Dose–response relationship and effect modifier of stabilisation exercises in nonspecific low back pain: a project-wide individual patient data re-analysis on 1483 intervention participants

1. Introduction

Exercise therapies reduce pain intensity and disability in low back pain.7 However, the overall effect is only low to moderate and often not even clinically relevant.7 These considerably small mean effects likely result from a high interindividual and intertrial variability.14

Such a heterogeneity and the (resulting) low-to-moderate group effects are also evident in interventions that aim to restore functional deficits.24 These types of interventions rank amongst the most established therapy regimen to treat low back pain24,27 and have central components in common: musculoskeletal control by afferent sensory input (in particular proprioceptive), central nervous system integration, and optimal motor control to ensure functional dynamic joint stability during perturbative situations.23 Varieties of sensorimotor movement therapies, such as motor control,24 stabilisation,4,13 functional restoration,8 and core-stability30 exercises, are likely to be the most effective active (exercise) treatments for low back pain.21,22.

An important characteristic of exercise interventions is the dose–response relationship. Knowing the optimal dose–response relationship is crucial for deriving and controlling recommendations on how an intervention needs to be structured for the training type (ie, the subtype of functional restoration training and/or additional components), the duration of the intervention, frequency, and intensity. Beyond the practical relevance, knowledge of the dose–response relationship may help explain some of the interindividual heterogeneity of the exercise effects. Identifying nonresponders and tailoring the intervention on the basis of an optimal dosage to response may, vice versa, lead to an increase of the overall group-level treatment effect. At group (meta-) level, a dose–response relationship in functional restoration (motor control) training is known; the largest treatment effects on pain intensity and disability reduction can be expected when training frequencies of 3 to 5 times per week with a single-bout duration of 20 to 30 minutes are scheduled.16 At an individual level, a high (general exercise) training dosage of more than 20 hours of training (total, not per week) may be more effective than lower doses.28 More detailed dose–response relationships in the target intervention stabilisation exercises have not, as yet, been delineated.

There are numerous further exercise effect modifiers in low back pain that lead to altered responses in the participants in the exercise group. Such known effect modifiers are age, the severity of symptoms at baseline, pain medication, motor control deficits, habitual physical activity levels, job demands, perceived social support, social satisfaction, anxiety, and depressive symptomatology.9,11,18,32 Thus, in order to delineate the role of the dose in exercise, such further potential effect (response) modifiers must also be considered.

In the underlying studies, on which the present analysis is based on, we found an effect of the pooling analyses for pain intensity and disability. The interventions led to moderate significant effects (with a considerable between-effect size heterogeneity) favouring the interventions in comparison to the control groups with significant between-group effect sizes (Hedge's g) ranging between −0.19 for pain and −0.24 for disability in the short term (3 weeks), between −0.26 for pain intensity and −0.27 for disability in the mid term (12 week intervention), and between −0.25 for pain intensity and −0.27 for disability in the long term (6-month follow-up).18 Given this mean effect in favour of the interventions in comparison to the control groups and the potential literature-based effect modifier and dose–response relationships in the target intervention stabilisation exercises, the overall objective of this project-wide individual patient data reanalysis was to derive the dose–response relationships of the sensorimotor stabilisation exercises, with respect to the contribution of further effect modifiers. We hypothesised that a dose–response relationship is verifiable, even when further potential effect modifiers are considered. More specifically, we hypothesised that irrespective of the type of sensorimotor exercise, a greater adherence to higher frequency training and a longer intervention duration leads to a more beneficial response on pain intensity, disability, and the number of disability days.

2. Methods 2.1. Design

The analyses and underlying studies were performed within the MiSpEx Network (Medicine in Spine Exercise—network).15 In this planned secondary analysis, data from all participants from 2 major multicentre studies were pooled: DRKS00004977 (05/16/2013, study 1) and DRKS00010129 (03/03/2016, study 2).20 Refer to the supplementary file 1 (available at https://links.lww.com/PAIN/B735) for further details of the data collection and pooled analysis.

2.2. From recruitment to randomisation

To be included in one of the underlying studies, representative volunteers were recruited and screened for eligibility. The main eligibility criteria were 1) being male or female aged 18-65 years and 2) having nonacute, nonspecific low back pain. Further inclusion and exclusion criteria are described elsewhere.20 Each participant signed informed consent prior to study enrolment.

Participants in the underlying studies were allocated to the intervention or control group using a 1:1 (study 1) or 2:1 (study 2) ratio, respectively. Both studies adopted a block-randomised procedure with nblock = 18 each.

2.3. Intervention type and dose

The control groups were real controls with no additional exercises beyond usual care. All 3 interventions were guided (centre based for the first 3 weeks of the intervention) and instructed (home based for weeks 4-12) by experienced medically trained sports therapists. Following the randomisation, the intervention group participants undertook a three-week centre-based guided intervention, followed by a nine-week home-based individual training; follow-up observations were made until 6 months after randomisation. Up to 12 weeks after randomisation, the intervention participants trained (scheduled) 3 times per week, with a mean total duration of 30 minutes per session. After 12 weeks, all participants were encouraged to continue further training by self-administered exercise at home (after completion of the intervention).

The core intervention consisted of 4 different sensorimotor exercises, each comprising 12 different levels of difficulty; 2 of the 4 exercises directly targeted the core stabilising muscles, whilst the other 2 sought to train motor control indirectly through the upper and lower extremities. All exercises were dynamic and can be described as (1) quadrupedal or all-fours stability, (2) deadlift or rowing, (3) double leg–single leg heel-pad stance, and (4) side planks. The exercises consisted of 3 series of 10 repetitions each. Further information on the different exercises, the set breaks, repetitions, start levels and progression determinations, home-counselling, and further training characteristics can be found in the study by Niederer et al.20 and in the supplementary file 2 (Table S1, available at https://links.lww.com/PAIN/B735).

In addition to these general exercise characteristics, 3 different additive training regimens were implemented in the respective study arms: (1) a stretching module (intervention arm stabilisation and stretching of study 1, performed after the stabilisation exercises), (2) self-initiated extra perturbative motor tasks such as ball tapping (one-handed, on the floor or against a wall) (intervention arm stabilisation and perturbation of study 2, during the stabilisation exercises), and (3) a behavioural therapy module (intervention arm stabilisation and behavioural therapy of study 1 during the stabilisation exercises, described in the study by Wippert et al.31.

Concerning the type and dose measure, a standardised training log was completed by the therapist (centre-based phase) and by the patient (home-based phase), respectively. In the training log, the type of exercises, exercise level, additional weights (if applied), and the detailed exercise dose were prospectively monitored. Beyond the exercise type, the exercise dose was quantified according to the duration of the intervention (weeks since randomisation) and training frequency (week−1). The total training duration (minutes per training) was kept constant.

2.4. Effect response

The dependent outcomes (effect estimators) were the characteristic pain intensity, disability, and disability days observed by means of the corresponding subscales of the Chronic Pain Grade questionnaire.12 The outcomes were observed at baseline, 3 weeks, 12 weeks, and 6 months post randomisation. Pain intensity and disability subscales were rated on a numeric rating scale from 0 (no pain/disability) to 10 (worst pain/disability). The subscales were then transformed into 0 to 100 pain intensity and disability sum scores. The disability days were processed as reported (number of days in the previous 3 months).

To interpret the potential intervention effects, clinically important treatment effects1 were defined as being (always scaled on a 0-10 points numeric rating scale) larger than a 1.5‐point decrease in pain (ie, larger than 1.5 points and a decrease in disability outcome larger than a 1.0‐point decrease) derived from a trial on the observed smallest worthwhile effect.3,7 These thresholds, usually observed in comparisons of the treatments vs no treatment, usual care, or placebo, thus indicate a likely beneficial response for the specific treatment.

2.5. Effect modifiers

Beyond basic patient characteristics, several potentially relevant effect modifiers were assessed across the studies; anxiety and depressive symptoms were quantified by the Hospital Anxiety and Depression Score,33 perceived social support was assessed by the Berlin Social Support Scales,26 and job position and lifestyle outcomes (alcohol use, smoking, analgesic medication, general satisfaction with health and sleep) were quantified by standardised questions.29 The job position was dichotomised into blue-color and white-collar workers in order to reflect the participant's job demands.

Motor control was quantified by postural control in the single-leg stance using balance boards (Wii Balance Board, Nintendo, Kyoto, Japan; modified by CSMi Computer Sports Medicine Inc, Stoughton, MA). Postural sway was estimated by the trace length [in millimeters] of the excursion of the centre of pressure. Participants maintained an upright, single-leg stance (eyes open focussing straight ahead on a target, a contralateral knee bent at 90°, barefoot and hands on hips) for 10 seconds (study 1) and 30 seconds (study 2). This setup and device shows sufficient validity and reproducibility.2,17 Values were processed as z-transformed normalised values. The z-transformation was made for each study separately to ensure between-studies comparability.

2.6. Statistical analyses

An alpha error of 5% was considered as a valid threshold for significance testing; all P values below are interpreted as being statistically significant. All analyses were performed using SPSS (Version 28, IBM SPSS).

The main data were descriptively displayed as means and standard deviations (or 95% confidence intervals).

Main analyses were performed using the (characteristic pain intensity, disability, and disability days) gain scores (baseline to 3 weeks, 3 weeks to 12 weeks, 12 weeks to 6 months of follow-up) of the intervention groups as dependent variables. Linear mixed models with repeated measures were calculated, while the time effects (intervention duration, first dose variable) were modelled as random slope and intercept effects (factor); intervention type and the remaining treatment dose variable (frequency) were modelled as fixed effects; the covariates were also modelled as fixed effects. The impact of the regressors (type and dose (frequency, duration) of the intervention) and the covariates on the response (outcomes) were fitted separately for each outcome. For the sensitivity analysis, several potential effect modifiers were included as covariates (fixed effects). All models were calculated following the checking for potential violations of the underlying assumptions for linear mixed models (heteroscedasticity, nonlinearity, normal distribution of the residuals).

Derived from the model estimates for the exercise frequencies and the (mean) training frequencies leading to the (mean) intervention effects, comparison of the intervention effects to the control group findings were also undertaken. The mean change scores were compared between the control and intervention groups with regards to a minimal exercise dose (i. e. frequency) that leads to an effect larger than the upper limit of the 95% confidence interval of the control group (at the different time points).

For further sensitivity of the analysis, we fitted a logistic regression with the same predictors (treatment, intervention duration, frequency) on the dichotomised outcomes for pain intensity and disability (change > minimal important change). The gain scores were subsequently transformed into clinically important treatment effects (yes or no); these dichotomised values were the dependent variables of the binary logistic regression models calculated afterwards. Again, the type and dose of the intervention were the regressors. Nagelkerkes R2 and the (pseudo-) odds ratio for a likely beneficial response (odds to reduce the symptoms by a clinically relevant amount) were calculated.

All models were calculated following the checking for potential violations of the underlying assumptions for linear mixed models (heteroscedasticity, nonlinearity, normal distribution of the residuals). Both mixed models and logistic regressions were modelled using all potentially relevant outcomes in a single model, and no regressor selection was performed.

Finally, for explorative purposes and in order to deduce a practically relevant dose recommendation, we calculated receiver operating characteristics analyses to identify a statistically optimal cutoff for the exercise frequency. The target variables were, again, likely to be beneficial responses to the treatment or not (as quantified by the minimal clinically important treatment effect). Areas under the curve values, including their 95% confidence intervals, were calculated as statistical outcomes; the highest Youden indices (Youdens J = sensitivity + specificity − 1) were selected for cutoff determination.

3. Results 3.1. Participant flow and characteristics

From the MiSpEx-project, all participants who were randomised into one of the intervention groups (n = 1483) were included in this analysis. The detailed project-wide participant flow and documented adverse events can be seen in the study of Niederer et al.18 The characteristics of the participants, stratified by the type of sensorimotor intervention, are displayed in Table 1. The dropout numbers were n = 204 (14%) until the third week visit, 237 (19%) between week 3 and week 12, and another n = 81 (8%) between week 12 and the 6 months of follow-up.18 In the control group, from the 767 included participants, n = 156 participants (20%) dropped out until week 3, another n = 56 (7%) until week 12, and another n = 16 (2%) until the 6-month follow-up.

Table 1 - Baseline characteristics of the participants included in this analysis and from the control group. Dimension Outcome Total sample Control SMT+P SMT+BT SMT+STR Number n (%) 2250 (100%) 767 (34%) 1049 (49%) 213 (9%) 221 (10%) Categorical data % Sex/gender Females/males/non-binary 57/43/0.1 58/42/0 56/44/0 58/41/0.4 59/41 Job position Blue/white collar worker 10/90 12/88 8/92 13/87 14/86 Pain medication intake No/yes 75/25 74/26 75/25 72/28 76/24 Interval scaled data Mean (standard deviation) Age y 39.7 (13.4) 37.4 (12.3) 41.7 (13.4) 38.5 (15.5) 39.29 (14.5) Body mass index kg/m2 24.2 (4.9) 24.0 (3.8) 24.8 (4.1) 23.4 (6.7) 22.8 (7.0) Physical activity/exercise/sport [minutes per week] 254 (318) 237 (194) 281 (444) 226 (164) 208 (150) Characteristic pain intensity* [Points] 34.5 (18) 32.1 (18.3) 36.5 (17.6) 32.4 (20.2) 35.5 (19.3) Disability* [Points] 24.4(20) 19.9 (19.8) 26.9 (18.9) 25.4 (21.2) 26.3 (20.4) Disability days* [per 3 mo] 2.04 (2.2) 1.8 (2.1) 2.20 (2.2) 2.11 (2.5) 2.11 (2.4) Depression† [Points] 5.730 (3.4) 5.32 (3.5) 5.8 (3.4) 5.36 (3.5) 5.31 (3.3) Anxiety† [Points] 3.64 (2.9) 3.62 (2.7) 3.57 (2.9) 3.81 (3.2) 3.93 (3.1) Perceived social support‡ [Points] 3.79 (0.3) 3.8 (0.2) 3.7 (0.4) 3.9 (0.1) 3.96 (0.1) Motor control—balance§ Centre of pressure trace [mm] 1132 (649) 834 (587) 1586 (554) 535 (282) 576 (298)

*Chronic Pain Grade questionnaire.

†Hospital Anxiety and Depression Scale.

‡Berlin Social Support Scales.

§CoP was processed as z-transformed (normalised per study) values.

Anthropometrics, disease-specific, and selected baseline values (potential effect modifiers) are displayed. All values are displayed for the total sample and are separated by the group allocation: control group, sensorimotor training plus perturbative motor task (SMT+P), sensorimotor training plus behavioural therapy (SMT+BT), and sensorimotor training and stretching (SMT + STR).

The chronic pain grades of the total intervention sample at baseline were grade 0: n = 83 (6%), grade 1: 1019 (69%), grade 2: 265 (18%), grade 3: 86 (6%), and grade 4: 30 (2%). The pain grades were not different between the intervention types (percentage share, P < 0.05): stabilisation and perturbation: 3%; 71%; 19%; 6%; 2%; stabilisation and behavioural therapy: 14%; 65%; 13%; 6%; 2%; and stabilisation and stretching: 8%; 64%; 20%; 5%; 3%, respectively, nor to the control groups: 4%; 77%; 13%; 5%, 0%.

3.2. Effects in comparison to the control group

In comparison to the control group, the stabilisation exercises improved symptoms with low-to-median mean effect sizes at 3 weeks (standardised mean difference between the control and intervention change scores for pain (Hedge g) = −0.19, 95% confidence interval [95% CI] −0.35 to −0.04) and disability (Hedge g = −0.24, [95% CI] −0.33 to −0.15). The same was found at the 12-weeks measurement (pain = −0.26 [−0.38 to −0.14]; disability = −0.27 [−0.40 to −0.13]) and at the 6-month follow-up (pain = −0.25 [−0.38 to −0.11]; disability = −0.25 [−0.39 to −0.11]).

The corresponding change scores from baseline to 3 weeks, always displayed as the pooled control vs the pooled intervention groups, were found to be −3.23 points [95% confidence interval −4.07 to −2.39] in the control group vs −7.02 points in the intervention group for pain and −3.41 [−4.36 to −2.45] vs 7.24 points for disability. The 12-week change scores were −4.65 [−5.75 to −3.56] vs −8.07 for pain intensity, and −4.95 [−6.17 to −3.73] vs −10.57 points for disability. At 6 months, these values were −5.00 [−6.28 to −3.72] vs −7.01 for pain intensity and −5.7 [−7.03 to −4.37] vs −10.8. The interventions led to small but significantly larger symptom reductions than the control at any measurement time in patients with low back. Detailed results are depicted in the study by Niederer et al.18

3.3. Descriptive response and dose values

Figure 1 shows the main data for the effect response taken for the dose–response analyses. Mean and confidence intervals of (response) pain intensity, disability, and disability days show a decrease in the symptom severity with increasing duration (descriptively).

F1Figure 1.:

Pain intensity, disability, and disability days during the 6-month duration, separated by the type of intervention. The data are displayed as means and 95% confidence intervals.

The number of (cumulated) trainings (dose) are depicted in the supplemental file 3 (Fig. S1, available at https://links.lww.com/PAIN/B735). The mean exercise frequencies were 2.7 (standard deviation 0.8) week−1 during the 3 weeks of centre-based intervention and 2.0 (1.1) week−1 during the home-based intervention, leading to a total mean frequency of 2.2 (1.2) week−1. The frequencies were broadly similar between the groups (stabilisation and perturbation vs stabilisation and behavioural therapy vs stabilisation and stretching) during the centre based (2.7 (0.9) vs 2.6 (0.6) vs 2.6 (0.6) week−1) nor during the total intervention time (2.0 (1.1) vs 2.0 (1.0) vs 2.0 (1.0) week−1), respectively.

3.4. Dose–response and effect modification analyses

In general, a dose–response relationship existed (Table 2). Detailed outcomes of mixed models with standard error and 95% confidence intervals are shown in the supplemental file 3 (Table S2, available at https://links.lww.com/PAIN/B735). A decrease in pain intensity, disability, and the number of disability days were all positively affected by the duration of the intervention (ie, lower or better values at 12 weeks after randomisation than at 3 weeks). Furthermore, higher training frequencies yielded greater symptom relief regarding pain intensity and days of disability. The type of exercise had an impact on the effect on pain intensity and disability (the largest decreases occurred in the stabilisation and perturbation group). All effects (intercept, slope, covariance parameters) were highly variable, as displayed by the unstructured estimates of the covariance parameters.

Table 2 - Outcomes of the linear mixed models on the dose–response relationship of the stabilisation exercises on pain, disability, and disability days. Characteristic PAIN intensity Disability Disability days Dose–response without effect modifiers and covariates  Intercept 5.50 1.00 0.24  Intervention duration: 3 wk −7.42 −7.23 −0.65  12 wk −2.96 −2.03 −0.23  6 mo Reference Type of intervention: stabilisation and perturbation −3.07 −1.68 −0.14 Stabilisation and behavioural 0.40 −1.19 −0.09 Stabilisation and stretching Reference Exercise frequency [week−1] −0.93 −0.09 −0.07 Dose-response with effect modifiers and covariates  Intercept 11.81 6.03 0.53  Intervention duration: 3 wk −7.35 −7.25 −0.66  12 wk −2.82 −2.26 −0.26  6 mo Reference Type of intervention: Stabilisation and perturbation −3.03 −1.05 −0.05 Stabilisation and behavioural −0.31 −1.32 −0.16 Stabilisation and stretching Reference Exercise frequency [week−1] −0.71 0.01 −0.07 Baseline pain grade [0-4] −3.48 −4.24 −0.45 Age [y] 0.03 0.02 0.01 Blue/white collar worker 0.33 −0.68 0.04 Exercise total [minutes/wk] <0.01 <0.01 <0.01 Postural control [mm] <0.01 <0.01 <0.01 Depression [HADS] 0.33 0.24 0.02 Anxiety [HADS] −0.02 −0.06 <0.01 Perceived social support [BSSS] −1.13 −0.06 <0.01 Painkiller intake (no/yes) 0.99 0.22 −0.12

For each outcome, the simple dose (type, duration, frequency) response model and a more complex model including the further effect modifiers are displayed. The fixed effects' estimates are depicted. Bold letters indicate significant contributors. (p < 0.05)

HADS, Hospital Anxiety and Depression Score.

The dose–response relationships remained robust when further potential effect modifiers were considered (Table 2 and Table S1, https://links.lww.com/PAIN/B735). Besides the effect of the intervention duration, type, and frequency, a better intervention outcome occurred in patients with higher baseline pain grades or with less depressive symptoms (less pain intensity) or those with painkiller medication (fewer disability days). The other potential regressors (age, blue-collar vs white-collar workers, and the total amount of exercise per week, anxiety, perceived social support, and postural control) had no modifying effect on the intervention effect.

For pain intensity, a training frequency of more than 0.0 times per week already led to an effect for pain in the 3-week centre-based period larger than the upper limit of the confidence interval of the change score in the control group. This value was statistically derived from the estimates for the training frequencies and the mean training frequencies leading to the mean effects. For the total 12-week intervention, exercising more than 0.2 times per week led to an effect when randomised in an intervention group. Most of the between-group effect occurred at the centre-based period; for an effect statistically larger than the control group's effect at week 3, one has to exercise at least 3.57 times per week.

3.5. Binary logistic regressions to predict a likely beneficial response

In n = 436 participants (29.4%), the pain intensity was, although the changes were small, reduced by the interventions larger than the minimal clinically important effect threshold. In addition, when considering disability, the share was comparable (again a small but clinically important reduction in 472 of the participants (31.9%)). The prediction models showed that the odds for a likely beneficial response for pain decreases in the short term but increases at the mid term and the follow-up with higher mean exercise frequencies (Table 3). For both outcomes of pain intensity and disability, the stabilisation and behavioural and the stabilisation and stretching participants were less likely to benefit (smaller odds ratio) at mid term (12 weeks after randomisation) and the 6-month follow-up.

Table 3 - Outcomes of the binary logistic regression models on the dose–response relationship of stabilisation exercises on the characteristic pain intensity and disability. Likelihood of having a clinically important effect Characteristic PAIN intensity Disability Odds ratio P Wald Odds ratio P Wald Value CI (LL to UL) Value CI (LL to UL) 3 wk of intervention Nagelkerkes R2 0.027 0.003 Intercept 1.088 0.661 0.192 0.356 0.001 16.837 Stabilisation and perturbation* 1 0.205 3.17 1 0.56 1.154 Stabilisation and behavioural 0.848 0.621 to 1.157 0.298 1.08 0.88 0.634 to 1.233 0.47 0.53 Stabilisation and stretching 0.748 0.523 to 1.069 0.111 2.541 0.85 0.592 to 1.216 0.37 0.802 Exercise frequency [week −1 ] 0.709 0.618 to 0.813 0.001 24.239 1.126 0.949 to 1.336 0.174 1.845 12 wk of intervention Nagelkerkes R2 0.041 0.014 Intercept 0.451 0.001 30.826 0.684 0.005 8.03 Stabilisation and perturbation * 1   0.001 31.481 1   0.002 12.041 Stabilisation and behavioural 0.377 0.256 to 0.553 0.001 24.746 0.672 0.488 to 0.926 0.015 5.925 Stabilisation and stretching 0.544 0.373 to 0.792 0.001 10.104 0.608 0.429 to 0.861 0.005 7.878 Exercise frequency [week −1 ] 1.13x 1.01 to 1.27 0.039 4.248 1.055 0.946 to 1.177 0.335 0.93 6 mo follow-up Nagelkerkes R2 0.043 0.017 Intercept 0.553

留言 (0)

沒有登入
gif