Abstract
Background It has been suggested that low back pain (LBP) is a condition with an unpredictable pattern of exacerbation, remission, and recurrence. However, there is an incomplete understanding of the course of LBP and the determinants of the course.
Objective The purposes of this study were: (1) to identify clusters of LBP patients with similar fluctuating pain patterns over time and (2) to investigate whether demographic and clinical characteristics can distinguish these clusters.
Design This study was a secondary analysis of data extracted from a randomized controlled trial.
Methods Pain scores were collected from 155 participants with chronic nonspecific LBP. Pain intensity was measured monthly over a 1-year period by mobile phone short message service. Cluster analysis was used to identify participants with similar fluctuating patterns of pain based on the pain measures collected over a year, and t tests were used to evaluate if the clusters differed in terms of baseline characteristics.
Results The cluster analysis revealed the presence of 3 main clusters. Pain was of fluctuating nature within 2 of the clusters. Out of the 155 participants, 21 (13.5%) had fluctuating pain. Baseline disability (measured with the Roland-Morris Disability Questionnaire) and treatment groups (from the initial randomized controlled trial) were significantly different in the clusters of patients with fluctuating pain when compared with the cluster of patients without fluctuating pain.
Limitations A limitation of this study was the fact that participants were undergoing treatment that may have been responsible for the rather positive prognosis observed.
Conclusions A small number of patients with fluctuating patterns of pain over time were identified. This number could increase if individuals with episodic pain are included in this fluctuating group.
It is a widely held belief that persistent low back pain (LBP) is a condition in which pain levels vary over time.1,2 However, because of the difficulty in collecting long-term and frequent pain measurements, there is an incomplete understanding of the course of persistent LBP as well as of how the condition is influenced by clinical and demographic characteristics.3,4 For instance, although it has been suggested that LBP is recurrent and fluctuates over time,1 there is still little evidence to support this suggestion.
The progression of pain over time, specifically fluctuation of pain, has been described and assessed in recently published longitudinal studies.1,5,6 The results of these studies revealed that the majority of patients with LBP have a rather positive prognosis, with most having a recovering pattern of pain over time, although many patients maintained low levels of pain over the study period.6,7 However, the results regarding fluctuation patterns are still controversial; some studies suggest that fluctuating patterns of pain are common,5,6 whereas others suggest that the number of patients with pain fluctuation may be smaller than first hypothesized.1,7 These conflicting results are likely due to the fact that there is a lack of definition of what constitutes a fluctuation. Furthermore, many of these studies used different methods, different outcome measures (eg, days of bothersomeness,8 numeric pain rating scale score1), different follow-up times (eg, 6 months,1 1 year7), and different statistical methods (eg, cluster analysis,1 graphical categorization6) and did not clearly define what represents a fluctuation.
A recent and ongoing debate in the literature and among back pain researchers is the differentiation between patients with chronic persistent fluctuating pain and those with episodic LBP. The former would be patients who have persistent pain, but with levels that fluctuate over time, while the latter would be those with episodic/recurrent LBP who recover from an episode, but then experience a recurrence. Because there are different definitions of recovery and recurrence,4,9 identifying those with episodic or recurrent LBP becomes challenging. Using de Vet and colleagues'10 definition of recovery (pain=0 for 1 month), a recently published study identified that 20% (n=16) of the patients had recovered from their episode; however, it was not clear how many of those had a recurrence and thus had episodic back pain and how many patients actually had fluctuating pain.7
Thus, the primary purpose of the present study was to conduct a cluster analysis to identify clusters of participants with similar pain fluctuation patterns over a 12-month period. Participants provided monthly pain intensity measurements for 12 months as well as baseline clinical and cognitive measures. These data were used to: (1) identify clusters of participants with similar pain fluctuation over time, (2) evaluate the relationship between fluctuating pattern and episodic pain, and (3) evaluate whether cluster membership was influenced by participant's baseline characteristics (such as age, duration of pain, coping strategies, pain anxiety, pain beliefs, clinical interpretation of the potential presence of lumbar “instability”) and the treatment condition (motor control and graded activity) to which they were assigned in the randomized controlled study. A secondary purpose was to further explore the group of patients without fluctuating pain patterns and investigate whether they could be clustered into groups with similar levels of severity of pain over the 12-month period of the study.
Participants and Method
This study involved a secondary analysis of the data collected as part of a randomized controlled trial evaluating the effects of exercise therapy in patients with chronic nonspecific LBP.11,12 A total of 172 participants seeking care for their back pain were recruited by general practitioners in Sydney and Brisbane or drawn from the waiting list of an outpatient physical therapy department at a public hospital in Sydney. All participants signed a consent form.
Participants were screened and excluded if they were unsuitable for exercise due to significant comorbidity (eg, serious spinal pathology or contraindication to exercise) or specific pathology (eg, fracture, inflammatory, or infective disease or cancer). Participants were included if they met all of the following inclusion criteria:
Nonspecific LBP with or without leg pain persisting for at least 3 months13
Currently seeking care for LBP
Aged between 18 and 80 years
English speaker
A score of moderate or greater on item 7 (“How much bodily pain have you had during the past week? [Answers include “None,” “Very mild,” “Mild,” “Moderate,” “Severe,” and “Very severe”]) or item 8 (“During the past week, how much did pain interfere with your normal work, including both work outside the home and housework?” [Answers include “Not at all,” “A little bit,” “Moderately,” “Quite a bit,” and “Extremely”]) of the 36-Item Short-Form Health Survey (SF-36)14
Data Collection
Demographic characteristics, including age, sex, weight, height, pain, function (measured with the Patient-Specific Functional Scale [PSFS]), disability (measured with the 24-item Roland-Morris Disability Questionnaire [RDQ-24]), and a clinical impression of the presence of lumbar instability (measured with the Lumbar Spine Instability Questionnaire),15 were collected at baseline. Additionally, psychological questionnaires, including the Örebro Musculoskeletal Pain Questionnaire (OMPQ) (also known as the Acute Low Back Pain Screening Questionnaire16), the 20-item Pain Anxiety Symptom Scale (PASS-20),17 and the Pain Self-Efficacy Questionnaire (PSEQ),18 were completed at baseline.
Participants were asked about their pain at baseline and once per month during a 1-year period using an automated short message service (SMS) system.19 All messages consisted of the following text: “Please rate on a 0–10 scale (0=no pain, 10=worst pain): (a) average pain over the last 24 hours; (b) average pain over the last week.” The automated system was scheduled to send an SMS message at 10 am every first, second, third, or fourth Monday of the month over a 1-year period. The week chosen depended on the week the participant entered the study. If a participant did not reply to the first message, another message was sent the next day (again at 10 am). The same procedure was performed on the following day. Therefore, each participant had 3 chances to answer the week's message sent via SMS. The automated system also sent an e-mail to the project coordinator with the names of the participants who had not responded after the third message. Because the responses to the SMS messages were an important outcome of the randomized controlled trial in which the participants of this study were involved, participants that did not respond to the third message were contacted via telephone to respond. For participants who did not have a mobile phone or did not know how to use text messages, the pain outcomes were collected by telephone. At the end of the study, all participants received AUD$10.00 to cover the cost of text messaging. For the purposes of this study, only the average pain over the last week (and not the average pain over the last 24 hours) was assessed because this is the most common method of measuring pain intensity in people with chronic LBP.
Additionally, as part of the follow-ups for the trial, we asked participants at 2, 6, and 12 months if they had experienced a recovery or a recurrence. De Vet et al10 defined a recovery as a pain-free period lasting at least 30 days and a recurrence as pain lasting for at least 24 hours following a recovery.
Although, as mentioned above, the participants in this study were originally from a randomized controlled trial and received either a motor control or graded activity intervention, they were included in this study as a single patient group for the cluster analyses. The randomized controlled trial demonstrated that treatment groups did not significantly differ for any of the outcomes evaluated, including pain, at any of the time points.12
Data Analysis
Because cluster analysis requires complete data, it was necessary to impute data for some participants and exclude others. If a participant had 2 or fewer missing pain measurements over the 12 months of the study, the missing scores were replaced by the mean pain score of that participant. However, if 3 or more pain measurements were missing, the participant was excluded from the analysis. Of the 172 participants included in the randomized controlled trial, 17 were excluded because they had more than 2 missing values for the monthly pain scores. Of the remaining 155 participants, 131 had complete data; imputation was needed for 24 participants. In addition to the imputation of pain measurements, when responses were missing for items included in the questionnaires administered at baseline, missing scores were replaced by the mean questionnaire score of that participant (total or subscale score when applicable). All statistical analyses were performed using PASW Statistics version 18 (SPSS Inc, Chicago, Illinois).
Identification of number of fluctuations.
The first step of the analyses involved the determination of a measure of pain fluctuation across time. For this step, separate linear regression analyses were conducted for each participant to predict pain as a function of log time. The R-square and predicted pain values for each time point for each participant were extracted. The number of clinically important fluctuations in pain was defined as the number of times the predicted pain value from the linear regression was 2 or more points above or below the actual pain intensity reported by the participant. Variations of pain equal to or greater than 2 on a 0-to-10 numerical scale are considered to be clinically important for patients with LBP.20
Cluster analysis to identify pain fluctuation.
A hierarchical cluster analysis21 with complete linkage was next conducted on the full sample of 155 participants. Complete linkage was used to identify distinct patterns of fluctuating pain over time. The variables for this analysis were R-square and number of clinically important fluctuations (0–12). These 2 variables were included because they provided information on how much the actual values varied from a linear or nonfluctuating model. A dendogram was generated to identify the number of clusters. Methods of selecting the best number of clusters vary in the literature and in most cases include subjective analysis.21 The method used in this study included evaluation of the amalgamation coefficients, which provide an indication of the nature of the composition of 2 clusters at one stage of the cluster analysis being combined into 1 cluster at the next stage. A sudden jump in value implies that 2 relatively dissimilar clusters have been merged; thus, the number of clusters prior to the jump is the most reasonable estimate of the number of clusters. Sudden jumps can be seen on the dendogram. Large distances between sequential vertical lines in the dendogram represent jumps in the coefficient.21 Amalgamation coefficients and the dendogram were visually inspected to identify the optimal number of clusters.
Confirmation of the number of clusters.
As recommended in the literature, a one-way analysis of variance (ANOVA) with the Brown-Forsythe test for unequal sample sizes and unequal variance among groups was conducted after identification of the number of clusters to confirm whether the clusters differed significantly with respect to the clustering variables.21 The level of significance was set at .05.
Identification of clusters with pain fluctuation.
Plots of pain values, descriptive statistics for R-square, and the number of fluctuations of members of each cluster were calculated to better characterize the obtained clusters. Participants who had significantly lower R-square values, higher numbers of fluctuations, and clear fluctuating patterns on the plots were considered to have a fluctuating pain pattern.
Although R-square and number of fluctuations were used to identify these clusters, the primary aim of this study was to classify the course of pain over time. Therefore, a generalized linear model for repeated measures was used to evaluate whether clusters significantly differed from each other in terms of pain over time. The level of significance was set at .05.
Cluster characteristics.
We used an independent-samples t test to evaluate whether age, duration of back pain, baseline pain, function, disability, clinical impression of the presence of lumbar instability (with the Lumbar Spine Instability Questionnaire), and psychological characteristics at baseline (with the OMPQ, PASS-20, PSEQ, and Coping Strategies Questionnaire) were significantly different among clusters. The chi-square test was used to evaluate whether there was a difference among clusters in relation to the dichotomous variables: sex, treatment group (motor control or graded activity), and smoking status (smoker or nonsmoker).
Episodic pain.
We used descriptive analysis and the chi-square test to identify the number of participants in each cluster who had recovered over the period of the study. Because there is no consensus in the literature about a definition of recovery, 2 methods were used in this study. The first method used de Vet and colleagues' definition of recovery: a pain-free period lasting for at least 30 days.10 The second, more relaxed, method classified participants as recovered if they reported pain less than or equal to 1 at any of the SMS responses referring to pain in the preceding week. We also evaluated durable recovery, which we defined as a report of no pain for at least 1 month and no recurrence of pain over the 12 months of the study.
Secondary Analysis
The goal of this step of the analysis was to further explore the group of patients who did not have a fluctuating pattern of pain over time. Based on the results of previous studies,1,8 it is likely that this group of patients contains individuals with different pain levels over time, which would be relevant for future prognostic studies. All participants who were identified in step 1 as having a nonfluctuating pain pattern (ie, in cluster 1) were included from the second step of the analysis.
Cluster analysis to identify pain patterns over time.
A hierarchical cluster analysis with complete linkage was performed with different clustering variables than the variables used in stage 1. In stage 2, we used data on pain patterns, gathered via SMS, from baseline to 12-month follow-up as clustering variables (13 values from baseline through the 12th month). A dendogram was generated and amalgamation coefficients were computed in order to identify the number of clusters.
Confirmation of the number of clusters.
A generalized linear model for repeated measures was used to evaluate whether clusters significantly differed from each other with respect to pain intensity over time. The level of significance was set at .05. A plot of the mean pain values for each time point was generated for each identified cluster to evaluate the differences in the course of pain over time.
Definition of clusters.
Plots of mean pain values over time for the 5 clusters were constructed, and clusters were defined with denominations similar to those used by Dunn et al.1 Graphical representations of both studies were used to compare identified clusters.
Characteristics of clusters.
Similarly to the main analysis of this study, we evaluated whether baseline characteristics were different among the clusters identified. We used a one-way ANOVA with the Brown-Forsythe test for unequal sample sizes and unequal variance between groups. Similarly, chi-square analysis was used to evaluate whether there was a difference among clusters in relation to the dichotomous variables.
Role of the Funding Source
The original trial was funded by the Australian National Health and Medical Research Council. Dr Macedo holds a postdoctoral fellowship from the Canadian Institutes for Health Research and receives a stipend incentive from Alberta Innovates Health Solutions. Dr Latimer's and Dr Maher's research fellowships are funded by the Australian Research Council. Dr Hodges holds a research fellowship funded by the Australian National Health and Medical Research Council.
Results
Identification of Number of Fluctuations (155 Participants)
The mean (SD) values for the terms from the regression analysis conducted for each participant were the following: intercept=5.0 (1.7), slope=−1.0 (1.4), and R-square=.5 (.3).
Cluster Analysis to Identify Pain Fluctuation
The cluster analysis based on R-square values and number of fluctuations yielded the dendogram shown in Figure 1. The values of the amalgamation coefficients suggested the existence of 3 main clusters. There were 134 participants in cluster 1, 16 participants in cluster 2, and 5 participants in cluster 3.
Dendogram of hierarchical cluster analysis based on R-square and number of fluctuations as clustering variables (n=155). Each short vertical line on the x-axis represents an individual participant.
Confirmation of the Number of Clusters
The mean (SD) of the R-square values for the participants in the 3 clusters were: cluster 1=.52 (.27), cluster 2=.33 (.23), and cluster 3=.31(.13). A one-way ANOVA revealed that the R-square values were significantly different among the 3 identified clusters (F=5.03, df=2, P<.05). The mean (SD) number of pain fluctuations for the clusters were: cluster 1=1.0 (1.0), cluster 2=4.5 (0.8), and cluster 3=8.8 (1.9). A one-way ANOVA revealed that the number of fluctuations was significantly different among the 3 identified clusters (F=217.5, df=2, P<.05).
Identification of Clusters With Pain Fluctuation
A generalized linear model with repeated measures demonstrated that the 3 clusters were not significantly different in terms of pain over time (F=1.23, df=2, P>.05), possibly because of the fluctuating nature of some of the clusters and the small sample size of 2 clusters. The within component of the repeated measures design was not of interest given the purpose of the present study. Therefore, the results for this component have not been provided for each of the repeated measure analyses that were performed.
The low R-square values (ie, due to lack of fit with a linear regression model) and high number of fluctuations over the period of the study demonstrated that clusters 2 (16 participants) and 3 (5 participants) included the patients who had a fluctuating pain course, which was confirmed by plotting individual participant's pain scores (Fig. 2 represents a random sample of 5 participants from clusters 1 and 2 and all 5 participants from cluster 3).
Pain scores from a random sample of 5 participants included in clusters 1, 2, and 3 from the first step of the cluster analysis: (A) cluster 1 (5 of 134 participants), (B) cluster 2 (5 of 16 participants), and (C) cluster 3 (5 of 5 participants).
Cluster Characteristics
Because clusters 2 and 3 represent patients with a fluctuation pain pattern over time, we merged the 2 clusters for the purpose of this analysis; this is the most logical clinical approach. However, we did run the analysis with all 3 clusters separately, and the results were very similar. The independent-samples t test demonstrated that disability at baseline (RMDQ-24) was the only continuous variable that significantly differed between fluctuating cluster (2 and 3) and nonfluctuating cluster (1) (t=−2.13, df=153, P<.05), with the fluctuating group having greater disability at baseline. However, the mean disability difference between clusters (ie, 2.5) was smaller than the minimal important difference (ie, 5) on a 24-item scale for the RMDQ-24.20 The chi-square test demonstrated that only treatment group was significantly different between clusters (χ2=5.86, df=1, P<.05), with more patients with a fluctuating pattern in the graded activity group. All results are presented in Table 1.
Demographic and Baseline Characteristics of Patients With Fluctuating Pain and Patients With Nonfluctuating Pain
Episodic Pain
Recovery (using de Vet and colleagues' definition) occurred in 47 participants (35.1%) with no fluctuations and in 6 (28.6%) of those with fluctuations. However, the difference between the 2 groups was not significant (F=0.14, df=1, P>.05). At the end of the study, durable recovery was present in 17 participants (12.7%) with no fluctuating patterns and in 3 participants (14.3%) with fluctuating patterns. Again, the difference between the 2 groups (F=0.101, df=1, P>.05) was not significant. Using the more relaxed definition of recovery (pain=1 on at least 1 monthly pain report), recovery was present in 58 participants (43.3%) with no fluctuations and in 12 participants (57.1%) with fluctuations. The difference between groups was not significant (F=2.298, df=1, P>.05).
Secondary Analysis
The 134 participants included in cluster 1 (ie, those without fluctuating pain) were included in the secondary analysis.
Cluster analysis to identify level of pain patterns over time.
The cluster analysis based on the SMS data collected from baseline to 12 months yielded the dendogram presented in Figure 3. The values of the amalgamation coefficients suggested the existence of 3 main clusters. There were 73 participants in cluster 1, 38 participants in cluster 2, and 23 participants in cluster 3.
Dendogram of the hierarchical cluster analysis of the secondary analysis of the study based on pain fluctuations from baseline to 12-month follow-up (n=134). Each short vertical line on the x-axis represents an individual participant.
Confirmation of the number of clusters.
The mean pain scores for each cluster are presented in Figure 4. A generalized linear model with repeated measures demonstrated that the 3 clusters from the second step cluster analysis were significantly different in terms of pain over time (F=310.87, df=2, P<.05).
Mean pain scores in each cluster of the secondary analysis of recovering mild pain (cluster 1 [n=73]), persistent moderate pain (cluster 2 [n=38]), and severe chronic pain (cluster 3 [n=23]).
Definition of clusters.
The plots of the clusters demonstrated 3 different patterns of level of pain over time.
severe chronic pain (23 participants [14.8%])
persistent moderate pain (38 participants [24.5%])
recovering mild low pain (73 participants [47.1%])
The term recovering was used in the name of the third cluster because 35% of participants in this group had no pain at 1 or more follow-ups and because the plot of the pain values demonstrated that pain decreased from baseline and was low for 12 months.
Cluster characteristics.
As shown in Table 2, the results of a one-way ANOVA demonstrated that the 3 clusters were statistically significantly different at the .05 level of significance in terms of baseline pain, function, disability, lumbar spine instability, OMPQ, PSEQ, coping strategies, and PASS-20 total score. In contrast, the clusters were not significantly different in terms of age and duration of back pain. The results of the chi-square analysis demonstrated that only smoking status for the dichotomous variables (sex, treatment group, and smoking status) were significantly different among clusters.
Baseline Characteristics of Patients With Nonfluctuating Pain Stratified by Clusters Identified in the Secondary Analysis
Discussion
This study identified that only 13.5% of the trial participants had a fluctuating pattern of pain over a 1-year period. The small percentage of individuals with a fluctuating pattern of pain is similar to the findings of previously published studies.1,7 However, if the number of participants who had fluctuating patterns is added to the number of participants who had recovered over the period of the study, but had a subsequent recurrence of fluctuating pain, the number of participants with fluctuating patterns of pain becomes much larger. Thus, it is possible that clinical and research reports suggesting that back pain is largely of a fluctuating nature may not be distinguishing between these 2 groups of patients. The clinical implications of combining or not combining these 2 patient groups are still unknown and need further investigation.
It has been reported that the results of different cluster analysis methods (eg, hierarchical cluster, latent class analysis) lead to different results and potentially even to different numbers of clusters.22 The differences in results among the previously published studies on the fluctuating nature of LBP may reflect some of these statistical differences. Nevertheless, when clear clustering methods were used, the number of individuals with fluctuating pattern reported in the literature was never larger than 35%,1,5 which, as mentioned above, may be low.
The t tests evaluating baseline differences among clusters demonstrated that disability (measured with the RMDQ-24) was the only statistically significant variable among clusters, although the magnitude of the difference is probably not clinically important. The only other variable that was different among clusters was treatment group (tested using chi-square statistics), which had more patients with fluctuation being from the graded activity group. The implications of the latter needs to be further investigated, given that the results of the randomized controlled trial from which the data for this study were drawn demonstrated that the 2 treatments were not significantly different among groups for any of the investigated outcomes at any of the follow-ups.
Theories for the mechanisms of action of each of the treatments provided in the trial (motor control and graded activity) suggest different explanations as to why each treatment may prevent fluctuations in pain. Motor control exercise is suggested to correct movement patterns, which, in the long-term, could reduce the number of flare-ups,23 whereas in graded activity, the restoration of fitness and conditioning, along with the use of self-management techniques, could be a different mechanism that leads to the same effect.24 However, apart from the collection of pain outcomes at usual time points (2, 6, and 12 months) the mechanisms underpinning the effect of these interventions on fluctuation and flare-ups have not been investigated.25,26
When a more strict definition of recovery was used, only a few participants with fluctuating pain had recovered over the period of the study. The results, therefore, support the idea that these participants had fluctuating LBP, not recurrent LBP. On the other hand, a similar number of patients from the nonfluctuating cluster also recovered during the study (13% to 43%, depending upon the method used to define recovery). Using de Vet and colleagues' definition, 16% of participants from the nonfluctuating group had a recovery with a subsequent recurrence. Some authors and clinicians would define this as a fluctuating pattern. It is clear that better definitions and delimitations as to what constitutes fluctuating and episodic pain are warranted before we can gain further insight into the prognosis, mechanisms, and management of these patient groups.
The secondary analysis of this study, which was of an exploratory nature, was aimed at further classifying the severity of pain of those participants without a fluctuating pattern. The results demonstrated that these groups could be subdivided into 3 different clusters: severe chronic pain, persistent moderate pain, and recovering mild low pain. Furthermore, the majority of the baseline characteristics were significantly different among clusters, demonstrating that psychological characteristics and clinical outcomes may play an important role in how the pain progresses over time.
An interesting finding of this study was that most participants in the secondary analysis (54.5%) were classified into the recovering mild pain cluster. This result suggests that although most participants did not fully recover from their pain, many participants seem to have a positive prognosis, at least after receiving the treatments that were studied in the randomized controlled trial from which the participants were drawn.
Future studies should evaluate whether the interventions provided as part of the randomized controlled trial had any influence on participant classification and whether these optimistic results occur with participants not undergoing treatment. The results of this study may have implications for understanding the mechanisms responsible for different prognoses of patients with chronic LBP as well as the responses to treatment among people with chronic LBP. Supporting the evidence from previous studies, individuals with poorer psychological profiles to begin with tended to have a poorer or less positive pain prognosis over time. It would be interesting to see whether providing interventions designed to modify these characteristics would provide a different pattern of pain over time. To date, most studies have evaluated the efficacy of an intervention by examining individuals at a few points in time, with little or no assessment of the course of an individual patient's pain. Thus, the effect of interventions that target a decrease in the number of fluctuations may not be captured when using current methods of outcome assessment.
A specific strength of this study is the fact that the results were extracted from a rigorously designed randomized controlled trial. Additionally, robust methods for identifying the clusters were used. A limitation of this study was the fact that pain outcomes were self-reported and measured only once per month over the 1-year period and, therefore, may not reflect the total variation occurring over the year.
It is also important to consider 4 issues related to the patient population used in this study. First, observations were collected from participants who were seeking treatment for their LBP after a period of pain that had persisted for at least 3 months. Thus, the data may be biased toward individuals who have fewer fluctuations of pain; participants would not have been eligible for inclusion in the study if their pain had recovered within 3 months prior to their assessment for inclusion. Second, the proportions of participants classified into each cluster were based on the methods we selected to define fluctuations (ie, deviation of the actual pain intensity by 2 or more points above or below pain value predicted from the linear regression). The prevalence of fluctuating pain would be different if other criteria were used. For example, other definitions of fluctuating pain based on disability measures and number of episodes were not included in this study. Third, the prevalence of recovery in the participants included in the present study may not reflect the natural course of recovery of patients with LBP, because all participants received treatment that was aimed at reducing pain and preventing the recurrence of pain. Thus, the results may overestimate the positive prognosis for this group. Finally, guided by advice from a statistician, we conducted a second cluster analysis on data already clustered, namely, the group with no fluctuating pain, to determine whether the patients in this group could be grouped by persistent but different levels of severity of pain. We acknowledge, however, that this approach is not universally accepted, and readers need to consider that issue when interpreting the results.
The Bottom Line
What do we already know about this topic?
Low back pain is a condition with an unpredictable pattern of exacerbation, remission, and recurrence.
What new information does this study offer?
This study identified that only a small number of patients had a fluctuating pattern of pain over 1 year. The majority of patients maintained a steady level of pain and this variance, or lack thereof, was predicted by some baseline characteristics.
If you're a patient or a caregiver, what might these findings mean for you?
Although these findings need to be further tested and replicated, your physical therapist should be aware of these results to identify potential baseline characteristics that could be associated with fluctuating or high but steady levels of pain. He or she may then adjust your prognosis.
Footnotes
Dr Macedo, Dr Maher, Dr Latimer, Dr McAuley, and Dr Hodges provided concept/idea/research design. Dr Macedo, Dr Maher, Dr McAuley, and Dr Hodges provided writing and data collection. Dr Macedo, Dr Maher, Dr Hodges, and Dr Rogers provided data analysis. Dr Macedo, Dr Maher, and Dr Latimer provided project management. Dr Latimer and Dr Hodges provided fund procurement. Dr Latimer provided facilities/equipment and institutional liaisons. Dr Macedo and Dr Latimer provided consultation (including review of the manuscript before submission).
The study design, procedures, and informed consent were approved by the human research ethics committees of the University of Sydney and the University of Queensland.
The original trial was funded by the Australian National Health and Medical Research Council. Dr Macedo holds a postdoctoral fellowship from the Canadian Institutes for Health Research and receives a stipend incentive from Alberta Innovates Health Solutions. Dr Latimer's and Dr Maher's research fellowships are funded by the Australian Research Council. Dr Hodges holds a research fellowship funded by the Australian National Health and Medical Research Council.
The trial was registered at the Australian New Zealand Clinical Trial Registry (ACTRN12607000432415).
- Received October 22, 2012.
- Accepted September 23, 2013.
- © 2014 American Physical Therapy Association