, and reduce pain perception [20]. Moreover, Pilates training can enhance neuromuscular coordination and activation patterns of the muscles surrounding the knee joint, thereby improving joint stability and functional performance [21]. In recent years, there have been reports suggesting potential benefits of Pilates exercise for patients with KOA, including pain reduction, improved joint function, and enhanced kinesthesia and position sense [22,23]. Compared with Tai Chi, which emphasizes dynamic balance, weight shifting, and continuous flow, and yoga, which focuses on flexibility, static postures, and mindfulness, Pilates facilitates neuromuscular re-education through slow, precise, and repetitive movements [24]. In addition, its low-impact and intensity-adjustable nature minimizes additional loading on the knee joint during exercise. These features confer unique and irreplaceable advantages of Pilates in the functional rehabilitation of KOA. Lastly, Pilates is considered a safe, low-cost, and widely accessible form of exercise that requires minimal environmental conditions—often needing only a mat for practice [25]. These practical and convenient features may help enhance adherence among older adults and mitigate the common issues of interruption and dropout associated with home-based self-training. However, the number of studies in this area is limited in number and exhibits notable variations in sample size, methodological quality, and intervention protocols. Therefore, it is crucial to establish reliable evidence regarding the efficacy of Pilates in treating KOA. To the best of our knowledge, no systematic reviews have specifically addressed the efficacy of Pilates for treating KOA.

The objective of our study is to conduct a systematic review and meta-analysis of relevant randomized controlled trials (RCTs) to determine the efficacy and safety of Pilates in alleviating pain and functional impairment in patients with KOA. We hope our results will provide critical evidence for clinicians, patients, and policymakers, and offer recommendations for future clinical trials.

2. Materials and methods

2.1. Protocol and registration

This systematic review and meta-analysis was reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement [26], Assessing the methodological quality of systematic reviews (AMSTAR) guidelines [27] and the methodology of the European Association of Urology (EAU) for systematic review [28]. To ensure comprehensive analysis and transparency of methods and results, this protocol was prospectively registered in the International Prospective Register of Systematic Reviews (PROSPERO) under the registration number CRD42024533708. The study was conducted in strict accordance with the registered protocol, with no major deviations.

2.2. Search strategy

We conducted the literature search in PubMed, Web of Science, EMBASE, Cochrane Library, and Google Scholar from inception to May 21, 2024, to identify RCTs investigating the use of Pilates exercise for treating KOA patients. We used the following MeSH terms and keywords: ‘Pilates,’ ‘Pilates Training,’ ‘Pilates-Based Exercises,’ ‘Training, Pilates,’ ‘Knee Osteoarthritis,’ ‘Osteoarthritis, Knee,’ ‘randomized controlled trial,’ etc. We initially conducted our search on PubMed and then replicated the search strategy in other databases (Supplementary Material 1). Additionally, a specific manual search was also conducted by screening major journals in the field of sports medicine (e.g. British Journal of Sports Medicine, Sports Medicine, American Journal of Sports Medicine) and reviewing the reference lists of the selected articles to further minimize the risk of missing relevant studies.

2.3. Inclusion and exclusion criteria

Articles were included for this research based on the following criteria: (1) Participant: Adults aged ≥40 years who met the diagnostic criteria for KOA according to the ACR criteria or the Kellgren-Lawrence grading system; (2) Intervention: The experimental group received either Pilates exercise alone or in combination with other therapies; if additional treatments were included in the experimental group, the control group was required to receive the same adjunct interventions; (3) Comparison: the control group received treatment other than Pilates or served as a blank control group; (4) Outcomes: Visual analogue scale (VAS), Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC), range of motion (ROM) of the knee; (5) Study design: RCT. Exclusion criteria were (1) Traumatic arthritis, neurological or rheumatic diseases; (2) Both the experimental and control groups received Pilates interventions; (3) Case reports, cross-sectional studies, letters, editorials, review articles, meta-analyses, and retrospective studies; (4) Animal studies; (5) Studies with insufficient data or data that cannot be extracted; (6) Articles not published in English.

2.4. Study selection

Two researchers independently reviewed all titles and abstracts. If the abstract was relevant to the purpose of this review, the full text was further assessed and evaluated against the inclusion and exclusion criteria to determine eligibility. Discrepancies between the two reviewers were initially resolved through discussion; if unresolved, a third reviewer was involved to make a final decision, ensuring consensus among all three investigators.

2.5. Data extraction

Data extraction from the included studies was independently performed by two researchers. They used a standardized form to extract information, including study identification, publication year, first author, participant characteristics (such as gender and age), sample size, intervention details, outcome measures (VAS, WOMAC, ROM), and adverse events. For studies reporting multiple follow-up time points, data from baseline and the final follow-up were consistently extracted. If the mean and standard deviation (SD) of change scores were explicitly reported, they were directly used. If not, the mean change was calculated by subtracting the baseline mean from the post-intervention mean, and the SD of the change was estimated using the formula recommended by the Cochrane Handbook: $S{D}_{{}_{E,\hspace{0.17em}\mathit{\text{Change}}}}=\sqrt{S{D}^2{}_{E,\mathit{\text{baseline}}}+S{D}^2{}_{E,\hspace{0.17em}\mathit{\text{final}}}-(2\times \mathit{\text{Corr}}\times S{D}_{E,\mathit{\text{baseline}}}\times S{D}_{E,\hspace{0.17em}\mathit{\text{final}}})}$, where r represents the correlation coefficient between baseline and post-intervention scores. In this study, we adopted r = 0.5, which is a commonly used assumption when the exact correlation is not reported. If continuous variables were reported as median and interquartile range (IQR) due to non-normal distribution or other reasons, we converted the data using the method proposed by Wan et al. [29], as recommended in the Cochrane Handbook. If any data were missing, we attempted to contact the corresponding authors to obtain the missing data. The two researchers coordinated in advance to ensure consistency in the extraction method and level of detail. After data entry, any discrepancies were discussed to reach a consensus. If disagreements persisted, a third researcher made the final decision.

2.6. Risk of bias and quality assessment

Two researchers independently assessed the risk of bias in the included studies using the Cochrane Risk of Bias Tool (RoB 1.0) for RCTs [30]. They evaluated the following criteria: selection bias, performance bias, detection bias, attrition bias, reporting bias, and other biases. The risk of bias was categorized as ‘low risk,’ ‘high risk,’ or ‘unclear.’ The definitions, judgment criteria, and risk levels for each domain of the RoB 1.0 are provided in Supplementary Material 2. Discrepancies were resolved by a third researcher. The level of agreement between the reviewers was calculated using the Kappa coefficient, which was interpreted as follows [31]: poor (<0.00), slight (0.00-0.20), fair (0.21–0.40), moderate (0.41–0.60), substantial (0.61–0.80), or almost perfect (0.81–1.0).

We assessed the quality of evidence for the outcomes using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) system. The certainty of the evidence was categorized into one of four levels: (a) high—very confident that the true effect is close to the estimated effect, and further research is unlikely to change this; (b) moderate—moderately confident in the effect estimate, with a possibility that the true effect is substantially different; (c) low—limited confidence in the effect estimate, and further research is likely to have an important impact and may change the estimate; (d) very low—very little confidence in the effect estimate. The quality rating was based on five domains: risk of bias, inconsistency, indirectness, imprecision, and publication bias. Evidence from RCTs was initially considered high quality, but could be downgraded due to the following factors: (a) risk of bias, such as lack of allocation concealment, lack of assessor blinding, or other methodological limitations; (b) inconsistency, referring to substantial heterogeneity or variability in results; (c) indirectness, such as differences in population, intervention, comparator, or outcomes; (d) imprecision, including small sample sizes, wide confidence intervals, or limited number of studies; and (e) publication bias, assessed by visual inspection of funnel plots or the potential existence of unpublished studies. The GRADE assessments were independently conducted by two reviewers. Any disagreements were resolved by consultation with a third reviewer, who made the final decision.

2.7. Data synthesis and analysis

All statistical analyses were performed using Review Manager 5 software (Version 5.3, The Nordic Cochrane Centre, Copenhagen, Denmark). Treatment effects were defined as the difference in score changes from baseline to post-intervention between the intervention and control groups. As the efficacy outcomes were continuous variables, either mean differences (MDs) or standardized mean differences (SMDs) with 95% confidence intervals (CIs) were calculated to estimate the effect size for each outcome measure. MDs were used when outcome measures were reported using the same scale across all included studies, while SMDs were applied when different scales or scoring systems were used across studies. Negative differences in VAS and WOMAC scores indicated favourable outcomes for Pilates-based exercise, while positive differences favoured alternative interventions; ROM results were interpreted conversely. Statistical heterogeneity was assessed using the I² test to identify differences in results, categorized as follows: (1) I²≤25%, low heterogeneity; (2) 25% < I² < 50%, moderate heterogeneity; (3) 50% < I² < 75%, substantial heterogeneity; (4) I² ≥ 75%, high heterogeneity [32]. If I² < 50% and p > 0.01, a fixed-effect model was used to evaluate the summary model of SMD with 95% CI; otherwise, a random-effects model was applied [33]. To explore potential sources of heterogeneity, we performed subgroup analyses based on control group type (health education, blank control, other exercises), intervention duration, participant age, geographic region and ethnicity, and sample size. If a subgroup included only one study, meta-analysis was not performed. Instead, a narrative approach was used to report the SMD and its 95%CI for the pre- and post-intervention change. For such subgroups, pooled effect estimates and heterogeneity statistics were not provided, and no diamond summary symbol was displayed in the forest plot. Sensitivity analyses were also conducted by sequentially removing individual studies to assess the stability and robustness of the pooled results. Meta-regression analysis was planned if more than 10 studies were included. Significance for all analyses was set at p < 0.05. Additionally, subgroup analyses were conducted to investigate efficacy differences among different age groups, geographical locations, sample sizes, exercise duration time, and the impact of these factors on heterogeneity. Egger’s test [34] and funnel plots [35] were used to examine potential publication bias in the RCTs included in this meta-analysis.

3. Results

3.1. Search result

As shown in the PRISMA flow diagram (Figure 1), we retrieved 67 articles through searches in five different electronic databases. Twenty-seven duplicates were excluded. After reviewing the titles and abstracts, we excluded 4 articles not published in English, 6 reviews, 1 case report, 2 non-RCT trials, 2 trials that did not involve patients with KOA, and 11 trial registration protocols. Subsequently, we thoroughly read the full text of the remaining 14 articles for further evaluation. Based on our inclusion and exclusion criteria, we excluded 3 trials that did not meet the criteria of intervention, 2 trials did not use the outcome measures we intended to collect, 1 trials with duplicated data (details were shown in Supplementary Material 3). Finally, this meta-analysis included 8 RCTs [36–43].

Figure 1.

PRISMA Flow diagram of study selection process.

3.2. Study characteristics

Table 1 summarizes the characteristics of the 8 included RCTs, detailing the first author, geographical location, publication year, participant information, intervention protocols, outcome assessment indicators, and adverse events. These 8 RCTs, which included a total of 322 participants, were published in English between March 2014 and January 2023. The geographical locations of these studies include India, Iran, Pakistan, Brazil, and Nigeria. Participants were diagnosed with KOA based on diagnostic criteria from the ACR, Kellgren and Lawrence scale (K-L) and imaging evidence. The sample sizes of the included studies ranged from 17 to 68, and the average age of participants ranged from 51.9 to 65.8 years. The intervention duration ranged from 6 to 8 weeks. Seven studies used the WOMAC to evaluate the efficacy of the interventions, 5 studies used VAS scores, and 3 studies used ROM. No adverse events were reported in any of the studies.

Table 1.

Trials characteristics.

				Gender	Mean age(years)		Mean BMI(kg/m2)		Sample	Intervention			Adverse
First Author	Nation	Year	L Grade	(M/F)	Trail	Control	Trail	Control	size(T/C)	Trail	Control	Outcomes	events
Rêgo T.A.M [41]	Brazil	2023	2-3	0/17	65.8 ± 2.8	64.8 ± 2.2	28.9 ± 1.4	32.0 ± 2.4	8/9	PE (60 min, 2/w, 7w)	Blank Control	WOMAC, SF-36	None
Bakki A [43]	Egypt	2023	NA	30/0	51.9 ± 3.7	51.9 ± 3.7	NA		15/15	PE (20 min, 3/w, 8w) + IR, US and routine physical therapy exercises	IR, US and routine physical therapy exercises (30 min, 3/w, 8w)	VAS, WOMAC, ROM	None
Nadia Saleem [39]	Pakistan	2022	2-3	0/40	57.6 ± 6.3	55.7 ± 7.3	25.8 ± 4.2	26.9 ± 4.3	20/20	Isometrics and PE (60 min, 2/w, 8w)	Isometrics (2/w, 8w)	WOMAC	None
Meenakshi C[38]	India	2021	1-2	NA	NA		NA		34/34	PE (60min, 3/w, 6w)	Closed kinematic chain exercises	VAS, WOMAC, HHD	None
Karimi N [42]	Iran	2021	NA	0/30	60.0 ± 4.6	61.4 ± 4.9	NA		10/10	PE (60 min, 3/w, 8w)	Suspension training	WOMAC, ROM, Y-test	None
					60.0 ± 4.6	65.6 ± 4.7			10/10		Blank Control
Chinta Meenakshi [40]	India	2021	3-4	NA	53		NA		32/32	PE (60 min, 3/w, 8w)	Neuromuscular Exercises	VAS, WOMAC	None
Akodu AK [36]	Nigeria	2017	1-3	5/28	55.5 ± 9.7	52.3 ± 14.9	30.7 ± 6.0	28.7 ± 4.5	13/11	PE(2/d, 8w) and TENS	Isometrics and TENS	VAS, WOMAC, ROM	None
					55.5 ± 9.7	63.2 ± 4.0	30.7 ± 6.0	29.7 ± 5.8	13/9		Education and TENS
Marziye Taleb [37]	Iran	2014	NA	0/40	54.8 ± 7.1	56.0 ± 5.2	NA		20/20	PE (60 min, 3/w, 8w)	Isometrics	VAS, SF-36	None

PE: Pilates Exercise; IR: Infrared; US: US application; TENS: Transcutaneous Electrical Nerve Stimulationl; NA: None Announced.

3.3. Risk of bias assessment

Agreement among the 2 researchers was substantial (κ = 0.749, p < 0.001). The final summarized risk of bias assessment results for all included studies are presented in Figures 2 and 3. In terms of selection bias, 6 studies showed a low risk in random sequence generation, but only 2 studies exhibited a low risk in allocation concealment, with the remaining studies not mentioning methods for allocation concealment. Among these studies, 7 studies were assessed as having a high risk of performance bias. Regarding detection bias, only 1 study employed blinding of outcome assessment and was assessed as having a low risk, while 6 studies were assessed as having a high risk. Furthermore, all studies were at low risk for attrition bias and reporting bias, while 6 studies were unclear regarding other biases.

Figure 2.

Risk of bias summary for included studies.

Figure 3.

Risk of bias graph across all included studies.

3.4. VAS scores

A total of 5 studies involving 127 participants in the Pilates group and 121 in the control group reported pain outcomes using the VAS. Since all studies utilized the same 0–10 cm VAS instrument to measure pain, MD with 95% CI was calculated. As shown in Figure 4, due to high heterogeneity (I² = 91% >50%, Tau² = 1.19, Chi² = 42.73, df = 4, p < 0.00001), a random-effects model was applied. Only 1 study compared Pilates with health education. A descriptive synthesis was conducted, and the reported MD was −1.74 [95% CI: −2.51 to −0.97], favouring Pilates. Four studies compared Pilates with other forms of exercise, and the pooled result showed a MD of −0.40 (95% CI: −1.41 to 0.61, p = 0.44), indicating no significant difference in pain relief between Pilates and other exercise modalities. The weights assigned to each study ranged from 18.3% to 21.1%, suggesting that no single study dominated the overall analysis.

Figure 4.

Forest Plot of the effects of pilates on VAS scores.

3.5. WOMAC scores

A total of 6 studies involving 135 participants in the Pilates group and 130 in the control group were included. As shown in Figure 5, due to variability in the scoring metrics of the WOMAC total score across studies, SMD were used for data synthesis. Given the observed heterogeneity (I² = 83% >50%, Tau² = 0.67, Chi² = 35.46, df = 6, p < 0.00001), a random-effects model was employed to pool the results. Only 1 study compared Pilates with health education. A descriptive synthesis was conducted, and the reported SMD was −1.42 [95% CI: −2.39 to −0.45], favoring Pilates. Five studies compared Pilates with other forms of exercise. The pooled result showed an SMD of 0.10 (95% CI: −0.52 to 0.71; p = 0.76), with substantial heterogeneity (I² = 77%), indicating no significant difference in physical function improvement between Pilates and other exercise modalities. Two studies compared Pilates with blank control group. The pooled result showed an SMD of −1.70 (95% CI: −3.14 to −0.25; p = 0.02), with substantial heterogeneity (I² = 68%), suggesting that Pilates may lead to clinically meaningful improvements in physical function compared to no intervention. The overall pooled analysis across all control groups (n = 121) revealed no statistically significant effect of Pilates (SMD = −0.33; 95% CI: −1.01 to 0.35; p = 0.34), with substantial heterogeneity (I² = 83%). The heterogeneity of the combined results for Pilates versus other exercises (I² = 77%) was higher than that for Pilates versus blank control groups (I²= 68%), suggesting that differences in the control interventions contributed to the heterogeneity. Study weights ranged from 10.5% to 16.6%, suggesting relatively balanced influence among included trials.

Figure 5.

Forest Plot of the effects of pilates on WOMAC total scores.

3.5.1. WOMAC - pain

Two trials reported changes in WOMAC pain scores following Pilates intervention, involving 28 participants in the Pilates group and 29 in the control group, and were included in the meta-analysis (Figure 6). Since the included studies adopted different scoring scales for the WOMAC pain subscale, SMD was used to estimate pooled effect sizes. Given the observed heterogeneity (I² = 73%, Tau²=0.67, Chi² = 3.67, p = 0.06), a random-effects model was employed. The pooled analysis indicated a trend toward reduced pain with Pilates compared to control interventions (SMD = −1.27; 95% CI: −2.59 to 0.05; p = 0.06), although this result did not reach statistical significance. In terms of individual study contributions, Nadia Saleem 2022 contributed a higher weight (57.9%) and showed a small-to-moderate effect size (SMD = −0.70; 95% CI: −1.34 to −0.06), while Régo T A M 2023 reported a larger effect (SMD = −2.06; 95% CI: −3.29 to −0.82), though with greater variability and wider confidence intervals.

Figure 6.

Forest Plot of the effects of pilates on WOMAC - pain scores.

3.5.2. WOMAC - stiffness

As only one trial reported pre- and post-intervention changes in the WOMAC stiffness subscale, involving 8 participants in the Pilates group and 9 in the control group, the result was presented narratively. A significant improvement in stiffness was observed in the Pilates group compared to the control, with an SMD of −1.33 (95% CI: −2.41 to −0.25; p = 0.02), indicating a meaningful reduction in stiffness among patients with KOA. This effect, although statistically significant, should be cautiously interpreted due to the absence of replication.

3.5.3. WOMAC - function

Two studies reported changes in WOMAC function subscale scores following Pilates interventions, involving 28 participants in the Pilates group and 29 in the control group, and were included in the meta-analysis (Figure 7). Due to differences in scoring scales across studies, SMD was used for effect size estimation. A random-effects model was adopted owing to moderate heterogeneity (I² = 66%, Tau² = 0.53, Chi² = 2.96, df = 1, p = 0.09). The pooled results showed a statistically significant improvement in physical function among participants receiving Pilates compared to control interventions (SMD = −1.49; 95% CI: −2.71 to −0.27; p = 0.02). In terms of individual study contributions, Nadia Saleem 2022 accounted for a slightly greater proportion of the pooled weight (59.8%), showing a small-to-moderate improvement in function (SMD = −0.98; 95% CI: −1.64 to −0.32). In contrast, Régo T A M 2023 contributed a lower weight (40.2%) but reported a large effect size (SMD = −2.25; 95% CI: −3.53 to −0.96), with wider confidence intervals, indicating greater variability. These findings suggest that Pilates may lead to meaningful improvements in physical function among individuals with KOA.

Figure 7.

Forest Plot of the effects of pilates on WOMAC - function scores.

3.6. ROM of the knee

Five studies involving 66 participants in the Pilates group and 60 in the control group reported changes in knee joint ROM before and after intervention. As shown in Figure 8, due to variations in measurement tools—some studies used digital goniometers while others employed manual protractors—SMD was used to synthesize the results across studies. Given the low heterogeneity (I² = 0%, Chi² = 1.70, df = 2, p = 0.43), a fixed-effects model was applied. Only 1 study (Akodu AK 2017) compared Pilates with health education. A descriptive analysis yielded an SMD of 0.56 (95% CI: −0.26 to 1.39), suggesting a potential trend toward improved ROM, though the wide confidence interval indicates high uncertainty. Another one study (Nahid Karimi 2021) conducted a comparison with no intervention. The SMD was 0.14 (95% CI: −0.58 to 0.86), indicating no significant difference in ROM improvement. Three studies compared Pilates with other forms of exercise. The pooled result showed an SMD of 0.12 (95% CI: −0.35 to 0.59; p = 0.62), indicating no significant advantage of Pilates over other exercise forms. There was a notable variation in study weights, ranging from 28.5% to 42.7%. The study with the greatest weight reported a smaller effect size (SMD = −0.13) and a relatively wide confidence interval, which may have diluted the overall effect. Although statistical homogeneity was observed, inconsistency in effect sizes across studies was still present.

Figure 8.

Forest Plot of the effects of pilates on ROM.

3.7. Adverse events

Although none of the 8 included RCTs reported adverse events associated with Pilates intervention, only two studies—Rêgo T.A.M [41] and Karimi N [42], involving a total of 18 patients receiving Pilates—explicitly stated that adverse events were actively monitored during the intervention and follow-up periods. The remaining studies did not mention adverse event monitoring or reporting, making it impossible to conclude definitively on the safety profile of Pilates for KOA patients.

3.8. Sensitivity analysis

Considering the high heterogeneity and risk of bias in this study, we conducted leave-one-out sensitivity analyses for VAS, WOMAC, and ROM outcomes to determine the robustness of the overall results, as presented in Supplementary Material 4–6. For VAS scores, exclusion of any single study did not substantially alter the overall effect size, which remained centered around −0.60. The 95% CIs of all iterations overlapped with the original pooled estimate (range: −1.37 to 0.16), indicating the findings were relatively stable. For WOMAC total scores, the pooled SMD was also robust across studies. Omitting any individual study did not result in substantial variation in the pooled estimates, with values consistently near −0.49 and CIs ranging from −1.17 to 0.19, confirming the stability of the results. For ROM outcomes, the pooled estimates were more dispersed, ranging around 2.00 with 95% CI from 0.67 to 3.34. The direction and magnitude of the effect remained stable across different iterations, supporting the robustness of the findings despite the use of different measurement tools.

3.9. Evaluation of publication bias

We assessed the presence of publication bias in the RCTs included in this study by creating funnel plots and conducting Egger’s test. The funnel plots for VAS and WOMAC showed evidence of asymmetry (Supplementary Material 7–8), and the slopes of the regression lines from Egger’s test were significantly non-zero (p = 0.0039 for VAS; p = 0.0154 for WOMAC), suggesting the presence of publication bias. This may be related to the non-publication of small sample size trials or studies with negative results, as well as the fact that only 8 trials were included in our study and all were published in English. The funnel plot for ROM was largely symmetric (Supplementary Material 9), and Egger’s test (p = 0.222) did not support the presence of publication bias.

3.10. Subgroup analysis

The results of subgroup analyses are presented in Table 2 and Supplementary Materials 10–21. Subgroup analyses were performed according to participant age, geographic region, sample size, and intervention duration to explore potential sources of heterogeneity for VAS, WOMAC, and ROM outcomes. Meta-analyses were conducted only when at least two studies were available per subgroup; otherwise, a narrative synthesis was reported.

Table 2.

Summary of subgroup analyses.

Subgroups	Outcomes	No. of studies	MD / SMD (95% CI)	P-Value for overall effect	I2 for Heterogeneity	P-Value for Heterogeneity
Age (years)
Age < 60	VAS	5	−0.62[−1.53, 0.28]	p = 0.18	90%	p < 0.01
	WOMAC	4	−0.24[−0.97, 0.49]	p = 0.52	84%	p < 0.01
	ROM	2	0.40[−0.10, 0.89]	p = 0.12	0%	p = 0.12
Age ≥ 60	VAS	0	None	None	None	None
	WOMAC	2	−0.95[−2.58, 0.68]	p = 0.25	86%	p < 0.01
	ROM	1	0.01[−0.50, 0.51]	p = 0.98	0%	p = 0.60
Duration time (weeks)
<8	VAS	1	−0.82[−1.34, −0.30]	None	None	None
	WOMAC	2	−1.43[−3.36, 0.50]	p = 0.15	86%	p < 0.01
	ROM	0	None	None	None	None
≥8	VAS	4	−0.56[−1.72, 0.59]	p = 0.34	90%	p < 0.01
	WOMAC	4	−0.19[−0.91, 0.54]	p = 0.61	81%	p < 0.01
	ROM	3	0.21[−0.15, 0.56]	p = 0.25	0%	p = 0.63
Geographical location
Asian populations	VAS	3	0.00[−1.10, 1.11]	p = 0.99	86%	p < 0.01
	WOMAC	3	−0.02[−0.92, 0.87]	p = 0.96	86%	p < 0.01
	ROM	1	0.01[−0.50, 0.51]	p = 0.98	0%	p = 0.60
Non − Asian populations	VAS	2	−1.24[−2.46, −0.01]	p = 0.05	87%	p < 0.01
	WOMAC	3	−0.96[−1.94, 0.02]	p = 0.05	78%	p < 0.01
	ROM	2	0.40[−0.10, 0.89]	p = 0.12	0%	p = 0.56
Sample size
n ≤ 30	VAS	1	0.05[−0.77, 0.87]	None	None	None
	WOMAC	3	−0.66[−1.77, 0.44]	p = 0.24	82%	p < 0.01
	ROM	2	0.00[−0.43, 0.44]	p = 0.98	0%	p = 0.87
n > 30	VAS	4	−0.74[−1.75 0.27]	p = 0.15	90%	p < 0.01
	WOMAC	3	−0.31[−1.25, 0.62]	p = 0.51	88%	p < 0.01
	ROM	1	0.58[−0.02, 1.18]	p = 0.06	0%	p = 0.95

In the subgroup analysis based on age (<60 vs. ≥60 years), no significant reduction in VAS scores was observed in the <60 years group compared to the control (MD = −0.62, 95% CI: −1.53 to 0.28; p = 0.18), and the heterogeneity level (I² = 90%) was comparable to that of the overall VAS meta-analysis (I² = 91%). Similarly, no significant reduction in WOMAC total scores was found (SMD = −0.24, 95% CI: −0.97 to 0.49; p = 0.52), with heterogeneity (I² = 84%) similar to the overall WOMAC analysis (I² = 83%). Improvements in ROM were also not significant (SMD = 0.40, 95% CI: −0.10 to 0.89; p = 0.12), and heterogeneity remained at 0%, consistent with the overall ROM analysis (I² = 0%). In the ≥60 years group, no significant improvement in WOMAC total scores was observed compared to the control (SMD = −0.95, 95% CI: −2.58 to 0.68; p = 0.25), with heterogeneity (I² = 86%) similar to the overall WOMAC analysis (I² = 83%). Likewise, ROM improvement was not significant (SMD = 0.01, 95% CI: −0.50 to 0.51; p = 0.98), with heterogeneity at 0%, consistent with the overall ROM analysis (I² = 0%). These findings suggest that current data do not support a differential effect of Pilates across age groups in patients with KOA, and that age is unlikely to be a major source of heterogeneity in this study.

In the subgroup analysis based on intervention duration (<8 weeks vs. ≥8 weeks), no significant reduction in VAS scores was observed in the ≥8 weeks group compared to the control (MD = −0.56, 95% CI: −1.72 to 0.59; p = 0.34), with substantial heterogeneity (I² = 90%), similar to the overall VAS meta-analysis (I² = 91%). For the <8 weeks group (n = 1), a descriptive result showed a significant reduction in pain (MD = −0.82, 95% CI: −1.34 to −0.30). Similarly, no significant improvement in WOMAC total scores was observed in either subgroup. In the <8 weeks group (2 studies), the pooled SMD was −1.43 (95% CI: −3.36 to 0.50; p = 0.15), with high heterogeneity (I² = 86%). In the ≥8 weeks group (4 studies with 6 comparisons), the pooled SMD was −0.19 (95% CI: −0.91 to 0.54; p = 0.61), also showing high heterogeneity (I² = 81%). These heterogeneity levels are comparable to the overall WOMAC analysis (I² = 83%). For ROM, In the ≥8 weeks group (5 studies with 6 comparisons), the pooled SMD was 0.21 (95% CI: −0.15 to 0.56; p = 0.25), with no heterogeneity (I² = 0%), consistent with the overall ROM analysis (I² = 0%). The <8 weeks group did not report ROM outcomes. These findings suggest that current data do not support a differential effect of Pilates across intervention durations in patients with KOA, and that intervention duration is unlikely to be a major source of heterogeneity in this study.

In the subgroup analysis based on geographic location (Asian vs. non-Asian populations), no significant reduction in VAS scores was observed in either subgroup. In the Asian subgroup (3 studies), the pooled MD was 0.00 (95% CI: −1.10 to 1.11; p = 0.99), with high heterogeneity (I² = 86%). In the non-Asian subgroup (2 studies with 3 comparisons), the pooled MD was −1.24 (95% CI: −2.46 to −0.01; p = 0.05), also with high heterogeneity (I² = 87%). The overall test for subgroup differences was not statistically significant (p = 0.14), indicating no clear regional difference in pain reduction. The heterogeneity levels across all subgroups were comparable to that of the overall meta-analysis for VAS scores (I² = 91%). For WOMAC total scores, the Asian subgroup (3 studies with 4 comparisons) showed no significant effect (SMD = −0.02, 95% CI: −0.92 to 0.87; p = 0.96) with high heterogeneity (I² = 86%), slightly higher than the overall WOMAC analysis (I² = 83%). The non-Asian subgroup (3 studies with 4 comparisons) showed a non-significant but more favorable trend (SMD = −0.96, 95% CI: −1.94 to 0.02; p = 0.05), with slightly lower heterogeneity (I² = 78%). For ROM, the Asian subgroup (1 study with 2 comparisons) showed no significant improvement (SMD = 0.01, 95% CI: −0.50 to 0.51; p = 0.98), and the non-Asian subgroup (2 studies with 3 comparisons) showed a trend toward improvement (SMD = 0.40, 95% CI: −0.10 to 0.89; p = 0.12), both with no heterogeneity (I² = 0%), identical to the overall ROM meta-analysis (I² = 0%). These results suggest that the effects of Pilates on KOA do not significantly differ by geographic region, and region is unlikely to be a major source of heterogeneity in this analysis.

In the subgroup analysis based on sample size (n ≤ 30 vs. n > 30), no significant improvement in VAS scores was observed in either subgroup. In the n ≤ 30 group (1 study), only a descriptive analysis was conducted, and the result (MD = 0.05, 95% CI: −0.77 to 0.87) showed no benefit of Pilates over control. In the n > 30 group (4 studies with 5 comparisons), the pooled MD was −0.74 (95% CI: −1.75 to 0.27; p = 0.15), with substantial heterogeneity (I² = 90%), similar to the overall VAS analysis (I² = 91%). For WOMAC total scores, neither subgroup showed significant improvement. In the n ≤ 30 group (3 studies with 4 comparisons), the pooled SMD was −0.66 (95% CI: −1.77 to 0.44; p = 0.24), with high heterogeneity (I² = 82%). In the n > 30 group (3 studies with 4 comparisons), the pooled SMD was −0.31 (95% CI: −1.25 to 0.62; p = 0.51), with slightly higher heterogeneity (I² = 88%). Both values were comparable to the overall WOMAC heterogeneity (I² = 83%). In terms of ROM improvement, no significant difference was found in either subgroup. The n ≤ 30 group (2 studies with 3 comparisons) yielded a pooled SMD of 0.00 (95% CI: −0.43 to 0.44; p = 0.98), with no heterogeneity (I² = 0%). The n > 30 group (1 study with 2 comparisons) showed a pooled SMD of 0.58 (95% CI: −0.02 to 1.18; p = 0.06), also with no heterogeneity (I² = 0%). Although the n > 30 group suggested a non-significant trend toward greater ROM improvement, the test for subgroup difference was not significant (p = 0.13). These results further confirm that the effects of Pilates on KOA do not significantly differ by sample size, and it is not a major source of heterogeneity in ROM outcomes.

In summary, although subgroup analyses based on patient age, intervention duration, geographic region, and sample size revealed only minor variations in heterogeneity across subgroups, these differences are insufficient to confirm these factors as the primary sources of heterogeneity in this study. However, their potential contribution to the overall between-study inconsistency cannot be ruled out.

3.11. Quality of evidence evaluation

As shown in Table 3, the certainty of evidence was rated as very low for VAS and WOMAC, and low for ROM. For VAS (6 RCTs, n = 248), the certainty was downgraded to very low due to serious risk of bias (unclear allocation concealment and blinding), substantial heterogeneity across studies (I² = 86%), imprecision in effect estimates (confidence interval crossing the line of no effect), and potential publication bias. The pooled SMD was −0.58 [95% CI: −1.32 to 0.16]. For WOMAC (8 RCTs, n = 165), similar issues were identified, including high heterogeneity (I² = 83%) and unclear methodological quality, resulting in a very low certainty rating. The pooled SMD was −0.46 [95% CI: −1.12 to 0.20]. For ROM (5 RCTs, n = 126), although heterogeneity was acceptable, the certainty was downgraded to low due to risk of bias and imprecision in the effect estimate. The pooled SMD was 0.21 [95% CI: −0.15 to 0.56]. Taken together, the overall quality of evidence regarding the efficacy of Pilates for KOA remains limited, and the results should be interpreted with caution.

Table 3.

Certainty of the evidence using GRADE approach for main outcomes.

Outcome	No. of Patients (studies)	Quality assessment					Relative effect (95% CI)	Quality
		Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations
VAS	248 (6 RCTs)	seriousa	seriousb	No concerns	seriousc	Suspected publication bias	SMD = −0.58 [−1.32, 0.16]	⊕ ⊖⊖⊖ Very low
WOMAC	165 (8 RCTs)	seriousa	seriousb	No concerns	seriousc	Suspected publication bias	SMD = −0.46 [−1.12, 0.20]	⊕ ⊖⊖⊖ Very low
ROM	126 (5 RCTs)	seriousa	Not serious	No concerns	seriousc	Undetected	SMD = 0.21 [−0.15, 0.56]	⊕⊕ ⊖⊖ Low

⊕ Upgrade. ⊖Downgrade.

^a Deficiencies in allocation concealmentand blinding.

^b High heterogeneity.

^c CI Crosses the equivalence margin.

4. Discussion

To the best of our knowledge, our study is the first meta-analysis to evaluate the efficacy and safety of Pilates for treating KOA. In recent decades, the importance of exercise therapy in the treatment of chronic diseases has become increasingly prominent, and it is now considered a core component of evidence-based management for all patients with KOA [44^,45]. Exercise is primarily categorized into three types: aerobic exercise, resistance training, and mind-body exercises [46]. Mind-body exercise is a form of low-intensity exercise that enhances coordination between the body and consciousness, focusing on the interactions among the mind, brain, behavior, and body [47]. Compared to traditional resistance and aerobic exercises, mind-body exercises have been proven to provide additional physiological and psychological benefits [48]. For middle-aged and elderly KOA patients who are susceptible to drug side effects and potential drug interactions, incorporating mind-body exercise as a complementary therapy offers numerous advantages. In the realm of mind-body exercises, Tai Chi and yoga, which are among the most representative forms, have been recommended in the latest KOA guidelines published by ACR [6]. Pilates, as a unique form of mind-body exercise, is becoming increasingly popular worldwide. In Brazil, it has become the second most popular exercise among people over 60 years old [49]. Joseph Hubertus Pilates, the creator of the method, defined it as the coordination and balance between body and mind during exercise execution, emphasizing posture and breathing [50]. When practicing Pilates, the focus is more on the quality of the exercise rather than the intensity typically associated with strenuous physical exercises. This is reflected in the relatively low total energy expenditure during a Pilates session, which ranges from 64 to 213 kcal [^,,51^,52]. Previous trials have confirmed that Pilates offers many benefits for middle-aged and elderly individuals, such as improved physical balance [53], enhanced cardiopulmonary function [54], better sleep quality [55], and strengthened core muscle strength [56]. Although previous studies have reported improvements in joint pain and function in patients with KOA after Pilates exercises [^,36^,37], the number of related studies remains limited and the results vary significantly between studies. Therefore, conducting a meta-analysis on the efficacy and safety of Pilates-based interventions for KOA in RCTs, and exploring the impact of different factors on treatment outcomes, is crucial for promoting the widespread adoption of Pilates and recommending more suitable exercise prescriptions.

Our results suggest that Pilates may be more effective in alleviating pain than health education, but not more so than other forms of exercise (isometric, closed-chain, and neuromuscular exercises). Pain, often the primary reason KOA patients seek medical help, is primarily related to inflammatory stimuli and nerve transmission [57]. It frequently causes patients to avoid moving the affected joint, thereby reducing walking and other activities, which in turn accelerates the decline in muscle strength. Muscle weakness is a significant risk factor for the progression of KOA, playing a critical role in joint stability and closely relating to the patient’s perception of pain [58]. Exercise can alleviate pain by reducing inflammation and enhancing muscle strength. In osteoarthritis rat models, resistance training has been shown to decrease MMP-2 activity in the quadriceps tendons [59], while treadmill and wheel exercises can reduce levels of IL-1β, IL-6, and TNF-α, and modulate JNK/NF-κB signalling to suppress inflammation [60]. Exercise training can also effectively increase muscle cross-sectional area and reduce muscle fiber density, improve tendon structure, and thus absorb more forces acting on the joint [61]. The 5 trials included in our study that used the VAS score as an outcome consistently reported significant reductions in pain scores in the Pilates group after treatment. Meenakshi C [38] noted that Pilates exercises may stimulate the release of endorphins by alpha-associated mechanoreceptors at the spinal level and inhibit pain by continuously stimulating A-β muscle fibres, thereby improving proprioceptive mechanisms in the knee joint. Rêgo T.Am [41] suggested that Pilates improves neuromuscular control and adjusts body posture to relieve pain, a finding consistent with the results of Patti A [62]. Nahid Karimi [42] believes there is a bidirectional relationship between muscle pain and weakness. If the force applied to a joint during activity is not adequately absorbed, it can lead to microfractures in the subchondral tissue, further exacerbating the pain. Pilates can enhance strength, endurance, and flexibility, mobilizing both the body and the brain to slow the progression of joint damage and manage various primary symptoms of KOA. Some Pilates exercises, such as single-leg circle movements, incorporate elements of isometric exercise; the shoulder bridge itself falls within the category of closed-chain exercises, and balance training falls within the scope of neuromuscular training. This may explain the similarities in effectiveness between Pilates and other exercises found in our study.

In terms of improving physical function, our study’s pooled data on the WOMAC total score show that Pilates performed better compared to blank control. The pooled data for WOMAC function further illustrate Pilates’ significant advantages in enhancing physical function. These findings are consistent with previous meta-analysis results by Fernández-Rodríguez R on the impact of Pilates on the physical function of the elderly [12], demonstrating that Pilates exercises can improve the flexibility and strength of the back and limbs in middle-aged and elderly individuals, thereby positively affecting the function of various body parts. Previous research indicates that the weakening of the quadriceps muscle strength in older adults is closely associated with a decline in joint position sense and lower limb function, which are critical determinants of the severity of disability [63]. Therefore, improving muscle strength and joint proprioception is essential for enhancing physical function. Pilates exercises focus on increasing the strength, endurance, and flexibility of core muscles, which play a crucial role in stabilizing peripheral joints. Moreover, Pilates may enhance body function by focusing efforts on correctly activating specific muscles at the right speed and controlling movements, thereby influencing proprioception [16].

Despite the 3 trials included in this study reporting significant improvements in ROM for the Pilates group before and after treatment, our pooled analysis indicate that Pilates does not have a significant advantage over other forms of exercise. This result may be attributed to the biomechanical focus of Pilates, which emphasizes core stability, neuromuscular control, and muscle strengthening, rather than specifically targeting joint flexibility or contracture release. In contrast, traditional stretching and flexibility training have been shown to effectively improve ROM by restoring the muscle-tendon length–tension relationship and reducing periarticular stiffness [^,64^,65]. Although Pilates theoretically activates the musculature surrounding the knee joint to improve ROM, its training protocols rarely include eccentric stretching or joint mobilization techniques. In knees with structural alterations or soft tissue contractures, such components may be essential for effectively improving ROM. Additionally, we speculate that the large standard deviations in ROM observed in these 3 studies are partly responsible for the negative results, and these deviations are closely associated with the measurement methods and the size of the sample. In each of these trials, the number of patients per group did not exceed 15, making individual variations more impactful on the means and standard deviations. Therefore, conducting larger, high-quality studies in the future is crucial for this field. Differences in measurement instruments, precision of joint angle positioning, and inter-rater consistency may have also significantly influenced the accuracy of the data. In the trial by Akodu AK and colleagues [36], the method of measuring joint ROM was not specified, and the standard deviation reached up to 16.3°; Nahid Karimi and colleagues [42] used a goniometer, reporting a reliability coefficient of 0.93 by a single evaluator, indicating good consistency; Essam A. Abd El Bakk and colleagues [43] also used a goniometer but did not specify the exact measurement method, personnel involved, or how consistency was ensured, with the standard deviation reaching up to 8.12°. Future research on the efficacy of exercise therapy should emphasize the accuracy and methodology of measuring ROM. It should also be noted that the commonly recognized benefits of Pilates—such as enhanced neuromuscular control and dynamic postural stability—are often not captured by conventional static ROM assessments. Therefore, ROM may not be the most sensitive or appropriate indicator for detecting the functional improvements achieved through Pilates training. Future studies should consider incorporating objective measures such as muscle strength testing, gait analysis, or proprioceptive control to more accurately evaluate the biomechanical effects of Pilates interventions.

In terms of safety, although none of the 8 RCTs included in this study reported adverse events related to Pilates interventions, only two studies explicitly stated that adverse events were actively monitored during the intervention and follow-up periods. The limited data restrict our ability to draw definitive conclusions regarding the safety of Pilates for patients with KOA. Given the overall suboptimal quality of the included research, caution is still advised for patients with poor physical condition and older age who undertake some of the more intense movements without professional supervision and guidance, due to the potential for exercise-induced injuries. In an observational study by Segal N.A. and colleagues [66], one out of 47 patients experienced mid-back pain during supine exercises, and two patients new to Pilates reported lower back muscle pain after the initial two months of classes, no joint pain was reported. According to previous meta-analyses related to Pilates [67], the incidence of adverse events and reactions during Pilates exercises is very low, and no severe adverse events caused by Pilates have been reported before.

The high heterogeneity observed in our meta-analyses, particularly for VAS (I² = 91%) and WOMAC total scores (I² = 83%), highlights the complexity of interpreting pooled effects in Pilates interventions for KOA. Our subgroup analyses provide some insights into potential contributors to this heterogeneity. According to our meta-analysis results, heterogeneity was notably higher in comparisons of Pilates with other forms of exercise (I² = 77%) and with no-intervention control groups (I² = 68%) than in the overall WOMAC meta-analysis (I² = 83%). This suggests that differences in the type of intervention between experimental and control groups may be one of the contributing factors to the observed heterogeneity. However, subgroup analyses based on age, intervention duration, geographic region, and sample size did not provide sufficient evidence to support a differential effect of Pilates across these factors. Although one study with an intervention duration of less than 8 weeks reported a significant reduction in VAS scores compared to controls, while pooled results from studies with durations of 8 weeks or more showed no such difference, we believe these positive findings may be influenced by confounding factors and should not be interpreted as indicating a decline in the effectiveness of Pilates with longer durations. In fact, more frequent or longer-term exercise may exert cumulative physiological benefits [^,68]. While subgroup analyses revealed some changes in heterogeneity, suggesting that the examined factors might potentially contribute to between-study inconsistency, these differences were not substantial enough to confirm any of them as the primary source of heterogeneity in this study. The high level of heterogeneity and the negative findings from subgroup analyses may be attributed to the limited number of available studies in this field and the methodological inconsistencies across studies. Moreover, although no differential effects were detected in our predefined subgroup analyses, we acknowledge that numerous potential factors may warrant further investigation to identify the optimal target population and dosing strategy for Pilates. These factors may have collectively contributed to the substantial heterogeneity observed in this study. Such factors include, but are not limited to, participants’ BMI, K-L grade, comorbidities, the mode of Pilates training (mat-based vs. equipment-based), training dosage (frequency × duration × intensity), supervision modality (remote guidance/on-site instruction, individual vs. group sessions), intervention duration, and adherence. However, due to the limited number of available studies and inconsistent reporting, further exploration of these aspects was not feasible. Future trials should strive to standardize intervention protocols and incorporate stratified or subgroup analyses based on these factors to generate higher-quality evidence for the use of Pilates in managing KOA.

In clinical practice, poor adherence remains a major challenge in exercise-based interventions, particularly in terms of long-term compliance [69]. There is a complex relationship between exercise adherence and therapeutic benefit. Although Pilates is generally considered an accessible and well-tolerated form of exercise, its long-term adherence among patients with KOA remains unclear. None of the eight RCTs included in this study reported data on continued participation after the intervention or on patients’ adherence to the prescribed programs. A meta-analysis of Pilates interventions for chronic musculoskeletal conditions indicated that adherence to Pilates also varies, with group-based interventions generally achieving higher attendance rates, while home-based formats are associated with lower adherence [70]. This aligns with existing views that structured and supervised programs improve participation, whereas self-directed interventions are more prone to discontinuation and dropout. These findings highlight the need to address the sustainability and real-world applicability of Pilates. To better understand the long-term value of Pilates in chronic KOA management, future RCTs should include follow-up assessments of adherence beyond the intervention period. The use of standardized tools such as the Exercise Adherence Rating Scale (EARS) or objective monitoring methods (e.g. session attendance records, wearable devices) is recommended to quantify participation. Only by confirming sustained adherence can the effectiveness and practical utility of Pilates in KOA be reliably evaluated.

It is noteworthy that the trials included in our study did not involve efficacy indicators related to mental health, the nervous system, or states of anxiety/depression. Given that KOA is a psychosomatic disease, changes in mental state and the nervous system are often closely linked to long-term therapeutic effects [71]. Mind-body exercises such as Tai Chi and yoga have been quantitatively evaluated for their effects on mental health outcomes; however, the impact of Pilates on psychological well-being in KOA patients has not yet been addressed in existing research. This remains an important gap that our study was also unable to fill. Future research should include relevant indicators to clarify the effectiveness of Pilates in mental and emotional aspects and to identify its potential advantages.

Moreover, one methodological consideration that may have influenced the interpretation of our findings is the use of SMDs in cases where outcome scales differed across studies. In our meta-analysis, we used MDs for VAS outcomes because all included studies employed a comparable 0–10 pain scale, allowing for direct comparison of absolute treatment effects. In contrast, SMDs were employed for WOMAC total scores and ROM, as different studies used varying versions of the WOMAC scale (e.g. 0–96, 0–100) and different methods of measuring ROM, necessitating normalization of effect sizes. While SMD enables the synthesis of heterogeneous measurement scales, it introduces several limitations. SMDs are influenced by between-study variation in standard deviations, which may reflect underlying heterogeneity in population characteristics (e.g. baseline variability, disease severity) or measurement precision. Moreover, SMD values are less intuitive to interpret clinically, as they do not correspond to meaningful clinical units such as the minimal clinically important difference (MCID). Therefore, although the use of SMD was necessary to ensure comparability across studies, it may have contributed to the observed heterogeneity and limited the direct clinical interpretability of our pooled results. Future trials should consider using consistent, validated measurement tools and reporting raw mean differences when feasible to improve the applicability of meta-analytic findings in clinical settings.

Another important methodological consideration in our meta-analysis is the potential presence of publication bias, particularly for VAS and WOMAC outcomes. This was indicated by asymmetrical funnel plots and statistically significant results from Egger’s tests, suggesting that studies with small sample sizes or negative results may be underrepresented in the published literature. Given that all included RCTs were published in English, language bias may also have contributed to the observed asymmetry. Publication bias can lead to an overestimation of treatment effects and reduce the credibility of pooled results, particularly in fields with a limited number of studies such as ours. Future meta-analyses should consider including gray literature and non-English publications where available, and researchers conducting primary studies are encouraged to publish null or negative findings to improve transparency and reduce bias in evidence synthesis.

A substantial body of systematic reviews and meta-analyses has provided robust evidence supporting the efficacy of exercise therapy for KOA. Fransen et al. reported that land-based exercises produced moderate improvements in pain and physical function in KOA patients, with benefits persisting for at least 2–6 months after cessation of formal treatment—an observation that has been widely accepted [72]. Wang et al. summarized the effects of mind-body exercises such as Tai Chi, yoga, and Baduanjin, and found that these interventions significantly improved pain, stiffness, physical function, and quality of life [73]. However, their analyses were limited by variability in study quality and high heterogeneity. In comparison, our meta-analysis is the first to systematically evaluate the efficacy of Pilates specifically in patients with KOA. Although our findings demonstrated comparable improvements in pain and physical function, no significant benefits were observed in ROM, contrasting with prior evidence that mind-body practices such as Tai Chi and yoga are often associated with ROM enhancement. Furthermore, meta-analyses of these other modalities typically include a larger number of high-quality RCTs (e.g. 18 trials involving ≥1,100 participants), whereas our analysis incorporated only 8 RCTs, most of which were limited by small sample sizes, high heterogeneity, and low to very low certainty of evidence as assessed by GRADE. Therefore, while the potential benefits of Pilates on pain and function appear to be in line with existing evidence for other mind-body exercises, it falls short in terms of overall evidence strength, sample size, and coverage of diverse outcomes. Finally, it is important to emphasize that the underlying mechanisms of Pilates differ somewhat from those of Tai Chi and yoga. Pilates focuses on core muscle activation, body control, and breath coordination, emphasizing trunk stability and neuromuscular re-education. Future high-quality research should further investigate and compare these mechanistic differences.

5. Limitations

We must acknowledge that this study has some potential limitations due to objective conditions. First, the literature search was limited to English-language publications, which may have contributed to the publication bias observed in this study and led to the omission of important evidence that could potentially alter the study’s conclusions. Second, although we included only randomized controlled trials to enhance the reliability of the evidence, the quality of the included studies is concerning. The overall effect size is likely influenced by unreliable randomization methods, the inability to blind doctors, participants, or statistical analysts, and other aspects of the study designs. Third, there was substantial heterogeneity among the included studies. Although our subgroup analyses identified some potential sources of heterogeneity, they did not significantly reduce the overall level of heterogeneity, and many contributing factors remain unclear. Therefore, the results of this article should be treated with caution.

6. Conclusion

Pilates exercise may improve pain and physical function in patients with KOA, showing comparable effects to other exercise interventions. However, no significant difference was observed in terms of ROM compared to the control group, and its clinical relevance remains uncertain. Importantly, these findings should be interpreted with caution due to the very low certainty of evidence, high heterogeneity, and methodological limitations of the included studies Given that only a few studies conducted active monitoring of adverse events, the safety profile of Pilates remains uncertain. Lastly, current clinical studies on Pilates lack assessments of mental and psychological health. Future studies should adopt standardized protocols, assess long-term adherence, and place greater emphasis on evaluating mental health outcomes.

Funding Statement

This research received no external funding.

Ethical approval

This study has no ethical implications.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Provenance and peer review

Not commissioned, externally peer-reviewed.

Patient and public involvement

Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.

Data availability statement

All authors confirm that all data used in this study were obtained from published randomized controlled trial articles in public databases, and all data and results are presented in the main text and supplementary files. Reasonable requests for the original statistical analysis data can be addressed to the corresponding author.