A Comparison of the Effects of 2 Types of Massage and Usual Care on Chronic Low Back PainFREE
Few studies have evaluated the effectiveness of massage for chronic low back pain.
To compare the effectiveness of 2 types of massage and usual care for chronic back pain.
Parallel-group randomized, controlled trial. Randomization was computer-generated, with centralized allocation concealment. Participants were blinded to massage type but not to assignment to massage versus usual care. Massage therapists were unblinded. The study personnel who assessed outcomes were blinded to treatment assignment. (ClinicalTrials.gov registration number: NCT00371384)
An integrated health care delivery system in the Seattle area.
401 persons 20 to 65 years of age with nonspecific chronic low back pain.
Structural massage (n = 132), relaxation massage (n = 136), or usual care (n = 133).
Roland Disability Questionnaire (RDQ) and symptom bothersomeness scores at 10 weeks (primary outcome) and at 26 and 52 weeks (secondary outcomes). Mean group differences of at least 2 points on the RDQ and at least 1.5 points on the symptom bothersomeness scale were considered clinically meaningful.
The massage groups had similar functional outcomes at 10 weeks. The adjusted mean RDQ score was 2.9 points (95% CI, 1.8 to 4.0 points) lower in the relaxation group and 2.5 points (CI, 1.4 to 3.5 points) lower in the structural massage group than in the usual care group, and adjusted mean symptom bothersomeness scores were 1.7 points (CI, 1.2 to 2.2 points) lower with relaxation massage and 1.4 points (CI, 0.8 to 1.9 points) lower with structural massage. The beneficial effects of relaxation massage on function (but not on symptom reduction) persisted at 52 weeks but were small.
Participants were not blinded to treatment.
Massage therapy may be effective for treatment of chronic back pain, with benefits lasting at least 6 months. No clinically meaningful difference between relaxation and structural massage was observed in terms of relieving disability or symptoms.
Primary Funding Source:
National Center for Complementary and Alternative Medicine.
Massage therapy is widely used for back pain.
Patients with chronic low back pain who received 10 weekly sessions of structural or relaxation massage therapy experienced clinically important improvements in symptoms and disability compared with patients who received usual care. There was no apparent difference in effect between the 2 types of massage.
Participants assigned to usual care knew that they were not receiving massage therapy and may have been more likely to report less improvement.
Massage therapy seemed to be an effective short- and longer-term treatment for chronic low back pain. There was no apparent difference between relaxation and structural massage, the latter of which may be more expensive and requires specialized techniques.
Massage is one of the most popular complementary and alternative medical therapies for neck and back pain (1), conditions that account for more than one third (2) of the more than 100 million annual visits to massage therapists in the United States (3). Almost all massage therapists in the United States use Swedish massage techniques aimed at relaxation, but only a minority take courses in such techniques as structural massage for treatment of chronic low back pain. Recent reviews have found limited evidence that massage is an effective treatment of chronic back pain (4, 5), and no studies have compared relaxation massage with structural massage, which focuses on correcting soft-tissue abnormalities. We therefore conducted a trial to determine whether relaxation massage reduces pain and improves function in patients with chronic low back pain and compared relaxation and structural massage for treating this condition.
Study procedures are described in detail elsewhere (6). In brief, we used advertisements in the health plan magazine of the Group Health Cooperative and mailings to invite members of the plan who were aged 20 to 65 years and had outpatient visit diagnoses suggesting nonspecific chronic low back pain to participate in the trial. Invitation letters were mailed 3 to 12 months after visits. Study staff phoned respondents to determine their eligibility; criteria were low back pain lasting at least 3 months without 2 or more pain-free weeks and pain bothersomeness rated at least 3 on a scale of 0 to 10. Exclusion criteria were 1) specific causes of back pain (for example, cancer, fractures, or spinal stenosis), 2) complicated back problems (for example, sciatica, back surgery in the past 3 years, or medicolegal issues), 3) conditions making treatment difficult (for example, paralysis or psychoses), 4) conditions that might confound treatment effects or interpretation of results (for example, severe fibromyalgia or rheumatoid arthritis), 5) inability to speak English, 6) massage within the past year, or 7) plans to visit a provider for back pain. Eligible respondents were administered a baseline questionnaire and randomly assigned to a treatment group. All participants gave written informed consent, and the study was approved by the institutional review board of Group Health Cooperative.
Randomization and Interventions
The randomization schedule was created by a biostatistician and was blocked on massage therapist; allocation was centrally generated and concealed. Participants were randomly allocated in a 1:1:1 ratio to receive usual care, relaxation massage, and structural massage. Therapists were not blinded to the type of massage that they provided. Participants knew whether they received massage but were blinded to type; usual care recipients were aware that they had enrolled in a trial of massage. Study personnel assessing trial outcomes were blinded to study assignment.
Massage was provided by 27 licensed therapists with at least 5 years of experience who were comfortable following the study protocol and had experience in the permitted techniques. Therapists received 1.5 days of protocol training (6). Treatment fidelity was promoted by monitoring treatment forms that the therapists completed at each visit, along with corrective feedback and a midstudy meeting.
Both massage techniques were provided in the therapists' offices at no cost. Both massage protocols prescribed 10 weekly treatments, with first visits lasting 75 to 90 minutes and follow-up visits lasting 50 to 60 minutes. We defined “adherence” as completion of at least 8 visits. At each visit, therapists could recommend up to 3 home exercises from a predefined list of 7 exercises, 6 of which were common to both treatments (6).
Relaxation massage, which is intended to induce a generalized sense of relaxation, comprised effleurage, petrissage, circular friction, vibration, rocking and jostling, and holding. Therapists were given time limits for each body region, including 7 to 20 minutes on the back and buttocks (6). Therapists could provide a compact disk of a 2.5-minute relaxation exercise to be done at home to enhance and prolong treatment benefits.
Structural massage, which is intended to identify and alleviate musculoskeletal contributors to back pain, comprised myofascial, neuromuscular, and other soft-tissue techniques (6). Myofascial techniques are intended to engage and release identified restrictions in myofascial tissues. Neuromuscular techniques are used to resolve soft-tissue abnormalities by mobilizing restricted joints, lengthening constricted muscles and fascia, balancing agonist and antagonist muscles, and reducing hypertonicity. The areas of the body that were treated varied across patients and treatment sessions. Therapists could recommend a home exercise consisting of psoas stretch to enhance and prolong any benefits of structural massage.
Usual care participants received no special care but were paid $50. Actual care was determined from medical records and interviews.
Outcomes and Follow-up
Outcomes were measured at baseline and at 10, 26, and 52 weeks by interviewers who were masked to treatment assignment. Prespecified primary outcomes were back pain–related dysfunction and symptoms at 10 weeks, immediately after treatment completion.
Dysfunction was measured by using the modified Roland Disability Questionnaire (RDQ), a reliable, valid, and sensitive measure with substantial construct validity (7, 8). A between-group difference in improvement in mean values of at least 2 points on this scale of 0 to 23 points was considered clinically meaningful (8, 9).
Symptoms were assessed by asking participants to rate the bothersomeness of their pain during the past week from 0 (“not at all”) to 10 (“extremely”). A between-group difference in improvement in mean values of at least 1.5 points was considered clinically meaningful (8, 9).
Secondary outcomes were 1) 26- and 52-week measures of the primary outcomes; 2) percentage of participants with prespecified clinically meaningful, individual-level reductions in dysfunction (≥3-point decrease in RDQ score) and symptoms (≥2-point decrease in symptom bothersomeness score); 3) physical and mental health Short Form-12 Health Survey component summary scores (10); 4) self-reported medication use for back pain in the previous week; 5) days spent in bed, home from work or school, or cutting down on usual activities because of back problems during the past week (11); 6) global improvement in back pain–related dysfunction, rated on a 7-point scale from “completely gone” to “much worse”; 7) the patient's feelings about spending the rest of his or her life with the back pain experienced in the past week (7-point scale ranging from “delighted” to “terrible”); 8) satisfaction with back care, rated by using a 5-point Likert scale (ranging from “very satisfied” to “very dissatisfied”); and 9) total costs of back pain–related visits, imaging studies, and medications during the follow-up year, as ascertained from electronic medical records. Visits not covered by the health plan were identified from interviews. Finally, participants were asked open-ended questions about adverse experiences on the 10-week interview and at each massage visit.
Sample size was calculated to ensure adequate power at 10 weeks to detect a clinically meaningful 2-point mean difference between the structural massage and usual care groups in RDQ scores, assuming that the structural massage group would have scores that were 1 point better than the relaxation massage group. Assuming 10% loss to follow-up and SDs that were derived from our pilot study of 92 participants (SD, 4.65 for RDQ score and 2.71 for symptom bothersomeness score), we estimated that a target sample of 399 persons (133 per group) was needed. For RDQ scores, this provided 85% power to detect a significant difference among the 3 treatment groups by using an omnibus Wald test and 91% power to detect a pairwise difference of 2 points; for symptom bothersomeness scores, it provided greater than 99% power to detect a significant difference among the 3 treatment groups and 99% power to detect a pairwise difference of 1.5 points.
Analyses were conducted according to the original randomized treatment assignment regardless of adherence to protocol. Analyses were conducted by using regression through generalized estimating equations (12) with an independent working correlation structure and robust SE estimates taking into account multiple outcomes per participant. Follow-up times were treated as categorical variables using dummy variables for each treatment, each time point, and all 2-way interactions between follow-up time and treatment. Adjusted models included baseline covariates that were prespecified, were imbalanced at baseline (that is, potential confounders), or were associated with a primary outcome (that is, precision variables): age, group, sex, baseline RDQ and symptom bothersomeness scores, education level, body mass index, type of work, original cause of back pain, more than 7 days of reduced activities because of back pain, and medication use in the previous week. We prespecified the adjusted analysis as the primary analysis. For continuous and binary outcome measures, we applied linear and modified Poisson regression, respectively, with robust SEs. Modified Poisson regression allows estimation of relative risks for nonrare outcomes using Poisson regression and corrects the misspecification of the variance using robust SEs in a generalized estimating equation framework (13).
To control for multiple comparisons, we used the least-significant-difference approach, in which pairwise treatment comparisons were evaluated at a given time only if the overall omnibus P value was statistically significant at 0.05. Mean differences, 95% CIs, and omnibus P values for treatment group effect and pairwise significance are presented.
To assess effects of individual providers on the RDQ outcome, we fitted an adjusted mixed-effects model with a random intercept for each provider by using only data from the 2 massage groups. The intraclass correlation coefficient was calculated to quantify the degree of variability due to providers relative to the overall variability of the outcome. Analyses were performed by using SAS software, version 9.2 (SAS Institute, Cary, North Carolina), and 2-sided P values were calculated.
The proportions of missing data for the primary outcome measures were 5% at 10 weeks, 7% at 26 weeks, and 9% at 52 weeks. We used Stata, version 10.1 (StataCorp, College Station, Texas) (14), to perform multiple imputation by chained equations to account for missingness. We imputed both missing outcomes and baseline covariates. Estimates from each imputed data set were combined by following rules outlined by Rubin (15, 16). Imputed analysis did not yield appreciable differences from the unadjusted and adjusted complete-case analyses (results not shown). Specifications for missing data imputation are shown in the Appendix.
Role of the Funding Source
This trial was funded by the National Center for Complementary and Alternative Medicine (NCCAM) and was approved their Office of Clinical and Regulatory Affairs. The NCCAM did not participate in the research.
Study Recruitment and Follow-up
We mailed 9127 invitations and evaluated 1161 responses; 402 persons (35%) were eligible and were randomly allocated (Figure 1). The main reasons for ineligibility were massage in previous year (22%), less than 3 months of back pain (20%), sciatica (14%), medical contraindications (12%), and planned visits for back pain (10%). One participant with an abdominal aortic aneurysm was excluded after randomization. Analyses included 401 participants assigned to relaxation massage (n = 136), structural massage (n = 132), or usual care (n = 133). Follow-up was 95% at 10 weeks, 93% at 26 weeks, and 91% at 52 weeks. Interviewers reported that treatment group (massage or usual care) had been revealed to them by fewer than 10% of participants.
Participants were mostly middle-aged, female, and white; 51% were college graduates (Table 1). Mean scores of 10.8 on the RDQ and 5.7 for symptom bothersomeness indicated moderately severe back problems. Participants assigned to relaxation massage had greater dysfunction (RDQ score >1 point higher), but there were no other clinically or statistically significant differences among the groups at baseline.
Treatment adherence was 93% in the relaxation massage group and 88% in the structural massage group. Therapist recommendations for the 6 home exercise options available to both treatments were similar (Appendix Table 1). For relaxation massage participants, therapists recommended the relaxation CD at 41% of first visits. For structural massage participants, therapists recommended the psoas stretch exercise at 8% of first visits. Therapists rated participants' mean level of adherence to self-care recommendations (mostly stretching exercises) in both groups as 7.0 on a scale of 0 to 10. After the first visit, therapists had similar expectations of the helpfulness for participants of relaxation and structural massage (6.5 and 7.2, respectively, on a scale of 0 to 10).
All groups showed improved function and decreased symptoms at 10 weeks (Figure 2 and Table 2), but improvement was greater with either type of massage. Compared with those of usual care recipients, RDQ scores were 2.9 points (95% CI, 1.8 to 4.0 points) lower among relaxation massage recipients and 2.5 points (CI, 1.4 to 3.5 points) lower among structural massage recipients (P < 0.001 for both adjusted estimates). There was no difference in function between the 2 types of massage (adjusted difference, 0.5 point [CI, −0.5 to 1.5 points]; P = 0.35), and the confidence bounds around the adjusted estimate of absolute difference between massage groups excluded values large enough to be considered clinically relevant (2-point difference in RDQ score and 1.5-point difference in symptom bothersomeness score [8, 9]). Similar results were found for symptom bothersomeness.
Effects decreased after the 10-week treatment (Figure 2), although differences in functional improvement among the groups remained statistically significant at 26 and 52 weeks (Table 2). At 26 weeks, participants in the usual care group continued to function less well than recipients of relaxation massage (by 1.8 RDQ points [CI, 0.6 to 3.0 points]) and structural massage (by 1.4 RDQ points [CI, 0.3 to 2.6 points]). There were no clinically or statistically significant differences between types of massage. At 52 weeks, relaxation massage was modestly more effective than structural massage (by 1.1 RDQ points [CI, 0.02 to 2.2 points]). No significant differences in symptoms were observed among the 3 groups at 26 or 52 weeks.
Results were similar when primary outcomes were measured as the percentage of participants who had clinically meaningful improvement. At 10 weeks, 62% to 65% of massage recipients and 38% of usual care recipients had clinically meaningful improvement (adjusted overall P < 0.001) (Figure 3). The benefits of massage decreased over time and were not statistically significant at 52 weeks. Massage recipients were more likely than participants in the usual care group to experience clinically meaningful reductions in symptom bothersomeness at 10 weeks.
At 10 weeks, the groups differed significantly in limitations in activities (days in bed, reduced activity, and days off work), patient global rating of improvement, use of nonsteroidal anti-inflammatory medications, satisfaction with current level of back pain, satisfaction with back pain care, and mental health (Short Form-12 Health Survey) (Appendix Table 2). The 2 massage treatments had similar effects that were generally greater than those of usual care. Most notably, 36% to 39% of participants receiving massage and only 4% receiving usual care claimed that their back pain was much better or gone at 10 weeks. Massage did not affect use of narcotic analgesics. Massage benefits persisted at 52 weeks for days of reduced activity, global improvement, and satisfaction.
Analysis of massage therapist effect yielded an intraclass correlation coefficient of 0.007, indicating little variability among therapists compared with the overall variability of the RDQ outcome.
Co-interventions and Subsequent Use of Massage
At 10 weeks, 33% of usual care recipients and 18% to 20% of massage recipients reported visiting a provider for back pain since randomization. Eight percent of usual care participants reported a massage visit. Between 10 and 26 weeks, 13% of usual care recipients, 21% of relaxation massage recipients, and 8% of structural massage recipients reported back pain–related massage visits. At 52 weeks, use of massage for back pain during the previous 6 months was 18%, 26%, and 22%, respectively, for these groups. Two participants receiving usual care and 1 receiving structural massage had back surgery.
Cost of Back Pain–Related Health Care After Randomization
The massage treatments received by the average participant would have cost about $540 in the community. There is no evidence that these treatments reduced costs of back pain–related health care services during the 1-year posttreatment period. Median health care costs were $25 (range, $0 to $8082) for usual care recipients, $78 (range, $0 to $3764) for relaxation massage recipients, and $38 (range, $0 to $1443) for structural massage recipients.
Five of 134 (4%) relaxation massage recipients and 9 of 131 (7%) structural massage recipients reported adverse events possibly related to massage, mostly increased pain. One event in the structural massage group (nausea, shortness of breath, and chest pain) was classified as serious and considered unrelated to treatment.
We found that massage therapy improved function and decreased pain more than usual care in patients with uncomplicated chronic lower back pain after 10 weeks. A course of relaxation massage, using techniques commonly taught in massage schools and widely used in practice, had effects similar to those of structural massage, a more specialized technique. Both types of massage were associated with improved function at 26 weeks; benefits at 1 year were of questionable clinical significance. We found no evidence of differential effectiveness among the massage therapists, and both relaxation and structural massage had very low rates of adverse effects.
A recent review identified 13 trials published before May 2008 (5) that evaluated the effect of massage on nonspecific back pain. We identified 2 additional trials (17, 18) in MEDLINE as of February 2011. Four trials (19–22) were conducted in North America and evaluated only techniques that were similar to the structural massage treatment in our trial; all found positive results. None of the previous trials evaluated relaxation massage techniques that are commonly taught in massage schools and widely practiced in North America.
The mechanisms explaining the beneficial effects of relaxation and structural massage remain unclear. These distinct forms of massage may trigger similar physiologic effects (for example, through local stimulation of tissue or a generalized central nervous system response) or may work through different mechanisms (for example, structural massage may foster beneficial changes in the treated soft tissues, whereas relaxation massage may operate through the central nervous system). It is also possible that improvements in pain and function are due to nonspecific effects, such as time spent in a relaxing environment, being touched, receiving care from a caring therapist, being given self-care advice, or increased body awareness (19). A combination of these explanations is also possible.
Our study has limitations. Participants assigned to usual care were told that they were enrolling in a trial of massage therapy, and they often received no additional treatment. This potential failure of blinding to treatment assignment may have led to less favorable self-assessments of function and symptoms, making massage therapy seem more superior than it really is. The exercises recommended in the 2 massage groups differed slightly, which could have contributed to differences in outcomes even though we could not detect a clinically meaningful difference between the 2 types. The massage therapists in our study may have been atypical, having practiced for at least 5 years and learned structural massage techniques. We believed that the study therapists might have provided treatment in a way that favored structural massage because that form of therapy requires more specialized training, but the apparent absence of a difference between massage types makes this possibility unlikely. The trial included mostly women with nonspecific chronic low back pain who were enrolled in a single health care system that serves a mostly white and employed population, which limits the generalizability of our findings. Finally, persons with specific causes of back pain, such as disc herniation, who conceivably could benefit more from structurally focused massage were excluded.
The uncertainties associated with these limitations make it difficult to determine the true magnitude of the benefits of massage observed in this trial. However, our trial also has strengths, including a large sample size, comparison of 2 massage techniques, inclusion of a control group, having the same therapists deliver both treatments, high rates of treatment adherence and follow-up, and long-term follow-up.
In summary, our findings suggest that both relaxation massage and structural massage are reasonable treatment options for persons with chronic low back pain. The findings may suggest a relative advantage for relaxation massage because it is based on techniques that are taught in almost all massage schools and is thus more readily accessible and slightly less expensive than structural and other more specialized forms of massage, which require additional training. Future research should explore the relative contributions of nonspecific context effects and specific treatment effects on outcomes for patients with low back pain who are receiving practitioner-based therapies, such as massage; whether different forms of massage produce benefits through the same or through different physiologic pathways; whether less experienced therapists would produce similar results; whether fewer treatments could have achieved the same outcomes; and whether education and self-care recommendations contribute to the effectiveness of massage.
Wolsko PM, Eisenberg DM, Davis RB, Kessler R, Phillips RS. Patterns and perceptions of care for treatment of back and neck pain: results of a national survey. Spine (Phila Pa 1976). 2003;28:292-7. [PMID: 12567035] CrossrefMedlineGoogle Scholar
Cherkin DC, Deyo RA, Sherman KJ, Hart LG, Street JH, Hrbek A, et al. Characteristics of visits to licensed acupuncturists, chiropractors, massage therapists, and naturopathic physicians. J Am Board Fam Pract. 2002;15:463-72. [PMID: 12463292] MedlineGoogle Scholar
Eisenberg DM, Davis RB, Ettner SL, Appel S, Wilkey S, Van Rompay M, et al. Trends in alternative medicine use in the United States, 1990-1997: results of a follow-up national survey. JAMA. 1998;280:1569-75. [PMID: 9820257] CrossrefMedlineGoogle Scholar
Chou R, Qaseem A, Snow V, Casey D, Cross JT, Shekelle P, et al; Clinical Efficacy Assessment Subcommittee of the American College of Physicians. Diagnosis and treatment of low back pain: a joint clinical practice guideline from the American College of Physicians and the American Pain Society. Ann Intern Med. 2007;147:478-91. [PMID: 17909209] LinkGoogle Scholar
Furlan AD, Imamura M, Dryden T, Irvin E. Massage for low back pain: an updated systematic review within the framework of the Cochrane Back Review Group. Spine (Phila Pa 1976). 2009;34:1669-84. [PMID: 19561560] CrossrefMedlineGoogle Scholar
Cherkin DC, Sherman KJ, Kahn J, Erro JH, Deyo RA, Haneuse SJ, et al. Effectiveness of focused structural massage and relaxation massage for chronic low back pain: protocol for a randomized controlled trial. Trials. 2009;10:96. [PMID: 19843340] CrossrefMedlineGoogle Scholar
Bombardier C. Outcome assessments in the evaluation of treatment of spinal disorders: summary and general recommendations. Spine (Phila Pa 1976). 2000;25:3100-3. [PMID: 11124724] CrossrefMedlineGoogle Scholar
Patrick DL, Deyo RA, Atlas SJ, Singer DE, Chapin A, Keller RB. Assessing health-related quality of life in patients with sciatica. Spine (Phila Pa 1976). 1995;20:1899-908. [PMID: 8560339] CrossrefMedlineGoogle Scholar
Hägg O, Fritzell P, Nordwall A; Swedish Lumbar Spine Study Group. The clinical importance of changes in outcome scores after treatment for chronic low back pain. Eur Spine J. 2003;12:12-20. [PMID: 12592542] CrossrefMedlineGoogle Scholar
Ware J, Kosinski M, Keller SD. A 12-Item Short-Form Health Survey: construction of scales and preliminary tests of reliability and validity. Med Care. 1996;34:220-33. [PMID: 8628042] CrossrefMedlineGoogle Scholar
Riess PW. Current Estimates from the National Health Interview Survey: United States, 1984. Vital and Health Statistics Series 10, No. 156. Hyattsville, MD: National Center for Health Statistics; 1986. Google Scholar
Liang KY, Zeger SL. Longitudinal data analysis using generalized linear models. Biometrika. 1986;73:13-22. CrossrefGoogle Scholar
Zou G. A modified poisson regression approach to prospective studies with binary data. Am J Epidemiol. 2004;159:702-6. [PMID: 15033648] CrossrefMedlineGoogle Scholar
van Buuren S, Boshuizen HC, Knook DL. Multiple imputation of missing blood pressure covariates in survival analysis. Stat Med. 1999;18:681-94. [PMID: 10204197] CrossrefMedlineGoogle Scholar
Rubin DB. Multiple Imputation for Nonresponse in Surveys. New York: J Wiley; 1987. Google Scholar
Li KH, Raghunathan TE, Rubin DB. Large sample significance levels from multiply imputed data using moment-based statistics and an F reference distribution. J Am Stat Assoc. 1991;86:1065-73. Google Scholar
Cherkin DC, Eisenberg D, Sherman KJ, Barlow W, Kaptchuk TJ, Street J, et al. Randomized trial comparing traditional Chinese medical acupuncture, therapeutic massage, and self-care education for chronic low back pain. Arch Intern Med. 2001;161:1081-8. [PMID: 11322842] CrossrefMedlineGoogle Scholar
Hernandez-Reif M, Field T, Krasnegor J, Theakston H. Lower back pain is reduced and range of motion increased after massage therapy. Int J Neurosci. 2001;106:131-45. [PMID: 11264915] CrossrefMedlineGoogle Scholar
Preyde M. Effectiveness of massage therapy for subacute low-back pain: a randomized controlled trial. CMAJ. 2000;162:1815-20. [PMID: 10906914] MedlineGoogle Scholar
Field T, Hernandez-Reif M, Diego M, Fraser M. Lower back pain and sleep disturbance are reduced following massage therapy. J Bodyw Mov Ther. 2007;11:141-5. CrossrefGoogle Scholar
Kumnerddee W. Effectiveness comparison between Thai traditional massage and Chinese acupuncture for myofascial back pain in Thai military personnel: a preliminary report. J Med Assoc Thai. 2009;92 Suppl 1S117-23. [PMID: 21299184] MedlineGoogle Scholar
Little P, Lewith G, Webley F, Evans M, Beattie A, Middleton K, et al. Randomised controlled trial of Alexander technique lessons, exercise, and massage (ATEAM) for chronic and recurrent back pain. BMJ. 2008;337:a884. [PMID: 18713809] CrossrefMedlineGoogle Scholar
Royston P. Multiple imputation of missing values. Stata J. 2004;4:227-41. CrossrefGoogle Scholar
Royston P. Multiple imputation of missing values: update. Stata J. 2005;5:188-201. CrossrefGoogle Scholar
Royston P. Multiple imputation of missing values: update of ice. Stata J. 2005;5:527-536. CrossrefGoogle Scholar
Appendix: Specifications for Missing Data Imputation
Description of Missing Data
Overall, data were complete at all 3 follow-up time points for 356 (89%) participants (Appendix Table 3). However, 385 (96%) participants had complete data for at least 1 follow-up time point. Of the 45 participants with missing values for primary outcomes, 21 were missing values only at 1 time point, 8 were missing values at 2 time points, and 16 were missing values at all 3 time points (Table 1). Only 3 participants were missing covariate information for the subset of variables used in adjusted models; none of those participants was missing data on primary outcomes. Missing outcome data seemed to be primarily related to age, education, assigned treatment group, and baseline RDQ score (Appendix Table 4).
Imputation Methods and Models
We used imputation by chained equations, which was originally proposed by van Buuren and colleagues (14) and coded for use in Stata 10 (23–25). Imputation was performed for both outcome variables and covariates used for adjustment. Twenty imputed data sets were generated by using 100 cycles per imputation and the bootstrap option, which relaxes the assumption of normality on the variable estimates. The 20 imputed data sets were imported into SAS, version 9.2, for analysis by using the rules of Rubin for combining results from separate analyses on each imputed data set. Omnibus hypothesis tests were conducted by using an F reference distribution as specified by Li and colleagues (16). All scalar variables were tested by using a t reference distribution as specified by Rubin (25).
Author, Article, and Disclosure Information
Daniel C. Cherkin,
From Group Health Research Institute and University of Washington, Seattle, Washington; University of Vermont, Burlington, Vermont; and Oregon Health & Science University, Portland, Oregon.
Acknowledgment: The authors thank the following persons for their contributions to the successful completion of this project: Marissa Brooks, Alexis Williams Coatney, Erika Holden, Juanita Jackson, Christel Kratohvil, Mary Lyons, Melissa Parson, Dawn Schmidt, Diana Thompson, the 27 licensed massage therapists who provided the study treatments, the members of the Massage Therapy Research Consortium who helped develop the treatment protocols, and the managers and interviewers of the Group Health Research Institute Survey Program.
Grant Support: By the NCCAM (grant R01 AT 001927).
Disclosures: Disclosures can be viewed at www.acponline.org/authors/icmje/ConflictOfInterestForms.do?msNum=M10-2933.
Corresponding Author: Daniel C. Cherkin, PhD, Group Health Research Institute, 1730 Minor Avenue, Suite 1600, Seattle, WA 98101.
Current Author Addresses: Drs. Cherkin, Sherman, and Cook; Mr. Wellman; Mr. Johnson; Ms. Erro; and Ms. Delaney: Group Health Research Institute, 1730 Minor Avenue, Suite 1600, Seattle, WA 98101.
Dr. Deyo: Department of Family Medicine, Oregon Health & Science University, Mail Code FM, 3181 SW Sam Jackson Park Road, Portland, OR 97239-3098.
Author Contributions: Conception and design: D.C. Cherkin, K.J. Sherman, J. Kahn.
Analysis and interpretation of the data: D.C. Cherkin, K.J. Sherman, J. Kahn, R. Wellman, A.J. Cook, E. Johnson, J. Erro, K. Delaney, R.A. Deyo.
Drafting of the article: D.C. Cherkin, K.J. Sherman, R. Wellman, A.J. Cook, K. Delaney.
Critical revision of the article for important intellectual content: D.C. Cherkin, K.J. Sherman, J. Kahn, R. Wellman, A.J. Cook, E. Johnson, J. Erro, R.A. Deyo.
Final approval of the article: D.C. Cherkin, K.J. Sherman, J. Kahn, R. Wellman, E. Johnson, R.A. Deyo.
Statistical expertise: R. Wellman, A.J. Cook, E. Johnson.
Obtaining of funding: D.C. Cherkin, K.J. Sherman.
Administrative, technical, or logistic support: J. Kahn, J. Erro, K. Delaney.
Collection and assembly of data: K.J. Sherman, R. Wellman, J. Erro, K. Delaney.