Trunk postural control during unstable sitting among individuals with and without low back pain: A systematic review with an individual participant data meta-analysis

Introduction Sitting on an unstable surface is a common paradigm to investigate trunk postural control among individuals with low back pain (LBP), by minimizing the influence lower extremities on balance control. Outcomes of many small studies are inconsistent (e.g., some find differences between groups while others do not), potentially due to confounding factors such as age, sex, body mass index [BMI], or clinical presentations. We conducted a systematic review with an individual participant data (IPD) meta-analysis to investigate whether trunk postural control differs between those with and without LBP, and whether the difference between groups is impacted by vision and potential confounding factors. Methods We completed this review according to PRISMA-IPD guidelines. The literature was screened (up to 7th September 2023) from five electronic databases: MEDLINE, CINAHL, Embase, Scopus, and Web of Science Core Collection. Outcome measures were extracted that describe unstable seat movements, specifically centre of pressure or seat angle. Our main analyses included: 1) a two-stage IPD meta-analysis to assess the difference between groups and their interaction with age, sex, BMI, and vision on trunk postural control; 2) and a two-stage IPD meta-regression to determine the effects of LBP clinical features (pain intensity, disability, pain catastrophizing, and fear-avoidance beliefs) on trunk postural control. Results Forty studies (1,821 participants) were included for the descriptive analysis and 24 studies (1,050 participants) were included for the IPD analysis. IPD meta-analyses revealed three main findings: (a) trunk postural control was worse (higher root mean square displacement [RMSdispl], range, and long-term diffusion; lower mean power frequency) among individuals with than without LBP; (b) trunk postural control deteriorated more (higher RMSdispl, short- and long-term diffusion) among individuals with than without LBP when vision was removed; and (c) older age and higher BMI had greater adverse impacts on trunk postural control (higher short-term diffusion; longer time and distance coordinates of the critical point) among individuals with than without LBP. IPD meta-regressions indicated no associations between the limited LBP clinical features that could be considered and trunk postural control. Conclusion Trunk postural control appears to be inferior among individuals with LBP, which was indicated by increased seat movements and some evidence of trunk stiffening. These findings are likely explained by delayed or less accurate corrective responses. Systematic review registration This review has been registered in PROSPERO (registration number: CRD42021124658).


Introduction
Sitting on an unstable surface is a common paradigm to investigate trunk postural control among individuals with low back pain (LBP), by minimizing the influence lower extremities on balance control.Outcomes of many small studies are inconsistent (e.g., some find differences between groups while others do not), potentially due to confounding factors such as age, sex, body mass index [BMI], or clinical presentations.We conducted a systematic review with an individual participant data (IPD) meta-analysis to investigate whether trunk postural control differs between those with and without LBP, and whether the difference between groups is impacted by vision and potential confounding factors.

Introduction
Low back pain (LBP) is a multifactorial condition [1] and the leading cause of disability globally [2][3][4].Recurrence of LBP episodes is common [1,5,6], and in some cases LBP becomes chronic [7,8].The quality of trunk postural control has been suggested as a risk factor for LBP development, recurrence, and/or perpetuation, mediated by effects of suboptimal loading on spine tissue health [9][10][11].Although this proposal is plausible, LBP is heterogeneous and findings from many (small samples) studies are inconsistent and inconclusive.
Trunk postural control is critical for executing human motion and completing everyday activities [12].Such control can be reflected in the capacity to maintain both postural equilibrium (control of the centre of mass over the base of support) and postural orientation between segments (within spine regions and between the spine and other body regions) [13][14][15].Trunk postural control requires motor skill [16], involving the integration of kinematic (position and movement) feedback from visual, vestibular, and proprioceptive systems [17,18], and the generation of coordinated motor output using an array of muscles [12,19,20].
An unstable sitting paradigm has been developed to assess the contribution of the trunk to postural control [21], by limiting contributions from the legs and arms [22].This paradigm (Fig 1) involves sitting on an unstable surface attached to a hemisphere [21] or on a chair (aka "wobble chair") that moves about a central pivot and is supported by four adjustable springs [23].Typically, the seat is placed over a force platform, and the seat movement is measured by calculating time series of the centre of pressure (CoP)-the location of the point of contact of the seat hemisphere [21,22], or the barycentre of the forces under the wobble chair with springs [24].To maintain balance in this paradigm, the global position of the upper body is maintained via dynamic movements at the base/seat [25].Movements at the base/seat are attenuated by coordinated movements of the hip and spine [26,27] to limit upper body movements, and maintain the overall centre of mass close to the CoP.
Many outcome measures from this paradigm have been reported to have acceptable to excellent test-retest reliability for assessing trunk postural control among individuals with or without LBP [21,24,28,29].These outcome measures are usually related to the amplitude of A seat is attached to (A) a hemisphere or (B) a central pivot surrounded by four springs.Task difficulty depends on the seat apparatus and its build characteristics.For the hemisphere-based seat, task difficulty depends on the radius of a hemisphere and the seat height from the support surface.For the springs-based seat, task difficulty depends on the stiffness of springs and the distance of springs (R spring ) from the pivot in percentage.The percentage of R spring is relative to the gravitation gradient (e.g., 100% = the location of springs that would fully balance the mass of the participant as if sitting on a stable chair).Both seats are commonly positioned on a force platform.A foot plate is usually attached to the seat that maintains the knees at 90f lexion and reduces the contribution of the lower limbs to balance control.In most applications, participants are asked to keep their arms crossed at their chest, to minimize the contribution of the upper limbs.https://doi.org/10.1371/journal.pone.0296968.g001 CoP/seat motion, and include root mean square (RMS) displacement, mean velocity, range, mean frequency, and other measures related to CoP/seat dispersion dynamics (stabilogram diffusion analysis).In most cases, greater CoP/seat motion is interpreted as inferior or impaired trunk postural control [19,22,30].
Many individuals with and without LBP have been studied using the unstable sitting paradigm.Results from some studies suggest that trunk postural control is inferior among those with LBP [19,22,[30][31][32], whereas other results indicate no difference from those without LBP [29,33,34].These conflicting findings might be explained by differences in LBP presentation (e.g., LBP clinical features), experimental setup (e.g., seat apparatus and its build characteristics) and protocol (e.g., visual condition), or statistical approaches.Regarding the latter, there is variability in whether covariates/confounders were included, such as age, sex, and body mass index (BMI) [21,34,35].Progress in understanding trunk postural control in LBP would be aided by a systematic review of available studies and an individual participant data (IPD) meta-analysis.
In the current study, we performed a systematic review with an IPD meta-analysis (when data were available) and a narrative review (when IPD were not available) of studies that used an unstable sitting paradigm to investigate trunk postural control among individuals with LBP and/or those without LBP (pain-free controls/individuals).Our aims were to: 1. Identify characteristics and summarize the main findings of studies that investigated trunk postural control among individuals with and/or without LBP when seated on an unstable surface.
2. Summarise the experimental methods used.
3. Assess the comprehensiveness of reporting and methodological quality.
4. Use IPD meta-analysis to determine whether trunk postural control (primary outcome measures included root mean square displacement [RMS displ ] and mean velocity [M vel ]) differs between individuals with and without LBP, by using studies that included both individuals with and without LBP.
5. Use IPD meta-analysis to determine whether similar conclusions are derived from alternative (secondary) outcome measures (e.g., stabilogram diffusion analysis) that are available for fewer studies, by using studies that included both individuals with and without LBP.
6. Use IPD meta-analysis to investigate the interaction between participant-level variables (age, sex, and BMI) and the difference between groups on trunk postural control, by using studies that included both individuals with and without LBP.
7. Use IPD meta-analysis to determine whether trunk postural control differs between individuals with and without LBP when vision is removed, by using studies that included both individuals with and without LBP in conditions with eyes open and closed.
8. Use IPD meta-regression to identify whether trunk postural control among individuals with LBP depends on pain intensity, pain duration, disability, and psychological features, by using all studies that included individuals with LBP. 9. Use visual inspection of mean plots to compare the outcomes of studies that tested both individuals with and without LBP (IPD meta-analysis) with: (a) the outcomes of studies that tested only LBP or pain-free individuals using IPD analysis (standardised statistical methods); and (b) a narrative review of studies with outcomes for which IPD could not be obtained.

Design, registration, and ethics
This systematic review with IPD meta-analyses was planned according to the Preferred Reporting Items for Systematic Review and Meta-Analyses of Individual Participant Data (PRISMA-IPD) guidelines [56].The study was registered in PROSPERO (registration number: CRD42021124658), and the study protocol has been published [57].Ethics exemption for this systematic review with IPD analysis was obtained from the Institutional Human Research Ethics Committee (The University of Queensland: 2019003026).Other ethical considerations (e.g., ownership and confidentiality of the IPD set of included studies) have been described previously [57].

Inclusion criteria.
Studies were included if they investigated trunk postural control using an unstable sitting paradigm among participants aged �18 years.Cross-sectional studies, clinical trials, and cohort studies with baseline data were considered.Studies had to include any of the following: (1) individuals with and without LBP (LBP versus pain-free control); (2) individuals with only LBP; or (3) only pain-free individuals.Participants with LBP of any duration were considered (e.g., acute, subacute, or chronic).Participants had to perform trunk postural control tasks using an unstable sitting paradigm that required individuals to control balance in the sagittal and frontal plane.Trunk postural control must have been measured by quantifying seat motion, either from CoP (force platform) or seat angle (from motion capture systems/sensors).

Exclusion criteria.
Studies with insufficient details to determine eligibility were excluded if the authors did not respond to requests (at least two attempts via email) to provide the required information.Non-English reports were excluded.For individuals with only LBP, studies were excluded if they included participants with: neurological disorders (e.g., stroke) except for sciatica (pain due to sciatic nerve compression); spinal structure deformities (e.g., scoliosis); cancer or infection; spine surgery; or a major injury/pain in any other body region within the preceding 12 months.For pain-free controls/individuals, studies were excluded if they included participants with: a history of LBP in the previous year; neurological disorders; structure deformities; cancer or infection; spine surgery; a major injury/pain in any body region within the preceding 12 months.In addition, studies were excluded if they: investigated disorders or diseases other than LBP; involved different seated balance tasks, such as provision of visual feedback or moving the seat to specific target locations; involved sitting on a hemisphere/springs but with feet supported on the floor; involved sitting tasks with perturbations; involved sitting on soft surfaces (e.g., a ball, foam, or air cushion); or studies that referred to an already identified dataset (e.g., studies that referred to secondary use of the same data presented in another study), these studies were excluded from the quantitative analysis but included in the descriptive analysis.

Outcome measures.
Table 1 presents additional information about the outcome measures.The primary outcome measures were RMS displ and M vel of the CoP/seat angle during trials in which participants balanced with eyes open or closed in the forward-to-backward (anteroposterior) and side-to-side (mediolateral) directions.Greater RMS displ and M vel are generally considered to reflect impaired postural control of the trunk [22,30].These measures are those most often used to quantify postural control.RMS displ [21,24] and M vel [21,24,28] measures have high test-retest reliability during unstable sitting tasks.These measures of postural control have also successfully differentiated individuals with and without pathologies [58].Secondary outcome measures included those less commonly reported/extracted, including range, mean power frequency (MPF), and measures related to stabilogram diffusion analysis [21,24].

Identifying studies
The following electronic databases were searched from their date of establishment to 25 March 2022 (the original search): MEDLINE and CINAHL via EBSCO, Embase and Scopus via Elsevier, and Web of Science Core Collection via Clarivate.An updated search was performed on the 7 th September 2023 to retrieve new records only (from March 2022 to September 2023).Reference lists of included studies were screened for other relevant studies.Corresponding authors of included studies were contacted and asked if they had other studies on the same topic.Search terms were determined based on the inclusion criteria (see S1-S8 Tables for details on our original and updated search strategies).

Study selection
EndNote software (version X9) was used to collect the search results, and to remove duplicates automatically and manually.Titles and abstracts of articles were screened for potential inclusion by two independent reviewers (MAA and HA) familiar with systematic reviews and meta-analyses.For articles that potentially met eligibility criteria, full texts were reviewed for final decisions.Disagreement between reviewers was resolved by consensus or a third reviewer (PWH).Cohen's kappa (inter-rater reliability) analysis [59] was performed to assess the degree of agreement between reviewers.The number of included and excluded articles, and reasons for exclusion, were recorded.

IPD collection
The corresponding authors of included studies were asked to share their IPD using author information reported in the article, or profiles on their university websites.If no response was received, co-authors were contacted.Authors were informed about the proposal IPD metaanalysis study/methods, and were asked if they were willing to provide their IPD sets.Any eligible studies for which IPD could not be obtained (e.g., authors did not respond or did not have access/authorization to provide IPD set) were retained for narrative analysis.

Data items
Study-and individual-level data were extracted using a standardised form.Extracted data were [57]: study characteristics, participant characteristics, LBP clinical features, inclusion and exclusion criteria, experimental setup, experimental protocol, any reported adverse effects, outcome measures, and main findings.Collected IPD sets were stored in a master spreadsheet and were screened in terms of presentation of overall data and available variables.The unit of measurement for each outcome measure was unified.For instance, some outcome measures from CoP data, such as RMS displ , were reported in two different units (cm or mm) and the unit of RMS displ was unified as "mm" before applying the IPD meta-analysis.

Comprehensiveness of reporting and methodological quality
Comprehensiveness of reporting and quality of methods were assessed with an adapted checklist [57], which includes components from a quality checklist developed by Mazaheri et al. [46] and Ruhe et al. [45] for systematic reviews of balance measures.This checklist has 25 items with three options to assess the comprehensiveness of reporting and quality of methods across five main domains: participant characteristics, LBP characteristics, experimental setup/protocol, confounding effects control, and statistical information (S9 Table presents the checklist content and item descriptions).Items were scored as '1' (yes), '0.5' (partially; some information was provided) or '0' (no) by two independent reviewers (MAA and HA).The overall reporting/quality score was obtained from the sum of all scores converted to a percentage.Separate reporting/quality scores were calculated for each domain.Reporting/Quality scores ranged from 0 to 100%, with higher scores indicating higher reporting/quality.Discrepancies between reviewers were settled by consensus and a third reviewer (PWH) when necessary.

Descriptive analysis.
A descriptive analysis (see S10 Table for a detailed listing of items included for the descriptive analysis) was used to identify characteristics and to summarize main findings of studies that investigated trunk postural control among individuals with and/or without LBP when seated on an unstable surface (Aim 1), and to assess the comprehensiveness of reporting and quality of methods (Aim 3).All studies were included in the descriptive analysis, including those that reported data from only LBP or pain-free individuals, and studies for which IPD sets were unavailable.Outcomes of each specific study were summarised in tabular format (Aim 1) and were discussed narratively only for studies that did not provide IPD (Aim 9).A detailed descriptive analysis was also undertaken to describe the experimental setup and protocol (Aim 2).Comprehensiveness of reporting and quality of methods before and after obtaining the IPD were summarised in tabular format (Aim 3).A quantitative analysis was conducted using a two-stage IPD meta-analysis to investigate trunk postural control among individuals with and without LBP when sitting on an unstable surface, while considering the characteristics of each individual participant (Aims 4-5).The two-stage IPD meta-analysis was obtained by (1) analysing IPD from each study separately to calculate aggregate data of interest using multilevel mixed-effects models, then (2) combining the results using conventional meta-analysis methods.The advantage of this approach is that it applies a standardised statistical method [38][39][40] and enables to control for covariates and confounding factors in the analysis [36][37][38][39][40]. Potential confounding variables of age and BMI (as covariates) and sex (as a fixed factor) [21,34,35] were included in the models.This analysis was limited to studies that included both individuals with and without LBP to identify between-group differences and was performed for each identified outcome (RMS displ , M vel , range, MPF, D short , D long , CP time and CP dist ; for definitions see Table 1), visual condition (eyes open and closed), and direction (anteroposterior and mediolateral directions).
A random effects model, fitted using the restricted maximum likelihood (REML) method, was used [60] to avoid misleading effect estimates and potentially inappropriate conclusions [61].Overall and individual study (observed) effect sizes were estimated using standardised mean differences (SMDs, Hedge's g), since different studies used different units to measure the same outcome [62].SMD effect sizes can be classified as trivial (>0.2), small (0.2-0.5), medium (0.5-0.8) or large (>0.8) [63].The I 2 index [64] was calculated to assess the percentage of total variability due to between-study heterogeneity rather than sampling error (withinstudy variability) [65,66].Prediction interval (PI) can provide a predicted range for the true effect (without sampling error) in a new (similar) study [67][68][69].This was calculated (when having a sufficient number of studies) to estimate how much the effect size varies across studies included in the IPD meta-analysis [70,71].The potential presence of small-study effects (also known funnel plot asymmetry or publication bias) in the IPD meta-analysis was tested using Egger's test (random-effects model with the REML method) and visualized using funnel plots.
A two-stage IPD analysis was performed to investigate the interaction between participantlevel variables (age, sex, and BMI) and the difference between groups (Aim 6).In the first stage, group and participant-level variables and their interactions were entered using multilevel mixed-effects models.An interaction of one variable (age) with group was entered and the remaining variables (sex and BMI) were included without interaction terms in the models.This method was applied for each of the participant-level variables, and was performed for each study separately.In the second stage, a conventional meta-analysis was performed to pool interaction effect (adjusted) coefficients using a random-effects model with the REML method.Both I 2 index and PI were calculated.This analysis was limited to studies that included both individuals with and without LBP, and was performed for each identified outcome, visual condition, and direction.
A two-stage IPD analysis was performed to investigate the interaction between vision (eyes open and closed) and the difference between groups (Aim 7).In the first stage, group and visual condition and their interaction were entered using multilevel mixed-effects models (for each study separately), including age and BMI as covariates, and sex as a fixed factor in the models.In the second stage, a conventional meta-analysis was performed to pool interaction effect (adjusted) coefficients using a random-effects model with the REML method.Both I 2 index and PI were calculated.This analysis was limited to studies that included both individuals with and without LBP and that investigated the effect of vision on trunk postural control, and was performed for each identified outcome and direction.
A two-stage IPD meta-regression was performed on data from studies that included individuals with LBP to identify the relationship between LBP clinical features and trunk postural control (Aim 8).In the first stage, each LBP clinical feature (pain intensity, disability, pain catastrophizing, and fear-avoidance beliefs) was entered using multilevel mixed-effects models (for each study separately), including participant-level variables (age, sex, and BMI).In the second stage, a standard meta-regression was performed to pool regression (adjusted) coefficients using a random-effects model with the REML method.This analysis was performed for each identified outcome, visual condition, and direction.
2.8.3.Other analysis.Descriptive statistics (e.g., means and standard errors) of the outcomes from studies with data from only LBP or pain-free individuals were plotted with the outcomes from other studies with both groups (individuals with and without LBP) using IPD (if available) or aggregate data (Aim 9).We did this by applying standardised statistical methods (multilevel mixed-effects models), including confounding variables (age, sex, and BMI).This analysis was performed for each identified outcome, visual condition, and direction.We compared between groups using the visual inspection of mean plots.Outcomes from studies for which IPD could not be obtained were summarised briefly in a narrative manner and were contrasted with the findings of the IPD meta-analysis.

Comprehensiveness of reporting and methodological quality.
The comprehensiveness of reporting and quality of methods were assessed twice (see S15-S20 Tables for details): first using the study-level data as reported in the published version of all included papers (n = 40), and second using the IPD obtained from authors (n = 24/40).Using published versions (n = 40), the overall reporting/quality score of all domains was 66.3% for studies with LBP data (n = 15/40) and 63.4% for studies with only pain-free data (n = 25/40).The lowest and highest reporting/quality domain score for studies with LBP data were obtained for the confounding effects control domain (39.4%) and the experimental setup/protocol domain (76.7%), respectively.For studies with only pain-free data, the lowest and highest reporting/quality domain score were obtained for the confounding effects control domain (18.7%) and participant characteristics (80%), respectively.
When reporting/quality was assessed using the IPD set (n = 24), there was a marked improvement in reporting/quality scores.The overall reporting/quality score of all domains increased from 67.5% to 90% for studies with LBP data (n = 10/24) and from 63.3% to 88.2% for studies with only pain-free data (n = 14/24).).There was an increase in the number of calculated measures when IPD were obtained.Many outcome measures were calculated in both the anteroposterior and mediolateral directions, but some measures were only calculated in the resultant direction.Outcomes and main findings of each study are presented in Tables 5  and 6.The following sections describe the outcomes and results of the IPD analysis and then compare them with the findings of studies that could not be included in the IPD analysis.
IPD for some participants (102 participants) were excluded from the analysis because of missing demographic information (23 participants) [19,33,80,85,90], or because they included participants who had no current pain (79 participants) but had a LBP history within the last 12 months [33] (S21 Table provides details of the excluded participants).For studies with multiple levels of seat difficulty, the level of seat difficulty was selected that matched closest with the seat build characteristics of other studies that included a single seat difficulty level (S22 Table provides details of the excluded seat difficulty levels).
Studies (n = 10/40) that referred to an already identified IPD set were excluded [21, 24, 26, 27, 74-76, 86, 89, 94].Data for four studies that were considered for the quantitative analysis (either as IPD or aggregate data) were excluded from quantitative analysis because they did not report the required values (e.g., means) of any of the identified primary and secondary outcome measures [81,93], or they only reported analysis of measures in the resultant direction [72,82].

3.3.3.
Comparison between the outcomes of single-group studies with the outcome of two-group studies using IPD or aggregate data.Mean plots of the outcomes of studies with data from only LBP or pain-free individuals (using either IPD or aggregate data) are presented with mean plots of the outcomes of studies included in the IPD meta-analysis (individuals with and without LBP).For primary outcome measures, the visual inspection of mean plots of RMS displ (Fig 4) and M vel (Fig 5) revealed substantial variation in outcomes between studies renders comparisons between groups unclear for the eyes open condition in both directions.Few studies with data from only pain-free individuals were available for RMS displ (n = 3) [25,77,92] and M vel (n = 4) [25,73,78,92] during eyes closed, and the results of outcomes differed considerably from studies with data including both individuals with and without LBP.One study with data from individual with only LBP [73] showed that M vel was higher during eyes closed compared to most studies with data either from only pain-free individuals or individuals with versus without LBP in both directions (Fig 5).Similar observations can be made for secondary outcome measures (range, MPF, D short , D long , CP time , and CP dist ), for which there were differences in outcomes when comparing studies with data from only pain-free individuals and studies with data from individuals with versus without LBP (see S8-S13 Figs for mean plots).
There was only one study [72] that included individuals with and without LBP for which IPD could not be obtained.That study reported that RMS displ and D short were higher among individuals with than without LBP in the resultant direction during the most difficult level of seat instability (50% R spring ), which is similar to the results of the IPD meta-analysis.There were five studies [79,81,82,93,95] that included data from only pain-free individuals where IPD could not be obtained, but aggregate data were obtained for RMS displ [79], M vel [95], and range [79] from two studies [79,95] (Figs 4 and 5 and S8 Fig) .Neither IPD nor aggregate data could be obtained for the other studies [81,82,93], as two of these studies [81,93] investigated different outcome measures, and one study [82] reported outcome measures only in the  resultant direction.As such, it was it challenging to compare the results of these studies with studies included in the IPD meta-analysis.After standardising the statistical analysis and including covariates in the models for all studies with available IPD, there were significant between-group differences for some outcome measures (RMS displ , range, and D long were higher among individuals with than without LBP) in two studies [31,34] that reported no differences between groups for the same outcome measures in the original published versions.

IPD meta-analysis: Effects of vision.
IPD meta-analyses also revealed significant interaction effects with vision on some outcome measures.For primary outcome measures (Fig 6), the difference between groups was greater with eyes closed than eyes open for RMS displ in both the anteroposterior and mediolateral directions (all P<0.05).For secondary outcome measures (see S22-S24 Figs for forest plots), the difference between groups was also greater with eyes closed than eyes open for D short and D long in the anteroposterior direction (all P<0.05; S23 Fig) .3.3.6.IPD meta-regression: Effects of LBP clinical features.IPD meta-regressions revealed no significant associations (see S23 and S24 Tables) between LBP clinical features (VAS/NPRS, RMDQ, PCS, FABQ-PA, FABQ-W and FABQ) and either primary (RMS displ and M vel ) or secondary (range and MPF) outcome measures.Results of individual analysis for each study using IPD are presented in S25-S34 Tables.

Discussion
This systematic review with IPD meta-analyses had several major findings.First, IPD metaanalyses indicate that trunk postural control differs between individuals with and without LBP.A pooling of results from multiple studies showed that RMS displ , range and D long were higher, and MPF was lower, among individuals with LBP, which can be concluded as indicating poorer postural control of the trunk.Second, IPD meta-analyses revealed that trunk postural control deteriorates more from removal of vision among individuals with than without LBP.Third, IPD meta-analyses revealed that older age and higher BMI have greater adverse

A two-stage individual participant data (IPD) meta-analysis of interactions between visual condition and the difference between groups on primary outcome measures of trunk postural control. (A) root mean square displacement (RMS displ ) and (B) mean velocity (M vel )
. The results are presented as interaction (interact.)effect coefficients with 95% confidence intervals (95% CIs) using forest plots.Significant overall interact.effects with their respective P-values are highlighted in bold font.Greater effect in either directions indicates worse effect on trunk postural control for individuals with versus without low back pain.Sizing of circles reflects the weight (WT) of the contribution of a study on the pooled meta-analysis (weighted average) in percentage.I 2 reflects the percentage of total variability due to heterogeneity between studies.95% prediction interval (95% PI) reflects how much the effect size varies across studies.https://doi.org/10.1371/journal.pone.0296968.g006impacts on trunk postural control among individuals with than without LBP.Fourth, IPD meta-regressions indicated that the limited clinical features that we could evaluate for the LBP group are not associated with poorer trunk postural control.Fifth, the visual inspection of mean plots indicated that the comparison between groups using the data from single-group studies (only LBP or pain-free data) with the data from two-group studies (LBP and pain-free data) was challenging, due to substantial variation in outcomes between studies, which suggests that the comparison between groups is only compelling when both groups are tested identically in the same study.

Experimental methods
Differences in experimental methods between studies comparing individuals with and without LBP that were included in the IPD meta-analysis can explain some of the inconsistency in study outcomes [19,22,[29][30][31][32][33][34][35].Important variation between studies includes differences in the seat apparatus-greater differences between groups were observed in studies that used the hemisphere-based seat [19,22,30,31,35] than the springs-based seat [29,32,34].This is likely explained by differences in seat dynamics imposed by hemisphere-based seats (more challenging) versus springs-based seats (less challenging) when balancing.A lesser task difficulty might limit the potential to identify differences between groups [22,72,96].The research group using the unstable springs-based sitting paradigm chose to use a not too difficult task (60% R spring ) to help participants maintain balance long enough (60 s) to obtain reliable summary measures of trunk postural control [24].It might be possible to slightly increase the difficulty of the task (e.g., from 60% to 50%), without affecting reliability too much, to increase the discriminative power of this test.Another possibility is to find the threshold of stability by finding the maximum task difficulty (e.g., . . .50, 55, 60, 65, 70% R spring , . ..) in which stability can be maintained over a given time [97].Removal of vision (eyes open versus eyes closed) also increases the challenge for trunk postural control, though this effect was not investigated in many studies [29,[32][33][34].One study described the LBP group in limited detail (e.g., no information about pain intensity, disability, or other clinical features) [33].These limitations reduced the total sample size that could be included in the IPD meta-analysis/meta-regression.As suggested by the analysis of covariates (see below), differences in participant characteristics might also explain some of the conflicting results between studies.
There are also several differences in experimental methods for studies with data from only LBP or pain-free individuals (single-group studies) that might explain the high differences in mean plots of the same outcome, including sample size, participant characteristics (e.g., sex, or anthropometrics), experimental setup (e.g., seat apparatus) and protocol (e.g., trial duration), and the data pre-processing (e.g., filters, data sampling for data reduction, etc.).These differences are greatly impacted on the comparison between groups, when we compared (visually using the mean plots) the outcomes of studies with data from only LBP or pain-free individuals with the outcomes of studies included in the IPD meta-analysis (individuals with and without LBP).This suggests that between-group comparison demands measurement within the same study with the same experimental parameters.

Differences in trunk postural control between individuals with and without LBP
IPD meta-analyses revealed that individuals with LBP had higher RMS displ , range, and D long than those without LBP, suggesting greater CoP/seat movements and less corrective long-term behaviour [98][99][100].Further, IPD meta-analyses indicated that some individuals with LBP had lower MPF, which could be explained by greater trunk stiffness [33].Taken together, these findings suggest the presence of compromised trunk postural control among individuals with LBP, and which might be explained by multiple plausible mechanisms.For instance, less proprioceptive sensitivity would likely reduce the ability to coordinate the muscle responses (e.g., delayed trunk muscle responses) required to maintain balance in the unstable sitting task [22,101], affecting the accuracy and precision of trunk movement [102], and also affecting the detection of movement errors to execute required postural adjustments [19,103].Together, these features would tend to destabilize trunk postural control as reflected by higher RMS displ , range, and D long .Alternatively, some individuals with LBP might adopt a distinct muscle activation strategy to control balance, such as increased muscle co-activation [78,104,105], either to reduce reliance on proprioception or to protect the spine from threat.Trunk stiffening would result in moving the trunk more as a whole and increase the moments of inertia on an unstable seat [79], resulting in lower sway frequency, and thus greater movements and stochastic activity of the CoP/seat.Decreased sense of lumbar proprioception [106][107][108][109], increased trunk stiffness [105,[110][111][112], and delayed onset or offset of trunk muscles activity [22,42,105,[113][114][115] have been widely documented in the LBP population.Regardless of the origin, impaired trunk postural control could be problematic for individuals with LBP and might expose them to the risk of sustaining or aggravating a back injury if maintained.For instance, A delay in response time would increase the vulnerability of spine to injury under sudden loading conditions [116], such as increasing tissue strain (further displacement) and stress (greater muscle force) of the spine [101].Accordingly, an altered muscle recruitment pattern would be used as a compensation strategy [105].However, adopting a co-contraction strategy also would increase spinal loads [117,118].

Effects of vision
Corrective adjustments for postural control require contributions of the visual, vestibular, and proprioceptive (sense of positioning and movements) senses [119][120][121].The demand for proprioceptive feedback (e.g., sensory receptors in spinal muscles, joints, and other tissues [73,122]) is even greater in the absence of visual feedback.In line with this, IPD meta-analyses revealed that RMS displ , D short , and D long were each increased by removal of vision, but this effect was greater for individuals with than without LBP, indicating greater CoP/seat movements and less tightly regulated behaviour [98][99][100].This implies that, without visual feedback, the quality of trunk postural control is reduced to a greater extent with LBP, as it potentially cannot be adequately compensated for by other alternative sensory sources.Control of CoP/seat diffusion rate relies on adequate and timely sensory information [123].Moving the CoP/seat further away from a relative equilibrium point over shorter periods of time in the LBP group might be a strategy to overcome higher sensory thresholds caused by LBP injury, and thus would enhance feedback [22,98], or a strategy to increase muscle stiffness to promote rapid postural corrections [77,78,124].Further, fewer corrective adjustments for CoP/seat movements over long-term intervals of time observed in the LBP group may due to lower proprioceptive sensitivity [22,35,101] or deficits in sensory reweighting [35,77].This conjecture is consistent with previous data that show individuals with LBP could not modulate and reweight their sensory information to other alternative sources (most likely proprioceptive feedback) when visual feedback was unavailable [22,35].

Effects of participant characteristics
IPD meta-analyses revealed that older age and higher BMI more negatively impacted trunk postural control for those with than without LBP.The findings of higher D short and longer CP dist and CP time indicate that the CoP/seat moved further within the short-term region, with a later critical point (the point at which corrective responses take place to slow CoP/seat diffusion rate and limit further displacement of CoP/seat [98,99]) among individuals with LBP who were older or had higher BMI.These findings might be explained by age-related changes in muscle mass [125,126], proprioception [127], muscle response [103], or sensory thresholds [103].Higher body mass that is placed above the radius of a hemisphere creates a mechanically more challenging system to balance [21].Higher body mass is also likely to impose biomechanical constraints that require greater muscular activity and torque [128] and interfere with the coordination of joints and muscles due to changes in anthropometrics [129].The impacts of these factors might be accentuated for the LBP group, who already have impaired balance.Although anthropometric characteristics (height and weight) may influence measures of trunk postural control during unstable sitting [21,24], IPD meta-analyses confirmed differences between groups while adjusting for these confounding variables.

Effects of LBP clinical features
Although we had limited data available, IPD meta-regressions indicated that LBP intensity, disability, pain catastrophizing, and fear of movement were not associated with poorer trunk postural control among individuals with LBP.It should be noted, though, that individuals with LBP in most studies had low scores on disability, catastrophizing, and fear avoidance.It is unlikely that our findings are explained by insufficient time for these features to develop [35,130], since several studies included participants who suffered from LBP for long periods [22, 29-32, 34, 73].Greater detail and consistent collection of clinical features would have made findings more robust.

Implications
Our IPD meta-analyses confirm that trunk postural control is different between individuals with and without LBP.This justified future evaluation whether interventions that address trunk postural control might be helpful in the management of LBP.The current review identified some weaknesses in the available IPD sets that should be addressed in future research.For example, the evaluation of clinical and related features of LBP such as pain intensity, disability level, and psychological features should be included in studies of postural control-many studies included in this review did not asses these features and this limited the capacity to evaluate the impact of these features on trunk postural control.Longitudinal studies including investigation of different stages of LBP (time) are required to determine whether the differences in trunk postural control are adaptive or maladaptive.This review also identified that the outcome measures used to evaluate trunk postural control provide limited insight into the interpretation of differences in trunk postural control and new research could focus on how the trunk is controlled (e.g., coordination of different spine regions/segments), not just a general output that is included in this IPD meta-analysis (e.g., CoP/seat motion).This could provide more information on potential modifiable factors for treatment.

Limitations
This review has several limitations that require consideration.First, IPD could not be obtained for six studies, although the impact of this will be limited as only one of those studies included data from individuals with versus without LBP.Second, overall SMD effect sizes for betweengroup differences were small or close to medium (maybe due to the small number of included studies), and 95% PIs were wider in some IPD meta-analyses (likely due to lack of precision).Third, there was evidence of small-study effects (funnel plots asymmetry) which suggests some publication bias.However, the results should be interpreted with caution as the number of included studies in all IPD meta-analyses was <10 and in many cases with secondary outcome measures there were <5 studies.This is just below the threshold of 10 studies that is considered necessary to detect funnel plots asymmetry [131].The funnel plots asymmetry detected for several measures might be explained by the small number of included studies and betweenstudy heterogeneity [131] rather than publication bias.Fourth, the effect of LBP duration (e.g., acute, sub-acute, chronic) on trunk postural control could not be evaluated in the IPD metaanalysis because of between-study variation in how these features were defined (or not at all), and an overall very few studies included individuals with acute LBP.Fifth, the effect of task difficulty (based on seat apparatus and its build characteristics) on trunk postural control between groups could not be examined because of the limited number of studies that used springs-base seats.Sixth, although some data from our IPD meta-analysis might have violated the normality assumptions, we used multilevel mixed-effects models in the first stage of our two-stage IPD meta-analysis which are robust to violations of normality assumptions [132].Seventh, we limited our review to include only those studies that were published in English.

Conclusions
This IPD analysis provides robust evidence that trunk postural control is compromised among individuals with LBP.The findings are likely explained by delayed or less accurate corrective responses, for which there are several plausible mechanisms-impaired proprioception, altered sensory processing/reweighting, or increased muscle co-activation.We cannot discriminate whether changes in trunk postural control are a cause or consequence of LBP.Our investigation highlights the value of IPD analysis to draw robust conclusions from biomechanical data, but also brings to light the affect variation in available (limited) data has on addressing important questions related to the association with LBP clinical and related features.results are presented as standardised mean differences (SMDs) with 95% confidence intervals (95% CIs) using forest plots.Significant overall effect sizes with their respective P-values are highlighted in bold font.Sizing of squares reflects the weight (WT) of the contribution of a study on the pooled meta-analysis (weighted average) in percentage.I 2 reflects the percentage of total variability due to heterogeneity between studies.95% prediction interval (95% PI) reflects how much the effect size varies across studies.The results are presented as interaction (interact.)effect coefficients with 95% confidence intervals (95% CIs) using forest plots.Significant overall interact.effects with their respective P-values are highlighted in bold font.Greater effect in either directions indicates worse effect on trunk postural control for individuals with versus without low back pain.Sizing of circles reflects the weight (WT) of the contribution of a study on the pooled meta-analysis (weighted average) in percentage.I 2 reflects the percentage of total variability due to heterogeneity between studies.95% prediction interval (95% PI) reflects how much the effect size varies across studies.(JPG)

S15 Fig. A two-stage individual participant data (IPD) meta-analysis of interactions between participant characteristics and the difference between groups on the mean velocity (M vel ) of CoP/seat angle. (A) age, (B) sex, and (C) body mass index (BMI)
. The results are presented as interaction (interact.)effect coefficients with 95% confidence intervals (95% CIs) using forest plots.Significant overall interact.effects with their respective P-values are highlighted in bold font.Greater effect in either directions indicates worse effect on trunk postural control for individuals with versus without low back pain.Sizing of circles reflects the weight (WT) of the contribution of a study on the pooled meta-analysis (weighted average) in percentage.I 2 reflects the percentage of total variability due to heterogeneity between studies.95% prediction interval (95% PI) reflects how much the effect size varies across studies.(JPG)

S16 Fig. A two-stage individual participant data (IPD) meta-analysis of interactions between participant characteristics and the difference between groups on the range of
CoP/seat angle.(A) age, (B) sex, and (C) body mass index (BMI).The results are presented as interaction (interact.)effect coefficients with 95% confidence intervals (95% CIs) using forest plots.Significant overall interact.effects with their respective P-values are highlighted in bold font.Greater effect in either directions indicates worse effect on trunk postural control for individuals with versus without low back pain.Sizing of circles reflects the weight (WT) of the contribution of a study on the pooled meta-analysis (weighted average) in percentage.I 2 reflects the percentage of total variability due to heterogeneity between studies.95% prediction interval (95% PI) reflects how much the effect size varies across studies.The results are presented as interaction (interact.)effect coefficients with 95% confidence intervals (95% CIs) using forest plots.Significant overall interact.effects with their respective P-values are highlighted in bold font.Greater effect in either directions indicates worse effect on trunk postural control for individuals with versus without low back pain.Sizing of circles reflects the weight (WT) of the contribution of a study on the pooled meta-analysis (weighted average) in percentage.I 2 reflects the percentage of total variability due to heterogeneity between studies.95% prediction interval (95% PI) reflects how much the effect size varies across studies.The results are presented as interaction (interact.)effect coefficients with 95% confidence intervals (95% CIs) using forest plots.Significant overall interact.effects with their respective P-values are highlighted in bold font.Greater effect in either directions indicates worse effect on trunk postural control for individuals with versus without low back pain.Sizing of circles reflects the weight (WT) of the contribution of a study on the pooled meta-analysis (weighted average) in percentage.I 2 reflects the percentage of total variability due to heterogeneity between studies.95% prediction interval (95% PI) reflects how much the effect size varies across studies.(JPG)

S19 Fig. A two-stage individual participant data (IPD) meta-analysis of interactions between participant characteristics and the difference between groups on the long-term diffusion (D long ) of CoP/seat angle. (A) age, (B) sex, and (C) body mass index (BMI).
The results are presented as interaction (interact.)effect coefficients with 95% confidence intervals (95% CIs) using forest plots.Significant overall interact.effects with their respective P-values are highlighted in bold font.Greater effect in either directions indicates worse effect on trunk postural control for individuals with versus without low back pain.Sizing of circles reflects the weight (WT) of the contribution of a study on the pooled meta-analysis (weighted average) in percentage.I 2 reflects the percentage of total variability due to heterogeneity between studies.95% prediction interval (95% PI) reflects how much the effect size varies across studies.The results are presented as interaction (interact.)effect coefficients with 95% confidence intervals (95% CIs) using forest plots.Significant overall interact.effects with their respective P-values are highlighted in bold font.Greater effect in either directions indicates worse effect on trunk postural control for individuals with versus without low back pain.Sizing of circles reflects the weight (WT) of the contribution of a study on the pooled meta-analysis (weighted average) in percentage.I 2 reflects the percentage of total variability due to heterogeneity between studies.95% prediction interval (95% PI) reflects how much the effect size varies across studies.The results are presented as interaction (interact.)effect coefficients with 95% confidence intervals (95% CIs) using forest plots.Significant overall interact.effects with their respective P-values are highlighted in bold font.Greater effect in either directions indicates worse effect on trunk postural control for individuals with versus without low back pain.Sizing of circles reflects the weight (WT) of the contribution of a study on the pooled metaanalysis (weighted average) in percentage.I 2 reflects the percentage of total variability due to heterogeneity between studies.95% prediction interval (95% PI) reflects how much the effect size varies across studies.(JPG)

S22 Fig. A two-stage individual participant data (IPD) meta-analysis of interactions between visual condition and the difference between groups on secondary outcome measures of trunk postural control. (A) range and (B) mean power frequency (MPF)
. The results are presented as interaction (interact.)effect coefficients with 95% confidence intervals (95% CIs) using forest plots.Significant overall interact.effects with their respective P-values are highlighted in bold font.Greater effect in either directions indicates worse effect on trunk postural control for individuals with versus without low back pain.Sizing of circles reflects the weight (WT) of the contribution of a study on the pooled meta-analysis (weighted average) in percentage.I 2 reflects the percentage of total variability due to heterogeneity between studies.95% prediction interval (95% PI) reflects how much the effect size varies across studies.(JPG)

S23 Fig. A two-stage individual participant data (IPD) meta-analysis of interactions between visual condition and the difference between groups on secondary outcome measures of trunk postural control.
(A) short-term diffusion (D short ) and (B) long-term diffusion (D long ).The results are presented as interaction (interact.)effect coefficients with 95% confidence intervals (95% CIs) using forest plots.Significant overall interact.effects with their respective P-values are highlighted in bold font.Greater effect in either directions indicates worse effect on trunk postural control for individuals with versus without low back pain.Sizing of circles reflects the weight (WT) of the contribution of a study on the pooled meta-analysis (weighted average) in percentage.I 2 reflects the percentage of total variability due to heterogeneity between studies.95% prediction interval (95% PI) reflects how much the effect size varies across studies. .The results are presented as interaction (interact.)effect coefficients with 95% confidence intervals (95% CIs) using forest plots.Significant overall interact.effects with their respective P-values are highlighted in bold font.Greater effect in either directions indicates worse effect on trunk postural control for individuals with versus without low back pain.Sizing of circles reflects the weight (WT) of the contribution of a study on the pooled meta-analysis (weighted average) in percentage.I 2 reflects the percentage of total variability due to heterogeneity between studies.95% prediction interval (95% PI) reflects how much the effect size varies across studies.(JPG)

Fig 1 .
Fig 1. Illustration of the unstable sitting paradigm.A seat is attached to (A) a hemisphere or (B) a central pivot surrounded by four springs.Task difficulty depends on the seat apparatus and its build characteristics.For the hemisphere-based seat, task difficulty depends on the radius of a hemisphere and the seat height from the support surface.For the springs-based seat, task difficulty depends on the stiffness of springs and the distance of springs (R spring ) from the pivot in percentage.The percentage of R spring is relative to the gravitation gradient (e.g., 100% = the location of springs that would fully balance the mass of the participant as if sitting on a stable chair).Both seats are commonly positioned on a force platform.A foot plate is usually attached to the seat that maintains the knees at 90f lexion and reduces the contribution of the lower limbs to balance control.In most applications, participants are asked to keep their arms crossed at their chest, to minimize the contribution of the upper limbs.
for both individuals with (361 participants) and without LBP (369 participants) for 378 female (individuals with LBP = 189; individuals without LBP = 189) and 352 male (individuals with LBP = 172; individuals without LBP = 180) participants.Those with LBP were younger

Fig 3 .
Fig 3. A two-stage individual participant data (IPD) meta-analysis, comparing individuals with and without low back pain (LBP) on primary outcome measures of trunk postural control.(A) root mean square displacement (RMS displ ) and (B) mean velocity (M vel ).The results are presented as standardised mean differences (SMDs) with 95% confidence intervals (95% CIs) using forest plots.Significant overall effect sizes with their respective Pvalues are highlighted in bold font.Sizing of squares reflects the weight (WT) of the contribution of a study on the pooled meta-analysis (weighted average) in percentage.I 2 reflects the percentage of total variability due to heterogeneity between studies.95% prediction interval (95% PI) reflects how much the effect size varies across studies.https://doi.org/10.1371/journal.pone.0296968.g003

Fig 4 .
Fig 4. A quantitative (non-pooled) analysis of all studies on the root mean square displacement (RMS displ ) of CoP/seat angle using IPD or aggregate data (if IPD were not available).Mean plots of the results from studies that were included in the individual participant data (IPD) metaanalysis (individuals with versus without low back pain [LBP]) are presented with mean plots of the results from studies that tested only pain-free individuals [CTRL].The results are presented as means with standard errors.For studies with two groups, significant differences between individuals with and without LBP are shown with square bracket.No bars in some studies = no data available.https://doi.org/10.1371/journal.pone.0296968.g004

Fig 5 .
Fig 5.A quantitative (non-pooled) analysis of all studies on the mean velocity (M vel ) of CoP/seat angle using IPD or aggregate data (if IPD were not available).Mean plots of the results from studies that were included in the individual participant data (IPD) meta-analysis (individuals with versus without low back pain [LBP]) are presented with mean plots of the results from studies that tested either only LBP or pain-free individuals [CTRL].The results are presented as means with standard errors.For studies with two groups, significant differences between individuals with and without LBP are shown with square bracket.No bars in some studies = no data available.https://doi.org/10.1371/journal.pone.0296968.g005

Fig 6 .
Fig 6.A two-stage individual participant data (IPD) meta-analysis of interactions between visual condition and the difference between groups on primary outcome measures of trunk postural control.(A) root mean square displacement (RMS displ ) and (B) mean velocity (M vel ).The results are presented as interaction (interact.)effect coefficients with 95% confidence intervals (95% CIs) using forest plots.Significant overall interact.effects with their respective P-values are highlighted in bold font.Greater effect in either directions indicates worse effect on trunk postural control for individuals with versus without low back pain.Sizing of circles reflects the weight (WT) of the contribution of a study on the pooled meta-analysis (weighted average) in percentage.I 2 reflects the percentage of total variability due to heterogeneity between studies.95% prediction interval (95% PI) reflects how much the effect size varies across studies.
(JPG) S3 Fig.A two-stage individual participant data (IPD) meta-analysis, comparing individuals with and without low back pain (LBP) on secondary outcome measures of trunk postural control.(A) mean squared distance coordinate of the critical point (CP dist ) and (B) mean time coordinate of the critical point (CP time ).The results are presented as standardised mean differences (SMDs) with 95% confidence intervals (95% CIs) using forest plots.Significant overall effect sizes with their respective P-values are highlighted in bold font.Sizing of squares reflects the weight (WT) of the contribution of a study on the pooled meta-analysis (weighted average) in percentage.I 2 reflects the percentage of total variability due to heterogeneity between studies.95% prediction interval (95% PI) reflects how much the effect size varies across studies.(JPG) S4 Fig. Funnel plots to test the potential presence of small-study effects (funnel plot asymmetry) in the two-stage individual participant data (IPD) meta-analysis of primary outcome measures.(A) root mean square displacement (RMS displ ) and (B) mean velocity (M vel ).The vertical solid red lines represent the overall standardized mean differences (SMDs) from the IPD meta-analysis.The two diagonal dashed lines in both sides represent the pseudo 95% confidence intervals (95% CIs) around the overall SMDs for each standard error (precision).Each orange dot represents a SMD for an individual study.Small studies are scattered more widely at the bottom of the funnel plot (lower precision) and larger studies are scattered more at the top of the funnel plot (greater precision).P-values for the potential presence of significant small-study effects (funnel plot asymmetry) are highlighted in bold font.(JPG) S5 Fig. Funnel plots to test the potential presence of small-study effects (funnel plot asymmetry) in the two-stage individual participant data (IPD) meta-analysis of secondary outcome measures.(A) range and (B) mean power frequency (MPF).The vertical solid red lines represent the overall standardized mean differences (SMDs) from the IPD meta-analysis.The two diagonal dashed lines in both sides represent the pseudo 95% confidence intervals (95% CIs) around the overall SMDs for each standard error (precision).Each orange dot represents a SMD for an individual study.Small studies are scattered more widely at the bottom of the funnel plot (lower precision) and larger studies are scattered more at the top of the funnel plot (greater precision).P-values for the potential presence of significant small-study effects (funnel plot asymmetry) are highlighted in bold font.(JPG)S6 Fig. Funnel plots to test the potential presence of small-study effects (funnel plot asymmetry) in the two-stage individual participant data (IPD) meta-analysis of secondary outcome measures.(A) short-term diffusion (D short ) and (B) long-term diffusion (D long ).The vertical solid red lines represent the overall standardized mean differences (SMDs) from the IPD meta-analysis.The two diagonal dashed lines in both sides represent the pseudo 95% confidence intervals (95% CIs) around the overall SMDs for each standard error (precision).Each orange dot represents a SMD for an individual study.Small studies are scattered more widely at the bottom of the funnel plot (lower precision) and larger studies are scattered more at the top of the funnel plot (greater precision).P-values for the potential presence of significant small-study effects (funnel plot asymmetry) are highlighted in bold font.(JPG) S7 Fig. Funnel plots to test the potential presence of small-study effects (funnel plot asymmetry) in the two-stage individual participant data (IPD) meta-analysis of secondary outcome measures.(A) mean squared distance coordinate of the critical point (CP dist ) and (B) mean time coordinate of the critical point (CP time ).The vertical solid red lines represent the overall standardized mean differences (SMDs) from the IPD meta-analysis.The two diagonal dashed lines in both sides represent the pseudo 95% confidence intervals (95% CIs) around the overall SMDs for each standard error (precision).Each orange dot represents a SMD for an individual study.Small studies are scattered more widely at the bottom of the funnel plot (lower precision) and larger studies are scattered more at the top of the funnel plot (greater precision).P-values for the potential presence of significant small-study effects (funnel plot asymmetry) are highlighted in bold font.(JPG) S8 Fig.A quantitative (non-pooled) analysis of all studies on the range of CoP/seat angle using IPD or aggregate data (if IPD were not available).Mean plots of the results from studies that were included in the individual participant data (IPD) meta-analysis (individuals with versus without low back pain [LBP]) are presented with mean plots of the results from studies that tested only pain-free individuals [CTRL].The results are presented as means with standard errors.For studies with two groups, significant differences between individuals with and without LBP are shown with square bracket.No bars in some studies = no data available.(JPG) S9 Fig.A quantitative (non-pooled) analysis of all studies on the mean power frequency (MPF) of CoP/seat angle using IPD or aggregate data (if IPD were not available).Mean plots of the results from studies that were included in the individual participant data (IPD) meta-analysis (individuals with versus without low back pain [LBP]) are presented with mean plots of the results from studies that tested only pain-free individuals [CTRL].The results are presented as means with standard errors.For studies with two groups, significant differences between individuals with and without LBP are shown with square bracket.No bars in some studies = no data available.(JPG) S10 Fig.A quantitative (non-pooled) analysis of all studies on the short-term diffusion (D short ) of CoP/seat angle using IPD or aggregate data (if IPD were not available).Mean plots of the results from studies that were included in the individual participant data (IPD) meta-analysis (individuals with versus without low back pain [LBP]) are presented with mean plots of the results from studies that tested only pain-free individuals [CTRL].The results are presented as means with standard errors.For studies with two groups, significant differences between individuals with and without LBP are shown with square bracket.No bars in some studies = no data available.(JPG) S11 Fig.A quantitative (non-pooled) analysis of all studies on the long-term diffusion (D long ) of CoP/seat angle using IPD or aggregate data (if IPD were not available).Mean plots of the results from studies that were included in the individual participant data (IPD) meta-analysis (individuals with versus without low back pain [LBP]) are presented with mean plots of the results from studies that tested only pain-free individuals [CTRL].The results are presented as means with standard errors.For studies with two groups, significant differences between individuals with and without LBP are shown with square bracket.No bars in some studies = no data available.(JPG) S12 Fig.A quantitative (non-pooled) analysis of all studies on the mean squared distance coordinate of the critical point (CP dist ) of CoP/seat angle using IPD or aggregate data (if IPD were not available).Mean plots of the results from studies that were included in the individual participant data (IPD) meta-analysis (individuals with versus without low back pain [LBP]) are presented with mean plots of the results from studies that tested only pain-free individuals [CTRL].The results are presented as means with standard errors.For studies with two groups, significant differences between individuals with and without LBP are shown with square bracket.No bars in some studies = no data available.(JPG) S13 Fig.A quantitative (non-pooled) analysis of all studies on the mean time coordinate of the critical point (CP time ) of CoP/seat angle using IPD or aggregate data (if IPD were not available).Mean plots of the results from studies that were included in the individual participant data (IPD) meta-analysis (individuals with versus without low back pain [LBP]) are presented with mean plots of the results from studies that tested only pain-free individuals [CTRL].The results are presented as means with standard errors.For studies with two groups, significant differences between individuals with and without LBP are shown with square bracket.No bars in some studies = no data available.(JPG) S14 Fig.A two-stage individual participant data (IPD) meta-analysis of interactions between participant characteristics and the difference between groups on the root mean square displacement (RMS displ ) of CoP/seat angle.(A) age, (B) sex, and (C) body mass index (BMI).
(JPG) S17 Fig.A two-stage individual participant data (IPD) meta-analysis of interactions between participant characteristics and the difference between groups on the mean power frequency (MPF) of CoP/seat angle.(A) age, (B) sex, and (C) body mass index (BMI).
(JPG) S18 Fig.A two-stage individual participant data (IPD) meta-analysis of interactions between participant characteristics and the difference between groups on the short-term diffusion (D short ) of CoP/seat angle.(A) age, (B) sex, and (C) body mass index (BMI).
(JPG) S20 Fig.A two-stage individual participant data (IPD) meta-analysis of interactions between participant characteristics and the difference between groups on the mean squared distance coordinate of the critical point (CP dist ) of CoP/seat angle.(A) age, (B) sex, and (C) body mass index (BMI).
(JPG) S21 Fig.A two-stage individual participant data (IPD) meta-analysis of interactions between participant characteristics and the difference between groups on the mean time coordinate of the critical point (CP time ) of CoP/seat angle.(A) age, (B) sex, and (C) body mass index (BMI).
(JPG) S24 Fig.A two-stage individual participant data (IPD) meta-analysis of interactions between visual condition and the difference between groups on secondary outcome measures of trunk postural control.(A) mean squared distance coordinate of the critical point (CP dist ) and (B) mean time coordinate of the critical point (CP time )

Table 1 . Primary and secondary outcome measures.
Abbreviations: CoP, centre of pressure; M vel , mean velocity.*Some studies calculated seat angle (as a surrogate of CoP) from motion capture systems/sensors to assess trunk postural control.https://doi.org/10.1371/journal.pone.0296968.t001

Table 3 . Participant characteristics for studies with data from only pain-free individuals.
Abbreviations/Symbols: IPD, individual participant data; BMI, body mass index; NA, not available; ♂, male; ♀, female.Statistics for IPD analysis: Overall summary measures for studies that provided IPD sets.*Number of participants included in the IPD analysis.† Studies that were only included in the descriptive analysis but excluded from quantitative/IPD analysis as these studies referred to an already identified IPD set.‡ IPD were not available as authors did not have access/authorisation to provide the IPD set.§ IPD were not available as authors did not respond to IPD request.**Overall summary measures for only studies with IPD.https://doi.org/10.1371/journal.pone.0296968.t003

Table 5 . Outcome measures and main findings for studies with data from individuals with LBP. Study Primary measures Secondary measures Main findings IPD RMS displ M vel Range MPF D short D long CP time CP dist
ML) was higher in LBP than control.LBP grabbed the safety rail more often than control.RMS displ , M vel , & D short increased with EC.

RMS displ M vel Range MPF D short D long CP time CP dist
RMS displ , M vel , D short , D long , CP dist ), & associated with higher BMI (RMS displ , M vel , D short , CP dist ), & more bar touches (RMS displ , M vel , MPF, D short , CP time , CP dist ).LBP, low back pain; RMS displ , root mean square displacement; M vel , mean velocity; MPF, mean power frequency; D short , short-term diffusion coefficient; D long , long-term diffusion coefficient; CP time , mean time coordinate of the critical point; CP dist , mean squared distance coordinate of the critical point; IPD, individual participant data; AP, anteroposterior; ML, mediolateral; R, resultant; SIL, seat instability level (higher level = higher difficulty); EC, eyes closed; CoP, center of pressure.*IPD were not available as authors did not have access/authorisation to provide the IPD set.† Studies that were only included in the descriptive analysis but excluded from quantitative/IPD analysis as these studies referred to an already identified IPD set https://doi.org/10.1371/journal.pone.0296968.t005

RMS displ M vel Range MPF D short D long CP time CP dist
Studies that were only included in the descriptive analysis but excluded from quantitative/IPD analysis as these studies referred to an already identified IPD set.† IPD were not available as authors did not have access/authorisation to provide the IPD set.
*‡ IPD were not available as authors did not respond to IPD request.https://doi.org/10.1371/journal.pone.0296968.t006 ).For the effects of age, the difference between groups was greater (worse effect for LBP group versus control group) in those with older age than those with younger age for D short (eyes open; both directions; all P<0.05; S18 Fig) and CP dist (eyes open; both directions; all P<0.05; S20 Fig).For the effects of BMI, the difference between groups was greater in those with higher BMI than those with lower BMI for CP dist (eyes closed; mediolateral direction; P<0.01; S20 Fig) and CP time (eyes open; mediolateral direction; P<0.05; S21 Fig).In contrast, for M vel , the difference between groups was greater in those with lower BMI than those with higher BMI (eyes closed; both directions; all P<0.05; S15 Fig).