Background:

Lumbar spinal stenosis (LSS) is a highly prevalent condition among older adults and the most frequent indication for spinal surgery in patients older than the age of 65. In this past decade the fastest growth in lumbar surgery in the United States has occurred in older adults with LSS, and the rate of complex fusion procedures has significantly increased. These operations are associated with significant health care costs, risks, complications, and rehospitalization rates. Yet, evidence is lacking for the effectiveness of the various nonsurgical treatment options offered to patients with LSS. This study was designed to help bridge this evidence gap.

Objective:

Compare the clinical effectiveness of 3 common nonsurgical approaches to the management of patients with LSS: (1) medical care (MC) provided by a physiatrist; (2) nonspecific group exercise (GE) classes provided by certified exercise instructors; or (3) a combination of manual therapy and individualized exercises (MTE) provided by chiropractors and physical therapists.

Methods:

Randomized controlled clinical trial of 259 patients with LSS. Patients were community-dwelling older adults (≥60 years of age) recruited from the Pittsburgh metro area. We confirmed diagnosis of LSS by both diagnostic imaging (MRI or CT) and symptoms of neurogenic claudication. Participants were randomized into 1 of the 3 groups described above and treated for a total of 6 weeks. Participants in the GE and MTE groups had a total of 12 treatment sessions; those in the MC group had a total of 3 treatment sessions. The primary outcome measures were self-reported pain/function measured by the Swiss Spinal Stenosis (SSS) questionnaire and walking performance measured by the Self-paced Walking Test (SPWT). The secondary outcome measure was daily physical activity measured by accelerometry. We took outcome measures at baseline as well as 2 months and 6 months from baseline. The primary end point was at 2 months. The primary analysis used linear mixed models to compare changes in each outcome measure between the groups. The secondary analysis was a comparison of the proportion of responders (≥30% change) in each outcome measure by group, using the chi-square test.

Results:

No serious adverse events were reported in any of the groups. At 2 months, there was a statistically significantly greater reduction in adjusted mean SSS score (range, 12-55) in the MTE group compared with MC (2.1; 95% CI, 0.3-3.9) or GE (2.4; 95% CI, 0.6-4.3). The minimum clinically important difference (MCID) for the SSS is 3.02 points; therefore the between-group SSS differences were not clinically significant. The adjusted mean differences in SPWT scores at 2 months favored MTE compared with MC (135.1; 95% CI, −17.2 to 287.4) or GE (46.2; 95% CI, −110.9 to 203.4), but these between-group SPWT differences were not statistically significant. GE showed significantly greater improvement in adjusted mean physical activity at 2 months compared with MC (30.5; 95% CI, 3.1-57.9), but clinical significance is unknown due to the lack of an established MCID for physical activity. The MTE group had significantly more SSS (20%) and SPWT (65.3%) responders at 2 months compared with MC (7.6%; 48.7%) or GE (3%; 46.2%) (P = .002 and P = .04, respectively). We prespecified responders as those participants who showed ≥30% improvement from baseline on the measured outcome. At 6 months, there were no longer significant between-group differences on any outcome measures. There was a general trend toward short-term improvement in SSS and physical activity that was not sustained over time; however, all groups maintained their improvements in walking performance (SPWT) at 6 months.

Study Limitations:

There were a greater of proportion of GE dropouts immediately after randomization and a potential attention bias due to the greater amount of individualized attention given to the MTE group.

Conclusions:

The combination of manual therapy and individualized exercise led to significantly greater improvement in SSS and SPWT at 2 months, whereas GE led to significantly greater improvement in physical activity at 2 months. The clinical significance of these short-term improvements is unknown.

Specific Aims (as Presented in the Original Research Application)

The study was a pragmatic comparative effectiveness trial designed as a 3-group randomized clinical trial. The main goal of this study was to provide research evidence that could better inform the choices between these 3 types of nonsurgical treatment options for patients with LSS. The 3 comparison groups in this trial were the following:

• Group 1: Usual MC*
• Group 2: Community-based GE classes
• Group 3: Clinic-based manual therapy and individualized exercise

*Note: In the design phase of this study, we interviewed primary care physicians (PCPs) and asked them about how they managed patients with LSS. They told us that they commonly followed a 2-step approach: Step 1: oral medications and advice to stay active; Step 2: referral for epidural injection. Therefore, we considered this 2-step approach to be reflective of “usual medical care.” However, the physician providing the MC in our trial was board certified in physical medicine and rehabilitation. Although we basically followed the same 2-step approach, we allowed him to follow a more pragmatic and tailored approach (described in more detail later). This led to the recognition that the words “usual medical care” were no longer the best terminology to describe this intervention group. Therefore, we decided to delete the adjective “usual” and simply use the term medical care in this report and future publications.

Primary Aim

To compare the clinical outcomes between these 3 interventions using 2 validated primary outcome measures of pain and physical function: Swiss Spinal Stenosis (SSS) questionnaire (self-report measure) and Shuttle Walk Test (performance-based measure).

Hypothesis: LSS subjects in groups 1 and 2 will demonstrate greater improvement in self-reported pain/function and walking performance compared with group 3.

*Note: We substituted the Self-paced Walking Test (SPWT) for the Shuttle Walk Test and notified PCORI of this change before we began recruitment. This change in our choice of walking performance measure occurred for 2 reasons. First, the developer of the SPWT consulted with us about our research design and presented evidence about the reliability and accuracy of the SPWT. Secondly, feedback from focus groups with stenosis patients indicated that the SPWT was a more patient-centered outcome that replicated real-life walking performance.

Secondary Aim: To compare changes in PA between these 3 treatment groups using the SenseWear Armband (real-time activity measure).

Hypothesis: Subjects in groups 1 and 2 will demonstrate a greater change in PA compared with group 3.

Exploratory Aim 1: To explore relationships between 6-month attrition rates, number of adverse events, adherence rates, number of falls, and co-interventions across treatment groups.

Hypothesis: For all 3 treatment groups there will be a low incidence of adverse events and similar treatment adherence rates. The 6-month attrition rate, number of falls, and co-interventions will be lower in groups 1 and 2 than with group 3.

Exploratory Aim 2: To explore treatment effects and responses by subgroup.

Hypothesis: A group of baseline physical, psychosocial, and demographic measures will be associated with treatment response and nonresponse in each group.

Methodological Standards Study Design

This was a 3-arm, randomized controlled trial that compared the clinical effectiveness of GE with manual therapy/individualized exercise, and each of these interventions with MC. The study was approved by the University of Pittsburgh IRB (PRO 12120422) and registered on ClinicalTrials.gov (NCT01943435).

Sample Size

We based sample-size estimation on hypothesized changes in our primary outcome measure, the SSS questionnaire. We originally calculated a total sample size of 180 subjects with an anticipated dropout rate of 15%, which would give us a final sample size of N = 150. This would give us 80% power to detect a difference as small as 3.6 points (SD, 6.1) on the total 12-item SSS score (30). In addition, this sample size also yields sufficient power to fit a regression model that detects a statistical difference in the proportion of variability explained.

More specifically, a sample size of 150 achieves 81% power to detect an R-square of 5% attributed to 2 independent variables (representing treatment) and assuming the control variables account for an additional 20% of the variability. Due to an early unanticipated success in enrollment and supplemental PCORI funding, we were able to continue recruitment for an additional 6 months and achieve a larger final sample size (N = 259). We performed no interim data analysis after reaching our original sample size of N = 180; we conducted only a single final data analysis using data from the final sample size of N = 259.

Inclusion/Exclusion Criteria

Participants were required to have been previously diagnosed with LSS by a physician and to supply an MRI or CT report in which the radiologist confirmed the presence of anatomical narrowing of the central canal, lateral recess, and/or foramen. We also confirmed the presence of the following clinical signs of LSS: leg symptoms worsened by walking/relieved by sitting, symptoms worsened by lumbar extension/relieved by flexion, and leaning forward on a shopping cart while walking to relieve leg pain. To be eligible for the study, participants were also required to (1) be aged 60 or older, (2) be able to read and write English, (3) be able to walk at least 50 feet without an assistive device, (4) have a limitation of walking due to LSS, (5) be able to engage in mild exercise, and (6) be willing to be randomized.

We excluded patients who (1) had a history of metastatic cancer, (2) had previous surgery for LSS or lumbar fusion, (3) had cauda equina symptoms, (4) had a known history of severe peripheral artery disease or exhibited an ankle-brachial index (ABI) of <0.8, (5) were told by a physician that they should not engage in physical exercise, (6) had severe hypertension (systolic >200 mm Hg or diastolic >110 mm Hg), (7) could not complete a self-paced walking test for any reason other than symptoms related to LSS, or (8) had a history of a neurological or neurodegenerative disease other than LSS that affected their ability to walk.

Recruitment

We used several strategies to recruit potentially eligible research participants from the general population of older adults in the Pittsburgh metro area. The principle investigator and research coordinator set up informational booths and attended local community health fairs that were targeted to older adults. Trifold informational brochures were produced and placed in the waiting rooms of many primary care clinics affiliated with UPMC. We also produced a full-page advertisement that we ran regularly in the Pittsburgh Senior News, a free monthly newspaper with a circulation of 20 000 that is available in public venues such as major supermarkets in the Pittsburgh region.

Our 2 most productive recruitment methods were the use of the University of Pittsburgh Clinical and Translational Science Institute (CTSI) research registry and a series of direct postcard mailings to the general public. The CTSI research registry is a voluntary database of more than 100 000 patients who consented to be contacted for potential participation in research studies. We also produced an oversize color postcard with information about our study, an conducted a targeted direct mailing to Pittsburgh residents ≥60 years of age, living within a 5-mile radius of our research facilities. Appendix B is a list of our direct mailings by ZIP code and number of residents, which totaled more than 60 000 mailings over a period of 2 years.

Our recruitment strategies were extremely successful; we achieved our goal of enrolling 180 research participants about 6 months ahead of schedule—by June 1, 2015. Because of this unanticipated early recruitment success, our program officer strongly suggested that we continue recruitment for as long as possible to increase our total sample size. We were subsequently awarded supplemental funding by PCORI, which allowed us to increase our final sample size from N = 180 to N = 259 subjects. We achieved this new recruitment milestone by the end of November 2015. In order maintain statistical integrity, we did not conduct any interim data analysis after reaching our original recruitment goal of N = 180 subjects. We conducted all analyses only after reaching the final total of N = 259 subjects.

We conducted a 2-phase screening process with those patients who contacted us expressing interest in our research study (n = 710). The first phase consisted of a telephone screening designed to filter out patients who clearly did not meet our inclusion criteria—for example those who were younger than the age of 60 years or did not have MRI evidence of LSS. Only those patients who passed the phone screening process (n = 298) were scheduled for the second phase, a baseline physical examination at our research clinic. We designed this examination to further screen out patients who had physical problems that could not be established with a phone screen, including abnormal ankle brachial index, abnormally high blood pressure, inability to walk 50 feet without an assistive device, or stopping the self-paced walking test for reasons other than LSS (eg, chest pain, shortness of breath).

Randomization

Randomization occurred immediately after the baseline screening examination confirmed that the patient was eligible to participate in the trial. We used a computerized adaptive allocation randomization algorithm to balance the 3 groups on 3 important baseline variables: (1) SSS score, (2) SPWT score, and (3) age. We based the randomization scheme on a minimization algorithm proposed by Stigsby and Taves.³⁰ The approach is to use a rank-based method to balance on multiple baseline prognostic factors.

All self-report questionnaires were converted into electronic format and completed by patients on iPads. Research staff entered data from the baseline physical examination into iPads. All data entered into the iPads were encrypted and transferred wirelessly to a secure central database, where the randomization algorithm automatically processed the data and created the group assignment. Within a few minutes of completing the baseline examination and self-report questionnaires, our research assistant received an email from the research coordinator with the group assignment. Each participant was informed about which intervention they would be receiving, and the research assistant scheduled his or her first appointment. This electronic system of randomization automated the concealment of allocation sequence and eliminated the need for double data entry.

Interventions

The MC arm of the study involved 3 visits to a physical medicine physician over 6 weeks—1 initial evaluation and 2 follow-up office visits. We designed a pragmatic treatment protocol that was based chiefly on prescription of oral medications and epidural injections. The physician reviewed each patient's currently prescribed medications for LSS and was permitted to modify/prescribe any of the following oral medications:

Nonsteroidal anti-inflammatories: ibuprofen, celecoxib, diclofenac, or misoprostol

• Adjunctive analgesics: acetaminophen or gabapentin
• Antidepressant agents: nortriptyline, duloxetine, sertraline, trazodone, or mirtazapine

In addition to prescribing any of the above oral medications, the physician had the option of referring participants for epidural steroid injections (ESIs) at 2 cooperating pain clinics in Pittsburgh. Indications for ESI referral included inadequate pain relief with oral medications, severe neurogenic claudication, or positive nerve tension signs. The clinical protocol was pragmatic; the physician was allowed to tailor his recommendations to each patient's individual needs while staying within the parameters of the medications listed above, with the option of recommending ESI when considered appropriate. The principle of shared decision-making was used for all suggested medications and ESIs; in other words, research participants had an open discussion with the physician about his recommendations and were not coerced into receiving any prespecified regime of medications or injections. In addition to oral medications and/or spinal injections, all subjects were given advice to stay active and to perform 2 home exercises: hip-flexor stretching and prone, lumbar-extension stretching (yoga cobra pose). In addition, they were encouraged to walk as much as possible despite their walking limitations.

The community-based GE arm of our trial involved attendance by the research participants in supervised GE classes for older adults. These classes were held at 2 local senior community centers in Pittsburgh. Participants could choose to attend exercise classes at either community center, based on convenience of location. The centers were responsible for setting up the participants' memberships, enrolling them in the GE classes, and providing the approved schedule of classes for the study.

Participants were asked to attend 2 exercise classes per week for 6 weeks, for a total of 12 exercise classes. Each class was about 45 minutes in length and was taught by certified fitness instructors who were experienced with supervising exercise classes for older adults. The intensity and difficulty level of the exercise classes ranged from very easy to medium. The pragmatic nature of our trial allowed for participants to self-select the level of exercise class based on their perception of their fitness level. Community center staff monitored attendance at each class, compiled this information into monthly spreadsheets, and sent them to our research team. Although the reports were sent monthly, they had details about daily class attendance.

The clinic-based manual therapy/individualized exercise arm of the trial involved treatment provided by chiropractors and physical therapists at the Physical Therapy Clinical and Translational Research Center at the University of Pittsburgh. Participants were treated 2 times per week for 6 weeks; each treatment session lasted about 45 minutes. The clinicians followed a pragmatic treatment protocol that consisted of 3 basic interventions: (1) warm-up procedure using a stationary bicycle; (2) manual therapy procedures that included lumbar distraction mobilization, hip-joint mobilization, side-posture lumbar and/or sacroiliac-joint mobilization, and sciatic/femoral nerve mobilization; and (3) individualized exercise procedures with instruction in spinal-stabilization exercises and stretching exercises to be completed at home. Each subject was assessed by physical examination for which specific muscles required stretching and which other muscles required strengthening. The treating chiropractor or physical therapist then developed an individualized program of stretching/strengthening exercises for each patient; this was in contrast with the nonspecific nature of the GE classes and general exercise instructions given to all patients by the physician. Appendix A provides a detailed list of all the manual therapy and therapeutic exercise procedures utilized in our treatment protocol for this arm of the study.

Study Outcome Measures

Our primary subjective outcome measure of self-reported pain/function was the SSS questionnaire,³¹ which is a validated patient self-report of pain and physical function. We used the 12-item version of this form, which has 2 subscales; a 7-item Symptom Severity subscale and a 5-item Physical Function subscale. The SSS has been shown to possess adequate psychometric properties for use with LSS patients, with a minimal clinically important difference (MCID) of 3.02 points for the total 12-item SSS score, which is 0.36 and 0.10 points per item for the Symptom Severity and Physical Function subscales, respectively.³² The primary objective outcome measure of walking performance was the distance walked (meters) during the SPWT, which is a validated measure of walking performance in patients with LSS.33 The SPWT has been shown to be highly reproducible with a test–retest intraclass correlation coefficient (ICC) of 0.98.³⁴ No MCID has been established for the SPWT.

We selected these outcome measures based on feedback received from our patient and clinician stakeholders. Patients with LSS who participated in our community group discussions told us that their general concerns were about their level of pain and function; they had a specific concern about the distance they could walk before needing to sit down. The SSS was an obvious choice for a self-reported measure of pain and function, because it contains both Symptom Severity and Physical Function subscales and was validated for use within the LSS population.

We decided to use the SPWT as our objective measure of walking performance instead of the Shuttle Walk Test, because the SPWT more closely resembles the way LSS patients walk in real-life settings.³⁴ Performed on level ground, the SPWT instructs patients to walk at their normal pace and to continue walking until their symptoms rise to a level that they must sit down and rest. A physical therapist walks behind the patient, measures the distance walked in meters, and records the total time walked in minutes. The developer of the SPWT consulted with our research team about the specific details of conducting this pragmatic test of walking performance.

We also used the SenseWear armband (BodyMedia, Inc) to record PA for our secondary outcome measure.^35,36 This small, wireless accelerometer is worn over the triceps muscle. It combines information from skin temperature and a biaxial accelerometer, enabling the device to estimate energy expenditure from activities that do not require ambulation.

PA measured by the SenseWear has compared well with reference standards such as doubly labeled water (ICC, 0.48-0.81)^35,36 and indirect calorimetry (ICC, 0.11-0.92).^37-41

After finishing the testing procedures at baseline, subjects were fitted with the SenseWear armband and instructed to wear the monitor for 7 days. Data from the activity monitor were inspected to ensure they were sufficient, which we defined as at least 4 days with 10 hours of PA data per day.^42-44 Subjects without sufficient data were asked to wear the portable monitor for an additional week before starting study-related interventions. We used the same procedures to collect PA data at the follow-up examinations.

We also obtained several other self-reported measures using validated instruments widely employed in rehabilitation trials. A patient global index of change (PGIC) and treatment satisfaction (SAT) questionnaire were given to our research participants at the follow-up examinations. The PGIC asked patients to rate their level of “overall status change since first beginning treatment” on a 7-point scale, ranging from “very much worse” (−3) to “very much better” (+3). The SAT questionnaire asked patients to rate their level of satisfaction with the treatment on a 6-point scale, ranging from “extremely dissatisfied” (−3) to “extremely satisfied” (+3).

We used a modified comorbidity disease index⁴⁵ to gather self-reported information about the number of comorbid medical conditions that were diagnosed in addition to LSS. We used a Falls History Form to ask questions about the number of falls over the past year and level of current fear of falling.⁴⁶ The Activities Specific Balance Confidence scale⁴⁷ was administered to assess participants' level of confidence about not losing balance during routine activities of daily living. We used the 10-item Oswestry Low Back Pain Disability Index⁴⁸ and the 11-item Tampa Scale for Kinesiophobia⁴⁹ to assess self-reported disability from back pain and fear of movement, respectively. We screened for depressive symptoms utilizing the short-form version of the PROMIS depression scale.⁵⁰ Last, we used a 6-item treatment expectancy and credibility questionnaire specifically modified for use in patients with chronic low back pain.⁵¹ Table A provides a summary of all outcome measures and the times at which they were collected during the trial.

Table A

Outcome Measures and Other Information Gathered at Baseline and Follow-up.

In addition to these self-reported measures, a research physical therapist performed some physical examination procedures at baseline to capture additional clinical data. This therapist recorded range of motion of the lumbar spine as well as that of the hip and knee joints, and palpation of these structures was performed for local tenderness. Tests for balance and mobility included a modified version of the Short Physical Performance Battery⁵²: 4-meter gait speed, timed chair stands, and timed single-leg standing. Blood pressure measurements were taken on all participants, as well as height and weight for calculation of body mass index (BMI).

We also used a portable Doppler unit and sphygmomanometer to record the ABI. The ABI is a highly sensitive and specific screening test for peripheral artery disease and lower leg vascular claudication.^53,54 Any patient with an ABI below 0.8 was excluded from participating in our study, because of the possibility of vascular rather than neurogenic claudication. We also excluded patients with severe high blood pressure, using the American College of Sports Medicine criteria (systolic >200 mm Hg or diastolic >110 mm Hg).

Blinding

Blinding of the treating clinicians and research participants was not possible, because both group were obviously aware of the intervention they were giving or receiving. We attempted to minimize bias by having an independent physical therapist perform all baseline physical examinations and follow-up reassessments. Our primary outcome measures (SSS) were a patient-report questionnaire and an objective outcome measure of walking performance (SPWT), conducted in a manner to minimize examiner bias. Participants were instructed to walk as far as they could until they needed to sit down and rest. The physical therapist observing the subjects was not permitted to coach or provide any encouragement; he simply recorded the total distance and time walked. Because the SenseWear armband is worn at the subject's home, the exploratory measure of PA was also not subject to examiner bias; data are uploaded directly from the device.

Study Setting

All treatments were provided at no charge to the participants or their insurance carriers. The MC was provided by a physical medicine physician at his outpatient clinical practice setting. Participants who attended the GE classes were given complimentary access to those classes or a temporary membership to a local community center. A chiropractor or a physical therapist in a research-based PT clinic at the University of Pittsburgh treated participants in the manual therapy arm. All locations were chosen because of their real-world settings.

Follow-up

For all 3 arms of the study, the respective interventions were completed over a 6-week period. Participants returned for 2 follow-up research examinations, at 2 months (2 weeks after completion of care) and 6 months (4 months after completion of care).

Analytical and Statistical Approaches

The analysis of our primary aim consisted of a between-groups comparison of the changes in the primary subjective (SSS) and objective (SPWT) outcomes from baseline values to 2 months (primary end point). The analysis of our secondary aim was a comparison of the between-group changes in the average amount of PA from baseline to 2 months. We defined our measure of PA as the average number of daily minutes spent in light, moderate, or vigorous PA (>1.5 metabolic equivalents of task [METs]). We also performed these same analyses comparing the data obtained at 6 months with the baseline measures. Due to the constraints of our 3-year PCORI contract, we could not obtain any results beyond 6 months.

We also performed a series of secondary responder analyses for each of the outcome measures associated with our primary and secondary aims. We accomplished this by dichotomizing all subjects into either “responders” or “nonresponders” based on the prespecified threshold of a minimum 30% improvement in outcome from baseline. We performed separate responder analyses for each of the 3 outcome measures (SSS, SPWT, and PA), comparing the proportions of responders at 2 months and 6 months.

We summarized the outcomes and all baseline characteristics with descriptive statistics, separated by treatment group. We used linear mixed models as the primary analytic method to assess the significance of treatment for each group, while adjusting for the three baseline randomization balancing variables: SSS, SPWT, and age. We followed a modified intention-to-treat principle: All participants who were randomized (N = 259) were included in the analysis, using linear mixed models to account for missing data. We verified the normality assumptions for the outcome variables for the simple analysis as well the assumptions for the multiple regression models on the change scores.

For our primary and secondary aims, we also performed a series of secondary responder analyses using dichotomous outcomes, consistent with the recommendations published by the Initiative on Methods, Measurement and Pain Assessment in Clinical Trials (IMMPACT).^55,56 Per IMMPACT recommendations, we defined treatment responders as those who achieved at least a 30% improvement in the outcome measure of interest, which is considered “moderate improvement.” We believe that a 30% improvement in symptoms and/or functional performance is a clinically important change over 2 months, considering that LSS is a chronic degenerative disease that tends to deteriorate over time. Our analyses compared the proportions of subjects considered “responders” by treatment group on these measures—SSS, SPWT, and PA. We assessed differences in the proportion of responders/nonresponders between the groups using chi-square and Fisher exact tests.

Our first exploratory aim was to explore 6-month attrition and adherence rates, number of falls, adverse events, and co-interventions. We defined adherence as any participant who did not drop out after randomization and attended at least 2 of the 3 scheduled MC visits, or at least 9 of the 12 scheduled GE or manual therapy sessions. We defined attrition as any participant who did not drop out after randomization and received at least 1 treatment but who did not show up for the follow-up research examination. At baseline, participants were asked, “Have you fallen any time during the past year? If yes, how many times did you fall?” At the 6-month follow-up evaluation, participants were asked, “Have you fallen any time between now and when you finished your study intervention? If yes, how many times did you fall?”

We defined a serious adverse event as any unanticipated health care problem, directly related to a study intervention, that caused the participant to seek MC outside of the study. We asked all research participants to neither seek any new type of health care nor start any new exercises outside of the study until after their 2-month follow-up evaluation. Therefore, we asked participants only about which co-interventions they had utilized during the period between their 2-month and 6-month follow-up evaluations. We summarized these exploratory outcomes with descriptive statistics of their respective counts and rates, separated by treatment group. The analysis for this aim consisted of omnibus chi-square tests for the differences in the rates or proportions of each outcome across the 3 groups.

Our second exploratory aim was to explore potential baseline predictors of treatment response. We first performed univariate analyses to test for possible associations between baseline characteristics and response, based on the criterion of responders being defined as those participants who demonstrated ≥30% change on either the SSS or SPWT. We used 2-sample t-tests for continuous variables and chi-square tests for categorical variables to test the association between responders and nonresponders and each variable of interest. We decided a priori that statistical significance would be set at P < .10 for variables to be included in the multivariable logistic regression model. We used the same responder–nonresponder dichotomous variables for SSS, SPWT, and PA created for the secondary responder analyses associated with our primary and secondary aims.

We then used logistic regression models to test the association between responders and nonresponders, treatment groups, and other variables of interest. We put the variables that we found significant using univariate procedures (P < .10) in a multivariable logistic regression model, and applied a backward elimination procedure. The level of significance for removal was P < .10.

We evaluated heterogeneity of treatment effects by testing the interaction between treatment and 8 baseline variables using multiple logistic regression models, with responder status as the dependent variable. These 8 variables were age, sex, BMI, comorbidities, race, kinesiophobia, knee osteoarthritis, and depression. We dichotomized age at <75 years vs ≥75 years. We split comorbidities, depression, and the other continuous variables at their respective medians. We considered baseline variables as potential moderators of treatment effect due to our experience in other studies and the published literature. We compared the P value for the interaction between the moderator and the treatment variable to .05. Regardless of significance, we presented between-treatment group comparisons as odds ratios and stratified 95% CIs by the potential moderator. Finally, we explored with descriptive statistics and responder analyses the between-group differences in Patient Global Index of Change and Treatment Satisfaction at 2 months and 6 months.

Missing Data

To account for any missing data, we used linear mixed-effects models to study treatment differences over time for the primary and secondary outcomes. We used a compound symmetry structure for the correlation structure and the Kenward-Roger approximation method for the degrees of freedom.57 Linear mixed-effects models use all available data; if a subject had a measurement at 1 time point and the rest of the data were missing, then that subject was still used in the analysis. Therefore, these mixed models included data from all participants who had a baseline examination and were randomized (N = 259).

Ancillary Qualitative Study

Although not part of our original research design, we were able to include an ancillary qualitative study that involved focus-group discussions with participants from each of the 3 intervention arms. These discussions were recorded, transcribed, coded, and analyzed for themes using qualitative software. The results of this ancillary qualitative study are available as a supplemental report upon request, as they will be published separately from the main results of this trial.

Participant Flow

Figure 1 provides a CONSORT participant flow diagram.⁵⁸ We screened 710 people over the phone and excluded 412 for not meeting the minimum inclusion criteria; 298 people were brought into our research facility for a baseline physical examination screening to assess their eligibility for participation in the trial. Informed consent was obtained from all potentially eligible patients. A total of 259 people met our inclusion criteria, were enrolled, and were randomized. A total of 19 enrolled participants dropped out after randomization and never attended any of the intervention sessions, with a larger number of dropouts in the GE arm (N = 12). We included in our analysis only those patients who attended at least 1 intervention session.

Figure 1

CONSORT Flow Diagram.

Baseline Data

Table 1 contains the baseline demographic and clinical characteristics of our participants, both by study total and assigned treatment group. The randomization process worked well; no significant differences existed between the groups on any demographic or clinical characteristics. The average age of our participants was 72.4 years (SD, 7.8), with a range of 60 to 92 years. Many of our participants reported comorbid osteoarthritis of the hip (16.6%) and/or knee (31.7%), with an average BMI of 31.0 (SD, 6.6). We achieved good racial and socioeconomic diversity, with our research participants represented by 21.6% African American, 47.1% without a college degree, and 51.4% with an annual income below $40 000.

Table 1

Baseline Demographic and Clinical Characteristics, by Study Total and Each Treatment Group.

Our participants had an average of 4.7 (SD, 2.2) medical comorbidities: arthritis (85%); cataracts (52%); emotional problems (34%); hearing problems (31%); and joint replacement (30%). As listed in Table 1, the 2 most common types of arthritis reported by our participants were hip osteoarthritis (16.6%) and knee osteoarthritis (31.7%). At baseline, our participants had an average SSS Symptom Severity subscore of 20.3 (SD, 4.3; range, 7-35) and walked an average of 455.3 meters (SD, 480.0) during the SPWT.

PA was recorded from the SWA device, which captures data in terms of METs. One MET is defined as 1 Kcal/kg/hour and is roughly equivalent to the energy cost of sitting quietly. The amount of METs expended can be used to create general activity categories as follows: sedentary, ≤1.5 METs; light, >1.5-3.0 METs; moderate, ≥3.0 to 6.0 METs; and vigorous, ≥6.0 METs. Our subjects were very sedentary, spending an average of about 18 hours per day (1095.3 ± 150.3 min/day) in activities ≤1.5 METs. They spent less than 3 hours per day (165.5 ± 129.7 min/day) performing activities >1.5 METs. Table B provides a comparison of the baseline characteristics of all participants who completed their assigned treatment and those for whom we had missing data. This table shows that those who withdrew after randomization were mostly similar with minor exceptions (eg, marital status).

Table B

Baseline Characteristics for All Participants With and Without Complete Data at Each Time Point.

Results of Primary Analyses of Primary and Secondary Aims (Linear Mixed Models)

Table 2 provides the results from the primary analyses of all outcome measures related to the primary and secondary aims. The results are organized into 3 sections within the table: (1) group means at each time point; (2) unadjusted within-group changes from baseline; and (3) adjusted between-group differences from baseline. The models for the between-group analyses were adjusted for the baseline randomization variables used to balance the groups; these included baseline SSS, SPWT, and age.

Table 2

Primary and Secondary Outcome Measures: Unadjusted Within-Group Changes From Baseline (Outcome to Baseline) and Adjusted Between-Group Differences in Improvement From Baseline.

At 2 months, there was a statistical significantly greater reduction in adjusted mean SSS score (range, 12-55) in the manual therapy and individualized exercises (MTE) group compared with the MC group (2.1; 95% CI, 0.3-3.9) and the GE classes group (2.4; 95% CI, 0.6-4.3). The MCID for the SSS is 3.02 points; therefore, the adjusted between-group SSS differences were not clinically significant. There was also improvement in the adjusted mean SPWT score at 2 months in the MTE group compared with MC (135.1; 95% CI, −17.2 to 287.4) and GE (46.2; 95% CI, −110.9 to 203.4), but these adjusted between-group SPWT differences were not statistically significant. GE showed a statistically significantly greater improvement in adjusted mean PA at 2 months compared with MC (30.5; 95% CI, 3.1-57.9), but clinical significance is unknown due to the lack an established MCID for PA. The linear mixed models did not reveal any significant adjusted between-group changes in any of the outcome measures at 6 months.

Results of Secondary Analyses of Primary and Secondary Aims (Responder Analyses)

Table 3a presents the results of the omnibus test of 3-way comparisons of proportions of responders in each treatment arm at 2 months and 6 months. We performed 3 sets of responder analyses, 1 related to each of the primary and secondary outcome measures (SSS, SPWT, and PA).

Table 3a

Responder Analysis: 3-Way Comparisons of Responders in Each Group After Dichotomizing Within-Person Change in Outcomes (Omnibus Chi-Square Test).

We defined a responder as any participant who showed at least a 30% improvement in his or her outcome measure from baseline to time point of analysis (2 months or 6 months). There was a significant between-group difference at 2 months for SSS (P = .002) and SPWT (P = .04), but not for PA (P = .51). We found no significant between-group differences at 6 months for any of the 3 outcomes.

We also performed additional analyses of all 2-way comparisons between groups. Tables 3 b-d present the between-group comparisons at 2 months, and Tables 3 e-g show the results at 6 months. At 2 months, there was a significantly greater proportion of SSS responders in the manual therapy arm (20.0%) compared with the GE (3.0%) and MC (7.6%) arms. There was also a significantly greater proportion of SPWT responders at 2 months in the manual therapy arm (65.3%) compared with the GE (46.2%) and MC (48.7%) arms. However, these between-group differences in SSS and SPWT outcomes were no longer significant at 6 months. There were no significant between-group differences in the proportions of PA responders at either 2 months or 6 months. The proportions of responders by group are also depicted visually with bar graphs in Figure 2 (2 months) and Figure 3 (6 months).

Figure 2

Responder Analysis of Main Outcomes: Bar Graph of the Proportions of Responders by Group.

Figure 3

Responder Analysis of Main Outcomes: Bar Graph of the Proportions of Responders by Group.

Table 3b

Two-Way Comparisons of Manual Therapy and Medical Care, Baseline to 2 Months.

Table 3c

Two-Way Comparisons of Manual Therapy and Group Exercise, Baseline to 2 Months.

Table 3d

Two-Way Comparisons of Medical Care and Group Exercise, Baseline to 2 Months.

Table 3e

Two-Way Comparisons of Manual Therapy and Medical Care, Baseline to 6 Months.

Table 3f

Two-Way Comparisons of Manual Therapy and Group Exercise, Baseline to 6 Months.

Table 3g

Two-Way Comparisons of Medical Care and Group Exercise, Baseline to 6 Months.

Results of Analyses for Exploratory Aim 1

Table 4 lists the rates of anticipated unpleasant side effects and serious adverse events reported by our participants at the 2-month follow-up evaluation. We defined anticipated unpleasant side effect as any postintervention symptom that was minor and transient in nature (≤2 days duration). There was a significantly greater (P < .001) proportion of musculoskeletal side effects (transient muscle and joint soreness) reported by the participants in the GE and manual therapy arms. Participants in the MC arm reported a significantly greater proportion of transient nonmusculoskeletal side effects. Fortunately, there were no serious adverse events to report from any participant in any of the 3 treatment arms of this trial.

Table 4

Reported Adverse Events and Anticipated Unpleasant Side Effects, Study Total and by Treatment Group at 2 Months (2 Weeks After End of Intervention).

Table 5 provides a tabulation of the number of falls reported by participants at baseline and at the 6-month follow-up. We found no significant between-group differences in the number of falls reported at either time point. Table 6 summarizes the adherence and attrition rates for each treatment arm at 3 different time points. Although group sizes were balanced at baseline, a significantly greater proportion of subjects dropped out of the GE arm after randomization and never attended any GE class. There were no significant between-group differences in adherence rates for those participants who attended at least 1 intervention session. The adherence rates were above 90% for each of the 3 intervention groups. There were no significant between-group differences in attrition rates for the 2- or 6-month follow-up evaluations.

Table 5

Number of Falls Reported at Baseline and at 6-Month Follow-up Evaluation.

Table 6

Attrition (Dropout) and Treatment Adherence Rates by Group at 2- and 6-Month Follow-up Evaluations.

Table 7 provides a tabulation of the various co-interventions reported by participants who completed 6 weeks of each treatment at the 6-month follow-up evaluation. A significantly lower proportion of GE participants (49%) were exercising at home as compared with those in either the MC (72%) or manual therapy (77%) arms. However, a significantly greater proportion of GE (48%) participants were engaged in community-based exercises, compared with 18% and 23% of those in the MC or manual therapy arms, respectively.

Table 7

Co-interventions Reported by Participants From End-of-Care to 6-Month Follow-up.

Results of Analyses for Exploratory Aim 2 (Heterogeneity of Treatment Effect)

Our second exploratory aim explored potential baseline associations and predictors of treatment response for SSS and SPWT in each of the 3 treatment groups (heterogeneity of treatment effect). We used both simple univariate tests of association (chi-square) and multivariable logistic regression for predictors of treatment effect. The results of the unadjusted univariate analysis of baseline characteristics and responder status are displayed in Table 8. We found significant differences between mean SSS and SPWT scores for the responders and nonresponders with respect to their association with age and falls score.

Table 8

Univariate analyses of baseline characteristics and responder status, for the SPWT and SSS outcome measures.

The results of the multivariable logistic regression models of responders and nonresponders at 2 months are displayed in Tables 9a and 9b. The MTE group showed a higher rate of SSS response (20%) compared with either the GE group (3%) or the MC group (7.6%). Using the MC group as the reference group, the MTE group had greater odds of being SSS responders (OR, 3.5; 95% CI, 1.2-9.6), and the GE arm had lower odds of being responders (OR 0.4; 95% CI, 0.08-2.1). The MTE group also showed a higher rate of SPWT response (65.3%) compared with either the GE group (46.2%) or the MC group (48.7%). The MTE group had greater odds of being SPWT responders (OR, 2.1; 95% CI, 1.1-4.0) when compared with the MC reference group. The GE group had about the same odds (OR, 0.95; 95% CI, 0.5-1.9) of being SPWT responders as the MC reference group. We also found a significant association between age and responder status; younger patients were more likely to be SSS and SPWT responders regardless of group assignment. Falls score was associated only with SSS responder status, with responders more likely to report fewer falls at 6 months.

Table 9a

SSS: Primary Outcome Measure (N = 225 Participants) at 2 Months.

Table 9b

SPWT: Secondary Outcome Measure (N = 216 Participants) at 2 Months.

In Tables 9a and 9b, multivariable logistic regression models (using backward stepwise elimination) of study participants who responded and who did not respond to intervention based on the SSS and SPWT at 2 months are shown (MC is the reference group).

To assess if any baseline variables were moderating the relationship between treatment and responders and nonresponders, we tested the interaction between treatment and each of several variables using multiple logistic regression models, with SSS and SPWT responder status as the outcome. We chose the variables based on their biologically plausible relevance to LSS and the clinical experience of the investigators. Tables 10 and 11 depict the results of these moderator analyses for the SSS and SPWT outcomes, respectively. We found no significant moderators associated with the SSS scores; however, Table 11 shows that baseline depression (P = .06) and the number of comorbidities (P = .06) were 2 variables that showed a trend toward significance as potential moderators of SPWT outcome. Age, sex, race, BMI, fear avoidance, and knee osteoarthritis were not associated with responder status. The treatment effect of manual therapy vs MC appears to be stronger among patients with depression scores above the median and comorbidity scores below it. The P value for these interactions was .06; although not reaching .05, it suggests a potential interaction and moderating effect. While we were not powered for these stratified analyses, these results warrant further investigation in future studies, given their clinical plausibility of being potential treatment moderators.

Table 10

Analysis of Hypothesized Moderators at Baseline, With Treatment Response for SSS as the Dependent Variable (Primary Outcome Measure).

Table 11

Analysis of Hypothesized Moderators at Baseline, With Treatment Response for SPWT as the Dependent Variable (Secondary Outcome Measure).

Patient Global Index of Change and Satisfaction

The results of additional exploratory analyses of the PGIC and SAT scores obtained at 2 months and 6 months are presented in tables. Descriptive statistics of the item ratings and proportions of responders/nonresponders for PGIC scores are presented in Tables 12a and 12b; those for SAT scores are presented in Tables 13a and 13b. The manual therapy arm showed the highest level of PGIC and SAT scores at both time points, followed by GE and MC.

Table 12a

PGIC, Baseline to 2 Months.

Table 12b

Patient Global Index Change, Baseline to 6 Months.

Table 13a

Treatment Satisfaction, Baseline to 2 Months.

Table 13b

Treatment Satisfaction, Baseline to 6 Months.

Missing Data

At our primary outcome end point of 2 months we had a total of 33 participants (12.7%) with missing data. Participants from the GE arm had the largest proportion of missing data at 2 months (20.2%) compared with the MC (10.2%) and manual therapy (8%) arms. One reason for missing data was that some subjects withdrew immediately after randomization and did not receive any intervention. Significantly more participants withdrew from the GE (14.3%) arm of our study than from the MC (4.5%) and manual therapy (3.4%) arms. Some subjects did not adhere to their assigned intervention and withdrew at various times during the 6-week intervention period. Other subjects completed their assigned intervention but failed to return for their 2- or 6-month follow-up.

To account for missing data, we used linear mixed-effects models to study treatment differences over time using SSS and SPWT as the dependent variables in separate models. These models included (1) unadjusted fixed effect of time, treatment, and treatment by time interaction; (2) least-squares means by group over time; (3) differences between group means over time; and (4) adjusted for covariates of interest. The linear mixed models borrow information pertaining to the relationships in the outcome at multiple time points such that persons missing data at certain time points can still be used in the analysis.

For the SPWT models, the unadjusted linear mixed model included a fixed effect of time and treatment. We tested the interaction between time and treatment and not found it to be statistically significant (P > .05). We then adjusted the linear mixed model with main effect of time and treatment for the covariates of interest previously used in our primary linear regression models. We found no statistically significant differences between treatment groups in any of the models involving SPWT. This is consistent with the results of our primary analysis.

For the linear mixed models involving the SSS as the dependent variable, the first model included a fixed effect of time, treatment, and treatment by time interaction. The interaction term was significant (P = .04), suggesting that the SSS score differed between treatment groups over time. As with the SPWT models, we then performed an adjusted model with the previously used covariates. The interaction between treatment and time remained significant, even after adjusting for covariates of interest, suggesting differences in SSS group means over time adjusted for covariates of interest. Table 14 provides a summary of all missing data for each outcome measure and each time point.

Table 14

Missing Data for Each Outcome Measure by Assigned Intervention Group.

Context for Study Results

Our results were mixed with respect to confirming our specific aims and hypotheses. For our primary aim, we hypothesized that participants in both the GE and MTE arms would show better clinical outcomes in self-reported pain/function (SSS) and walking performance (SPWT) compared with those randomized to the MC arm. However, we found that all 3 groups showed modest reductions in SSS and improvement in SPWT at 2 months, but only the improvements in SPWT were sustained at 6 months.

The results of our regression analyses for the primary outcomes of pain/function found a statistically significant improvement in mean SSS scores in the MTE arm compared with the other 2 arms; however, the magnitude of this effect has only marginal clinical significance because the adjusted between-group differences did not exceed the MCID. The mean unadjusted within-group improvement of 4.1 points in SSS score from baseline for MTE group was only modest, considering that the MCID is 3.02 points. The greatest improvement in walking performance was found in the manual therapy arm, although this was not statistically significant when compared with the improvements found in the other 2 study arms. Also the clinical significance of this magnitude of walking improvement in unknown, as there is no established MCID for walking performance (SPWT).

Compared with those from the regression models, the results of the secondary responder analyses for our primary outcomes may be more clinically relevant and easier to interpret. The responder analyses revealed that there were significantly larger proportions of SPWT and SSS responders in the manual therapy arm at 2 months as compared with either the GE or MC arms. However, the short-term results favoring the manual therapy intervention arm at 2 months did not persist at 6 months. This may be explained in part by the lack of any follow-up treatment, or “booster sessions,” after the initial 6-week intervention period.

For our secondary aim, we hypothesized that participants in the GE and MTE arms would demonstrate greater changes in PA compared with participants in the MC arm; however, we found that only participants in the GE arm showed improvement in their level of PA, compared with either of the other 2 arms. The GE arm showed significantly greater improvement in PA compared with the MC arm at 2 months but not at 6 months. It was interesting to find that both the MC and MTE arms showed less PA at 6 months, with only the GE arm participants maintaining their baseline PA level. This is in direct contrast with the results showing an overall improvement in walking performance in all 3 groups. This is not surprising, considering that none of our research interventions provided a targeted treatment for improving PA. This reinforces the fact that measures of physical performance (walking capacity) are different from measures of overall PA (average time spent in activities >1.5 METs). It may also have been unrealistic in this age group to set a 30% increase in PA as the minimum threshold.

We also found some interesting results from our analyses of our first exploratory aim. One unexpected finding was that a significantly greater number of participants randomized to GE dropped out immediately after randomization, compared with the other 2 arms. Although no significant difference in attrition (dropout) rates existed between groups at 2 months or 6 months, participants who adhered to the GE intervention still had a higher number of individuals fail to show for their follow-up examinations (Table 6). Feedback received from those who dropped out of the GE arm after randomization indicated that lack of motivation to exercise and transportation issues were the major barriers to attending GE classes.

However, for those participants who attended at least 1 intervention session, adherence rates to the assigned interventions were excellent: above 90% in each of the study arms (Table 6). As noted in the Results section, we did not have any serious adverse events in any of the 3 treatment groups (Table 4). The higher rate of transient musculoskeletal side effects in the GE and MTE arms was expected due to the more physical nature of those interventions.

Conversely, the higher rate of gastrointestinal side effects in the MC arm was also expected and anticipated as normal reactions to the oral medications prescribed by our research physician (Table 4).

A little more than 40% of our participants in all 3 arms reported having experienced at least 1 fall in the past year, and this rate did not change significantly in any of the arms at 6 months (Table 5). We also analyzed the type and number of co-interventions reported by our participants between the 2- and 6-month follow-up. A significantly greater proportion of participants from the manual therapy and MC arms reported that they were doing home-based exercises as compared with the GE arm. This can be explained by the fact that in both of these intervention arms, participants were given instructions for home-based exercises.

Conversely, a significantly greater number of participants in the GE arm reported that they were exercising in a community setting as compared with the other 2 arms. This is not surprising, considering that patients had become accustomed to a routine of exercising in a group setting twice a week for 6 weeks, which may have led them into a new routine of regular exercise. In addition, participants in our focus groups told us that they enjoyed the socialization aspect of GE classes, which enhanced their motivation to continue exercising in the community setting. A total of only 4 research participants (2.1%) reported having spinal surgery at 6 months, with no significant differences in surgery rates between the 3 groups.

The largest RCT (SPORT trial) comparing surgical and nonsurgical treatments for LSS concluded that patients treated with surgery had better results for up to 4 years following surgery.^17,18 However, in this study there was no standardization of the interventions available to the subjects who were not randomized to surgery. A secondary analysis of the SPORT trial data was performed, which focused on the subset of patients in the nonsurgical arm who received PT treatment.⁵⁹ This study showed that receiving PT within the first 6 weeks after enrollment was associated with a lower rate of progression to surgery at 1-year follow-up.

Another RCT published after the SPORT trial compared surgical decompression and PT for LSS.¹⁹ This study showed that both surgical and nonsurgical approaches yielded similar results on self-reported pain and function at 2 years.

Considering the high prevalence of LSS in older adults, there is a surprising lack of evidence about the effectiveness of well-defined and standardized nonsurgical management approaches for this common condition. The results of our study should be viewed in the context of the results from these previous LSS trials. The secondary analysis of the SPORT trial concluded that future research studies were needed to compare the safety and effectiveness of clearly defined nonsurgical treatment protocols. Our study provides new evidence about the safety and effectiveness of 3 well-defined nonsurgical interventions, which may help clinicians provide patients with information about nonsurgical options that can then be discussed in a shared-decision-making process. Our results also support the view that patients with LSS can show some modest level of clinical improvement over time without surgical intervention.

It is also important to view the results of this study within the greater context of the current nonsurgical management approach to patients with LSS. The guidelines from leading professional organizations provide little or no guidance to clinicians or patients about which nonsurgical methods might benefit patients with LSS. There also seems to be a general feeling among patients and providers that nothing can be done to help a chronic degenerative condition like LSS—that a slow decline in function is simply to be expected as part of the natural progression of the condition. Although the natural history of LSS without any treatment or intervention is not well known, it appears that many patients with LSS remain stable or improve over time.⁶⁰ Participants in our focus groups expressed frustration about how little information was provided to them about viable nonsurgical treatment options by their PCPs. Many told us that their PCPs discouraged participation in GE classes, PT, and/or chiropractic care for 2 basic reasons: (1) fear of potential injury and (2) lack of evidence for effectiveness.

The results of our study provide new evidence for the safety of community-based GE classes as well as clinic-based manual therapy and individualized exercise provided by physical therapists and chiropractors. There were no serious adverse events associated with participation in either of these intervention groups. Although musculoskeletal side effects were very common in these 2 groups, they were minor and transient. In fact, patients in our focus groups said that they actually enjoyed feeling some muscle soreness after exercising in the group setting or after having their PT or chiropractic session; they interpreted this soreness as a positive sign that their muscles and joints had been worked out. Our study should provide physicians with some assurance that patients with LSS can safely participate in GE classes and will not necessarily be harmed by physical therapists or chiropractors.

With respect to effectiveness, although the amount of improvement in self-reported pain/function was only short lived and modest, the magnitude of improvement in walking performance was larger and sustained at 6 months in all 3 groups. Wide variability in our walking performance data and no established MCID for the SPWT made it difficult to draw conclusions about the clinical significance of the observed changes in mean distance walked; however, at 6 months, the responder analysis showed that about half of the patients in all 3 groups were walking at least 30% farther than they were at baseline.

This finding suggests that although LSS is a chronic degenerative condition, patients can still make some improvements in their physical function and should not be assumed to have a poor or guarded prognosis. This should be an important component of shared decision-making when patients with LSS explore nonsurgical treatment options with their PCPs. That said, these findings should be balanced with the amount of time and cost associated with each intervention. Most patients would be required to make a copay for each of 12 PT or chiropractic visits, which would be 4 times costlier to the patient than making just 3 copays with a physician. The cost of GE classes is free or just a few dollars each session at most community centers.

Uptake of Study Results and Generalizability of Findings

The demographic makeup of our study participants (Table 1) is very similar and comparable to the populations of LSS participants in the previously noted trials by Weinstein et al^17,18 and Delitto et al¹⁹ with respect to baseline age, BMI, medical comorbidities, and self-reported levels of pain and function. The socioeconomic and racial diversity of our participants was comparable to the general demographics of the urban Pittsburgh population, and thus our results should be generalizable to most other major metropolitan urban populations in the United States.

The pragmatic nature of this trial facilitates the implementation and uptake of our results by many of our stakeholders. Our MC protocol was straightforward and pragmatic.

Although administered by a physiatrist in our study, this approach to shared decision-making and tailoring medications to each individual patient could easily be implemented by most PCPs. PCPs could also provide more referrals to chiropractors and physical therapists for nonsurgical management of LSS; however, information obtained from focus groups of LSS patients revealed that most PCPs are hesitant to refer their LSS patients to chiropractors and physical therapists, citing lack of evidence for the safety and effectiveness of such interventions. To overcome this barrier to the uptake of our study results, our research physiatrist will actively disseminate our results at PCP grand rounds. We also plan to publish a commentary article about our MC protocol in a peer reviewed journal geared toward a PCP readership. In addition, our local UPMC Health Plan stakeholders will assist us in developing an active dissemination and implementation plan to increase the uptake of our study results, by increasing awareness among PCPs who are network providers in this local health plan.

The findings from our GE arm are clearly generalizable to the real-world setting. Most community centers that provide services to older adults offer some type of GE class. We did not have to create any new GE protocol for our study; we simply provided access to existing community-based GE classes. The cost of these classes is minimal or free to most older adults as a benefit under major health insurance plans, which removes the financial barrier to uptake of the study results. Our stakeholders include the directors of 2 large community centers that service older adults in Pittsburgh. They agreed that these findings should be disseminated widely to the other community centers in the Pittsburgh metro area. They will also work with the PI on an active dissemination plan to push these study results directly to their members and community-dwelling older adults residing in neighborhoods near their centers.

The findings from the manual therapy and individualized exercise protocol that we developed for this study are very relevant to the chiropractic and PT professions. Our results show that utilization of this protocol is a safe and effective intervention for the nonsurgical management of patients with LSS. Most of the individual procedures within our study protocol are commonly utilized by chiropractors and physical therapists, making the results potentially generalizable to clinicians in both professions, who are already familiar with at least some of these procedures. However, very few physical therapists or chiropractors are currently combining these individual procedures into a comprehensive “boot camp” approach, as we utilized in our research protocol. We also discovered a potential barrier to the uptake of these study results from focus groups and informal discussions; namely, that many chiropractors and physical therapists lack confidence in their ability to provide effective treatment to patients with LSS. They shared a collective belief that there was a lack of evidence for the safety and effectiveness of the nonsurgical methods available to them.

To help overcome this barrier, the PI and co-investigators are in the process of designing a weekend-based continuing education course for chiropractors and physical therapists; in it we will provide a review of the clinical examination, diagnosis, and treatment options for LSS. This course will also include a hands-on workshop to provide skills and training in the combination of procedures utilized in our research protocol. We believe that this active dissemination and implementation strategy will lead to greater uptake of our study results by the chiropractic and PT professions, who will have greater confidence and skills in managing patients with LSS. We also believe that this training will lead to a greater rate of referral to these providers by PCPs.

Subpopulation Considerations

As discussed in the Results section of this report, we explored the heterogeneity of treatment effect by performing an analysis of baseline associations and predictors of treatment response. We found a significant association between age and responder status: Younger patients were more likely to be SSS and SPWT responders, regardless of group assignment. Falls score was associated only with SSS responder status, with responders more likely to report fewer falls at 6 months. We did not find any significant moderators associated with either of our primary outcome measures of self-reported pain/function (SSS) or walking performance (SPWT).

Baseline depression and number of comorbidities showed a trend (P = .06) toward significance as possible moderators of SPWT outcome.

The treatment effect of manual therapy vs MC appears to be stronger among patients with depression scores above the median and comorbidity scores below it. Age, sex, BMI, fear avoidance, and knee osteoarthritis were not associated with responder status. It is important to note that we were not sufficiently powered for these stratified analyses; however, some of these associations warrant further investigation in future research trials given their clinical plausibility as potential treatment moderators.

Study Limitations

There was large heterogeneity in the clinical response found in our study, and that while some patients improved in each group and the interventions were shown to be safe, the overall levels of response (improvement) were low. In addition, our study was unable to identify which patient characteristics were predictive of treatment response. The large standard deviations associated with the walking performance and PA data reflect the large variation in the clinical status of patients with LSS. There is a need for further research in the area of nonsurgical interventions and how they may have differential effects on various subgroups of LSS patients.

Compared with either of the other intervention arms, a significantly greater proportion of subjects withdrew from the GE arm immediately after randomization. This may have created selection bias, in which those subjects who chose to accept randomization into GE were more motivated toward PA than subjects in the other 2 arms of the study. Increased motivation might be a confounding variable in those results showing greater PA in this group at 2 months. Also, subjects who received the MTE intervention spent about 45 minutes face to face with a physical therapist or chiropractor for 12 sessions. This increased personal attention might be a confounding variable in those results showing greater short-term improvement in self-reported pain/function with MTE.

We also found some limitations in the use of the SPWT. We noticed a sense of boredom in some of our subjects, which may have influenced their decision about when to stop the test even if they could have walked farther. We also received feedback from participants in our focus groups about how their walking performance could vary day to day, based on the natural fluctuations of a chronic degenerative disease such as LSS. They also told us that in real life, they had challenges with walking over uneven ground, steps, and curbs, which were not part of the SPWT. Some participants had other medical problems that precluded them from completing the SPWT, which led to some missing SPWT data at 2- and 6-month follow-ups.

However, subjects agreed that given the choice, the SPWT was still a better measure of their true walking performance compared with other outcome measures such as the Shuttle Walk Test or treadmill tests.

There were also some limitations with the use of the PA monitor. Although the device was rather small and unobtrusive, it was still large enough that some subjects found it uncomfortable to wear 24 hours per day for 7 consecutive days. As noted earlier, this led to the situation in which we had missing PA data for a subset of research participants at 2- and 6-month follow-ups. Participants in our focus groups also gave us important feedback about their perceptions of the challenges associated with measuring PA. They told us that, similar to the challenges with measurements of walking performance, their level of general PA varied by day and by week based on many factors other than their LSS. Many subjects told us that they wanted to be able to walk farther at any 1 time but that they did not really desire to increase their overall level of PA.

This trial allowed for some discretion on the part of our MC and manual therapy providers to modify their clinical interventions based on each patient's needs. The heterogeneity in providing this care is a limitation that prevented us from assessing which component(s) of these interventions may have moderated the treatment effects, reducing the internal validity of the results; however, this heterogeneity is found in the real-world setting and improves the external validity of these results.

Finally, in retrospect we realize that our original choice of the term usual MC may not have been the most accurate terminology. Patients in our focus groups who were randomized to the MC arm told us that our research physician gave them much more time and attention than what they generally had received previously from their other physicians. Also they told us how much they appreciated our physician's thoroughness in reviewing all of their medications, trying his best to minimize the dosages and number of medications and taking the time to more fully explain all treatment options. In short, our “usual” MC may have been “unusual,” which may partly explain our research participants' generally favorable response to this intervention.

Future Research

One potential area for future research would be to compare the effectiveness of multimodal “bundles” of different interventions, possibly with a factorial design. We designed this study as a comparative effectiveness trial seeking the most effective of these 3 interventions; however, in real clinical practice it might better serve patients if various combinations of these interventions are utilized in a multimodal or sequential manner. It is possible that a combination of 2 or 3 interventions used in our trial might have a synergistic or sequential effect and provide LSS patients with more clinical benefit than any one individual intervention. For example, patients with high levels of pain may get more benefit from manual therapy or be more tolerant of exercise when concurrently managed by a physician with appropriate medications or even an epidural steroid injection. Also, patients who are responding well to chiropractic or PT treatment might benefit from participating in concurrent GE classes, and vice versa.

Another research question is whether it is more effective to provide “booster sessions,” or periodic visits to a chiropractor or physical therapist spaced out over time, rather than in a single bolus of treatment typically given within a 6-week intervention. In the focus groups, participants told us that they realized LSS was a chronic disorder that could not realistically be “cured” with 6 weeks of any intervention. They told us that they felt they would benefit from more regular and periodic visits, or booster sessions, to a physical therapist or chiropractor as a type of long-term management of their chronic degenerative condition. They also told us that continuing with regular exercise over the long term was helpful in maintaining the short-term benefits they gained during 6-week intervention period.

It would also be a novel concept to create more collaboration and coordination of care between physical therapists, chiropractors, and the community centers that offer GE classes for older adults. After completing an intensive course of individualized PT or chiropractic care, LSS patients could be transitioned to GE classes for ongoing maintenance of their improved function. These are all testable hypotheses for future clinical comparative effectiveness research studies.

Appendix A.

Manual Therapy and Individualized Exercise Protocol Implemented by the Study Chiropractors and Physical Therapists (PDF, 204K)

Appendix B.

Spinal Stenosis Recruitment – Postcard Direct Mailings to Selected ZIP Codes Within the City of Pittsburgh (Nov 2013-Sept 2015) (PDF, 129K)