Empirically derived back pain subgroups differentiated walking performance, pain, and disability

Introduction

Chronic pain is a complex, multifactorial disease that leads to functional limitations and disability [41,60,76]. The largest subset of chronic pain disorders, musculoskeletal pain, is the leading cause of disability worldwide [76]. However, the underlying processes contributing to the development of pain-related disability are not well understood. The Model for Integrating Pain With Movement (Figure 1) is a conceptual model developed by our group that demonstrates how sensory, psychological, and motor domains within the nervous system interact to drive pain-related movement impairments and functional decline [9]. Using this conceptual framework to study multifactorial relationships in musculoskeletal pain conditions may improve our understanding of how disability develops.

Figure 1.

Model for integrating pain with movement. Yellow indicates nervous system processing. Blue indicates the results of adaptation and feedback. Gray shading indicates recovery spectrum. Butera KA, Fox EJ, George SZ. Physical Therapy, 2016, Volume 96, Issue 10, Pages 1503–1507, by permission of the American Physical Therapy Association.

Low back pain (LBP) is the most commonly experienced musculoskeletal pain condition, and evidence already exists supporting sensory (higher pain sensitivity) and psychological (increased distress) contributions to greater self-reported disability [5,10,18,28,35,36,69,80]. Further, prior work shows that individuals with LBP demonstrate changes in muscle activation during functional tasks, particularly walking [2,45,46,72,74]. However, in LBP studies sensory, psychological, and motor domains are often analyzed in isolation as separate systems and are rarely interpreted within the context of performance-based functional outcome measures. Thus, how different combinations of sensory, psychological, and motor domains contribute to movement and functional performance among individuals with LBP is unclear.

Therefore, the purpose of this study was to test the theoretical components of the Model for Integrating Pain With Movement [9] using a LBP sample. The study aims were: 1) to empirically derive LBP subgroups using a priori selected sensory, psychological, and motor factors; and 2) to validate the derived LBP subgroups using performance-based walking measures and self-report measures of pain and disability. Empirical derivation of multifactorial subgroups is responsive to United States pain research initiatives (e.g., Federal Pain Research Strategy [32] and National Pain Strategy [75])..

Methods

Study procedures were approved by the University of Florida Institutional Review Board and all participants provided informed consent prior to enrollment.

Participants

Adults age 18-75 with current or recent LBP were recruited from north central Florida communities via referrals from local health care providers, flyer advertisements in health care offices and local businesses, and social media advertising. Inclusion and exclusion criteria are detailed in Table 1. Eligibility was determined via a phone screen conducted 1-3 weeks prior to the study visit. On the day of the study visit we verified there were no changes in status or symptoms. Eligible participants were enrolled from December 2017 through October 2019.

Table 1.

Eligibility criteria.

Inclusion Criteria	Exclusion Criteria
Current LBP or recent LBP (i.e. an episode of LBP lasting 24 hours or longer within the past 3 months)	Current numbness, tingling, sensory loss, motor loss, or radiating pain below the knee Current knee or ankle joint pain requiring pain medication in past 48 hours
Age ≥18 years and ≤75 years	Widespread chronic pain syndrome diagnosis (e.g. fibromyalgia) or neuropathic pain syndrome diagnosis (e.g. diabetic neuropathy)
English-speaking	Recent trauma (≤12 months) resulting in a fracture to the spine or lower extremities Recent spinal or lower extremity surgery (≤12 months) Previous spinal surgery that included hardware placement Current psychiatric disorder (defined as currently under the care of a psychiatrist and/or taking 2 or more psychiatric medications) Current treatment for active cancer Neurological disorder diagnosis (e.g. stroke, spinal cord injury) Known pregnancy

Eligibility was determined via a phone screen conducted 1-3 weeks prior to the study visit. On the day of the study visit we verified there were no changes in status or symptoms.

General Study Procedures

This cross-sectional study included a single data collection for each participant. Briefly, participants underwent sensory, psychological, and motor factor testing (Aim 1) and completed performance-based walking measures and self-report measures (Aim 2).

Demographic and Historical Characteristics

Participants self-reported age, sex, race, ethnicity, employment status, and level of education. Two questions from the NIH Pain Consortium Task Force on Research Standards for Chronic LBP Minimal Data Set were included to determine presence of chronic LBP (i.e., ≥ 3 months and pain on at least half the days in the past 6 months) [20]. Participants also reported if this was their first episode of LBP and reported use of opioid medication, injections, and/or psychological counseling for LBP in the last 3 months. Comorbidities were assessed using the Functional Comorbidity Index [37]. Participants indicated yes/no for presence of 18 comorbid conditions and a sum score was generated. Higher Functional Comorbidity Index scores are associated with decreased physical function [37].

Procedures for Electromyography and Kinematic Assessments

The following procedures were utilized for recording trunk extensor muscle electromyograms (EMG) during study-related tasks (i.e., forward bending and walking). Bilateral surface EMG data from erector spinae muscles (L4 spine level) were collected at 2000 Hz using a Trigno wireless system (Delsys Inc.; Boston, USA). EMG data were high-pass filtered (30 Hz) using a fourth-order Butterworth filter, demeaned, rectified, and low-pass filtered (5 Hz) using a fourth-order Butterworth filter. Three dimensional kinematic data were recorded at 100 Hz using a 12-camera motion capture system (VICON, Los Angeles, USA) to quantify components of walking tasks (detailed in subsequent sections). A modified version of an established marker set was used to define 9 body segments (head, trunk, pelvis, and each thigh, shank, and foot) [42].

Subgroup Factors

Sensory Measures

High pain sensitivity is often associated with worse clinical outcomes [15,17,23,36,70]. Two dynamic pain sensitivity measures—Conditioned Pain Modulation and Temporal Summation—represented sensory factors in the planned analysis.

Conditioned Pain Modulation

An established Conditioned Pain Modulation testing paradigm [71] was utilized with pressure pain threshold testing [26] at the tibialis anterior via a handheld digital pressure algometer (Wagner Instruments; Greenwich, CT) being the test stimulus and immersion of the hand for 1 minute in cold water (8° C) being the conditioning stimulus. Conditioned Pain Modulation values were generated as a percent change in pressure pain threshold with higher, positive percent Conditioned Pain Modulation values indicating greater pain inhibition.

Temporal Summation

An established testing paradigm [69,71] was adapted to assess Temporal Summation and included delivery five consecutive heat pulses with a peak temperature of 49° Celsius to the dominant forearm using a computerized thermal stimulator (TSA II NeuroSensory Analyzer; Medoc Ltd.; Ramat Yishai, Israel). Specifically, a 3-centimeter by 3-centimeter contact thermode was applied to the dominant forearm continuously while the temperature increased from a baseline temperature of 41° C to a peak temperature of 49° C at a rate of 10° C/second; the peak temperature was maintained for 0.3 seconds, and returned to baseline again at a rate of 10° C/second. Participants rated the pain intensity for each peak pulse using a Numeric Pain Rating Scale (0-no pain to 100-worst pain imaginable) [38]. Temporal Summation was calculated by subtracting the 1^st pulse pain rating from the maximum pain rating obtained from the 2^nd-5^th pulses. Higher Temporal Summation values indicated greater pain facilitation.

Psychological Measures

Both positive and negative psychological factors (e.g., pain catastrophizing, fear-avoidance, optimism) demonstrate established relationships with pain conditions [3,33,34,52,54,61,62,64]. Participants completed the 10-item OSPRO-YF tool, which is a reliable and valid psychological assessment tool for evaluating positive affect/coping, negative coping, and negative mood [10]. Positive Affect/Coping and Negative Coping were selected to represent the psychological factors in the planned subgrouping analysis based on their associations with multiple established clinical measures (e.g., pain intensity, disability, and physical and mental quality of life) [10].

Motor Measures

Compared to healthy individuals, individuals with LBP often demonstrate greater, sustained trunk extensor muscle activation during forward bending [14,50,78] and walking tasks [2,45,46,72,74]. Two established ratio measures represented motor factors in the subgrouping analysis.

Flexion-Relaxation Ratio

The Flexion-Relaxation Ratio [14,50,78] was obtained using an established protocol involving trunk extensor muscle activation testing during forward bending [78]. Participants completed two trials of forward bending while bilateral surface EMG data from erector spinae muscles was recorded. The Flexion-Relaxation Ratio was calculated separately for the left and right erector spinae by dividing the maximum EMG amplitude during forward bending by the midpoint EMG amplitude during a 2-second static hold in full flexion. The mean Flexion-Relaxation Ratio across two trials was obtained; mean left and right Flexion-Relaxation Ratio values were then averaged to obtain a grand mean Flexion-Relaxation Ratio. A lower Flexion-Relaxation Ratio indicated less relaxation during the static hold phase relative to the flexion/forward bending phase (i.e., atypical pattern) [14,50,78].

Swing-Stance Ratio

The Swing-Stance Ratio was obtained using an established protocol for assessing trunk extensor activation during walking [73]. Participants completed at least three trials of walking 10 meters at a self-selected pace while bilateral surface EMG data from erector spinae muscles was recorded. The Swing-Stance Ratio was calculated by dividing the average EMG amplitude during the swing phase by the average EMG amplitude during preceding double stance phase. The mean Swing-Stance Ratio across all trials was obtained; mean left and right Swing-Stance Ratios were then averaged to obtain a grand mean Swing-Stance Ratio. A higher Swing-Stance Ratio indicated less relaxation during the swing phase relative to double stance phase (i.e., atypical pattern) [73].

Subgroup Validation Measures

Performance-Based Walking Measures

Performance-based walking measures were utilized to validate the derived subgroups and enhance their clinical application, as walking is an important functional task and critical to participation in life roles (e.g., work and social roles). Moreover, reduced walking function is a known, modifiable risk factor for increased disability and adverse health outcomes (e.g., hospitalization, falls) [11,12,30,48,57,58,79].

Self-Selected and Fastest-Comfortable Walking Speed

Participants performed three trials of walking 10-12 meters first at a self-selected pace and then at a fastest-comfortable pace. Self-selected and fastest-comfortable walking speeds were measured via Visual 3D software (C-Motion Inc., Germantown, MD, USA) using the previously described 9-segment experimental model; walking speeds were calculated based on kinematic events (i.e., heel strike). Cycles from all walking trials for each condition were combined to obtain average self-selected and fastest-comfortable walking speed.

Timed Up and Go and Timed Up and Go-Cognitive

Participants completed three trials the Timed Up and Go (TUG) and Timed Up and Go-Cognitive (TUG-Cog) and average completion time was calculated. The TUG and TUG-Cog are reliable and valid mobility measures [39,40,51,59].

Obstacle Negotiation

Participants performed three walking trials at a fastest-comfortable pace that included stepping over a 0.2 meter tall block. Obstacle negotiation parameters (obstacle crossing speed; obstacle approach speed; obstacle clearance) were measured via motion capture using methods similar to prior studies [55,56,63]; an average value for each obstacle parameter was obtained.

Self-Report Measures

Brief Pain Inventory (Average Daily Pain Intensity and Perceived Pain Interference)

The Brief Pain Inventory is a valid and commonly used clinical measure of pain intensity and pain interference with mood, activity level, and function [13,43,66]. Average daily pain intensity (from 0 to 10) was calculated using a composite rating including current pain, worst pain, and best pain during the last week. Additionally, sum scores from the pain interference section of the BPI were used to determine perceived pain interference; higher scores indicated higher perceived pain interference.

Oswestry Disability Index (Perceived Disability)

The 10-item Oswestry Disability Index is a valid and commonly used clinical measure of LBP-specific disability [19,24,29,31,49,53]. Higher percent scores indicated greater perceived disability.

Pain Sensitivity Measures

Local and Remote Pressure Pain Thresholds

Two pressure pain threshold testing trials using a handheld pressure algometer (Wagner Instruments; Greenwich, CT) were conducted over the L4 spinous process (local pain sensitivity) and the right tibialis anterior (remote pain sensitivity); average pressure pain threshold values were generated. Lower pressure pain threshold values indicate higher pain sensitivity and are associated with increased risk for poor clinical outcomes [15,16].

Statistical Analysis

All data analyses were conducted using IBP SPSS Statistics for Windows Version 25.0 (IBM Corp, Armonk, NY). Alpha was set at 0.05 to determine significance. Less than 5% of the sample had missing data for at least one measure and listwise deletion was used as appropriate.

Derivation of Subgroups

A hierarchical cluster analysis was conducted to empirically derive subgroups. Cluster analyses have been used previously by our research group to identify and validate subgroups of individuals in pain-related research investigations [6,7]. Clustering variables included the six previously described nervous system factors—Temporal Summation, Conditioned Pain Modulation, Positive Affect/Coping, Negative Coping, Flexion-Relaxation Ratio, and Swing-Stance Ratio. First, these clustering variables were standardized by converting raw values into z-scores. Second, a hierarchical agglomerative cluster analysis was performed using Ward’s method and the above six clustering variables [77]; clusters were formed by measuring the squared Euclidean distances between clustering variables and using this as the criteria for combining similar variables into a cluster. Finally, the optimal cluster solution (number of subgroups and associated factors) was determined through examination of the agglomeration coefficient table and visual inspection of the scree plot [25,44,65]. Clusters, or subgroups, were labeled based on the variables that comprised them. While z-scores were necessary for the cluster analysis, follow-up subgroup comparisons of nervous system factor raw values (i.e., clustering variables) are also reported to aid clinical interpretation of findings. Further, subgroup differences on demographic and historical characteristics were investigated via independent samples t-tests, or One-way ANOVAs with Bonferroni-corrected post-hoc comparisons as appropriate, for continuous variables, as well as chi-square tests for categorical variables. Finally, a discriminant function analysis was used to cross-validate the cluster analysis results; this follow up analysis was also used to determine which variables best distinguished cluster membership.

Subgroup Construct Validity

Subgroup differences were investigated via independent samples t-tests, or One-way ANOVAs with Bonferroni-corrected post-hoc comparisons as appropriate. Performance-based walking, self-report, and pain sensitivity measure values were treated as dependent variables, while group membership was treated as the independent variable.

Results

Sample Demographics and Historical Characteristics

Study enrollment is shown in Figure 2. Two enrolled participants were withdrawn at the time of the study visit due to inability to complete the protocol. Additionally, six participants did not complete Conditioned Pain Modulation testing and were omitted from the dataset, as complete clustering variable data was required for the analysis. Compared to the 70 included participants, the 6 omitted participants demonstrated lower Positive Affect/Coping (Cohen’s d=−1.04; p<0.05), but were otherwise similar on all other clustering variables, demographics, historical characteristics, and validation measures (p’s>0.05).

Figure 2.

Flowchart of study recruitment and enrollment.

The final analyzed study sample (N=70) exceeded the minimum recommended sample size (N=64) to conduct a cluster analysis with six clustering variables [27]. Demographic data for the analyzed sample (N=70) are reported in Table 2. The sample had a mean age of 44.59 years (standard deviation=17.02; range: 20-74 years) and included 36 females (51.4%). The majority of participants were White (77.1%), employed full-time (38.6%), and had finished college (60%). Most participants also reported this was not their first episode of LBP (87.1%). Additionally, 61.4% of the participants met the NIH Pain Consortium Task Force definition for chronic LBP [20].

Table 2.

Demographic and covariate data for the overall sample compared to the Maladaptive and Adaptive subgroups.

	Overall sample (N=70)		Maladaptive subgroup (n=21)		Adaptive subgroup (n=49)
	M or N	Sd or %	M or N	Sd or %	M or N	Sd or %	p-valuea
Age (years)	44.59	17.02	49.29	15.36	42.57	17.44	0.13
Sex
Female	36	51.4	9	42.9	27	55.1	0.35
Male	34	48.6	12	57.1	22	44.9
Race
White	54	77.1	12	57.1	42	87.5	0.02
Black/African American	13	18.6	8	38.1	5	10.4
Asian	1	1.4	1	4.8	0	0.0
American Indian/Alaska Native	1	1.4	0	0.0	1	2.1
Missing	1	1.4	0	0.0	0	0.0
Employment Status
Full-time	27	38.6	6	28.6	21	42.9	0.14
Part-time	13	18.6	4	19.0	9	18.4
Unemployed	7	10.0	5	23.8	2	4.1
Retired	13	18.6	4	19.0	9	18.4
Student	10	14.3	2	9.5	8	16.3
Education Level
Graduated from high school	3	4.3	1	4.8	2	4.1	0.66
Some college	25	35.7	10	47.6	15	30.6
Graduated from college	12	17.1	2	9.5	10	20.4
Some post-graduate work	7	10.0	2	9.5	5	10.2
Completed post-graduate degree	23	32.9	6	28.6	17	34.7
Household Income
Less than $20,000	15	21.4	7	35.0	8	17.0	0.48
$20,001 to $35,000	13	18.6	4	20.0	9	19.1
$35,001 to $50,000	9	12.9	3	15.0	6	12.8
$50,001 to $70,000	5	7.1	1	5.0	4	8.5
Greater than $70,000	25	35.7	5	25.0	20	42.6
Missing	3	4.3
Functional Comorbidity Index scoreb	2.0b	2.0b	2.0b	3.0b	1.5b	2.0b	0.37c
“Is this your first episode of low back pain?”
No	61	87.1	18	85.7	43	87.8	0.82
Yes	9	12.9	3	14.3	6	12.2
Reported chronic low back painc
No	27	38.6	5	23.8	22	44.9	0.10
Yes	43	61.4	16	76.2	27	55.1
Opioid medication for low back pain (prior 3 months)
No	61	87.1	17	81.0	44	89.8	0.31
Yes	9	12.9	4	19.0	5	10.2
Injections for low back pain (prior 3 months)
No	63	90.0	18	85.7	45	91.8	0.43
Yes	7	10.0	3	14.3	4	8.2
Psychological counseling for low back pain (prior 3 months)
No	68	97.1	19	90.5	49	100	0.09
Yes	2	2.9	2	9.5	0	0

M=Mean; N=Number/Frequency; Sd=Standard Deviation; %=percent

^ap-value reported for between subgroup comparisons from independent samples t-tests (age), Fisher’s Exact Test (psychological counseling), Mann-U Whitney test (Functional Comorbidity Index), and Pearson Chi-Square tests (all other variables)

^bas the Functional Comorbidity Index does not have a normal distribution, data are reported as the median and interquartile range rather than mean and standard deviation

^cMet the definition of chronic low back pain as set by the NIH Pain Consortium Task Force (i.e. reported low back pain for at least 3 months that had resulted in pain on at least half the days in the past 6 months) [20].

Derived Subgroups

The a priori selected nervous system factors (i.e., clustering variables) from the sensory, psychological, and motor domains are reported in Table 3.

Table 3.

Nervous system factor data for the overall sample compared to the Maladaptive and Adaptive subgroups.

Nervous System Factor	Overall sample (N=70)		Maladaptive subgroup (n=21)		Adaptive subgroup (n=49)
	Raw Values		Raw Values		Raw Values
	M	Sd	M	Sd	M	Sd	p-valuea
Positive Affect/Coping (score)	17.63	3.50	14.71	3.87	18.88	2.45	<0.001
Negative Coping (score)	7.03	4.11	10.81	4.04	5.41	2.91	<0.001
Temporal Summation (NPRS 0-100)	20.01	17.84	15.67	12.88	21.88	19.41	0.18
Conditioned Pain Modulation (% change)	10.42	32.26	− 2.76	23.61	16.07	33.99	0.02
Flexion-Relaxation Ratio	2.58	2.01	1.56	0.72	3.01	2.23	<0.001
Swing-Stance Ratio	0.59	0.22	0.78	0.23	0.50	0.15	<0.001
	z-score		z-score		z-score
	M	Sd	M	Sd	M	Sd	p-valuea
Positive Affect/Coping	0	1.0	− 0.83	1.11	0.36	0.70	<0.001
Negative Coping	0	1.0	0.92	0.98	− 0.39	0.71	<0.001
Temporal Summation	0	1.0	− 0.24	0.72	0.10	1.09	0.18
Conditioned Pain Modulation	0	1.0	− 0.41	0.73	0.18	1.05	0.02
Flexion-Relaxation Ratio	0	1.0	− 0.50	0.36	0.22	1.11	<0.001
Swing-Stance Ratio	0	1.0	0.90	1.07	− 0.39	0.67	<0.001

M=Mean; Sd=Standard Deviation; NPRS=Numeric Pain Rating Scale; %, percent

^ap-value reported independent samples t-test between subgroup comparison.

Cluster Analysis Results

The agglomeration coefficients from the cluster analysis revealed the largest change between the 1 and 2-cluster stages, indicating that a 2-cluster solution was optimal for this sample (absolute coefficient difference=84.89) [25,44,65]. This was confirmed by visual inspection of the dendrogram and plot of agglomeration coefficients. A 3-cluster solution was considered and determined to have less statistical support; moreover, exploratory analysis indicated little clinical value when isolating a third subgroup. Thus, results for the 2-cluster solution are reported.

Profiles for “Maladaptive” (n=21) and “Adaptive” (n=49) subgroups are reported as z-scores in Figure 3. Independent samples t-tests revealed that the Maladaptive subgroup had lower Positive Affect/Coping, higher Negative Coping, lower Conditioned Pain Modulation, lower Flexion-Relaxation Ratios, and higher Swing-Stance Ratios compared to the Adaptive subgroup (p’s<0.05); Temporal Summation was not different between groups (p>0.05) (Table 3).

Figure 3.

Z-scores for differences in Maladaptive and Adaptive subgroups based on sensory, psychological, and motor factors.

* indicates significant subgroup differences (p’s<0.05); error bars represent standard error of the mean

Discriminant Function Analysis Results

All factors contributed to a 2-cluster solution, except for Temporal Summation (Positive Affect/Coping: F(1,68)=29.39, Wilks’ λ=0.70, p<0.001; Negative Coping: F(1,68)=39.67, Wilks’ λ=0.63, p<0.001; Temporal Summation: F(1,68)=1.80, Wilks’ λ=0.97, p=0.18; Conditioned Pain Modulation: F(1,68)=5.32, Wilks’ λ=0.93, p=0.02; Flexion-Relaxation Ratio: F(1,68)=8.44, Wilks’ λ=0.89, p=0.005; and Swing-Stance Ratio: F(1,68)=36.82, Wilks’ λ=0.65, p<0.001). The overall test was significant (χ² (6)= 83.17; Wilks’ λ =0.28; p<0.001) indicating the set of factors adequately distinguished between the two subgroups. Function 1 accounted for 72% of the relationship between factors and subgroups (Eigenvalue=2.60; Canonical R=0.85) and was able to correctly classify 96% of cross-validated grouped cases.

The standardized canonical discriminant function coefficients, as well as pooled within-group correlations between discriminating variables and standardized canonical discriminant functions are reported in Table 4. Positive Affect/Coping demonstrated a strong positive relationship, while Negative Coping and Swing-Stance Ratios demonstrated strong negative relationships with the Adaptive subgroup. Conditioned Pain Modulation demonstrated a moderate positive relationship and Flexion-Relaxation Ratios demonstrated a weak positive relationship with the Adaptive subgroup. Consistent with the findings from the z-scores and the above discriminant function analysis results, Temporal Summation was not correlated with the Adaptive subgroup.

Table 4.

Coefficients of nervous system factors predicting the 2-cluster discriminant function.

Nervous System Factor	Discriminant Function 1 standardized canonical discriminant function coefficient	Pooled within-groups correlations between discriminating variable and standardized canonical discriminant functions
Positive Affect/Coping	0.51	0.41
Negative Coping	−0.61	−0.47
Temporal Summation	0.31	0.10
Conditioned Pain Modulation	0.35	0.17
Flexion-Relaxation Ratio	0.25	0.22
Swing-Stance Ratio	−0.77	−0.46

Subgroup Differences in Demographic and Historical Characteristics

Demographic and historical characteristics for the subgroups are reported in Table 2. The Maladaptive subgroup demonstrated a lower proportion of individuals identifying as White compared to the Adaptive subgroup (Pearson χ²=10.38, p<0.05), but all other demographic and historical characteristics were similar between subgroups.

Subgroup Construct Validity

Differences in Performance-Based Walking Measures

Walking data for the overall sample and for the Maladaptive and Adaptive subgroups are reported in Table 5. Briefly, compared to the Adaptive subgroup, the Maladaptive subgroup demonstrated slower self-selected walking speed, slower TUG completion, slower obstacle approach speed, and slower obstacle crossing speed; all of these differences showed moderate to large effect sizes (p’s<0.05). In contrast, the Maladaptive and Adaptive subgroups were not different on fastest-comfortable walking speeds, TUG-Cog completion, and obstacle clearance (p’s>0.05).

Table 5.

Walking and clinical data for the overall sample.

	Overall sample (N=70)		Maladaptive subgroup (n=21)		Adaptive subgroup (n=49)
	M	Sd	M	Sd	M	Sd	Cohen’s da	p-valueb
PERFORMANCE-BASED WALKING MEASURES
Self-selected walking speed (m/s)	1.18	0.18	1.10	0.23	1.22	0.15	−0.68	0.048
Fastest-comfortable walking speed (m/s)	1.88	0.33	1.76	0.43	1.94	0.26	−0.56	0.10
Timed Up and Go (s)	6.94	1.73	7.91	2.25	6.52	1.26	0.86	0.01
Timed Up and Go-Cognitive (s)	7.87	2.84	9.25	4.49	7.30	1.51	0.71	0.07
Obstacle clearance (m)	0.18	0.05	0.18	0.05	0.18	0.04	0.07	0.78
Obstacle approach speed (m/s)	1.63	0.33	1.46	0.39	1.69	0.28	−0.73	0.007
Obstacle crossing speed (m/s)	1.47	0.31	1.33	0.34	1.53	0.28	−0.67	0.01
SELF-REPORT MEASURES
Average Daily Pain Intensity (from BPI)	3.16	2.04	4.10	2.44	2.76	1.71	0.69	0.01
Perceived Pain Interference (from BPI)	2.75	2.32	4.29	2.72	2.09	1.78	1.05	0.002
Perceived Disability (% score from ODI)	18.74	14.33	28.95	18.42	14.37	9.41	1.14	0.002
PAIN SENSITIVITY MEASURES
Local Pressure Pain Threshold (Kg)	3.46	2.53	3.63	2.97	3.39	2.34	0.09	0.72
Remote Pressure Pain Threshold (Kg)	3.99	2.11	4.68	2.41	3.69	1.91	0.48	0.07

M=Mean; Sd=Standard Deviation; m/s=meters/second; s=seconds m=meters; BPI=Brief Pain Inventory; ODI=Oswestry Disability Index; Kg=kilograms

^aeffect size between subgroups

^bindependent samples t-test between subgroup comparison.

Differences in Self-Report Measures

Self-reported pain and disability data for the overall sample and for the Maladaptive and Adaptive subgroups are reported in Table 5, respectively. Briefly, compared to the Adaptive subgroup, the Maladaptive subgroup exhibited higher average daily pain intensity, higher perceived pain interference, and higher perceived disability; all of these differences showed moderate to large effect sizes (p’s<0.05).

Differences in Pain Sensitivity Measures

Local and remote pressure pain thresholds are reported in Table 5. There were no differences in comparisons between Maladaptive and Adaptive subgroups (p’s>0.05).

Discussion

The Model for Integrating Pain With Movement (Figure 1) proposes that interactions between sensory, psychological, and motor domains contribute to movement impairments and subsequent disability among individuals with pain [9]. The current study involving participants with LBP supports several aspects of this theoretical framework. First, two distinct subgroups were identified from a priori selected sensory, psychological, and motor factors. Five out of the six factors contributed statistically to subgroup classification. Second, the derived subgroups demonstrated substantial differences (moderate to large effect sizes) on seven out of 12 validation measures. Convergence of subgroup differences across several, diverse performance-based walking measures and self-reported pain and disability measures supports construct validity of the derived LBP subgroups.

The identified LBP subgroups were derived independent of common pain-related demographic or historical characteristics (e.g., age, sex, or pain duration). An innovative aspect of the current study was consideration of motor factors and a positive psychological factor as subgrouping variables. These were two of the strongest classification variables in deriving subgroups and provide novel evidence endorsing inclusion of motor and positive psychological measures in clinical pain studies and future updates to pain phenotyping guidelines. Additionally, there were subgroup differences for validation measures commonly used as outcomes in cohort studies or clinical trials. While other subgrouping approaches have been validated for use among individuals with LBP, the current subgrouping method may be poised to guide comprehensive assessment of underlying LBP-related sensory, psychological, and motor mechanisms related to worse functional performance and clinical presentation. Future longitudinal studies are needed to determine if the current subgrouping method is also predictive of increased risk for chronicity or persistent disability.

Prior work has established that individuals with LBP exhibit changes in walking function, including slowed walking speed, compared to healthy controls [1,47,74]. In the current study, the mean self-selected walking speeds for the overall sample and both subgroups were slower than expected based on age-matched normative values [8]. Moreover, the self-selected speed of the Maladaptive subgroup (1.1 meters/second) is notable because walking speeds less than 1.0 meters/second are associated with risk for long-term disability, mobility dependence, and adverse health outcomes, such as falls and hospitalization [11,12,30,48,57,58,79]. Thus, future work is warranted to establish walking speed risk values associated with adverse health outcomes in LBP populations.

Interestingly, however, the similar fastest-comfortable walking speeds between subgroups suggests that individuals in the Maladaptive subgroup have the capacity to walk as fast as those in the Adaptive subgroup when prompted or motivated by an external cue (i.e., verbal instruction). This finding reveals that slow walking speed is modifiable for individuals with LBP. Additionally, prior work indicates that individuals recently recovered from LBP have the capacity to increase their speed of movement, but those with higher pain-related fear exhibit altered motor control strategies to achieve the task at a faster pace [67]. Therefore, future work should consider how the derived subgroups’ characteristics may impact walking-related motor control strategies.

The Maladaptive subgroup also demonstrated some difficulty with complex walking tasks, as evidenced by slower TUG completion and slower obstacle crossing and approach speeds. In contrast, there were no subgroup differences on obstacle clearance, nor TUG-Cog completion. These findings underscore that variable task demands may be differentially impacted by the derived subgroups’ characteristics. Complex walking tasks have shared elements that make them more challenging than self-selected or fast walking (e.g., postural transitions, environmental demands), however, they also require different cognitive, attentional, and/or motor resources to execute [4,21,22,68]. Thus, the Maladaptive subgroup may have had difficulty with the physical-dual task (i.e., obstacle negotiation), while their cognitive-dual task (i.e., TUG-Cog) performance was not impaired. Additionally, the size of the obstacle may have had a limited impact on the Maladaptive subgroup’s performance (i.e., slower speed, but not an exaggerated step). Further research is required to establish whether variable task demands are associated with LBP-related performance outcomes.

Limitations

First, this study reports cross-sectional associations and there is a need to advance this work via longitudinal investigations to determine temporal relationships between the subgroups and clinical outcomes. Second, subgroups were derived using measures from sensory, psychological, and motor domains. While the a priori selected factors were selected strategically to represent key domains within the nervous system as guided by our conceptual model, we acknowledge that additional factors not included in this study also contribute to the LBP experience (e.g., social, cognitive, and genetic factors). Additionally, the wide age range could be considered a limitation; thus, the influence of age should be explored in future subgroup investigations, as it could impact the functional performance measures. This study also focused solely on LBP and therefore results may not be generalizable to other musculoskeletal pain conditions. Finally, this study considered the relationship between LBP subgroups and walking performance measures and findings may not be generalizable beyond walking activities. Because there are many postures and functional activities that may improve or exacerbate LBP, inclusion of diverse and comprehensive measures of function is needed to establish whether there is a relationship between subgroups and other specific functional tasks and/or general functional performance.

Conclusions

The current study described a novel approach that resulted in Maladaptive and Adaptive LBP subgroups. The Maladaptive subgroup was characterized by lower positive affect/coping, higher negative coping, lower pain modulation, and atypical trunk extensor activation during functional tasks compared to the Adaptive subgroup. Construct validity of these subgroups was supported by differences across several measures of walking performance, pain, and disability, with the Maladaptive subgroup demonstrating reduced walking performance and higher pain and disability. Longitudinal studies are needed to determine if poorer performance and clinical outcomes are associated with the Maladaptive subgroup. Finally, the Maladaptive subgroup demonstrated a preferred walking speed associated with potentially greater risk for persistent disability and adverse health outcomes. However, they also demonstrated a capacity to increase their speed when prompted. Thus, future longitudinal work is needed to determine whether preferred walking speed is a potential, and modifiable, risk factor for individuals with LBP.