Assessment of the Effects of Physiotherapy on Back Care and Prevention of Non-Specific Low Back Pain in Children and Adolescents: A Systematic Review and Meta-Analysis

Non-specific low back pain (NSLBP) in children and adolescents has increased in recent years, and the evidence of the physiotherapy interventions in back care needs to be updated. Our main goal was to quantify the effects of preventive physiotherapy interventions on improving behavior and knowledge related to back care and prevention of NSLBP in children and adolescents. Based on two previous meta-analyses, Cochrane Library, MEDLINE, PEDro, Web of Science, LILACS, IBECS, PsycINFO, and IME databases and several journals were searched. Two researchers independently extracted data and assessed the risk of bias in the studies using the RoB2 tool. Data were described according to PRISMA guidelines. A total of 24 studies (28 reports) were included. In the posttest, the behavior variable obtained an overall effect size of d+ = 1.48 (95%CI: 0.40 to 2.56), and the knowledge variable obtained an effect size of d+ = 1.41 (95%CI: 1.05 to 1.76). Physiotherapy has demonstrated beneficial impacts on behavior and knowledge concerning back care and to prevent NSLBP in children and adolescents. Interventions focusing on postural hygiene and exercise should be preferred, especially those that are shorter in number of weeks, more intense, and incorporate as many intervention hours as possible.


Introduction
Low back pain (LBP) has become a major public health concern, especially as its incidence among children and adolescents has increased in recent times [1,2].Research suggests that approximately 39% of people aged 9 to 16 years suffer LBP at some point during this stage and that the prevalence resembles that of adults by the age of 15 years [3].It should be noted that having LBP during childhood and adolescence increases the probability of having it in adulthood too [2,4].The most common form of LBP is non-specific low back pain (NSLBP) [1].
The appearance of LBP in childhood and adolescence can have serious consequences, such as limitations in activities of daily living, restrictions in participation in sports activities and in the school environment, and even cases of school absenteeism.Therefore, it is essential to prioritize preventive measures against this condition [5,6].
Physiotherapists use treatments for back care and prevention of NSLBP, given that adolescents [7,8] and parents [9] do not usually have knowledge about back care.The treatments developed focus on improving knowledge about back care based on postural hygiene

Material and Methods
The meta-analysis was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [16]; for more information, see Supplementary Table S1.Additionally, the study was registered with PROSPERO (CRD42024510058).

Eligibility Criteria
The inclusion and exclusion criteria were previously established using the participants, interventions, comparators, outcomes, and study design (PICOS) strategy.
The criteria applied were: (1) participants under 18 years of age, who did not have spinal pathologies or other pathologies causing LBP; (2) the interventions had to be preventive physiotherapy, including education, therapeutic exercise, and physical activity, alone or in combination; (3) the studies had to compare at least one experimental group with a control group, and there could be several experimental groups in the same study; (4) the results should measure the knowledge and/or behavior of the participants, in a pretest and posttest evaluation using the same tool, providing sufficient statistical information to perform the analyses (sample size, mean, and SD); (5) the studies had to be controlled clinical trials, randomized and non-randomized, including published and unpublished studies of any type (journal study, doctoral thesis, conference proceedings, etc.).

Data Sources and Search
The research included in this study comprises data gathered from two previously conducted meta-analyses [14,15].The search spanned from inception to May 2012 in metaanalysis 1 [14] and from May 2012 to May 2020 in meta-analysis 2 [15].Several databases were used for data collection, including Cochrane Library, MEDLINE, PEDro, Web of Science, and IME, along with journals from the Elsevier Iberoamerican database for metaanalysis 1.For meta-analysis 2, data were collected from Cochrane Library, MEDLINE, PEDro, Web of Science, LILACS, IBECS, and PsycINFO databases, as well as from specialized journals such as BMJ and Spine.Both meta-analyses also involved a search for unpublished studies and a thorough review of the bibliographies of the included studies.For more information, see Supplementary Table S2.

Study Selection
The studies were included based on the two meta-analyses previously mentioned [14,15].In addition, two reviewers (JMGM and ICM) independently screened the studies to confirm that they met the inclusion criteria; in case of discrepancy, a third author (AGC) would decide.Cohen's kappa was used to assess inter-rater agreement, with a result of 1.

Data Extraction Process
The data extracted from the two previous meta-analyses [14,15] were used.These data were extracted individually by two authors in their corresponding studies.For the present study, one author (JMGM) extracted the data from the two previous meta-analyses with the review of a second author (ICM) and, in the case of discrepancy, a third author would decide (AGC) but this was not necessary because there was total agreement.When required, additional data were requested directly from the corresponding authors.
Different variables were extracted from each study based on a previously established manual based on Lipsey's recommendations [17].The variables have been grouped into three different categories: substantive (treatment, context, and participant), methodological, and extrinsic variables.For more information about the coded variables, see Supplementary Table S3.

Assessment of Methodological Quality, Risk of Bias, and Certainty of Evidence
The Risk of Bias 2 (RoB2) tool was used to analyze the risk of bias [18].This tool allows a clinical trial to be judged as "low risk of bias", "some concerns", and "high risk of bias".Each study was analyzed in each of its domains and the final assessment was the worst of the domains, or if a study had several domains where "some concerns" where found it was rated as "high risk of bias", as recommended by the authors of the tool.
The GRADE system was applied for each of the variables studied in this meta-analysis.Two authors performed the analysis of risk of bias and certainty of evidence separately (JMGM and ICM) and, in case of discrepancy, a third author would decide (AGC) but this was not necessary since the agreement was total.

Statistical Analysis
A random-effects model was applied [19] with the correction proposed by Hartung [20].A forest plot, including 95% confidence intervals, was created to represent numerically and graphically the individual effects of each study, as well as to display the average effect size.The prediction interval was also added.To calculate the effect size, a standardized mean difference "d" was used for quantitative variables [21] (Equations ( 6) and (7) were used to quantify the standardized mean change in each group and its sampling variance, and then the d index was calculated as the difference using Equations ( 10) and (11) of ref. [21]).
Four included studies were three-arm trials with two experimental groups [22][23][24][25].For these, the effect size was calculated separately for each experimental group in relation to the control group.While this violates the independence assumption required when applying standard meta-analytic techniques (as each participant allocated to control groups in those studies contributed to two effect sizes), we note that the extent of multiplicity in our database-and consequently the potential for statistical dependency-is minimal given the reduced number of studies affected and the small size of the groups involved compared to the rest of the included studies (see Table 1).Therefore, standard meta-analytic models were preferred for this review over other methodological options that either involve loss of information or additional statistical complexity [26].
To assess heterogeneity, the I 2 index was used and interpreted in combination with 95% prediction intervals.
For the analysis of moderator variables, weighted ANOVA was used for qualitative moderator variables and meta-regression for continuous moderator variables.Both analyses were corrected as proposed by Knapp and Hartung [27].For the analysis of publication bias, Egger's test and a funnel plot were applied for all variables.All statistical analyses were performed using R [28] in conjunction with the metafor package [29].

Study Selection
After searching the articles following the previously mentioned search strategy, a total of 4107 articles were found.After elimination of duplicates, 4058 articles were chosen for examination.After review of the titles and abstracts, 121 articles were selected to be thoroughly analyzed to determine whether they met the inclusion criteria.Finally, 24 studies were included resulting in 28 reports.The flow chart (Figure 1) provides a detailed description of the article selection process.

Study Selection
After searching the articles following the previously mentioned search strategy, a total of 4107 articles were found.After elimination of duplicates, 4058 articles were chosen for examination.After review of the titles and abstracts, 121 articles were selected to be thoroughly analyzed to determine whether they met the inclusion criteria.Finally, 24 studies were included resulting in 28 reports.The flow chart (Figure 1) provides a detailed description of the article selection process.

Sample
The number of participants in the included studies varied, ranging from 32 [48] to 706 [33].In the initial assessment (pretest), the study included 5996 participants in total: 3241 in the experimental groups and 2755 in the control groups.Following the posttest, the total number of participants decreased to 5869, with 3155 in the experimental groups and 2714 in the control groups.The age mean ranged from 8 years [40] to 15.38 years [43].The age groups were mostly children, a few studies included only adolescents [43,47], and one study included both children and adolescents [38].The gender of the participants spanned from 0% [49] to 61% male [47], and in some studies it was not specified [22,36,37,39].

Intervention
Regarding the intervention, the experimental groups performed postural hygiene, exercise, or physical activity alone or in combination.Regarding the control groups, four studies included active control groups [23,31,32,48] and the rest, inactive control groups.Postural hygiene was the most common treatment, being present in all experimental groups.For the experimental group, the treatment was based exclusively on postural hygiene in fifteen studies [22][23][24][25]31,[33][34][35]37,38,[40][41][42]44,45], the combination of postural hygiene + exercise in six studies [30,32,39,46,47,49], the combination of postural hygiene + physical activity in one study [36], and the combination of postural hygiene + exercise + physical activity in two studies [43,48].Regarding the types of postural hygiene, the most common were knowledge acquisition (theoretical and/or practical) in combination with postural habits training.All studies applied a group treatment.
As for the active control groups, the interventions were based on regular academic activities on related topics [31], postural hygiene [23], postural hygiene + exercise [32], and treatment as usual [48].
Concerning treatment time, in the experimental groups, the duration in weeks ranged from 1 week [22,47] to 96 weeks [35,36].The intensity (hours/week of treatment) ranged from 0.16 [22,23] to 2.625 [48].Total treatment time ranged from 0.16 hours [22,23] to 19 hours [32].All experimental groups except one [40] had the number of sessions established before starting treatment, and all experimental groups except one [40] were based on homogeneous treatments to all participants.Only four studies included homework for participants to perform treatment at home [32,43,44,47].

Risk of Bias
All studies were judged to show some concerns except four studies with high risk of bias [22,43,45,48].Three of the studies had high risk of bias in the first domain (randomization) [43,45,48].One study obtained high risk of bias because it had some concerns in most of the domains [22].All studies obtained some concerns in the second domain (deviations from intended interventions) except one that obtained high risk of bias [48].One study obtained high risk of bias because it had some concerns in most of the domains [22].All studies obtained low risk of bias in the fifth domain.For more information, see Table 2.For the behavior variable, 19 studies were included, which resulted in 23 effect sizes.In the posttest, an overall effect size of d + = 1.48 (95%CI: 0.40 to 2.56), with I 2 = 98.94% of the total variability due to heterogeneity, was obtained.The 95% prediction interval ranged from −3.4 to 6.3.All studies except one yielded better results for the experimental group than for the control group [22].One study obtained an extremely large effect size d + = 13.03[32].Thirteen effect estimates revealed significant differences in favor of physiotherapy [24,30,32,34,35,[41][42][43][45][46][47][48][49] (Figure 2).The Egger test was significant (p = 0.0090), with the asymmetry in the funnel plot suggesting potential publication bias (Figure 3).For the behavior variable, 19 studies were included, which resulted in 23 effect sizes.In the posttest, an overall effect size of d+ = 1.48 (95%CI: 0.40 to 2.56), with I 2 = 98.94% of the total variability due to heterogeneity, was obtained.The 95% prediction interval ranged from −3.4 to 6.3.All studies except one yielded better results for the experimental group than for the control group [22].One study obtained an extremely large effect size d+ = 13.03[32].Thirteen effect estimates revealed significant differences in favor of physiotherapy [24,30,32,34,35,[41][42][43][45][46][47][48][49] (Figure 2).The Egger test was significant (p = 0.0090), with the asymmetry in the funnel plot suggesting potential publication bias (Figure 3).For the behavior variable, 19 studies were included, which resulted in 23 effect sizes.In the posttest, an overall effect size of d+ = 1.48 (95%CI: 0.40 to 2.56), with I 2 = 98.94% of the total variability due to heterogeneity, was obtained.The 95% prediction interval ranged from −3.4 to 6.3.All studies except one yielded better results for the experimental group than for the control group [22].One study obtained an extremely large effect size d+ = 13.03[32].Thirteen effect estimates revealed significant differences in favor of physiotherapy [24,30,32,34,35,[41][42][43][45][46][47][48][49] (Figure 2).The Egger test was significant (p = 0.0090), with the asymmetry in the funnel plot suggesting potential publication bias (Figure 3).Given that the study by Méndez and Gómez 2001 [32] obtained an extreme effect size (d + = 13.03), a sensitivity analysis excluding this study was performed.The new analysis resulted in a decrease in the effect size to d + = 1.04 (95%CI: 0.64 to 1.43), with I 2 = 91.54% of the total variability due to heterogeneity.The 95% prediction interval then ranged from −0.6 to 2.68.The Egger test for this analysis was not significant (p = 0.99).
Most studies that were included in meta-analysis 2 [15] reported statistically significant results in favor of physiotherapy, whereas studies included in meta-analysis 1 were more heterogeneous [14].This may be because primary studies that were published after metaanalysis 1 considered the findings of that study.
The two studies that included the most weeks of treatment (96 weeks) obtained better results in the experimental groups, with one reporting statistically significant results [35] while the other one did not [36].The only study that obtained better results in the control group compared to the experimental group based its treatment on a single week [22].Other studies that also applied single-week interventions obtained non-significant differences in favor of the experimental group [44] and another study did obtain significant differences in favor of the experimental group [46].
Regarding the number of hours/week of treatment, the study with the largest effect size was the study with the highest number of hours/week (2.375 h/week) [32].The study with the highest number of hours/week of treatment obtained a large effect size with statistical significance in favor of the experimental group [48].Regarding the studies with the lowest number of hours/week (0.16 h/week) [22,23], none obtained significant improvements and one study obtained better results for the control group [22].
Regarding the total number of hours, the study with the highest number of total hours of treatment (19 total hours) was the study with the largest effect size [32].The second study with the highest number of total hours of treatment obtained statistical significance in favor of the experimental group [43].In contrast, the two studies with the lowest total number of hours obtained non-significant results in favor of the control group [22] and non-significant results in favor of the experimental group [44].
The study with the largest effect size based its intervention on postural hygiene + exercise [32], the second study with the largest effect size only performed postural hygiene [24], and the third study with the largest effect size performed postural hygiene + exercise + physical activity [48].Conversely, the three studies with the smallest effect sizes only performed postural hygiene [22,25,44].

Moderator Analyses
Weighted ANOVA Due to the high heterogeneity observed in the meta-analysis, a search for potentially moderating qualitative variables was conducted using ANOVA.Initially, the effects of different treatments (postural hygiene vs. postural hygiene + exercise) were assessed.Studies that incorporated postural hygiene in addition to exercise obtained better results than those that focused only on postural hygiene, although the difference between them was only marginally significant (p = 0.064).Furthermore, an evaluation of the postural hygiene methods (theory vs. theory + practical) was undertaken.Results suggest that studies implementing a combination of theoretical and practical approaches to postural hygiene achieved better outcomes compared to those relying solely on theoretical frameworks, although the difference between them was again only marginally significant (p = 0.086).Age groups (children vs. adolescents) were also analyzed through ANOVA, with no evidence of statistically significant differences between them (p = 0.60).Lastly, in the analysis exploring the impact of risk of bias (some concerns vs. high), there was no evidence of a difference between studies with "some concerns" and those assessed at high risk of bias (p = 0.53), however, the studies that received some concerns of risk of bias obtained statistically significant results, and the high risk of bias studies did not.Further information is available in Table 3.

Meta-Regression Analysis
Quantitative moderator variables were also analyzed using meta-regression.Specifically, the number of treatment weeks was examined, revealing no significant relationship with effect sizes (b j = −0.0053,95%CI −0.051 to 0.04, p = 0.810).Similarly, the number of hours per week of treatment (intensity) was not found to influence the effect sizes (b j = 0.0735, 95%CI −0.277 to 1.747, p = 0.143).However, concerning the total number of treatment hours, we found evidence of a direct relationship (b j = 0.395, 95%CI 0.209 to 0.581, p = 0.0003).Further details can be found in Table 4. k = number of studies.b j = regression coefficient of each predictor.CI.LL = confidence interval of lower limit.CI.UL = confidence interval of upper limit.F = Knapp-Hartung's statistic for testing the significance of the predictor (the degrees of freedom for this statistic are 1 for the numerator and k − 2 for the denominator).p = probability level for the F statistic.

Knowledge
For the knowledge variable, 17 studies contributing 20 effect sizes were included.In the posttest, an overall effect size of d + = 1.41 (95%CI: 1.05 to 1.76), with I 2 = 86.03% of the total variability due to heterogeneity, was obtained.The 95% prediction interval ranged from −0.04 to 2.86.All studies obtained better results in the experimental group than in the control group.Twelve studies obtained significant results in the experimental group in comparison with the control group [22,25,[30][31][32][33]37,38,41,46,48,49] (Figure 4).The studies with the longest effect size compared the physiotherapy group with an inactive control group [30,38,41,49].These four studies were not the ones with the longest treatment duration, most hours of treatment per week, or highest total treatment hours compared to all others included.Of these, two included exclusively postural hygiene [30,38], while another two combined postural hygiene with exercise [41,49].However, these four studies included both theory and practice in their treatment of postural hygiene.The Egger test was non-significant (p = 0.96), and the symmetry in the funnel plot shows the absence of publication bias (Figure 5).
included both theory and practice in their treatment of postural hygiene.The Egger test was non-significant (p = 0.96), and the symmetry in the funnel plot shows the absence of publication bias (Figure 5).We observed that all studies that were included in the most recent meta-analysis 2 [15] reported statistically significant results in favor of physiotherapy, whereas the studies included in meta-analysis 1 were more heterogeneous [14].As for the behavior variable, this may be because the studies that were published after meta-analysis 1 were influenced by its findings.
included both theory and practice in their treatment of postural hygiene.The Egger test was non-significant (p = 0.96), and the symmetry in the funnel plot shows the absence of publication bias (Figure 5).We observed that all studies that were included in the most recent meta-analysis 2 [15] reported statistically significant results in favor of physiotherapy, whereas the studies included in meta-analysis 1 were more heterogeneous [14].As for the behavior variable, this may be because the studies that were published after meta-analysis 1 were influenced by its findings.
Conversely, within the subset of studies reporting non-significant results [24,35,36,39,40], two conducted the longest treatment periods, extending to 96 weeks We observed that all studies that were included in the most recent meta-analysis 2 [15] reported statistically significant results in favor of physiotherapy, whereas the studies included in meta-analysis 1 were more heterogeneous [14].As for the behavior variable, this may be because the studies that were published after meta-analysis 1 were influenced by its findings.
Conversely, within the subset of studies reporting non-significant results [24,35,36,39,40], two conducted the longest treatment periods, extending to 96 weeks [35,36] and 15 weeks of treatment [24].However, in these five cases, details on intensity (hours/week) and total time of treatment were not provided.Notably, all of these studies mainly incorporated postural hygiene into their treatment strategies, with two exceptions.One study combined postural hygiene with physical activity [36], and another combined postural hygiene with exercise [39].

Moderator Analyses Weighted ANOVA
As a result of the high heterogeneity found in the meta-analysis, an analysis of the qualitative moderator variables by ANOVA was performed.First, the type of treatment (postural hygiene vs. postural hygiene + exercise) was analyzed, with no evidence of a difference between the options (p = 0.52).The type of postural hygiene (theoretical vs. theoretical + practical) was also analyzed, and again no evidence of a difference between categories was found (p = 0.96).In addition, the risk of bias categories were also analyzed (some concerns vs. high) without significant differences between them (p = 0.81).Last, an analysis of age groups (children vs. adolescents) was not feasible since all studies analyzed only included children.Further information is available in Table 5.

Meta-Regression Analysis
For quantitative moderator variables, the number of weeks of treatment was analyzed, resulting in a statistically significant result which surprisingly points to an inverse relationship (b j = −0.013,95%CI −0.022 to −0.003, p = 0.012).The number of hours per week of treatment was not found to influence the effect sizes (b j = −0.083,95%CI −0.355 to 0.189, p = 0.522).Last, total treatment time was also analyzed and was found not to influence the results (b j = −0.003,95%CI −0.091 to 0.098, p = 0.934).For more information, see Table 6.CI.UL = confidence interval of upper limit.F = Knapp-Hartung's statistic for testing the significance of the predictor (the degrees of freedom for this statistic are 1 for the numerator and k − 2 for the denominator).p = probability level for the F statistic.

Certainty of Evidence
The GRADE system was applied by two researchers independently with full agreement.Both variables, behavior and knowledge, obtained moderate certainty of evidence.For more information, see Supplementary Table S4.

Discussion
The aim of this meta-analysis was to quantify the effects of preventive physiotherapy on knowledge and behavior regarding back care in children and adolescents, for which 24 studies resulting in 28 reports were included.
Although behavior and knowledge have been studied previously via several metaanalyses [14,15], until this study not all studies carried out on this subject had been combined in a single meta-analysis, which is why this study has additional relevance.The results of this meta-analysis followed the same direction as the previously published meta-analyses, since this study obtained an overall effect size for the behavior variable of d + = 1.48 (95%CI: 0.40 to 2.56) for the posttest, slightly higher than that of meta-analysis 1 (d + = 1.33 (95%CI: 0.76 to 1.90)) and higher than that of meta-analysis 2 (d + = 1.19 (95%CI: 0.62 to 1.76)).For the knowledge variable, in the present meta-analysis an overall effect size of d + = 1.41 (95%CI: 1.05 to 1.76) was obtained, higher than in meta-analysis 1, which obtained an effect size of d + = 1.29 (95%CI: 0.90 to 1.68), but lower than in meta-analysis 2, which obtained an effect size of d + = 1.84 (95%CI: 0.58 to 3.09).It should be noted that meta-analysis 2 only included four studies in its analysis of knowledge, hence the effect size has such a wide range.In addition, the present study also provides the prediction interval, which was not present in the previous meta-analyses and included a more extensive number of studies in all the variables analyzed.
Whether the current study had a larger effect size than previous studies in general may be explained by the fact that the studies published since 2012 took into account the findings of meta-analysis 1, such as including more practical examples in their sessions and combining postural hygiene and exercise, instead of only postural hygiene.
This study offered new insights into the behavior variable, for example, revealing that the studies that included postural hygiene in combination with exercise obtained better results, since in meta-analysis 1 the comparison was imprecise (only one study included postural hygiene in combination with exercise) and in meta-analysis 2 few studies were included and the result was not significant.Concerning the continuous moderating variables, meta-analysis 1 found that the number of weeks did not influence the effect size, intensity was close to statistical significance, and the total number of hours was included, and meta-analysis 2 also found that intensity was close to statistical significance, with the rest not being influential.However, the present study concluded that intensity is not significant, and neither is the number of weeks, although the total treatment time was.
Furthermore, the analysis of the moderator variables in the knowledge variable showed the same results in meta-analysis 1 and the present study that the combination of postural hygiene and exercise had better results than postural hygiene alone, but without significant differences.The present study showed more consistency in the different treatments (fourteen studies included postural hygiene and six studies included postural hygiene + exercise) compared to meta-analysis 1 (twelve studies included postural hygiene and three studies included postural hygiene + exercise).In addition, with respect to the type of teaching, meta-analysis 1 and the present one were the same, with a small difference in favor of theory only in comparison with theory + practical, with no significant differences between them.For quantitative variables, this meta-analysis discovered that the number of weeks negatively impacted effect size, a finding consistent with meta-analysis 1.Both studies agreed that neither treatment intensity nor total hours significantly affected knowledge effect size.Meta-analysis 2 did not perform moderator variable analysis due to the few studies included.
A recently published systematic review included eight studies based on back care intervention in children and adolescents, examining the variables knowledge and behavior, among others [50].The studies included in that review were also mostly included in the present study; those excluded did not have the statistical information needed for a meta-analysis or studied a different variable from those of interest in the present study.This systematic review defended that a back care education program was useful to improve children's knowledge and behavior, highlighting the need to incorporate it in education programs, which is also in line with the present study.Another recent systematic review [13] argues that exercise is the most useful way to promote spinal health in the short term, but the combination of exercise with education is necessary for long-term management.
Considering the findings acquired in this study and combining them with those of the previous ones [14,15,50], we can develop strategies to improve knowledge and behavior regarding back care in this population.In this way, in order to improve both variables as a conjunct, it would be convenient to combine postural hygiene and exercise with theory and practice, reducing to a minimum the number of weeks of treatment and increasing the intensity and the total time of treatment.Although it is true that there are differences depending on whether we aim to improve knowledge or behavior separately, we recommend that when interventions are carried out to improve back care to prevent LBP, these two variables be addressed at the same time, since an improvement in knowledge can benefit behavior, mainly in the long term.
By using the GRADE approach to evaluate the certainty of the evidence, our study provided a better understanding of the effects of physiotherapy in back care and prevention of non-specific low back pain in children and adolescents.Since a moderate level of certainty of evidence was obtained, physiotherapists may be confident in the findings, being able to apply the insights provided in this study to their clinical practice.
The main limitations of this study were the high heterogeneity found among the included studies, probably due to the variability among the studies (type and duration of treatment, etc.) as well as the fact that most of the studies obtained some concerns of risk of bias and a few had high risk of bias.Furthermore, the search strategy included studies up to 2020, as the aim of this study was to integrate the two previous meta-analyses to provide overarching conclusions in this field.Regarding the strengths, it is a study that encompasses all previously published studies, reaching a large sample size.In addition, two researchers separately performed the extraction of information, risk of bias analysis, and the application of the GRADE system.State-of-the-art statistical analysis techniques were used, and an exhaustive analysis of moderator variables was performed.
As far as we are aware, this is the most comprehensive meta-analysis to date, providing important insights that had not been discovered before.The results found in this study can serve physiotherapists, teachers, and policy makers to make precise interventions in the approach to back care and adapt their treatments to the study population.

Conclusions
Physiotherapy showed positive effects on behavior and knowledge related to back care to prevent NSLBP in children and adolescents.Postural hygiene and exercise were identified as the fundamental cornerstones of intervention.Treatments with the smallest number of weeks, the highest intensity, and the highest total number of treatment hours possible should be preferred.

Figure 1 .
Figure 1.PRISMA flow diagram.Process of identification and selection of studies.

Figure 1 .
Figure 1.PRISMA flow diagram.Process of identification and selection of studies.

Figure 3 .
Figure 3. Funnel plot for the behaviour in the posttest.

Figure 3 .
Figure 3. Funnel plot for the behaviour in the posttest.Figure 3. Funnel plot for the behaviour in the posttest.

Figure 3 .
Figure 3. Funnel plot for the behaviour in the posttest.Figure 3. Funnel plot for the behaviour in the posttest.

Figure 5 .
Figure 5. Funnel plot for the knowledge in the posttest.

Figure 5 .
Figure 5. Funnel plot for the knowledge in the posttest.

Figure 5 .
Figure 5. Funnel plot for the knowledge in the posttest.

Table 1 .
Characteristics of the studies.
Knowledge acquisition + postural habits training (picking up, carrying, and handling of weights, working positions and activities of daily living and resting positions), theoretical (guided discovery) and practical + stretching + strengthening + pelvic tilt + relaxing
•Based on guided discovery and active hands-on method, anatomy and pathology of the back and basic principles of correct posture in daily activities were taught.For teachers and parents, an information session was organized based on back care principles to enhance learning at home and in the classroom.The
Teachers completed the integration of the back postural principles learned weekly with the use of imagery.In addition, the physical education teacher taught the improvement of dynamic sitting (active and variable sitting), the interruption of prolonged static sitting, and the activating approach (changes in class organization, promotion of postural behavior, and activating methodology).The teacher'
•Ergonomics and postural hygiene based on activities of daily living such as lying down and getting out of bed, how to lie in bed, getting in and out of a car, washing hands, sitting on a high stool, using pillows to rest, carrying weight, lifting weight, lifting small objects, how to sit in a chair and how to get up, reading in bed, sitting to study, working on the computer, backpack use (preparation and carrying), carrying small cabinets, using ladders to pick up objects at height, watching TV, sitting to eat, working while standing, ironing while standing, and in the last session a general review of all of the above 5 weeks, 1 h per week,

Table 2 .
Risk of bias of included studies.

Table 3 .
Results of the weighted ANOVAs for the behavior, taking qualitative moderator variables as independent variables.= number of studies.d + = mean coefficient alpha.LL and LU = lower and upper 95% confidence limits for d + .F = Knapp-Hartung's statistic for testing the significance of the moderator variable. k

Table 4 .
Results of the simple meta-regressions for the behavior and magnitude measures, taking continuous moderator variables as predictors.

Table 5 .
Results of the weighted ANOVAs for the knowledge, taking qualitative moderator variables as independent variables.= number of studies.d + = mean coefficient alpha.LL and LU = lower and upper 95% confidence limits for d + .F = Knapp-Hartung's statistic for testing the significance of the moderator variable. k

Table 6 .
Results of the simple meta-regressions for the knowledge and magnitude measures, taking continuous moderator variables as predictors.