Optimal exercise dose and type for improving schizophrenia symptoms in adults: A systematic review and Bayesian network meta-analysis

A B S T R A C T Background: Exercise has shown promise in aiding patients with schizophrenia, potentially improving symptoms, quality of life, mood, and cognition. This review evaluates the effectiveness of different physical activities in this context, aiming to guide treatment and research towards the most beneficial exercise interventions for schizophrenia management. Methods: This systematic review and network meta-analysis searched PubMed, MEDLINE, Embase, PsycINFO, Cochrane CENTRAL, Web of Science, and CNKI up to March 30, 2024. We utilized the "gemtc," "MBNMAdose," "metafor," and "ggplot2" packages for data analysis in R. Additionally, our results were reported as standardized mean differences with 95% confidence intervals. To assess the risk of bias in the included studies, we used ROB2 and CINeMA tools. Results: In the results section of our systematic review and network meta-analysis, we analyzed data from 47 studies involving 4031 participants. Our findings indicate that among the various exercise interventions examined for schizophrenia patients, yoga emerged as the most effective in improving Total symptoms, positive and negative symptoms. RT + AE (Resistance Training + Aerobic Exercise) showed the greatest improvement in general symptoms, whereas Tai Chi (TC) was the least effective. In terms of improvement of cognitive functions and depression, AE proved the most beneficial, while TC was found to be the least effective. EX + PT (Physical Exercise + Psychological Nursing) was identified as the most effective approach for improved quality of life, with TC again showing the least efficacy. The optimal total exercise dose to improve the patient ’ s total symptoms was estimated to be 1200 METs-min/week. (SMD: (cid:0) 0.956, 95%Crl: (cid:0) 1.376 to (cid:0) 0.536). Conclusions: Yoga, RT + AE, EX + PT, and Baduanjin have demonstrated enhanced efficacy in alleviating clinical symptoms, quality of life, depressive states, and cognitive functions in patients with schizophrenia. We identified optimal overall exercise doses and provided exercise guidance recommendations for healthcare professionals.


Introduction
Schizophrenia is a leading cause of mental disability worldwide (Charlson et al., 2018).Approximately 30-40 % of individuals with schizophrenia exhibit treatment-resistant forms, commonly referred to as refractory schizophrenia (Visceglia and Lewis, 2011).Despite the widespread use of both typical and atypical antipsychotics as the primary treatment strategy, medication adherence among patients with schizophrenia remains a significant challenge, with partial or nonadherence rates reaching 60 %.This high rate of nonadherence is often attributed to a lack of insight into the illness (Prasko et al., 2011) and the severe side effects associated with antipsychotic medications (Marder et al., 2004;Newcomer and Hennekens, 2007;Taylor et al., 2017).Therefore, the search for alternative treatments for schizophrenia is urgent.Increasing scientific evidence suggests that physical activity and exercise should be highly considered in preventing and treating schizophrenia disorders (Bueno-Antequera and Munguía-Izquierdo, 2020).
A 20-week randomized controlled trial (RCT) has revealed that resistance training programs and combined aerobic and resistance exercises significantly relieve schizophrenia symptoms (Silva et al., 2015).Furthermore, investigations have demonstrated that comprehensive community-based interventions, such as the amalgamation of social skills training and tai chi, exhibit the potential to ameliorate clinical symptoms, bolster quality of life, and enhance medication adherence among individuals with schizophrenia (Kang et al., 2016).Similarly, a meta-analysis has underscored the association between aerobic training and advancements in language acquisition, memory retention, and reasoning and problem-solving abilities (Xu et al., 2022).Thus, exercise stands as an effective avenue for the amelioration of clinical symptoms in schizophrenia.However, determining the distinct superiority among various physical activities remains complex through the utilization of RCTs or pairwise meta-analyses.Consequently, selecting the most efficacious exercise intervention to enhance clinical outcomes in individuals with schizophrenia poses a challenging quandary for both clinicians and patients.
Network meta-analysis (NMA) is an advanced statistical method that simultaneously compares multiple interventions.NMA not only integrates the results of direct comparison studies, but also uses indirect comparisons to infer the relative effects of interventions that are not directly compared.The advantages of this method are its ability to handle multiple comparisons, provide a global perspective, and provide a ranking of treatments, thereby providing a more comprehensive basis for clinical decision-making.These characteristics give NMA a significant advantage, especially when evaluating multiple interventions in the absence of comprehensive head-to-head studies.In addition, understanding the dose-response relationship, which refers to the link between the dose (i.e., the amount or intensity of the intervention) and the resulting physiological or behavioral response, is critical to optimizing treatment outcomes (Martin, 2020).The meta-analysis result of Firth et.al.shows exercise of approximately 90 minutes of moderate to vigorous exercise per week significantly reduced psychiatric symptoms (Firth et al., 2015).Follow this, the meta-analysis result of Firth et.al.show that it is possible to improve the cognitive ability of schizophrenic patients, which means that it can be used to increase the amount of movement (Firth et al., 2017).However, there is a lack of evidence-based guidance on how to carefully formulate exercise prescriptions to optimize intervention outcomes while avoiding excessive or insufficient exercise dosages for individuals with schizophrenia.Current exercise prescription guidelines often extrapolate from studies involving the general population, specifically healthy, untrained middle-aged and elderly individuals.However, the literature notably lacks comprehensive and effective exercise dose-response recommendations tailored to individuals with schizophrenia.Therefore, a thorough examination of the intricate and multifaceted relationship between exercise type, dosage, and subsequent intervention effects is essential.
Recently, Bayesian framework-based model-driven dose-response network meta-analysis had been effectively utilized (Sadeghirad et al., 2023;Yuan et al., 2024).This approach integrates Bayesian statistics with dose-response models to investigate the effects of different exercise dosages on individuals with schizophrenia.In contrast to conventional meta-analyses, this methodology embeds dose-response relationships within statistical models, enabling a more accurate assessment of the relationship between exercise dosage and schizophrenia.While traditional meta-analyses primarily aggregate effect sizes across studies, Bayesian framework-based model-driven meta-analysis offers a broader scope of information.This includes credible intervals for dose effects, estimations of uncertainties, and probabilistic comparisons among different dosages, offering a more holistic foundation for decision-making.
This study employs an advanced network meta-analysis technique, specifically focusing on model-driven dose-response network metaanalysis within the Bayesian framework.It synthesizes evidence from existing randomized controlled trials to evaluate the efficacy of ten different exercise interventions in improving clinical symptoms, depression, cognition, and quality of life in individuals with schizophrenia.Moreover, it examines the dose-response relationship between the overall exercise dose and the comprehensive management of schizophrenia.

Methods
This pre-registered systematic review with network meta-analysis (PROSPERO reference number #CRD42023428287 was reported following the PRISMA checklist (Supplementary 1).

Search strategy
We conducted a systematic search in PubMed, MEDLINE, Embase, PsycINFO, Cochrane Central Register of Controlled Trials (CENTRAL), Web of Science, and China National Knowledge Infrastructure (CNKI) from their inception date to March 30, 2024, with no language restrictions.The reference lists of relevant articles and reviews were also screened for additional studies.Title/abstract and full-text screening were conducted independently and duplicated by investigators (HYZ and XYF), with disagreements resolved by discussion or adjudication by a third author (YY).

Study selection
The inclusion criteria were based on the PICOS (participants, interventions, comparators, outcomes, and study design) approach (Hutton et al., 2015).(1) the participants were diagnosed with schizophrenia at baseline according to a structured diagnostic interview, and the mean age ≥ 16 years; (2) the exercises were divided into 10 types according to their content; (3) the usual care group (UC), regular daily activities, wait-list, and health education.Besides, for head-to-head studies, the comparator may be any of the 10 exercise types; (4) Studies were required to include at least one of the outcome measures of interest: the primary outcome was total symptom changes, and secondary outcome included positive, negative, and depressive symptoms of schizophrenia, quality of life, and cognitive level.And we provide a measurement method for different outcomes in the Supplementary 2; (5) in the study design, we included published RCTs (individual design, cluster design, or the first half of crossover).
We followed exclusion criteria: (1) the acute effects of a single session on schizophrenia patients; (2) studies that did not clearly describe the types of exercise and the dose of treatment; (3) participants have not assessed schizophrenia at baseline; (4) the sample had a mean age below 16 years.

Data extraction
After all relevant articles were searched in the mentioned databases; they were stored in an EndNote X9 reference manager.Two authors independently extracted data from studies that met the inclusion criteria (HYZ and XYF) and disagreements were resolved by consensus between all authors.Relevant publication information (i.e., author, title, year, journal), number of patients, patient characteristics (e.g., age and sex), interventions considered, and outcome measures were extracted.The formulas utilized for the calculation of Mean change and SD change values were: According to the guidelines of the Cochrane Handbook, the correlation coefficient (Corr) was set to 0.5 (Higgins, 2008).In addition to meeting the data analysis requirements of the Dose-Response network meta-analysis Package in R, we converted the standard errors (SE).n as sample size.
If the data needed for the study could not be extracted from the above methods, we contacted the authors about the data at least four times within six weeks.

Data coding and management
We categorized interventions into three levels (Gallardo-Gómez et al., 2022): First, interventions were coded as "Exercise" and "Usual Care (UC)" At the second level, the interventions were coded according to their main exercise type: "Yoga", "Ping-Pong (PP)", "Resistance Training (RT)", "Aerobic Exercise (AE)", "Baduanjin", "Resistance Training + Aerobic Exercise (RT+AE)", "Tai Chi (TC)", "Balance and Tone training (BT)", "Usual Care (UC)", Physical Exercise + Psychological Nursing (EX+PT)", "Multi-component exercise (Mix)".Finally, interventions were coded at the intersection of overall exercise and dose-defined as the energy expenditure (i.e., Metabolic Equivalent of Task, MET) that results from the product of the duration, frequency, and intensity of a certain type of exercise (Ainsworth et al., 2011;Wasfy and Baggish, 2016); and expressed as METs-min per week.To better promote the connectivity of the network, we performed an approximate value of 0 (UC), 200, 300, 400, 600, 700, 800, 900, or 1200 METs-min per week for the exercise dose, which has been adopted in previous studies (Gallardo-Gomez et al., 2022), and which is a necessary step for the network meta-analysis (J.P. T. Higgins et al., 2012).

For network meta-analysis
We analyzed all the data based on the R statistical environment (V.4.2.2, www.r-project.org), and we used the "gemtc" package to perform Bayesian network meta-analysis combining direct and indirect comparisons (Shim et al., 2019).The effect sizes were measured using the standardized mean difference (SMD) of the change score due to the varying rating scales or units of the outcome across studies.Additionally, a random-effects model was employed to combine the data.and the surface under the cumulative ranking curve (SUCRA) was used to rank the treatments.
On the other hand, we utilized the tau-squared (τ 2 ) test and I 2 statistic to analyze the statistical heterogeneity between the studies.Inconsistency was assessed statistically using the design-by-treatment interaction test for global consistency, and we separated indirect from direct evidence using the SIDE (Separating indirect from direct evidence) test (Dias et al., 2010;J. P. Higgins et al., 2012).
Additionally, we used the "metafor" package to evaluate the robustness of the treatment effects for the main outcome (Total symptom) in the meta-regression, considering factors such as mean age, percentage of female participants, sample size, and exercise duration.Finally, we compared the adjusted funnel plot to assess the risk of publication bias under specific circumstances and Egger's test was suggestive of publication bias when p<0.05 in Stata software (Chaimani et al., 2013;Shim et al., 2017).

For dose-response analysis
We analyzed the relationship between the overall exercise dose and total symptom using the MBNMAdose package in the R statistical environment (v4.2.2, www.r-project.org).We employed various models, including the Emax function, restricted cubic spline (RCS), nonparametric model, and exponential model, to fit the data for both random and fixed effects models.To select the best-fitting model, we compared different fitting metrics such as the deviation information criterion (DIC), standard deviation, model parameters, and residuals (Wheeler et al., 2010).Based on these results, the restricted cubic spline model with random effects demonstrated the best fit and we plot a nonlinear relationship graph in Supplementary 8, Fig. 1.At the same time, the model-fitting comparison metrics results were provided in Supplementary 8, table 8.2.Therefore, we selected the restricted cubic spline model with random effects (DerSimonian and Kacker, 2007;Mawdsley et al., 2016;Pedder et al., 2019).We visualized the relationship and model fitting between overall exercise dose and total symptoms using line and box plots in Supplementary 8, table 8.3.
Finally, to visualize our model-fitting prediction results, we placed three knots at the 10th, 50th, and 90th percentiles of the treatment dose.
All key dose-response prediction results were estimated using the MCMC model of beta coefficients on the restricted cubic spline curves (Markov Chain Monte Carlo: 3 chains, 20,000 iterations, with the first 10,000 discarded, n. thin = 10).

Risk of bias and quality of evidence
Three reviewers (YY, HYZ, and XYF) assessed and rated the studies according to the Cochrane Risk of Bias 2.0 (ROB2) criteria with low risk, some concerns, and high risk (Sterne et al., 2019).According to RoB2, the following domains were considered for the assessment of the risk of bias: randomization process, deviations from intended interventions, missing outcome data, measurement of the outcome, and selection of the reported result.
We examined the confidence of evidence using the CINeMA (Confidence of Network Meta-Analysis) web application with judging each comparison for within study bias, reporting bias, indirectness, imprecision, heterogeneity, and inconsistency, according to recommended guidelines, which allows the confidence of the results to be graded as high, moderate, low, and very low (Nikolakopoulou et al., 2020).

Sensitivity analysis
In sensitivity analysis, we excluded studies assessed as high risk and reanalyzed the results of individual outcome variables.In addition, the effects of exercise intervention on overall schizophrenia symptoms were also included in the dose-response analysis.We compared the differences with the original analysis after excluding high-risk studies, focusing on assessing the robustness and consistency of the results.

Characteristics of included studies
Through initial electronic searches, we identified 468 potentially eligible studies.After removing duplicate studies, we selected 457 studies.Following the screening of titles and abstracts, we considered 220 studies that potentially met the inclusion criteria.Based on the PICOS inclusion and exclusion criteria, we finalized 47 articles and retrieved the full-text articles (Fig. 1).The characteristics of the included studies were shown in (Supplementary 3).The publication years ranged from 2005 to 2022.The average age of the reports' participants was ≥16 years.Among the 4031 participants, 2398 (59.48 %) were male, and 1633 (40.52 %) were female.The treatment duration ranged from 4 to 96 weeks (with a median treatment duration of 12 weeks).

Network geometry
Whether connectivity is met determines the basis of NMA.Lack of connectivity can lead to low statistical power and misleading results when direct comparison is not possible (Rouse et al., 2017).This network meta-analysis included multiple interventions, and Fig. 2 shows the number of comparative studies in different symptom domains.For total symptoms, a total of 31 studies (n = 1364) were included, of which eight interventions, including Ba Duan Jin, Tai Chi, and Yoga, participated in 1-18 comparative studies, respectively.In the analysis of positive symptoms and negative symptoms, 26 studies were included (positive symptoms n = 1256, negative symptoms n = 1276), and each intervention participated in 1-21 comparative studies in their respective fields.The general symptom analysis included 14 studies (n = 624), and each intervention participated in 1-12 comparative studies.The analysis of cognitive function included 17 studies (n = 655), and each intervention participated in 1-11 comparative studies.In the analysis of quality of life, 16 studies (n = 701) covered 1-7 comparative studies for each intervention.The analysis of depressive symptoms included 11 studies (n = 544), with each intervention involved in between 1 and 12 comparison studies.The figure further details the number of comparison studies for each intervention in different symptom domains.

Total symptom results
Compared to UC, Yoga, PP, RT, AE, RT+AE, TC, and BT demonstrated improvements in total symptoms, Although BT and Ba Duan Jing had rankings, they did not effectively improve participants' total symptoms.When SUCRA ranked by the degree of overall symptom changes, Yoga was considered the most effective (SMD: − 1.02, 95 % CI: − 1.38 to − 0.66) (Fig. 3 A

Cognitive function results
Compared to UC, only AE showed significant improvement in cognitive function (SMD: 0.56, 95 % CI: 0.09-1.03).Although Ba Duan Jin, Yoga, RT, BT, RT+AE, and TC methods were SUCRA ranked high, they did not significantly improve the cognitive level of the participants.(Fig. 3 B).

General symptoms results
Compared to UC, RT+AE, RT, and AE significantly improved general symptoms, while BT, Mix, Yoga, and TC exhibited no significant changes.When SUCRA ranked by the degree of general symptom changes, RT+AE was identified as the most effective (SMD: − 1.22, 95 % CI: − 1.99 to − 0.45) (Fig. 3 D).Fig. 4. : Heat of exercise ranked according the associated of alteration in total, positive, negative, and depressive symptoms of schizophrenia, quality of life, and cognitive.Numbers reflect SUCRA, which ranks exercise on a continuous from 0 to 1.The higher the P score, the better the improvement of symptoms of schizophrenia.

Negative symptoms results
Compared to UC, Yoga, RT+AE, and AE showed improvements in negative symptoms, while RT, Mix, TC, and BT exhibited no significant changes.When SUCRA ranked by the degree of negative symptom changes, Yoga was identified as the most effective (SMD: − 1.01, 95 % CI: − 1.50 to − 0.53), while BT was deemed the least effective (Fig. 3 E).

Positive symptoms results
Compared to UC, Yoga, RT+AE, and AE showed improvements in positive symptoms, when SUCRA ranked by these exercises, BT, Mix, RT, and TC exhibited no significant changes, meanwhile, Yoga was identified as the most effective (SMD: − 0.91, 95 % CI: − 1.38 to − 0.43) (Fig. 3 12.7.Quality of life results Compared to UC, AE, EX+PT, Mix, and RT+AE improved quality of life, when SUCRA was ranked these exercises, RT, Yoga, BT, and TC did not significantly change the quality of life, meanwhile, EX+PT was identified as the most effective (SMD: 1.02, 95 % CI: 0.77-1.28)(Fig. 3 G).

Treatment effective ranking results
Finally, we provide different league table rankings for various outcome indicators in Supplementary 5. Additionally, the SUCRA rankings of exercise interventions are concentrated and displayed in a heatmap (Fig. 4).

Dose-response relationships
We plotted the non-linear dose-response relationship between the overall exercise dose and total symptoms by ggpot2 package in R (Fig. 5).At the same time, we found a U-shaped between the overall exercise dosage and total system, and the optimal overall exercise dose peak was identified.We observed that the effective range of overall exercise dose was from 340 to 1200 METs-min/week, and the optimal exercise dose was estimated at 1200 METs-min/week (SMD: − 0.956, 95 %Crl: − 1.376 to − 0.536) for improving total symptoms.

Additional analyses
Our comparison-adjusted funnel plot had good symmetry, and the linear fitting line (green) was not perpendicular to quadrant 0, additionally, according to Egger test results of different outcomes, we found no significant risk of bias.Therefore, no small study effect was found in the outcomes (Fig. 6).
The heterogeneity most outcomes was moderate to high.The design of the total system handling interactive inspection results indicates insignificant global inconsistencies.(Q=9.82,τ 2 =0.1385, p=0.5463).The SIDE test of total symptoms showed that the percentage of comparisons with evidence of inconsistency was 0 %.The SIDE test of all outcomes showed that the percentage of comparisons with evidence of inconsistency ranged from 0 % to 9.09 %.For this part, we provide detailed analysis results in Supplementary 6.
Additionally, through meta-regression analysis of the primary outcome (Total symptom) baseline characteristics and exercise training doses, we addressed the heterogeneity sources related to the inclusion of baseline features and exercise training doses in this study (Supplementary 7).

Risk of bias and quality of evidence
Overall, 16 studies (34 %) were classified as having a low risk of bias, 18 studies (38 %) were classified as having an unclear risk of bias, and 13 studies (28 %) were classified as having a high risk of bias.Fig. 7 presents the results of the Cochrane Risk of Bias Tool 2 and the studylevel risk of bias assessments.We used the ggplot2 package to create Fig. 5. : Dose-response association between overall exercise dose and change in total symptoms in Schizophrenia.Dotted line: The optimal exercise dose.
Y. Yang et al. Neuroscience and Biobehavioral Reviews 167 (2024) heatmaps to visualize the risk of bias results.Additionally, we graded the evidence for total symptoms in the network meta-analysis using CINeMA.For the main results, we assessed the credibility of 97.2 % of the evidence as very low.CINeMA confidence ratings are shown in the appendix (Supplementary 9).

Sensitivity analysis
The results showed that after excluding studies with a high risk of bias, the effect size did not change significantly compared to the original analysis.This suggests that our main conclusions are robust, even when high-risk studies are excluded.Additionally, the dose-response results showed that the effective dose for improving schizophrenia was between 300 and 1000 METs-min/week with the optimal dose at 690 METs-min/ week (SMD: − 0.606, 95 %Crl: − 0.945 to − 0. 242).We provided detail sensitive analysis results in the Supplementary 10.

Main findings
This study for the first time reveals the optimal physical exercise interventions for clinical symptoms, depression, cognition, and quality of life in schizophrenia patients, as well as the nonlinear relationship between overall exercise and the total system symptoms.By conducting a comprehensive analysis of data from 47 randomized controlled trials involving 4031 schizophrenia patients, we discovered that physical exercise significantly improves clinical symptoms, quality of life, and cognitive function, and reduces depression indices.These findings are consistent with previous systematic reviews and meta-analyses (Firth et al., 2015), further confirming the efficacy of physical exercise in the comprehensive treatment of schizophrenia.However, unlike previous studies, this research not only distinctly categorized the effects of different types of exercise interventions but also provided an analysis of exercise dosages, filling a gap in the existing literature.As anticipated, Yoga showed significant improvement in clinical symptom indicators, although it did not meet the expectations for general symptoms.Notably, a nonlinear dose-response relationship between overall exercise and the total system symptoms in schizophrenia patients was observed, with the effective dosage range identified at 1200 MET-min/week, and we found the optimal physical exercise dose at 1200 METs-min/week to improve the total system symptoms.This finding was crucial for the future development of exercise guidelines for schizophrenia patients, especially in improving clinical symptoms, quality of life, and reducing depression indices.

Strengths and implications
In exploring the effects of different types of exercise interventions, Yoga has shown significant improvement in the clinical symptoms of schizophrenia patients.This finding aligns with previous research (Govindaraj et al., 2020), where Yoga is considered an effective adjunct therapy.Particularly for those schizophrenia patients who are stable on antipsychotic medications, Yoga can improve their psychopathology (Varambally et al., 2012).We attempt to explain the potential mechanism of Yoga in improving the clinical symptoms of schizophrenia patients: Yoga training stimulates an increase in brain-derived neurotrophic factor levels, which aids in enhancing neuroplasticity (Yang et al., 2016), improving neuronal connections, and information transmission.This may help alleviate issues related to cognition and emotional regulation, thereby indirectly reducing the severity of positive symptoms.Additionally, the meditation and relaxation techniques in Yoga can enhance psychological well-being, and increase social participation, and quality of life (Neelam et al., 2012), which are highly beneficial in alleviating negative symptoms.Yoga also affects the endocrine system by reducing the activity of the pituitary-hypothalamic-adrenal axis and slowing down the release of adrenaline and cortisol (Pascoe and Bauer, 2015), which is beneficial in reducing anxiety and stress, factors commonly associated with the overall symptoms of schizophrenia patients.
However, it is noteworthy that Yoga showed relatively less effectiveness in improving general symptoms.Our research found that compared to usual care, RT+AE is most effective in improving general symptoms, where Resistance Training primarily enhances muscle strength and endurance (Hare et al., 1999), and AE improves cardiopulmonary function and circulatory health (Kemi and Wisløff, 2010).These improvements in physical health can reduce anxiety and stress caused by physical discomfort, thereby positively impacting the general symptoms of patients.Moreover, our study also reveals the significant effect of EX+PT in enhancing patients' quality of life, including emotional management, social abilities, and daily functioning.
Further analysis found that AE improves various health indicators in schizophrenia patients.Especially in depressive symptoms, recently, a network meta-analysis published by Michael et al. in BMJ demonstrated that walking or jogging is the most effective way to improve depression (Noetel et al., 2024).This finding is consistent with our results, which confirm the effectiveness of AE in alleviating depression symptoms.Although these improvements are relatively limited.This difference may be related to the intensity of exercise, the duration of intervention, and the frequency of intervention, as these factors affect the specific impact of exercise on patients' physical functions.However, it is important to note that traditional forms of exercise may lead to boredom and fatigue over long periods, affecting participants' enthusiasm and compliance.Interestingly, among all treatment methods, only AE showed significant improvement in cognitive functions.This finding suggests that in designing treatment plans, special consideration should be given to the potential value of AE in improving cognitive abilities in schizophrenia patients.
More worthily, we performed regression analysis to consider covariates such as mean age, sample size, and exercise period.Although we mainly focused on the overall effect of different exercise interventions, this analysis helps to reveal the potential impact of these covariates on the intervention effect.We have partially considered the variability of individual characteristics when evaluating the effect of exercise intervention.Nevertheless, in order to formulate personalized exercise prescriptions, future studies still need to further explore the specific effects of different age groups and individual characteristics on exercise effects.
Additionally, existing research indicates that exercise dosage may be a crucial factor influencing intervention outcomes.Our study results suggest a non-linear improvement relationship between overall exercise and overall symptom relief in patients with schizophrenia.This finding underscores the dependence of the overall exercise dose on symptom alleviation in schizophrenia patients and highlights the necessity of appropriate training regimens.We noted that symptom relief in these patients began to improve significantly with exercise beyond 340 METsmin/week and the graph showed a non-linear improvement in symptoms with increasing exercise dose up to 1200 METs-min/week.with marked improvements occurring within specific dose ranges.Our statistical analysis pinpointed 1200 METs-min/week as the optimal dose for significant symptom relief.In the dose-response analysis, our sensitivity analysis results showed that the optimal dose changed from 1200 METs-min/week to 690 METs-min/week, and the effective range changed from 340 to 1200 METs-min/week to 300-1000 METs-min/ week.Although these changes show the adjustment of the doseresponse relationship after sensitivity analysis, the overall trend still supports the main conclusion that appropriate doses of exercise have a positive impact on health.These changes indicate that sensitivity analysis can help to define the effective exercise dose more accurately, but these adjustments still need to be carefully considered in practical applications to ensure the applicability and effectiveness of the intervention.
This finding not only offers a feasible training schedule but also provides robust guidance for developing tailored exercise intervention plans.In conclusion, our study provides empirical support for a specific dose range of physical exercise for patients with schizophrenia, underscoring the importance of incorporating exercise intensity into treatment plans.However, further research is required to elucidate the mechanisms and long-term effects of physical activity in treating schizophrenia.
It is worth noting that we systematically assessed the risk of bias using the Cochrane tool 2, and the results showed that 13 studies had a high risk of bias.These studies showed a high risk of bias in terms of blinding of participants and personnel, blinding of outcome assessment, and handling of incomplete data, which may have a certain impact on the results of the studies.In particular, the high risk of bias in blinding may lead to an overestimation of the intervention effect due to participants or assessors knowing the treatment allocation.To assess the impact of these biases on the results of the studies, we conducted a sensitivity analysis and excluded these studies with a high risk of bias.Overall, the effect estimates remained consistent, indicating that our main conclusions were robust.However, in the outcome indicators (Total symptoms, Negative), some minor changes did appear after the sensitivity analysis.These changes suggest that although the overall results are still reliable, individual effect sizes may be affected by bias.Therefore, caution should be exercised when interpreting these results, especially in areas with a high risk of bias.Future studies should further reduce bias and enhance the credibility of the results through stricter blinding procedures and improved handling of incomplete data.

Limitations and future directions
It is important to acknowledge the limitations present in our network meta-analysis.First, the inability to access the full texts of certain randomized controlled trials may introduce bias into our results.This limitation heightens the risk of publication bias, potentially causing the findings of our network meta-analysis to deviate from actual outcomes.Second, both our study and the original studies contain inherent limitations.Heterogeneity among the original studies is unavoidable, including differences in the proportion of male and female participants and variations across different regions and ethnicities.
Third, although our result shows the multifaceted positive effects of exercise on patients with chronic schizophrenia, it is important to note that patients face multiple physical barriers in participating in exercise.First, medication-related muscle pain and extrapyramidal symptoms may severely limit patients' ability to exercise, making it more difficult to execute exercise plans.In addition, neuroleptics often cause fatigue and cardiovascular side effects, which not only affect patients' physical fitness but also may increase the risk of cardiovascular events during exercise.Patients with chronic schizophrenia may also show poor handeye coordination, which is particularly evident in sports involving fine movements or complex movements.Furthermore, lack of motivation and obesity are common phenomena in patients with schizophrenia, which further limits their participation in exercise.Lack of motivation not only affects patients' compliance with exercise, but may also reduce the psychological and physical health benefits they gain from exercise.Given these barriers, future studies should consider developing and testing exercise intervention programs that target these special needs.For example, low-intensity, progressive exercise may be more suitable for patients with severe physical disabilities.In addition, given the patient's possible medication side effects, exercise plans should include monitoring and management of cardiovascular health, as well as close collaboration with the medical team to ensure the safety and effectiveness of the exercise program.
On the other hand, although we classified exercise intensity using metabolic equivalents (METs), we only explored the dose-response relationship of overall exercise on clinical symptoms of schizophrenia.Additionally, when conducting the dose-response analysis, we only fit and compared models without testing the three major hypotheses of dose-response.This introduces some methodological heterogeneity; however, considering that we classify exercise as overall movement, this step can be logically omitted.
Additionally, although METs are a widely used metric for assessing exercise intensity by quantifying the intensity of different exercises relative to resting-state energy expenditure, their practical application faces several significant limitations (Franklin et al., 2018).First, the calculation of METs is based on the average resting metabolic rate, without considering individual differences such as age, gender, weight, and physical condition that affect energy expenditure.This means that the same activity may represent different actual intensities for different individuals.Second, MET values provide a fixed estimate of energy expenditure for an activity but lack the sensitivity to capture subtle changes in activity intensity, especially when distinguishing between high-and low-intensity exercise.Additionally, METs assume a uniform resting metabolic rate (1 MET) for all individuals, ignoring actual differences in resting-state energy expenditure.METs are also challenging to apply to complex exercises that contain multiple levels of intensity and are difficult to measure accurately in everyday settings without specialized equipment.Finally, standard MET values do not apply to individuals with specific health conditions, as they do not reflect the unique responses of these individuals to exercise.Therefore, while METs provide a convenient metric for estimating exercise intensity, more accurate and personalized assessments are recommended.Methods such as heart rate monitoring, which considers the individual's specific health status and energy expenditure characteristics, are preferable.
Future research should include long-term studies of diverse populations, encompassing different age groups, genders, and ethnicities, to provide insights into the long-term effects of exercise interventions and their applicability across various demographic groups.Such research will help refine exercise recommendations to ensure they are inclusive and effective for all individuals with schizophrenia.Additionally, integrating wearable technology with mobile health applications should be promoted, as these tools can provide real-time monitoring and feedback on exercise intensity and compliance.This integration can facilitate personalized exercise planning and improve the precision of doseresponse analyses by capturing more granular physical activity data.

Conclusion
In conclusion, our systematic review and network meta-analysis have demonstrated the efficacy of physical exercises such as yoga, RT+AE, and EX+PT in managing schizophrenia.These interventions notably improve clinical symptoms, quality of life, depressive states, and cognitive functions.Our findings emphasize the importance of carefully planned exercise dosages, which yield significant clinical benefits.However, the complex nature of schizophrenia calls for further research to explore the long-term effects and mechanisms underlying the benefits of physical exercise in this context.This continued investigation will enhance treatment strategies and deepen our understanding of schizophrenia's multifaceted challenges.
Y.Yang et al.Neuroscience and Biobehavioral Reviews 167 (2024)  105896 Fig. 2. : Network graphs for total symptoms, positive, negative, general, depression, quality of life, and cognitive; Network diagram of the effects of different exercise interventions on schizophrenia: Network diagram of existing comparisons of the effects of different exercise interventions on sleep quality in older adults.The size of the nodes is proportional to the number of participants randomized to each intervention.The width of the connecting lines corresponds to the number of studies comparing the respective interventions.PP Ping-Pong; RT Resistance Training; AE Aerobic Exercise; RT+AE Resistance Training + Aerobic Exercise; TC Tai Chi; BT Balance and Tone training; UC Usual Care; EX+PT Physical Exercise + Psychological Nursing; Mix Multi-component exercise.

Fig. 3 .
Fig. 3. : Forest plots for mean differences of exercise compared with usual care.The shaded part represents a limited dose range; the gray part of the curve is the confidence interval; PP Ping-Pong; RT Resistance Training; AE Aerobic Exercise; RT+AE Resistance Training + Aerobic Exercise; TC Tai Chi; BT Balance and Tone training; UC Usual Care; EX+PT Physical Exercise + Psychological Nursing; Mix Multi-component exercise.

Fig. 6 .
Fig. 6. : The Funnel plot for positive, negative, general, depression, quality of life, cognitive, and total symptom.The results of Egger's test are attached in the figure (P<0.05:significant bias, P>0.05: significant bias); PP Ping-Pong; RT Resistance Training; AE Aerobic Exercise; RT+AE Resistance Training + Aerobic Exercise; TC Tai Chi; BT Balance and Tone training; UC Usual Care; EX+PT Physical Exercise + Psychological Nursing; Mix Multi-component exercise.