Cognitive Therapy, Mindfulness-based Stress Reduction, and Behavior Therapy for the Treatment of Chronic Pain

Randomized Controlled Trial

John W. Burns; Mark P. Jensen; Beverly Thorn; Teresa A. Lillis; James Carmody; Andrea K. Newman; Francis Keefe


Pain. 2022;163(2):376-389. 

In This Article



Figure 1 depicts the study structure and indicates the number of participants who completed each part of the study and who were ineligible, withdrew, or declined to participate. Participants were recruited through referrals from staff at the University Pain Center, Midwest Orthopedic Clinic and Neurosurgery Clinic at Rush University, and at Clinics at the Duke University Medical Center Pain Clinic (eg, Pain Clinic, Spine Clinic, Neurosurgery Clinic, and Orthopedic Clinic). Flyers were posted at community physician waiting rooms, and local newspaper ads were placed. All procedures were approved by the Rush University and Duke University IRBs. Recruitment ran from December 1, 2014, to March 1, 2019, and follow-up was completed October 18, 2019. Using medical clinic and community physician referrals and ads allowed for recruitment of a broad array of patients. Each participant received up to $300 reimbursement for time and effort on the assessments and to defray travel expenses. The Identifier is NCT02133976.

Figure 1.

CONSORT diagram.

The inclusion criteria were as follows: (1) significant daily chronic pain intensity (rated as being at least 4 on average on a 10-point scale; see below) and interference in performing daily activities due to pain (rated as being at least 3 on a 6-point scale; see below) for at least 6 months; (2) musculoskeletal pain of the low back and/or leg pain which may be related to the history of degenerative disk disease, spinal stenosis, or disk herniation (radiculopathy subcategory), or muscular or ligamentous strain (chronic myofascial pain subcategory); and (3) age between 18 and 80 years. The exclusion criteria were as follows: (1) meet criteria for alcohol or substance abuse problems, (2) meet criteria for past or present psychotic or bipolar disorders, (3) inability to understand English well enough to complete questionnaires or to participate in therapy, (4) active suicidal ideation with intent, and (5) pain because of cancer, rheumatoid arthritis, migraine or tension headache, fibromyalgia, or complex regional pain syndrome. Table 1 describes demographic characteristics of the sample.

Inclusion and exclusion criteria were assessed by the review of medical records by study physicians and a brief psychosocial history—including administration of relevant items from the Mood Disorder, Psychotic Screening, and Substance Use Disorders Modules of the Structured Clinical Interview for DSM-IV Axis I Disorders—Non-Patient Edition (SCID-IV/NP)[15] by study research assistants. General health was assessed, and information was gathered about the circumstances of the onset of low back pain, the sequence of events in medical intervention to date, exacerbating and ameliorating factors, medications used currently and in the past, previous diagnostic modalities used, and previous medical interventions and their impact.


Potential participants underwent a brief phone screen to determine eligibility. If they were eligible and agreed to participate, a baseline assessment appointment was made. During the assessment visit, an RA interviewed the participant regarding inclusion and exclusion criteria and the process of randomization was explained to the participant. Written consent was obtained, after which pretreatment assessments were completed. They were then randomly assigned to treatment condition using a customized program designed by our statistician. After these assessments, participants were informed of their treatment condition by an unblinded research staff member. It was explained to participants that phone assessments would occur within 48 hours of each session. For participants assigned to the TAU condition, they were told that they would also participate in the weekly phone assessments. The personnel conducting these assessments were blind to participant treatment condition. The 8 therapy sessions were scheduled for the CT, MBSR, and BT participants, and the 8 phone assessments were scheduled for the TAU participants.

Each session was 90 minutes in length, and they were conducted in an individual format. Sessions for all treatment conditions followed the same general format: (1) review of previous week's session, (2) homework review, (3) session treatment objectives, and (4) assign homework and conduct postsession assessments. Participants attended a brief visit at 6-month posttreatment.


All treatments consisted of 8 weekly 90-minute individual sessions. For each of the 3 active conditions, the content of each session was based on a standardized treatment manual. Manuals contained detailed session by session information, instructions for therapists, and patient handouts. Manuals were based on those used in our prior studies of patients with persistent pain.[5,22,36]

Behavioral Therapy. The BT protocol taught patients that when pain persists, learned, maladaptive patterns of daily activity often develop (ie, "pain behaviors")[16] such as spending much time resting or reclining and decreasing their involvement in pleasant and meaningful activities. The goals of BT were to help patients better understand how activity patterns can contribute to pain-related problems and to learn ways to increase their involvement in activities that are important and meaningful (ie, increase "well behaviors").[16] Patients were systematically taught how to work toward goals in areas of activity affected by pain. To enhance their awareness of activity patterns, they kept a daily activity diary. They learned to set and track completion of daily activity goals that were measurable and realistic. Patients were taught how to use a "quota system" (the activity-rest cycle) to increase "uptime" (time spent not reclining). An initial quota for uptime that they know they can manage despite pain was set (eg, 20 minutes walking and standing per hour). Participants were taught how to reward themselves with a rest break (eg, 5 minutes of sitting) after reaching the quota. During treatment, the uptime activity quota was increased. Second, pleasant activity scheduling was used to increase the level and range of daily reinforcing activities. Here, patients identified activities they enjoy (eg, visiting friends), find meaningful (eg, helping others), or that provided a sense of mastery (eg, learning to draw) and then learned how to use goal setting to schedule and track levels of pleasure related to these accomplishments. Patients were trained in brief relaxation and taught how to use brief relaxation sessions (eg, 2–5 minutes) to reward their accomplishment of their activity goals. To help patients deal with obstacles, they were taught problem-solving skills to meet behavior change goals. Patients were also trained in communication skills to enable them to communicate more effectively with others about the progress they were making and problems they were experiencing. Finally, patients developed a written maintenance plan that included a list of short-term and long-term activity goals and a plan for dealing with possible setbacks.

Cognitive Therapy. Cognitive therapy (also known as cognitive restructuring)[35] was used to help patients recognize the relationships between thoughts, feelings, and behaviors. These techniques helped patients (1) identify automatic pain-related thoughts; (2) evaluate automatic thoughts for accuracy, identify sources of distorted thoughts, and recognize connection between automatic thoughts and emotional/physical shifts; (3) challenge negative, distorted automatic thoughts using "evidence"; (4) develop new realistic alternative cognitive appraisals responses; and (6) practice applying new rational appraisals and beliefs. To help patients deal with obstacles in applying CT skills, they were taught to identify possible problems and develop solutions to meet CT goals. Finally, patients developed a written maintenance plan that included a list of short-term and long-term goals for applying cognitive restructuring and a plan for dealing with possible setbacks in their cognitive change efforts.

Mindfulness-based Stress Reduction. Participants assigned to the MBSR condition received training in mindfulness through (1) body scan meditation, a gradual moving of attention through the body, accompanied by awareness of bodily sensations including sensations of breathing while in a lying position; (2) sitting meditation, focusing on awareness of breathing, bodily sensations, thoughts, and emotions, while sitting on a chair or cushion; and (3) gentle movement exercises intended to develop awareness (mindfulness) during movement. Each session included practice of one or more of these techniques. In-class didactic material and discussion of patients' experiences of developing and applying mindfulness in everyday life were also part of each session. In-class activities included suggestions for application of mindfulness as a method for responding positively to stress, dealing with the challenges of pain, and exercises focusing on the challenges and achievements patients experience in integrating mindfulness into their lives and the stressful situations they encounter. Additional discussion focused on stress reactivity. Patients were taught problem-solving skills for identifying possible problems and developing solutions for dealing with obstacles related to practicing mindfulness. Finally, patients developed a written maintenance plan that included a list of short-term and long-term goals for applying mindfulness methods and a plan for dealing with possible setbacks.

Treatment as Usual. Participants assigned to TAU did not receive any additional psychosocial treatment beyond the other treatments they were receiving before enrolling in this study. Treatment as usual participants continued with their ongoing treatment regimens for pain, including psychotropic and analgesic medications.

Study Therapists, Therapist Training, Adherence, and Quality

Study Therapists. All study therapists were postdoctoral level clinical psychologists with prior experience delivering psychosocial interventions for pain. Over the course of the five-year study, 10 therapists were trained to administer all 3 active treatments. We chose not to assign one therapist to perform only one treatment to avoid confounding results of condition with results because of quality of therapists. Instead, we closely monitored treatment integrity with procedures described below.

Therapist Training. All study therapists received training before conducting treatment sessions with study subjects. Initial training consisted of a 4-day didactic and experiential course conducted by the authors. Therapists were provided detailed manuals and outlines of treatment protocols, and the treatment strategies were taught by didactic instruction and role-play of common scenarios. All instruction sessions were digitally videotaped for reference and/or education of new therapists. Therapists were certified to deliver each treatment by having supervisors (ie, the authors) rate audio-recordings of practice role-play sessions before their working with study participants. Mastery of each protocol was required for therapists to deliver treatment in the research protocol.

Procedures to Ensure Consistency of Treatment. To ensure that the study therapists consistently followed the appropriate treatment protocol, 4 steps were taken: (1) use of a detailed treatment manual; (2) weekly supervision sessions; (3) audio recording of sessions for treatment adherence ratings (see below), with these recordings and feedback from the adherence raters reviewed during the weekly supervision meetings conducted by the authors; and (4) provision of therapist feedback on treatment consistency and further didactics and role plays to correct "drift" if needed.

Treatment Adherence and Quality. Treatment adherence refers to the extent to which a therapist uses interventions prescribed by a protocol.[41] Treatment quality refers to therapist competence in delivering the treatment. A total of 72 sessions (20%) were rated by an author and a research assistant. The Therapy Adherence and Competence Scale is an adapted version of the Cognitive Adherence and Competence Scale[2] and was used to measure adherence and quality of treatment. Adherence ratings were provided on the basis of demonstrated delivery of unique elements and treatment components for each session. The mean adherence rating was 98%. Each session received a quality rating based on a 4-point scale with end points ranging from "0 = poor to 3 = excellent." The mean quality rating was 2.03 (SD = 0.13), indicating that ratings of all 3 conditions were in the good to excellent range.


We evaluated the effects of the treatments on 5 outcome domains: pain interference (primary outcome), pain intensity, depressive symptoms, physical function, and sleep disturbance (secondary outcomes). The Pain Interference Subscale of the Multidimensional Pain Inventory[24] assessed interference with general functioning due to pain. The scale has shown excellent psychometric characteristics in past research.[29] In the present sample, the Cronbach's alpha was 0.92 at the baseline assessment. Pain intensity was assessed with a 0 to 10 numerical rating scale of average pain over the past week.[20] It also shows excellent psychometric characteristics. Depressive symptoms were assessed with the Center for Epidemiologic Studies–Depression (CESD) Scale Short Form, also reported to have excellent psychometric qualities in past research.[8] In the present sample, the Cronbach's alpha was 0.83 at the baseline assessment. The level of physical activity was assessed with the PF-10 Physical Functioning scale from the SF36,[42] again, a psychometrically excellent measure. In the present sample, the Cronbach's alpha was 0.88 at the baseline assessment. Sleep disturbance was assessed with the 6-item the Patient-Reported Outcomes Measurement Information System Sleep Disturbance Scale. It has shown excellent psychometric characteristics in past research.[46] In the present sample, the Cronbach's alpha was 0.88 at the baseline assessment.

Power Analysis

Because a closed form approximation of power calculations would be difficult to derive, we conducted a series of mathematical simulations under a variety of situations (eg, different effect sizes, extent of missing data, and assuming each theoretical model is true). Simulations were informed by our experience with clinical trials and the outcomes under study. We originally planned to recruit 460 participants, assuming 15% attrition. Assuming that the TAU group could still exhibit a small reduction in pain interference (d = 0.10 SD pre–post change), whereas the other treatments would exhibit larger effects (d~0.85), the omnibus tests of intervention effectiveness had adequate statistical power (power ≥ 0.90) under all of the considered patterns of missing data, even taking into account a 15% attrition rate.

Statistical Analysis

Primary Analyses. Given the longitudinal and nested nature of the data (observations [level I] nested within participants [level II]), linear mixed models were conducted to examine (1) within-group and between-group changes in outcomes over time from baseline through posttreatment (Session #8), (2) within-group and between-group changes from posttreatment to 6-month follow-up, and (3) within-group and between-group changes from baseline through 6-month follow-up. Intent-to-treat procedures were conducted using linear mixed models. These models were estimated with full maximum likelihood estimation methods; a procedure which allows all randomly assigned participants to be included in analyses, thereby yielding unbiased parameter estimates for missing data (MCAR/MAR).[1,6,14,33]

Group × Time Analyses. To determine whether groups differed significantly in degrees of change in outcomes variables over time, linear mixed models that included a cross-level interaction between group (level II factor; BT, CT, MBSR, and TAU) and time (level I, covariate) were estimated with a diagonal covariance structure and included a fixed and random intercept for each outcome. Significant Group × Time interactions were dissected by estimating pairwise comparison linear mixed models for each group pairing (BT vs TAU; CT vs TAU; MBSR vs TAU; BT vs CT; BT vs MBSR; and CT vs MBSR). Those comparisons with a significant interaction were then plotted and probed to determine the significance of simple slopes and the region of significance for the interaction.[30]

Regions of Significance. Analyzing regions of significance allowed us to examine the relative rates in which the change occurred between the active treatments. Although the treatments may produce similar degrees of change by the end of treatment, a given treatment may produce the final degree of posttreatment change before the final session, whereas another treatment may not produce this degree of change until the final session. The former treatment may be seen as superior to the other, slower treatment for producing the final level of pretreatment to posttreatment effects sooner and thus incurring less cost and patient burden. We hypothesized that one or more treatments could produce more rapid change than the other treatment(s). "Rate of change" is defined here as how quickly a given treatment will produce significant differences compared with TAU on a given outcome. For instance, CT may significantly differ from TAU on pain intensity changes by session 5, whereas BT may not differ from TAU until session 8. Such effects could be taken as evidence that CT was superior to BT on this new metric.

Within-group Analyses. Any significant interactions were further dissected to determine changes in outcomes over time for each group. Individual linear mixed models were estimated for each group and outcome with a diagonal covariance structure and included a fixed and random intercept and a fixed effect of time (level I, covariate).

Between-group Effect Sizes. In addition, following recommendations by Feingold,[13] effect sizes for differences between treatments and TAU on outcome measures at session 8 were calculated by transforming the coefficient of the slope difference into a standardized mean difference (Cohen's d):

where b is the unstandardized coefficient of the slope difference (interaction term), duration is the length of the study based on units associated with the coefficient (# of sessions minus one), and SD is the pooled within-group standard deviation. Confidence intervals for these effect sizes were calculated following recommendations from Feingold[14] which substitute the lower and upper confidence interval of the unstandardized slope difference coefficient in the following equations:

where LCLb is the lower confidence interval of the unstandardized slope difference coefficient (interaction term), UCLb is the upper confidence interval of the unstandardized slope difference coefficient (interaction term), duration is the length of the study based on units associated with the coefficient (# of sessions minus one), and SD is the pooled within-group standard deviation.

Posttreatment to 6-month Follow-up Analyses and Baseline to Six-month Analyses. A second series of multilevel models were conducted to evaluate posttreatment to six-month follow-up effects. Linear mixed models that included a 2-way interaction between group (level II factor; BT, CT, MBSR, and TAU) and time (level II factor) were estimated with a diagonal covariance structure and included a fixed and random intercept for each outcome. Significant overall Group × Time interactions were dissected by estimating pairwise comparison linear mixed models for each group pairing (BT vs TAU; CT vs TAU; MBSR vs TAU; BT vs CT; BT vs MBSR; and CT vs MBSR). Significant pairwise interactions were plotted and probed in a means-as-outcomes model.[30,31] Within-group effects were estimated with linear mixed models for each group separately with a diagonal covariance structure, a fixed and random intercept, and a fixed effect of time (level II factor). Evaluation of significant within-group effects were based on comparing the estimated marginal means from these models at posttreatment to six-month follow-up.

A third series of multilevel models were conducted to evaluate baseline to 6-month follow-up effects. These analyses were conducted in a similar fashion to the baseline to posttreatment analyses.