The economic burden of musculoskeletal disorders is

The Effect of Surgical and Nonsurgical Treatment on Longitudinal Outcomes of Lumbar Spinal Stenosis Over 1 Years Yuchiao Chang, PhD, Daniel E. Singer, MD, Yen A. Wu, MPH, Robert B. Keller, MD, w and Steven J. Atlas, MD, MPH OBJECTIVES: To assess the relative effect of initial surgical and nonsurgical treatment on longitudinal outcomes of patients with lumbar spinal stenosis over a 1-year followup period. DESIGN: A prospective observational cohort study. SETTING: Enrollment from community-based specialist practices throughout Maine. PARTICIPANTS: One hundred forty-four patients with lumbar spinal stenosis who had at least one follow-up: 77 initially treated surgically and 67 initially treated nonsurgically. INTERVENTION: Initial surgical or nonsurgical treatment. MEASUREMENTS: Clinical data were obtained at baseline and outcomes followed at regular intervals over 1 years with mailed questionnaires including patient-reported symptoms of back pain, leg symptoms, back-specific functional status, and satisfaction. Longitudinal data were analyzed using general linear mixed models. In addition to treatment (initial surgical or nonsurgical care), time period, and the interaction between treatment and time, the models included baseline score, patient age and sex, and a timevarying general health status score. The effects of these covariates in explaining differences between treatment groups were also examined. The effect of subsequent surgical procedures was assessed using different analysis strategies. RESULTS: The 1-year rate of subsequent surgical procedures was 23% and 38% for patients initially treated surgically and nonsurgically, respectively, and the overall 1-year survival rate was 69%. Patients undergoing initial surgical treatment had worse baseline symptoms and functional status than those initially treated nonsurgically. For all outcomes and at each time point, surgically treated patients reported greater improvement in symptoms and functional status and higher satisfaction scores, indicative of better outcomes, than nonsurgically treated patients. However, the relative magnitude of the benefit diminished over time such that the relative differences for low back pain and satisfaction were no longer significant over long-term follow-up (both P 5.8 for treatment effect between 5 and 1 years after controlling for covariates). Regardless of initial treatment received, patients undergoing subsequent surgical procedures reported less improvement in outcomes over time than patients who did not undergo subsequent procedures, but the relative differences between treatment groups were similar in analyses that controlled for outcomes after subsequent procedures. CONCLUSION: After controlling for covariates, patients initially treated surgically demonstrated better outcomes on all measures than those initially treated nonsurgically. Although outcomes of initial surgical treatment remained superior over time, the relative benefit of surgery diminished in later years, especially for low back pain and satisfaction. Patients undergoing subsequent surgery had worse outcomes regardless of initial treatment received, but excluding them did not change overall treatment group comparisons. The analytical methods described may be helpful in the design and analysis of future studies comparing treatment outcomes for patients with lumbar spinal stenosis. JAm Geriatr Soc 53:785 792, 25. Key words: lumbar spinal stenosis; prospective cohort study; outcomes; lumbar disc surgery; natural history From the General Medicine Division and the Clinical Epidemiology Unit, Medical Services, Massachusetts General Hospital, Harvard Medical School, Boston, Massachusetts; and w Center for the Evaluative Clinical Sciences, Dartmouth Medical School, Hanover, New Hampshire. Supported by grants from the National Institute on Aging through the Harvard Older Americans Independence Center (P6 AG8812) and from the Agency for Healthcare Research and Quality (HS-6344, HS-8194, and HS- 984). Address correspondence to Yuchiao Chang, PhD, General Medicine Division, Massachusetts General Hospital, 5 Staniford Street, Boston, MA 2114. E-mail: ychang@partners.org DOI: 1.1111/j.1532-5415.25.53254.x The economic burden of musculoskeletal disorders is large and is increasing as the population ages. 1 Although rarely life-threatening, low back pain in older individuals can be disabling. Spinal stenosis, a degenerative condition of the spine first described in the 195s, 2 is an increasingly recognized cause of low back pain and radiculopathy in older individuals. An aging population, advances in imaging techniques, 3 and increased surgical treatment JAGS 53:785 792, 25 r 25 by the American Geriatrics Society 2-8614/5/$15.

786 CHANG ET AL. MAY 25 VOL. 53, NO. 5 JAGS options have led to a surge in Medicare spine surgery rates, with significant variation across geographic regions. 4 6 Although validated outcome measures have been developed, 7,8 few published studies of spinal stenosis have incorporated them. 9 The natural history of spinal stenosis is poorly understood, 1 resulting in substantial uncertainty regarding the indications for surgery, long-term treatment outcomes, and predictors of outcome. 11 Despite this, surgery is an accepted treatment for symptomatic patients. A meta-analysis of randomized surgical studies for degenerative spondylosis (including spinal stenosis) found no evidence to support any form of surgery over conservative care. 9 A small, randomized study that compared surgically treated and 18 nonsurgically treated patients over a 1-year period had a high crossover rate to surgical treatment and no formal comparison of the relative benefits of surgical versus nonsurgical treatment. 12 Other randomized studies of patients with spinal stenosis are ongoing.,14 Preliminary short-term results favor surgical treatment, 14 but long-term results will not be available for a number of years. For individuals treated in contemporary clinical practice in the United States, the only prospective outcomes comparing surgical and nonsurgical treatment are from a prospective observational study, the Maine Lumbar Spine Study. 15 The Maine study and others have demonstrated a high rate of subsequent surgical procedures that complicate the assessment of outcomes based upon initial treatment, regardless of whether the allocation is randomized or not. 12,16 18 In an aging population with other medical problems, it is unclear how changes in general health status over time influence back-related symptoms and their effect and whether the influence is different in patients who previously underwent spine surgery and those who did not. These issues will complicate the assessment of longitudinal outcomes regardless of the intervention or study design. Efforts are needed to examine their potential effect. Outcomes from the Maine study have previously been reported and demonstrate better outcomes associated with surgical than nonsurgical treatment. 16,19,2 These and other studies have shown that the benefit of surgery declines over time, 21 23 presumably related to the ongoing degenerative process and subsequent changes at the surgical site or adjacent spinal levels, but the timing and extent of change in the relative outcomes over time remains uncertain because most prior studies have been qualitative, without the use of prospectively assessed validated outcome measures. The goal of the current study was to assess the relative benefits of initial surgical and nonsurgical treatment over a 1-year follow-up period using longitudinal analytic techniques that control for baseline differences between treatment groups, subsequent surgical procedures, and general health status over time. Validated patient-reported outcome measures were used that incorporated a broad range of healthrelated quality of life. METHODS The Maine Lumbar Spine Study prospectively followed patients treated by orthopedic surgeons and neurosurgeons in community-based practices throughout Maine. The patient and the physician determined treatment, surgical or nonsurgical, in a routine clinical manner. Almost all surgically treated patients with lumbar spinal stenosis had a decompression laminectomy with findings consistent with a diagnosis of spinal stenosis. 19 Fusion was uncommon, and internal fixation devices were not used. Nonsurgically treated patients received a variety of treatments, with back exercises, bed rest, physical therapy, spinal manipulation, narcotic analgesics, and epidural corticosteroid being most frequently used. 19 The diagnosis of lumbar spinal stenosis was based upon physician assessment of appropriate symptoms and examination and radiographic findings. Patients with spinal stenosis on advanced imaging studies could also have a herniated lumbar disc. Patients were excluded if they had had prior lumbar spine surgery, cauda equina syndrome, developmental spine deformities, vertebral fractures, spine infection or tumor, inflammatory spondylopathy, pregnancy, or severe comorbid conditions. Patients were enrolled from 199 to 1992, with enrollment stratified to obtain approximately equal numbers of surgical and nonsurgical patients. Patients who initially chose nonsurgical treatment but underwent surgery before the first follow-up evaluation at 3 months were included in primary analyses as having been surgically treated (n 5 1, %). 19 For eligible patients, physicians completed a detailed baseline questionnaire that included history, physical and neurological findings, results of diagnostic procedure, and planned treatment. Baseline imaging studies (computed tomography, magnetic resonance imaging, or myelogram) were ordered as directed by the treating physicians, and 54% were available for independent review. 19 Baseline interviews were conducted in person with mailed follow-up questionnaires at 3, 6, and 12 months and then yearly through 1 years. The institutional review boards at the participating institutions approved all study activities. Outcome Measures Outcomes assessed at each follow-up included the frequency of low back and leg pain symptoms, back-specific functional status, and satisfaction with the current state. The frequency (from 5 not at all to 6 5 always) of low back pain, leg pain, leg or foot weakness, leg numbness, and leg pain after walking in the previous week were assessed at baseline and follow-up. A leg symptom frequency index, with scores ranging from to 24, was created by summing the four leg-related questions. 24 Back-specific functional status was measured using the modified Roland disability scale. 24 Satisfaction with current state was assessed at each follow-up with the following question, If you were to spend the rest of your life with your back symptoms just the way they are now, how would you feel about that? Responses ranged from delighted to terrible on a 7-point scale. For each variable, higher scores indicate more-severe symptoms or worse function and worse satisfaction. Analysis Longitudinal data were analyzed using general linear mixed models. Upon evaluation, a spatial power covariance structure was used to model the correlations from the repeated measures data. Except for the satisfaction outcome, the change from baseline scores at each follow-up was used as the primary outcome in the analysis. Because the satisfaction

JAGS MAY 25 VOL. 53, NO. 5 TREATMENT OUTCOMES OF SPINAL STENOSIS 787 outcome represents the absolute state at each follow-up, the primary outcome was coded so that higher scores represented greater levels of satisfaction (terrible 5 to delighted 5 6). The models included treatment group, time period, and the interaction between treatment and time. Responses over the first year were defined as short term, whereas responses from Years 2 to 4 were defined as midterm. Responses from Year 5 and beyond were defined as long term. Baseline score for each outcome (except satisfaction) was also included in the models to control for regression to the mean. In addition, patient sex and age at baseline and a time-varying general health status score were included in the model. The patient s general health status was assessed at each follow-up with a question from the Medical Outcomes Study 36-item Short Form (SF-36) questionnaire, In general, would you say your health is excellent (1) to poor (5). 25 The effect of these covariates was examined in explaining the relative differences between treatment cohorts. First surgery for patients initially treated nonsurgically and reoperations for those treated surgically were assessed for all patients using physician office records, state hospital discharge data, and patient responses to follow-up surveys and telephone contact. The primary analysis used the patient s initial treatment to define a treatment group over time, which is similar to the intention-to-treat analysis for randomized, clinical trials (Figure 1). Specifically, unadjusted outcomes of patients according to initial treatment group and whether they underwent a subsequent surgical procedure were examined. Because there was a high rate of subsequent surgical procedures, several secondary analyses were performed to address how actual treatment may affect the relative benefit of surgical and nonsurgical care. These analyses examined the effect of group assignment over time when managing the issue of subsequent surgical procedures. Two secondary analyses are presented. First, follow-up data after a patient received any subsequent surgical procedure (initial nonsurgical patients who crossed over to the surgical group or initial surgical patients who received a reoperation) were excluded. Second, follow-up data after initial surgical patients received a reoperation were excluded and follow-up data were reassigned to the surgical group after initial nonsurgical patients crossed over, using the date of surgery as a new baseline time period. RESULTS Of 148 eligible, enrolled patients, 144 completed at least one follow-up survey (77 of 8 initially treated surgically and 67 of 68 initially treated nonsurgically) and are included in these analyses. Of these 144 patients, 41 died during follow-up. The 1-year survival rate after taking into account censoring was 69%. Of surviving patients, 86 (83%) (51 surgical and 35 nonsurgical) provided 1-year follow-up. Baseline Patient Characteristics and Findings Patient sociodemographic characteristics of those initially treated surgically were similar to those treated nonsurgically (Table 1). Most patients had symptoms for more than 6 months. Half had unilateral leg symptoms, abnormal physical examination findings were uncommon, and most had moderate or severe imaging findingsfall similar between treatment groups. However, there were important differences between those treated surgically and nonsurgically. Surgically treated patients had more-severe low back and leg pain, worse back-related function, and worse generic physical function (Table 2). Despite worse back-related symptoms and function, surgically treated patients had better SF-36 general health scores than nonsurgically treated patients. However, even for individual variables with significant differences between surgically and nonsurgically treated patients, there was considerable overlap. 19 Outcomes Over 1 Years Longitudinal outcomes over 1 years according to initial treatment group are presented in Figure 2. For low back pain, leg symptoms, and modified Roland back-specific functional status, unadjusted outcomes are presented as the change in score from baseline to follow-up, so that positive scores indicate improvement compared to baseline (Figure 2A C). For satisfaction, the unadjusted outcome is the absolute response reported at each follow-up, with higher scores indicating greater satisfaction (Figure 2D). For all outcomes and at each time point, surgically treated patients reported larger changes, indicative of greater improvement than nonsurgically treated patients. The figures demonstrate narrowing of the differences between the treatment groups over time. Although there Primary Analysis Secondary Analyses Baseline Date of Most recent enrollment subsequent surgery follow-up Initial surgical Exclude Follow-up excluded (2 analysis 1) Follow-up included: Reset to baseline and assign to surgical group (2 analysis 2) Initial nonsurgical Follow-up excluded (2 analysis 1) Exclude Figure 1. Diagram of analytic plan.

788 CHANG ET AL. MAY 25 VOL. 53, NO. 5 JAGS Table 1. Patient Characteristics at Baseline Evaluation Treatment Group Surgical Nonsurgical Characteristic (n 5 77) (n 5 67) P-value Age, mean SD 67.9 11.6 65.3 15.3.27 Female, % 64.6 59.7.63 Highest level of education, % High school 22.1 21.2.73 High school graduate 53.3 59.1 College graduate 24.7 19.7 Comorbid illnesses, yes, % w 59.7 68.2.3 Retired, % 57.1 45.5.16 Receiving or applying for workers compensation, % 6.5 15.2.12 Past episodes of back or leg pain, none, % 27.3 18.2.2 Length of current episode46 month, % 64 54.23 Past treatment, Epidural steroids, % 33.8 45.5.15 Abnormal exam findings, mean SD z 1.1.9 1.1.92 1. Radiographic image reviewed, n 48 3 Moderate or severe findings, % k 83.3 66.7.9 Note: Denominators differ slightly between variables because not all patients answered each question on the survey. P-values compare surgical and nonsurgical treatment groups using Fisher exact test or t test. w Any self-reported chronic pulmonary disease, heart disease, stroke, cancer, or diabetes mellitus. z Mean number of positive physical examination findings for a patient, including unilateral strength, sensation, or reflex abnormality (range 3 findings). Any computerized tomography, magnetic resonance imaging, or myelogram available for independent review. k Global rating from normal to severe by study neuroradiologist blinded to treatment group and clinical information. SD 5 standard deviation. were significant differences between surgical and nonsurgical patients for all outcomes at short-term and mid-term periods, the relative differences diminished over long-term follow-up. For example, the difference between treatment groups in low back pain was 1.9 at 3 months,.8 at 5 years, and 1.3 at 1 years (Figure 2A). The long-term (between 5 and 1 years) treatment effect was not statistically significant (P 5.1). Similarly, Figure 2D shows that the difference between treatment groups in satisfaction was 1.8 at 3 months,. at 5 years, and.4 at 1 years (P 5.9 for Table 2. Clinical Features of Patients at Baseline Evaluation Clinical Feature Treatment Group Surgical Nonsurgical (n 5 77) (n 5 67) P-value Symptoms Unilateral leg pain, % yes 49.4 52.3.73 Pain frequency in past week (range 6), mean SD w Low back pain 4.4 2. 3.6 2.2.2 Leg pain 4.9 1.5 3.2 2. o.1 Stenosis index (range 24), mean SD w 15.7 5.7 1.6 6.7 o.1 Functional and Disability Status Modified Roland score (range 23), mean SD w 16 4.2 12.5 6.2 o.1 Short Form-36 score (range 1), mean SD z Physical function 28.3 23.4 44.7 27.1 o.1 Bodily pain 17.2 16. 35.2 23. o.1 Role emotional 51.6 45.9 5.5 43.6.89 Mental health 67.7 2.8 64.9 22.3.44 General health 75. 2.1 66.1 21.2.1 Note: Denominators differ slightly between variables because not all patients answered each question on the survey. P-values compare surgical and nonsurgical treatment groups using Fisher exact test or t test. w Higher values indicate worse outcomes. z Higher scores indicate better function. SD 5 standard deviation.

Satisfaction with Current Status JAGS MAY 25 VOL. 53, NO. 5 TREATMENT OUTCOMES OF SPINAL STENOSIS 789 long-term treatment effect). These findings were confirmed by examining the interaction between treatment group and time period in models adjusting for covariates. With the exception of the modified Roland functional status score, outcomes showed a significant interaction between treatment group and time period (P 5.3 for back pain, P 5.3 for leg symptoms, P 5.8 for Roland functional status, and Po.1 for satisfaction). Effect of Subsequent Surgical Procedures on Outcomes Fifteen of 77 patients initially treated surgically underwent a second operation during follow-up: one during the first year, seven between Year 1 and Year 5, and seven after Year 5. The 1-year reoperation rate after taking censoring into account was 23%. Twenty-one of 67 patients initially treated nonsurgically underwent a surgical procedure (crossed over) during follow-up: four between 3 and 12 months, 11 between Year 1 and Year 5, and six after Year 5. The 1-year crossover rate after taking censoring into account was 38%. Unadjusted outcomes over 1 years according to whether initial treatment was maintained were examined. Regardless of initial treatment received, patients undergoing subsequent surgical procedures reported less improvement in outcomes over time than patients who continued with their original treatment cohort. For example, Figure 3 depicts change in leg symptom frequency scores over time according to initial treatment cohort and whether patients underwent a subsequent surgical procedure. Surgical patients who underwent a reoperation had less favorable outcomes than those who only had an initial operation. Because an intention-to-treat analysis for the surgical group includes these less-favorable responses after a reoperation, such analyses would understate the magnitude of benefit of successful initial surgical treatment. Similarly, the responses of patients initially treated nonsurgically who underwent subsequent surgical treatment were also less favorable than the responses of those who did not crossover. Therefore, including these crossover patients in the intention-to-treat analysis for the nonsurgical group would also reduce the magnitude of improvement observed in those responding to initial treatment. If data from patients after surgical reoperation or nonsurgical crossover are excluded (performing an as treated analysis), the improvement reported by both treatment groups is larger than in the intention-to-treat analysis. However, the relative benefit of surgical and nonsurgical treatment in such as-treated analyses was similar to the intention-to-treat analyses. Outcomes continued to favor surgical treatment for all outcome measures at the short, mid, and long term (all P.5). As with the intention-to-treat analysis, there was narrowing of the relative treatment group differences over time for leg symptom frequency and satisfaction with the current state (interaction between treatment group and time period significant at P 5.1 for both outcomes) but not for low back pain frequency and Roland functional status (interaction between treatment group and time period P 5.9 and P 5.24, respectively). Finally, analyses were performed that excluded data after reoperation for surgical patients who underwent a second surgery but reassigned follow-up data for nonsurgical patients after crossing over to the surgical group. The A Change in Back Pain Frequency 4 3 2.5 2.6 2.4 2 1 1 5 11.1 1.4 1.8 2.2 9.5 2.4 9.8 2.3 9.2 2. 1.2 8.8 1.8 1.7 9.2 8.6 2.1 2..6.7.8.8.6.7.7.6.5.4.4 1 2 3 4 5 6 7 8 9 1 B 15 Change in Leg Symptom Frequency 1 5 8.6 8.8 8.3 7.8 8.3 7.7 7.5 7.7 6.8 9.5 7.5 1.9 9.1 9.4 4.2 3.6 3.6 2.8 2.9 2.3 2.3 2.3 1.5 1.7 1.3 1.7 1 2 3 4 5 6 7 8 9 1 C 15 Change in Modified Roland Score 4. 3.9 4. 4 2 2.2 3. 2.7 3.8 3.7 3.7 3.6 3.6 3.5 3.1 3. 3.2 3.6 3.3 3.4 3.3 3.7 3.8 7.1 6.8 4.2 3.1 2.6 2.2 2.4 2.6 1.3 2. 1.6 1.8.3 1.8 1 2 3 4 5 6 7 8 9 1 D 6 1 2 3 4 5 6 7 8 9 1 Figure 2. Ten-year outcomes according to initial treatment group. Change in outcomes assessed at 3, 6, and 12 months and then yearly follow-up for initial nonsurgical ( ) and surgical ( ) treatment. Higher scores indicate improvement. Numbers are mean score at each follow-up point. A. Frequency of back pain symptoms. B. Leg symptom frequency index. C. Modified Roland back-specific functional status score. Change in unadjusted mean scores between baseline and each follow-up time point. D. Satisfaction with current state. Unadjusted satisfaction score at each follow-up time point. 2.9 3.6 3.2

79 CHANG ET AL. MAY 25 VOL. 53, NO. 5 JAGS Change in Leg Symptom Frequency 15 11 12 12 11 31 32 38 4 33 28 9 44 39 39 33 35 37 12 11 6 9 14 12 11 12 18 23 18 26 19 18 3 3 25 27 21 19 17 14 14 17 16 11 3 14 6 1 1 2 3 4 5 6 7 8 9 1 Figure 3. Effect of subsequent surgical treatment on change in leg symptom frequency index. Change in unadjusted mean scores between baseline and each follow-up time point for initial nonsurgical and surgical treatment according to whether (nonsurgical ( ) or surgical ( )) or not (nonsurgical cross over ( )or surgical reoperation ( ))a subsequent surgical procedure occurred. Higher scores indicate improvement. The numbers indicate patients providing data at each follow-up point. date of surgery was considered as the new baseline for the follow-up data. Nonsurgical patients who subsequently crossed over to surgery had less-favorable responses than those initially treated surgically; therefore, adding these patients to the surgical group reduced the overall effect of surgery. The relative benefit of surgical and nonsurgical treatment in this analysis continued to favor surgical treatment, but the differences in leg symptom frequency scores no longer reached statistical significance at the mid- and long-term time periods (P 5.8 and.82, respectively). As with the as-treated analysis, the interactions between treatment group and time period were significant at P 5.1 for leg symptom frequency and satisfaction with the current state but not for frequency of low back pain and Roland functional status (P 5.15 and P 5.33, respectively). Effect of Treatment Outcomes Over Time After Adjusting for Covariates Because treatment was not randomly allocated, the analysis adjusted for covariates that may be associated with outcomes in patients initially treated surgically or nonsurgically. Baseline variables included age, sex, and baseline score (except satisfaction outcome). The analysis also controlled for general health status at each follow-up because patients with spinal stenosis were known to have important comorbid conditions at baseline and a high mortality rate over the course of follow-up. Patients initially treated surgically were slightly older, more likely to be female, and had worse baseline symptoms and functional status but reported better general health scores over time (Tables 1 and 2). More-severe baseline low back pain and leg symptoms and worse functional status were highly associated with their respective outcomes. Regression to the mean was adjusted for by including a term for baseline severity. Because surgical patients generally had worse baseline findings, adjusting for baseline severity decreased the relative effect of initial surgical treatment. For example, the unadjusted mean difference A 15 Change in Leg Symptom Frequency 1 5 B 12 Change in Modified Roland Score 1 2 3 4 5 6 7 8 9 1 9 6 3 1 2 3 4 5 6 7 8 9 1 Figure 4. Effect of covariate adjustment on change in modified Roland back-specific functional status score. Change in unadjusted mean scores between baseline and each follow-up time point for initial nonsurgical ( ) and surgical ( ) treatment and after models adjusting for baseline score (nonsurgical ( ) and surgical ( )) and baseline score, age, sex, and follow-up general health perceptions (nonsurgical ( ) and surgical ( )). Higher scores indicate improvement. A. Leg symptom frequency index. B. Modified Roland back-specific functional status score. between the two treatment groups was about 1.7 at Year 3 for the low back pain score. After adjusting for the baseline score, the mean difference was reduced to.8. Figure 3 shows the effect of adjusting for baseline score on change in leg symptom frequency and Roland functional status for patients initially treated surgically and nonsurgically. The unadjusted mean difference between treatment groups at Year 4 was about 5.7 for the leg symptom frequency score. It was reduced to 2.5 after adjusting for the baseline scorefa 56% reduction. Over the 1-year follow-up, adjusting for baseline score reduced the treatment group difference from 32% at 6 months to 73% at 5 years. For the Roland functional status outcome, baseline score accounted for less of the mean difference between treatment groups, about 2%. Additional adjustment for general health status over time and for age and sex had a smaller effect on the relative treatment cohort outcomes (Figure 4). In general, older age was associated with smaller changes in symptom and functional status scores and higher levels of satisfaction. However, after adjusting for baseline score and general health

JAGS MAY 25 VOL. 53, NO. 5 TREATMENT OUTCOMES OF SPINAL STENOSIS 791 score, older age was a predictor for greater change in symptom and functional status scores in the multivariable models. Sex was not found to be a significant predictor of any outcome. As expected, worse general health status over time was associated with less improvement in symptoms, functional status, and satisfaction. Finally, even after adjusting for these variables, initial surgical treatment was associated with improved, albeit smaller, outcomes than nonsurgical treatment over time. As with unadjusted intention-to-treat analyses, adjusted outcomes showed a significant interaction between treatment group and time period, with the exception of the modified Roland functional status score (P 5.4 for back pain, P 5.2 for leg symptoms, P 5.8 for Roland functional status, and P 5.1 for satisfaction). DISCUSSION Better outcomes for initial surgical treatment of lumbar spinal stenosis at 1 and 4 years and for certain outcomes at 1 years than for initial nonsurgical treatment have previously been reported. 16,19,2 In this study, longitudinal analytical techniques were employed to examine the relative effect of treatment over the short, mid, and long term. Outcomes of interest to patients and providers included validated measures of low back pain, leg symptoms, backspecific functional status, and satisfaction with current state. Because this was an observational cohort study, analyses included adjustments for potential covariates to determine their effect on relative treatment group and time period comparisons. There was also a high rate of subsequent surgical procedures in this population; therefore their effect on the relative treatment group comparisons was assessed. The current results demonstrate that surgical treatment is associated with significantly better outcomes over time but that the magnitude of the benefit diminishes over long-term follow-up. After 5 years, change in leg symptoms and functional status continued to favor surgical treatment, but change in frequency of low back pain and satisfaction with the current state were no longer statistically significant. Despite being an increasingly common diagnosis in aging individuals, 4,5,26 few studies have assessed long-term outcomes of patients with lumbar spinal stenosis, and the authors are not aware of any that have used newer longitudinal analytical techniques. Although randomized studies are ongoing,,14 a single, small trial randomly assigned 31 patients ( to surgery and 18 to nonsurgical treatment) with lumbar spinal stenosis and followed them over 1 years. 12 Ten of 18 individuals randomly assigned to nonsurgical treatment had crossed over to surgery after a median of 3.5 months. Of 28 of 31 patients completing 1-year follow-up, good treatment results were reported for 1 of 11 (91%) surgical patients, eight of eight (1%) nonsurgical patients, and four of nine (44%) nonsurgical patients crossing over to surgery. The study did not use statistical testing to compare these results or assess longitudinal trends. Available long-term outcomes data come from observational studies that generally involved surgical treatment using unvalidated outcomes retrospectively assessed. These studies have generally found improvement with surgical treatment, but the absolute benefit has been highly variable. 12,17,21 23,27 31 Fewer studies have assessed trends over time, and these generally have reported declines in the magnitude of initial surgical benefit. 21 23 Fewer studies have included nonsurgical case series, 12,32 35 and their length of follow-up has generally been shorter than for surgical case series. Outcomes of nonsurgical treatment are also highly variable, along with rates of subsequent surgical treatment, but there has been little longitudinal assessment of outcomes. 2 Assessing long-term outcomes of aging patients with spinal stenosis is made more difficult because the disabling musculoskeletal condition is occurring in the context of other age-related problems. These other comorbid conditions likely account for the mortality rate seen in this study rather than being directly related to the back condition itself (the survival rate was similar in patients initially undergoing surgical and nonsurgical treatment). Although elderly patients may have comorbid conditions that increased their surgical risk and decreased their postoperative life span, highly symptomatic patients with impaired function should not necessarily be excluded from these interventions. Future studies should continue to explore the effect of other morbidity over time on outcomes of treatment using more-detailed comorbidity measures. The high rate of subsequent surgical procedures in this study has been found in some 12,17,18 but not all prior reports. 22 The multiple sources used, such as the physician, patient, and regional hospital discharge records, likely account for the high rates reported in this study. Because subsequent surgical procedures will occur regardless of whether a study is observational or experimental, methods to assess their potential effect are important. At first glance, the analyses may provide reassurance that these subsequent surgical procedures will have a similar effect on initial treatment groups, but if rates of or indications for subsequent surgery differ by initial treatment group, such as-treated analyses may take on a more important role. Additionally, these rates would need to be incorporated into the design of the study because of their potential influence on sample size. Because treatment was determined in a routine clinical manner, baseline differences showed that patients initially treated surgically had more-severe symptoms and greater functional impairment than those treated nonsurgically. To control for variables that may relate to outcomes observed in addition to initial treatment, linear models included terms for baseline score as well as age and sex. Because change in other comorbid conditions may influence the back-related outcomes assessed, the models also controlled for time-varying general health perceptions. Little is known about factors associated with longitudinal treatment outcomes in this population, so other important covariates may not be accounted for in these models. Larger studies involving various treatments and using prospectively assessed, validated measures are needed to better understand predictors of favorable outcomes. Strengths of this study include prospective follow-up from contemporary comparison groups treated in community-based practices using validated outcome measures. In addition, the multivariate models examined relative treatment group changes over time by assessing individual

792 CHANG ET AL. MAY 25 VOL. 53, NO. 5 JAGS patient data rather than traditional analyses that compare group means between two time points. The study s major limitation is its observational, nonrandomized design. 15,16,19,2 Because treatment assignment was not randomized, surgically treated patients had worse symptoms and function at baseline. Despite controlling for differences in potential covariates between treatment groups, the relative outcomes observed cannot be definitively stated to be due to the treatment received rather than to persistent or unmeasured confounders. The small sample size and high mortality rate may have diminished the ability to detect clinically important treatment group differences. Finally, because few patients underwent fusion procedures, this study could not compare outcomes of those treated with fusion with outcomes of those treated with decompression laminectomy alone or with nonsurgical care. In conclusion, for patients with lumbar spinal stenosis, initial surgical treatment provided greater improvement in low back pain, leg symptoms, functional status, and satisfaction than initial nonsurgical treatment over 1 years follow-up. However, the magnitude of the benefit of surgical treatment declined over time so that frequency of low back pain and satisfaction outcomes were no longer statistically different between surgical and nonsurgical groups in outcomes between 5 and 1 years. Leg symptom frequency and back-related functional status continued to favor those initially receiving surgical treatment even after 5 years. A large number of patients, regardless of initial treatment, had subsequent surgery over long-term follow-up and had worse outcomes than those who continued with their original treatment. Analyses that examined as-treated outcomes were similar to intention-to-treat comparisons. For patients and their healthcare providers, these results indicate that surgery may offer better outcomes than nonsurgical treatment over a number of years. For patients reluctant to undergo surgery, they can elect initial conservative care knowing that their symptoms will likely remain stable and that some long-term outcomes are similar. Better information from larger, controlled studies is needed to help patients and physicians make informed decisions about treatment options that are based upon their symptoms, functional impairment, other limiting medical conditions, and personal preferences. 36,37 REFERENCES 1. Yelin E, Herrndorf A, Trupin L et al. A national study of medical care expenditures for musculoskeletal conditions. Arthritis Rheum 21;44:116 1169. 2. Verbiest H. A radicular syndrome from developmental narrowing of the lumbar vertebral canal. J Bone Joint Surg 1954;36-B:23 237. 3. Kent DL, Haynor DR, Larson EB et al. Diagnosis of lumbar spinal stenosis in adults: A meta-analysis of the accuracy of CT, MR, and myelography. AJR Am J Roentgenol 1992;158:15 1144. 4. Ciol MA, Deyo RA, Howell E et al. An assessment of surgery for spinal stenosis: Time trends, geographic variations, complications, and reoperations. J Am Geriatr Soc 1996;44:285 29. 5. Taylor VM, Deyo RA, Cherkin DC et al. Low back pain hospitalization. Recent United States trends and regional variations. Spine 1994;19:127 12. 6. Dartmouth Atlas Working Group,Weinstein J. Dartmouth Atlas of Musculoskeletal Health Care. Wennberg JE, Cooper MM, eds. Dartmouth Atlas of Health Care Series. Chicago: American Hospital Association Press, 2. 7. Bombardier C. Outcomes assessments in the treatment of spinal disorders. Spine 2;25:31 3. 8. Deyo RA, Andersson G, Bombardier C et al. Outcomes measures for studying patients with low back pain. Spine 1994;19:232S 236S. 9. Gibson JNA, Grant IC, Waddell G. The Cochrane review of surgery for lumbar disc prolapse and degenerative lumbar spondylosis. Spine 1999;24:182 1832. 1. Fritz JM, Delitto A, Welch WC et al. Lumbar spinal stenosis: A review of current concepts in evaluation, management, and outcome measurements. Arch Phys Med Rehabil 1998;79:7 78. 11. Turner JA, Ersek M, Herron L et al. Surgery for lumbar spinal stenosis: Attempted meta-analysis of the literature. Spine 199;17:1 8. 12. Amundsen T, Weber H, Nordal HJ et al. Lumbar spinal stenosis: Conservative or surgical management? A prospective 1-year study. Spine 2;25:1424 1436.. Birkmeyer NJ, Weinstein JN, Tosteson AN et al. Design of the Spine Patient Outcomes Research Trial (SPORT). Spine 22;27:61 72. 14. Malmivaara A, Slatis P, Heliovaara P et al. Operative treatment for moderately severe lumbar spinal stenosis. A randomized controlled trial. Presented at the Annual Meeting of the International Society for the Study of the Lumbar Spine. Vancouver, Canada: May 23. 15. Keller RB, Atlas SJ, Singer DE et al. Maine Lumbar Spine Study, Part I background concepts. Spine 1996;21:1769 1776. 16. Atlas SJ, Singer DE, Wu YA et al. Long-term outcomes of surgical and nonsurgical management of lumbar spinal stenosis: 8 1 year results from the Maine Lumbar Spine Study. Spine, in press. 17. Hee HT, Wong HK. The long-term results of surgical treatment for spinal stenosis in the elderly. Singapore Med J 23;44:175 18. 18. Herno A, Airaksinen O, Saari T. Long-term results of surgical treatment of lumbar spinal stenosis. Spine 1993;18:1471 1474. 19. Atlas SJ, Deyo RA, Keller RB et al. The Maine Lumbar Spine Study Part III. 1- year outcomes of surgical and nonsurgical management of lumbar spinal stenosis. Spine 1996;21:1787 1795. 2. Atlas SJ, Keller RB, Robson D et al. Surgical and nonsurgical management of lumbar spinal stenosis: Four-year outcomes from the Maine Lumbar Spine Study. Spine 2;25:556 562. 21. Javid MJ, Hadar EJ. Long-term follow-up review of patients who underwent laminectomy for lumbar stenosis: A prospective study. J Neurosurg 1998; 89:1 7. 22. Katz JN, Lipson SJ, Chang LC et al. Seven to 1-year outcome of decompressive surgery for degenerative lumbar spine disorders. Spine 1996;21: 92 98. 23. Scholz M, Firsching R, Lanksch WR. Long-term follow up in lumbar spinal stenosis. Spinal Cord 1998;36:2 24. 24. Patrick DL, Deyo RA, Atlas SJ et al. Assessing health-related quality of life in patients with sciatica. Spine 1995;2:1899 199. 25. Ware JE, Sherbourne CD. The MOS 36-item short-form survey (SF-36): 1. Conceptual framework and item selection. Med Care 1992;3:473 483. 26. Jansson KA, Blomqvist P, Granath F et al. Spinal stenosis surgery in Sweden 1987 1999. Eur Spine J 23;12:535 541. 27. Cornefjord M, Byrod G, Brisby H et al. A long-term (4- to 12-year) follow-up study of surgical treatment of lumbar spinal stenosis. Eur Spine J 2;9:563 57. 28. Hurri H, Slatis P, Soini J et al. Lumbar spinal stenosis: Assessment of long-term outcome 12 years after operative and conservative treatment. J Spinal Disord 1998;11:11 115. 29. Iguchi T, Kurihara A, Nakayama J et al. Minimum 1-year outcome of decompressive laminectomy for degenerative lumbar spinal stenosis. Spine 2;25:1754 1759. 3. Jolles BM, Porchet F, Theumann N. Surgical treatment of lumbar spinal stenosis: Five-year follow-up. J Bone Joint Surg [Br] 21;83:949 953. 31. Rompe JD, Eysel P, Zollner J et al. Degenerative lumbar spinal stenosis. Longterm results after undercutting decompression compared with decompressive laminectomy alone or with instrumented fusion. Neurosurg Rev 1999;22:12 16. 32. Delport EG, Cucuzzella AR, Marley JK et al. Treatment of lumbar spinal stenosis with epidural steroid injections: A retrospective outcome study. Arch Phys Med Rehab 24;85:479 484. 33. Johnsson KE, Rosen I, Uden A. The natural course of lumbar spinal stenosis. Clin Orthop 1992;279:82 86. 34. Simontas AC, Dorey FJ, Hansraj KK. Nonoperative treatment for lumbar spinal stenosis. Clinical and outcome results and a 3-year survivorship analysis. Spine 2;25:197 24. 35. Swezey RL. Outcomes for lumbar stenosis: A five-year followup study. J Clin Rheumatol 1996;2:129 4. 36. Deyo RA, Cherkin DC, Weinstein J et al. Involving patients in clinical decisions: Impact of an interactive video program on use of back surgery. Med Care 2;38:959 969. 37. Goldberg HI, Deyo RA, Taylor VM et al. Can evidence change the rate of back surgery? A randomized trial of community-based education. Eff Clin Pract 21;4:95 14.