Determining One-Year Trajectories of Low-Back–Related Leg Pain in Primary Care Patients: Growth Mixture Modeling of a Prospective Cohort Study
Abstract
Objective
The clinical presentation and outcome of patients with back and leg pain in primary care are heterogeneous and may be better understood by identification of homogeneous and clinically meaningful subgroups. Subgroups of patients with different back pain trajectories have been identified, but little is known about the trajectories for patients with back-related leg pain. This study sought to identify distinct leg pain trajectories, and baseline characteristics associated with membership of each group, in primary care patients.
Methods
Monthly data on leg pain intensity were collected over 12 months for 609 patients participating in a prospective cohort study of adult patients seeking health care for low-back and leg pain, including sciatica, of any duration and severity, from their general practitioner. Growth mixture modeling was used to identify clusters of patients with distinct leg pain trajectories. Trajectories were characterized using baseline demographic and clinical examination data. Multinomial logistic regression was used to predict latent class membership, with a range of covariates.
Results
Four patient clusters were identified: improving mild pain (58%), persistent moderate pain (26%), persistent severe pain (13%), and improving severe pain (3%). Clusters showed statistically significant differences in a number of baseline characteristics.
Conclusion
Four trajectories of leg pain were identified. Clusters 1, 2, and 3 were generally comparable to back pain trajectories, while cluster 4, with major improvement in pain, is infrequently identified. Awareness of such distinct patient groups improves understanding of the course of leg pain and may provide a basis of classification for intervention.
Introduction
Significance & Innovations
- In primary care patients with low-back–related leg pain, using growth mixture modeling, we identified 4 distinct trajectories: improving mild, persistent moderate, persistent severe, and improving severe leg pain, with the majority of patients on average remaining in stable patterns.
- Three of the trajectories, improving mild, persistent moderate, and persistent severe leg pain, are generally comparable to back pain trajectories. The improving severe cluster represented a group with severe leg pain, whose symptoms improved over time. This group is less often identified in back pain patients.
- The identification of trajectory patterns of leg pain in patients who have low-back–related leg pain in primary care may potentially improve understanding of the course of leg pain and guide interventions.
- For the majority of this patient group, considering mainly conservative management options might be justifiable, such as medication and physiotherapy input. However, for those patients who have very severe pain that does not improve in the first few weeks, perhaps more invasive management options should be considered earlier in the course of pain, if these options are appropriate and desirable.
Studies on the clinical course of most musculoskeletal pain conditions 6-9 have mainly shown a marked improvement in pain within the first few weeks, but after that point improvement slows considerably. These findings are based on single growth trajectories, with the assumption that individuals are drawn from a single homogeneous population with common population parameters. However, the moderate-to-high person-to-person variability in pain at follow-up time points reported in these studies 7 clearly points to the heterogeneity in the clinical course of pain. This heterogeneity has led to a number of studies in the past decade focusing on investigation of the underlying averaged course of low-back pain and has demonstrated that different trajectory patterns exist 10, 11.
Despite this growing body of research focused on identifying distinct trajectory patterns of low-back pain over time, little is known about the temporal evolution of leg pain intensity for patients with back and leg pain. Identification of homogeneous and clinically meaningful subgroups of low-back–related leg pain patients would be important, as such pain trajectory better reflects individuals’ course patterns and may provide a basis of classification for intervention. The aim of this study was to identify distinct leg pain trajectory groups in primary care patients consulting with low-back–related leg pain, and to identify baseline patient characteristics associated with membership of each trajectory group.
Materials and Methods
Data source
This study used data from a prospective cohort study (Assessment and Treatment of Leg Pain Associated With the Spine [ATLAS]) of 609 patients ages ≥18 years, visiting their family doctor (general practitioner) with symptoms of low-back–related leg pain (including sciatica), of any severity and duration, at general practitioner practices in North Staffordshire and Stoke-on-Trent, UK. Details of the protocol and results have been published elsewhere 12-14. Briefly, potentially eligible patients were sent a letter with information about the study, an invitation to attend the initial research clinic, and baseline questionnaires capturing sociodemographic, pain, psychological, and health variables. At the research clinic, patients underwent a standardized clinical assessment by 1 of 7 musculoskeletal physiotherapists and were diagnosed as having sciatica (spinal nerve root involvement) or referred (nonspecific) leg pain, based on the examiner's clinical opinion. Providing that there were no clinical contraindications to the procedure, patients had a lumbar spine magnetic resonance imaging (MRI) scan within 2 weeks of their baseline assessment. As part of the study, monthly data for leg pain intensity were collected over 12 months, using brief postal questionnaires. Leg pain intensity was measured using the mean of three 0–10 numerical rating scale measurements for least, usual, and current leg pain over the previous 2 weeks 15. Most participants received physiotherapy treatments, and a small number were referred for a specialist's opinion and management. The ATLAS study care pathways are described in detail elsewhere 14.
Information on access to the individual patient data from studies hosted by the Arthritis Research UK Primary Care Centre is available at http://www.keele.ac.uk/pchs/publications/datasharingresources/. Ethical Approval for this study was obtained from the South Birmingham Research Ethics Committee (10/H1207/82).
Baseline patient characteristics
There are no known baseline factors associated with leg pain trajectory class membership. Therefore, based on previous research in other musculoskeletal pain conditions, a number of patient baseline sociodemographic, pain, psychological, and health variables were selected to describe the characteristics of participants in each of the trajectory groups. These included age, sex, employment status, currently smoking, body mass index, sleep disturbances due to back and/or leg pain, a clinical diagnosis of sciatica (made by clinician without knowledge of MRI findings), disability measured with the Roland Morris Disability Questionnaire leg pain version 16, 17, neuropathic pain measured using the Self-Administered Leeds Assessment of Neuropathic Symptoms and Signs 18, Sciatica Bothersomeness Index (SBI) composite score (range 0–24) 16, leg pain duration, anxiety, and depression measured using the Hospital Anxiety and Depression Scale (HADS) 19, whether pain extended below the knee, whether leg pain was worse than back pain, evidence of nerve root compression on MRI, and whether a patient was referred to secondary care for a spinal specialist opinion. Supplementary Table 1, available on the Arthritis Care & Research web site at http://bibliotheek.ehb.be:2356/doi/10.1002/acr.23556/abstract, shows these variables.
Statistical analysis
To identify possible homogeneous and clinically meaningful trajectory groups based on the observed longitudinal trend of pain over time, we applied growth mixture models (GMM) 20-22. GMM is a statistical approach that captures patients’ heterogeneity (individual differences in pain intensity over time) in terms of the growth intercept (individual differences in pain at the beginning of the study) and growth slope (individual differences with respect to their pain profile over time), by classifying individuals into unobserved groupings with more homogenous patterns, called latent trajectory classes, with each subject belonging exclusively to 1 latent class. We fitted a random-effects model, which allows for within-class variability, as opposed to assuming that all individual growth trajectories within classes are homogeneous.
To decide on the optimal number of classes, we fitted several sets of models successively (from 2-class to 6-class solution) and compared their fit by considering: a Bayesian information criterion (BIC) statistic, with a low BIC value indicating a well-fitting model; bootstrapped parametric likelihood ratio test, which compares the model with K classes to a model with K-1 classes; classification quality determined by the posterior probabilities, ensuring that the average of the posterior probabilities of group membership for individuals assigned to each group exceeds a minimum threshold of 0.7 23, 24; face validity of the clusters in terms of their clinical interpretability; and class size (the number of individuals in each class) 25. Baseline characteristics of the identified latent trajectory classes were described. Longitudinal plots of the raw individual-level leg pain data were shown as well as the overall trajectory smoothed mean curve, estimated using loess regression. Multinomial logistic regression models were used to determine the baseline factors independently associated with the latent trajectory class membership. The univariable association between each baseline characteristic and trajectory group was estimated, and those with P values less than 0.25 were selected for inclusion in the multivariable models. Tests of multicollinearity were performed between the predictors. Manual backward elimination was performed using likelihood ratio tests and the BIC statistic, to remove nonsignificant variables from the multivariable model, until only predictors with P values less than 0.05 were retained in the final model. Using the same modeling process, we performed a subgroup analysis comparing baseline characteristics between those assigned to the improving severe and persistent severe trajectory.
Latent class analyses were carried out by maximum likelihood estimation, using R 26 and MPlus 27 software. Subsequent analyses were carried out using Stata 14 software 28. The maximum likelihood estimation makes use of all available data points, so missing values are handled without need for imputation, assuming that missing data are missing at random, meaning that given the observed outcomes and covariates, missingness does not depend on unobserved outcomes. As sensitivity analysis, we repeated the analyses to determine the optimal number of latent trajectory classes by analyzing only subjects with complete follow-up data, and also by relaxing the assumption of within-class normality using a skew-T growth mixture model.
Results
Participants and monthly response rate
At baseline, 609 participants (mean ± SD age 50 ± 13.9 years, 63% women) were included in the study and completed the baseline questionnaire and clinical assessment. Characteristics of these participants have previously been reported 13. As described, responders and nonresponders to follow-up questionnaires showed reasonable comparability in key baseline characteristics (age, sex, and area-level deprivation). On average, leg pain intensity for the whole sample reduced over the first 3 months and thereafter remained almost unchanged (Figure 1). Monthly response rates ranged from 46% (282 of 609) at month 5 to 75% (455 of 609) at month 1, with month 12 having a 74% (450 of 609) response rate. Twenty-nine percent of participants (n = 176) had complete data for leg pain at all follow-up time points, while 61 participants (10%) did not provide any follow-up data. There were no systematic differences in follow-up rates across the clusters.
Trajectories of low-back–related leg pain
The individual-level patient leg pain profile (trajectories) over the 12 months revealed a heterogeneous population with a wide range of patterns in the clinical course of back and leg pain for individuals (Figure 1). The BIC statistics indicated that the 4-class model was the best fitting solution (see Supplementary Table 2, available on the Arthritis Care & Research web site at http://bibliotheek.ehb.be:2356/doi/10.1002/acr.23556/abstract). The bootstrapped parametric likelihood ratio test for 3 classes versus 4 classes also showed that 4 classes had a better fit than the 3 classes (P < 0.001). The 4-model solution also reflected good clinical interpretability and was chosen as the final model. The average posterior probability for each class ranged from 72% to 85% (see Supplementary Table 3, available on the Arthritis Care & Research web site at http://bibliotheek.ehb.be:2356/doi/10.1002/acr.23556/abstract), showing acceptable precision of classification of individuals into classes. Figure 2 shows the mean trajectories obtained from the 4-class model, along with Supplementary Figure 1, available at http://bibliotheek.ehb.be:2356/doi/10.1002/acr.23556/abstract, adding 95% confidence bounds. Similar results were obtained when normality assumptions were relaxed. Figure 2 shows 4 distinct trajectories that differ from each other in their mean levels and changes in pain.
Detailed individual-level raw data observed for each trajectory group (Figure 3) show that the groups identified are clearly different, but also that there are fluctuations around the mean within the groups. Based on the growth patterns (Figures 2 and 3), the largest trajectory class (cluster 1, n = 352 of 609 patients; 58%) was labeled improving mild pain. Members of this class began with mild-to-moderate leg pain averaging 4.2 at baseline that reduced gradually with time to no pain and had a total amount of growth across the entire time interval of −0.23 (time-averaged slope: P < 0.001). Cluster 2 contained approximately one-fourth of the sample (n = 161; 26%) and was named persistent moderate pain. Members of this class began with an average leg pain of 5.6 at baseline, with a total amount of growth across the entire time interval of −0.03 (slope: P = 0.23), indicating little change in leg pain intensity. Cluster 3 (n = 79; 13%) was named persistent severe pain. Members of this class began with an average leg pain of 8.1, had a total amount of nonsignificant growth across the entire time interval of −0.01 (slope: P = 0.65), i.e., almost no change over time; this group still had severe leg pain averaging 7.2 by 12 months. Cluster 4 (n = 17; 3%) was named improving severe pain. Members of this class began with an average leg pain of 8.4, which remained high for up to approximately 4 months and afterwards started to reduce, with a significant negative growth across the entire 12 months of follow-up time of −0.56 (slope: P < 0.001). The sensitivity analysis based on a subgroup of participants with complete leg pain data at all time points showed similar cluster structures (see Supplementary Figures 2 and 3, available on the Arthritis Care & Research web site at http://bibliotheek.ehb.be:2356/doi/10.1002/acr.23556/abstract), with n = 102, 48, 21, and 5 for clusters 1, 2, 3, and 4, respectively.
The characteristics of the latent trajectory groups
The baseline characteristics of the latent trajectory groups are shown in Table 1. Both the persistent severe and improving severe leg pain groups had higher scores for anxiety, depression, disability, and sciatica bothersomeness than the improving mild and persistent moderate groups. The proportion of patients clinically diagnosed with sciatica was highest in the persistent severe group (94%), followed by improving severe (85%) and persistent moderate (74%), and least among the improving mild group (71%). The persistent severe group participants were characterized by the highest level of possible neuropathic pain (73%). The improving severe group of participants was characterized by the highest proportion of women, self-reported sleep disturbance due to back and/or leg pain, sciatica clinical diagnosis, leg pain being worse than back pain, reporting having pins and needles and/or numbness, evidence of nerve root compression on MRI, referrals for spinal specialist opinion, and all having pain below the knee.
| Characteristics | Improving mild (n = 352, 58%) | Persistent moderate (n = 161, 26%) | Persistent severe (n = 79, 13%) | Improving severe (n = 17, 3%) |
|---|---|---|---|---|
| Age, mean ± SD years | 49.0 ± 13.7 | 51.4 ± 13.7 | 51.4 ± 14.8 | 56.9 ± 12.8 |
| Women | 213 (60.5) | 110 (68.3) | 48 (60.8) | 12 (70.6) |
| Body mass index, kg/m2 | ||||
| Normal (18.5 to <25) | 79 (22.5) | 36 (22.4) | 15 (19.2) | 6 (35.3) |
| Overweight (25 to <30) | 129 (36.8) | 61 (37.9) | 28 (35.9) | 5 (29.4) |
| Obese/morbidly obese (≥30) | 143 (40.7) | 64 (40.0) | 35 (44.9) | 6 (35.3) |
| Current smoker | 95 (27.0) | 58 (36.0) | 35 (44.3) | 7 (41.2) |
| Currently in paid job | 245 (70.0) | 88 (55.0) | 27 (34.6) | 7 (41.2) |
| Sleep disturbances due to back and/or leg pain | 228 (64.8) | 127 (78.9) | 58 (73.4) | 15 (88.2) |
| Comorbidities, ≥1 other health problemb | 123 (34.9) | 65 (40.4) | 41 (51.9) | 9 (52.9) |
| RMDQ disability score (range 0–23), mean ± SD | 11.4 ± 5.4 | 13.3 ± 5.6 | 16.4 ± 5.5 | 15.4 ± 4.8 |
| Sciatica clinical diagnosis | 250 (71.0) | 119 (73.9) | 67 (84.8) | 16 (94.1) |
| Sciatica Bothersomeness Index, mean ± SD | 12.4 ± 5.0 | 15.1 ± 4.6 | 19.3 ± 3.9 | 19.6 ± 3.3 |
| Leg pain duration | ||||
| <6 weeks | 167 (49.4) | 60 (39.7) | 21 (27.3) | 3 (17.7) |
| 6–12 weeks | 77 (22.8) | 25 (16.6) | 11 (14.3) | 7 (41.2) |
| >3 months | 94 (27.8) | 66 (43.7) | 45 (58.4) | 7 (41.2) |
| S-LANSS, possible neuropathic pain | 139 (39.6) | 89 (55.6) | 57 (73.1) | 8 (47.1) |
| Pain below the knee | 210 (62.1) | 105 (69.1) | 62 (80.5) | 17 (100.0) |
| Leg pain worse than back pain by patient report | 145 (41.2) | 79 (49.1) | 43 (54.4) | 13 (76.5) |
| HADS depression subscale score, mean ± SD | 5.7 ± 3.6 | 6.3 ± 3.9 | 8.9 ± 4.7 | 8.2 ± 3.7 |
| HADS anxiety subscale score, mean ± SD | 6.8 ± 3.8 | 8.3 ± 4.0 | 10.5 ± 4.5 | 9.6 ± 4.0 |
| Pins and needles and/or numbness by patient report | 205 (58.2) | 103 (64.0) | 60 (76.0) | 14 (82.4) |
| Mild or severe muscle weakness | 62 (17.6) | 26 (16.2) | 13 (16.7) | 4 (23.5) |
| Reduced or loss of pin-prick sensation | 135 (38.4) | 70 (43.5) | 39 (49.4) | 9 (52.9) |
| Patient referred to secondary care | 22 (6.3) | 26 (16.2) | 14 (17.7) | 8 (47.1) |
| Had surgery for back or leg pain over 12 months | 3 (0.9) | 2 (1.2) | 4 (5.1) | 5 (29.4) |
| MRI evidence of nerve root compression | 161 (50.8) | 78 (53.8) | 44 (57.9) | 14 (87.5) |
- a Values are the number (%) unless indicated otherwise (total n = 609). GMM = growth mixture model; RMDQ = Roland Morris Disability Questionnaire; S-LANSS = Self-Administered Leeds Assessment of Neuropathic Symptoms and Signs; HADS = Hospital Anxiety and Depression Scale; MRI = magnetic resonance imaging.
- b The health problems included chest problems, heart problems, raised blood pressure, diabetes mellitus, and circulation problems in the leg.
Relationships between baseline patient characteristics and the latent trajectory groups
The multinomial logistic regression model results comparing the baseline variables of interest among the latent trajectory groups, with the improving mild group (cluster 1) as the reference, are shown in Table 2. The table shows the risk of belonging to each cluster for a given characteristic compared to the reference cluster, expressed as a relative risk ratio. The final multivariable model included baseline measures of being employed full-time, SBI, leg pain duration, leg pain being worse than back pain, anxiety, and referral to a spine specialist for opinion. Controlling for other variables in the model, we found that patients with longer periods of leg pain, higher anxiety scores, and referral for a specialist opinion were more likely to be in the persistent moderate class than improving mild class. Patients were significantly more likely to be in the persistent severe class relative to improving mild if they were not in full-time jobs, had higher SBI scores, had longer pain duration, with leg pain worse than back pain, and had higher anxiety scores. Patients were more likely to be in the improving severe class relative to the improving mild class if they were in full-time jobs, had higher SBI scores, had leg pain worse than back pain, and were referred for a spinal specialist opinion.
| Baseline characteristicsc | Univariable, unadjusted | Multivariable, adjustedb | ||||
|---|---|---|---|---|---|---|
| Persistent moderate (n = 161) | Persistent severe (n = 79) | Improving severe (n = 17) | Persistent moderate (n = 161) | Persistent severe (n = 79) | Improving severe (n = 17) | |
| Age, years | 1.01 (0.99–1.03) | 1.01 (0.99–1.03) | 1.04 (1.00–1.08) | – | – | – |
| Women (men) | 1.41 (0.95–2.09) | 1.01 (0.61–1.67) | 1.57 (0.54–4.54) | – | – | – |
| BMI, kg/m2 (normal 18.5 to <25) | ||||||
| Overweight (25 to <30) | 1.04 (0.63–1.71) | 1.14 (0.57–2.27) | 0.51 (0.15–1.73) | – | – | – |
| Obese/morbidly obese (≥30) | 0.98 (0.60–1.61) | 1.29 (0.66–2.50) | 0.55 (0.17–1.77) | – | – | – |
| Current smoker (nonsmoker) | 1.52 (1.02–2.27) | 2.15 (1.30–3.56) | 1.89 (0.70–5.12) | – | – | – |
| Currently in paid job (not currently in paid job) | 0.52 (0.36–0.77) | 0.23 (0.13–0.38) | 0.30 (0.11–0.81) | 0.50 (0.32–0.76) | 0.24 (0.13–0.46) | 0.26 (0.09–0.79) |
| Sleep disturbances due to back and/or leg pain (no disturbance) | 2.03 (1.31–3.14) | 1.50 (0.87–2.59) | 4.08 (0.92–18.12) | – | – | – |
| Comorbidities: ≥1 other health problem (none) | 1.26 (0.86–1.85) | 2.01 (1.23–3.29) | 2.09 (0.79–5.57) | – | – | – |
| RMDQ disability score (range 0–23) | 1.06 (1.03–1.10) | 1.20 (1.14–1.26) | 1.15 (1.04–1.26) | – | – | – |
| Sciatica clinical diagnosis (referred leg pain) | 1.16 (0.76–1.76) | 2.28 (1.18–4.39) | 6.52 (0.85–49.8) | – | – | – |
| Sciatica Bothersomeness Index | 1.12 (1.07–1.16) | 1.41 (1.31–1.52) | 1.43 (1.24–1.66) | 1.10 (1.05–1.15) | 1.36 (1.25–1.47) | 1.35 (1.16–1.57) |
| Leg pain duration (<6 weeks) | ||||||
| 6–12 weeks | 0.90 (0.53–1.55) | 1.14 (0.52–2.47) | 5.06 (1.27–20.10) | 0.84 (0.47–1.47) | 1.22 (0.50–3.00) | 3.61 (0.80–16.33) |
| >3 months | 1.95 (1.27–3.01) | 3.81 (2.13–6.77) | 4.14 (1.04–16.41) | 1.62 (1.02–4.57) | 2.68 (1.32–5.42) | 2.56 (0.58–11.34) |
| S-LANSS, possible neuropathic pain (no) | 1.91 (1.31–2.79) | 4.13 (2.40–7.13) | 1.36 (0.51–3.60) | – | – | – |
| Pain below the knee (pain above the knee) | 1.36 (0.91–2.05) | 2.52 (1.38–4.61) | Perfect prediction | – | – | – |
| Leg pain is worse than back pain (back pain worse) | 1.38 (0.95–2.00) | 1.71 (1.04–2.79) | 4.64 (1.48–14.51) | 1.47 (0.96–2.24) | 1.99 (1.05–3.76) | 3.64 (1.04–12.81) |
| HADS depression subscale score | 1.04 (0.99–1.09) | 1.21 (1.14–1.28) | 1.16 (1.03–1.31) | – | – | – |
| HADS anxiety subscale score | 1.10 (1.04–1.15) | 1.25 (1.17–1.32) | 1.18 (1.05–1.33) | 1.06 (1.01–1.12) | 1.14 (1.05–1.23) | 1.11 (0.97–1.28) |
| Pins and needles and/or numbness by patient report (none) | 1.27 (0.87–1.87) | 2.26 (1.30–3.96) | 3.34 (0.94–11.85) | – | – | – |
| Mild or severe muscle weakness (normal) | 0.90 (0.54–1.49) | 0.94 (0.49–1.80) | 1.44 (0.45–4.56) | – | – | – |
| Reduced or loss of pin-prick sensation (none) | 1.24 (0.85–1.81) | 1.57 (0.96–2.55) | 1.80 (0.68–4.80) | – | – | – |
| Patient referred to secondary care (no) | 2.89 (1.58–5.28) | 3.23 (1.57–6.64) | 13.33 (4.69–37.93) | 2.05 (1.06–3.94) | 1.42 (0.58–3.50) | 5.40 (1.65–17.65) |
| Evidence of nerve root compression on MRI (none) | 1.12 (0.76–1.67) | 1.33 (0.80–2.21) | 6.78 (1.52–30.33) | – | – | – |
- a BMI = body mass index; RMDQ = Roland Morris Disability Questionnaire; S-LANSS = Self-Administered Leeds Assessment of Neuropathic Symptoms and Signs; HADS = Hospital Anxiety and Depression Scale; MRI = magnetic resonance imaging.
- b All variables in the univariable model except BMI and muscle weakness were significant (P < 0.25) and were included in the initial multivariable model. For the multivariable model, depression and anxiety were highly correlated and only anxiety was left in the model, as it had stronger univariable association with class membership.
- c The reference categories for all categorical variables are shown in parentheses.
Differentiation of the improving severe from the persistent severe groups at baseline
Table 3 shows a comparison of the baseline characteristics between those assigned to the improving severe and persistent severe trajectory groups for only significant predictors. Participants in the improving severe class were significantly more likely to have evidence of nerve root compression on MRI and be referred for a spinal specialist opinion than those in the persistent severe class, but were less likely to have neuropathic pain.
| Baseline characteristics (reference) | Unadjusted OR (95% CI)b | Adjusted OR (95% CI) |
|---|---|---|
| S-LANSS, possible neuropathic pain (no) | 0.34 (0.11–0.96) | 0.27 (0.08–0.87) |
| Referred for specialist opinion (no) | 4.13 (1.35–12.57) | 5.28 (1.59–17.47) |
| Evidence of nerve root compression on MRI (none) | 5.09 (1.08–23.98) | – |
- a Pain extending below the knee was a perfect predictor for being in the improving severe class, since all members of that cluster had pain extending below the knee. 95% CI = 95% confidence interval; S-LANSS = Self-Administered Leeds Assessment of Neuropathic Symptoms and Signs; MRI = magnetic resonance imaging.
- b All the baseline variables considered in Table 2 were examined for the univariable association but were all nonsignificant (P > 0.25), and thus are excluded from this table.
Discussion
We identified 4 distinct trajectories of leg pain over 12 months. To our knowledge, this is the first study demonstrating trajectories of leg pain. The first cluster with more than half of the participants, which we labeled improving mild leg pain, comprised patients who, on average, had mild-to-moderate leg pain at baseline that gradually improved over the 12-month follow-up. The second cluster, labeled persistent moderate leg pain, comprised patients with moderate leg pain at baseline that persisted throughout the 12 months. The third cluster, labeled persistent severe leg pain, consisted of patients whose leg pain was consistently severe over the year. The final cluster, labeled improving severe, though with few patients, had a very distinctive feature because the patients had very severe leg pain at baseline followed by slow recovery for up to approximately 4 months, then rapid recovery to almost no pain by 12 months.
The 4 trajectory groups differed significantly regarding specific patient sociodemographic, pain, psychological, and clinical characteristics (obtained from clinical examination data). Patients who had severe leg pain at baseline (clusters 3 and 4) had on average higher scores on anxiety, depression, disability, and sciatica bothersomeness, and were more likely to have a sciatica diagnosis than patients who had moderate-to-mild leg pain. In our final multivariable model examining the predictors of trajectory group membership, the baseline variables that significantly differentiated the other trajectory groups from the mild improving group included being employed full-time, SBI, leg pain duration, leg pain being worse than back pain, anxiety, and whether participants were referred for a spinal specialist opinion.
Since the first article reporting statistically derived trajectories in back pain was published in 2006 29, several studies have investigated trajectories of back pain and other musculoskeletal pain conditions but, to our knowledge, no study has investigated trajectories specifically in patients with low-back–related leg pain. A recent overview of previous studies of low-back pain trajectories, from 10 cohorts over the past decade 11 found that most cohorts demonstrated 4 or 5 patterns as the optimal number of trajectory patterns, with persistent mild, recovering mild, fluctuating, and severe chronic pain patterns as the common trajectory patterns. An overview of low-back pain studies also found that most people who experience low-back pain will have trajectories of either persistent or episodic pain rather than a single well-defined episode 30. Similar features, common between our study and those previous low-back pain cohort studies, include the improving mild, persistent moderate, and persistent severe trajectory patterns. Despite such similarity in patterns with the previous low-back pain studies, the proportion of patients in each trajectory differs significantly with our study. For example, the proportion of recoverers (improving mild) in our study (58%) was much higher than in most of the low-back pain studies (ranging from 7% to 54%) 11, or studies in other musculoskeletal pain conditions such as knee (12%) 31 and hip osteoarthritis (17%) 32.
Despite many low-back pain studies identifying trajectory groups of episodic/fluctuating patterns comprising between 15% and 34% of the sample 3, 29, 33, our study of low-back–related leg pain patients did not show a trajectory predominantly representing such a group of patients. However, since the identified trajectories allow for individual variations within trajectories, as shown in Figure 3, fluctuations are likely to be superimposed on these underlying trajectories, but are not the predominant patterns. Because we used data spanning only 12 months, how the patterns we identified may develop over a longer follow-up is not known, so we cannot determine whether the recoveries observed in 2 of the trajectory groups are definite recoveries with no future recurrences. However, a study that investigated the stability of low-back pain trajectories over time by following the same cohort over two 6-month periods that were 7 years apart 34 found that the majority of patients with back pain remain in a particular low-back pain trajectory over long periods of time.
Noteworthy in our study is the improving severe cluster, which represented a group with severe leg pain on average, whose symptoms improved over time. This cluster, however, contained only 17 participants, and thus should be interpreted with caution until the results are replicated in other studies. This group is less often identified in patients who have back pain of longer-term follow-up. However, studies on short-term follow-up 35, 36 have shown an early improvement group with a more rapid improvement than in our study.
The results from our study have important implications for the way we understand low-back–related leg pain. We have shown that distinct leg pain clinical course patterns exist; therefore leg pain may not be fully described by measuring pain intensity at only 1 or a few points in time, or by single growth trajectories. Identification of such trajectory patterns in low-back–related leg pain patients may potentially improve understanding of the course of leg pain and guide targeted interventions. More than half of our study sample showed improving mild-moderate pain. For the majority of participants in this group, considering mainly conservative management options might be justifiable, such as medication and physiotherapy input. Indeed, because the ATLAS study was a treatment cohort, the majority of patients did receive physiotherapy input. We also identified subgroups of patients with persistent moderate and persistent severe pain trajectories. Whether these patients may benefit from consideration of more aggressive treatment options for pain relief, early on, assuming these options are appropriate for the individual patient, we are not able to say.
Even though the persistent severe and the improving severe groups had severe leg pain at baseline and seemed to have similar characteristics compared to the other groups, there were a few characteristics that could distinguish them at baseline. Participants in the persistent severe group were more likely to report leg pain of a possible neuropathic nature than those in the improving severe group. Conversely, all participants in the improving severe group had leg pain extending below the knee, had a significantly higher likelihood of having nerve root compression on MRI, and were more likely to be referred to a spinal specialist. However, the results of MRI directly influence the decision to refer to spine specialists, in these cases with very severe pain that does not improve over time with conservative management, and who are, in principle, appropriate candidates for invasive management options, such as injections and spinal surgery. Disentangling the effects of treatment from those of natural course is not possible. We are unsure whether it is possible for clinicians to differentiate early in the course of patients’ conditions between the 2 groups with severe leg pain at baseline. However, it is normal clinical practice to re-assess patients regularly, especially those with more severe symptoms, and to consider further appropriate investigations for those with severe pain and lack of improvement.
Similarly to research in the low-back pain field 37, future studies may develop simple approaches easily used in a clinical setting to identify patients likely to belong to a particular pain trajectory at an early stage of leg pain symptoms. The ability to predict leg pain trajectories early could guide patient care in terms of not waiting for all conservative management options to be exhausted before opting for more invasive treatments, such as spinal injections and surgery, where appropriate.
This study benefits from the use of longitudinally collected data with monthly follow-up measurements up to a year. Moreover, we used a robust statistical method, GMM, for identifying the latent trajectory groups. A further novelty of this cohort is the availability of clinical examination data, including a clinical diagnosis of sciatica, as opposed to many studies that have relied purely on self-report.
A limitation of this study, similar to all prospectively collected observational data, is the high number of dropouts from the original sample. The problem of missing data could influence the selection and the pattern of trajectories, although our sensitivity analysis results showed similar patterns of trajectories, and the differences between the participants and the dropouts were minimal in terms of the key baseline characteristics. In addition, the possible biases arising from such a problem were minimized by the use of full-information maximum method. Given that the fourth cluster contained <5% of the participants, we would suggest obtaining further evidence from future studies on leg pain trajectories to confirm that this group is also identified in other data sets. Further, the small size of the fourth cluster may have inhibited our ability to detect differences in baseline characteristics between clusters 3 and 4.
In primary care patients with back-related leg pain, we identified 4 distinct and clinically meaningful trajectories of leg pain over 12 months and a number of baseline patient characteristics associated with membership of each trajectory class. Three of the trajectory classes, improving mild, persistent moderate, and persistent severe leg pain, are generally comparable to back pain trajectories. The improving severe cluster represented a group with severe leg pain, whose symptoms improved over time; this group is less often identified in patients with back pain. These findings could help to gain a better understanding of the nature of low-back–related leg pain in primary care. The findings also confirm that describing an entire low-back–related leg pain population using a single growth trajectory is oversimplifying the leg pain growth patterns. Identification of such distinct groups of patients could improve understanding of the course of leg pain and may provide a basis of classification for further diagnostic tests and treatment choice, from potential and appropriate interventions.
Acknowledgments
The authors thank the members of the ATLAS study research team, Elaine M. Hay, Martyn Lewis, Sue Jowett, Danielle A. W. M. van der Windt, Samantha L. Hider, and Steve Vogel, and all the participating patients, clinicians, and managers.
Author Contributions
All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be submitted for publication. Dr. Ogollah had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study conception and design
Ogollah, Konstantinou, Stynes, Dunn.
Acquisition of data
Ogollah, Konstantinou, Stynes, Dunn.
Analysis and interpretation of data
Ogollah, Konstantinou, Stynes, Dunn.
