Utilization and Clinical Outcomes of Outpatient Physical Therapy for Medicare Beneficiaries With Musculoskeletal Conditions

Background Medicare beneficiaries frequently receive physical therapy for musculoskeletal conditions. Little information is available about this care. Objective The purposes of this study were: (1) to describe characteristics, clinical outcomes, and utilization for Medicare beneficiaries receiving physical therapy in outpatient clinics within one integrated health care system; (2) to compare characteristics, outcomes, and utilization based on the body region affected; and (3) to examine factors predictive of outcomes and utilization. Design This was a prospective, longitudinal study. Methods Medicare beneficiaries aged 65 years or older (n=1,840 episodes of care) participated in the study. Descriptive statistics were calculated for patient characteristics and outcomes. Comparisons were made based on body region. Regression models evaluated factors associated with change in pain, improved outcome, and utilization. Results The patients' mean age was 74.2 years (SD=6.3), and 65.3% were female. The most common body regions were the lumbar spine, shoulder, and knee, collectively accounting for 71.3% of the episodes of care. Patients attended a mean of 6.8 visits (SD=4.7), and 63.9% experienced an improved outcome. Episodes of care for lumbar spine conditions had less reduction in pain, whereas shoulder conditions and foot/ankle conditions showed the greatest improvement. Care for hip conditions was least likely to result in an improved outcome. Knee conditions were most likely to have an improved outcome. Care for shoulder and knee conditions had the highest number of visits. Factors associated with greater reduction in pain and improved outcomes included greater initial pain or disability and attending more visits. Factors associated with greater utilization included a postsurgical condition and higher initial pain rating. Limitations The study was performed in one geographic region within a single health care delivery system. Conclusion The results provide information on outcomes of physical therapy for Medicare beneficiaries in one health care system. Further research is needed to examine optimal utilization and care for these patients.

M usculoskeletal conditions are highly prevalent in the United States and throughout the world. Musculoskeletal conditions can significantly diminish an individual's quality of life and impose a substantial economic burden on both individuals and society. 1 In the United States, approximately 1 in 4 individuals is affected by a musculoskeletal condition, and these conditions are a leading cause of work disability. 2 There is some evidence to suggest the prevalence of musculoskeletal conditions is growing. 3 The ubiquity and impact of musculoskeletal conditions have been recognized by the United Nations, the World Health Organization, and numerous government agencies through support for the Bone and Joint Decade initiative. 4 Musculoskeletal conditions are particularly prevalent among older adults. Results of the 2000 Health and Retirement Survey, administered to adults aged 65 or older across the United States, showed musculoskeletal conditions to be the most prevalent chronic disorder in this age group. 5 Pain related to musculoskeletal conditions has been reported to be present in about half of individuals older than age 70 years, with a higher prevalence in women. 6 Musculoskeletal conditions are the leading cause of disability in older adults 7 and have been associated with lower self-perceptions of general health, increased risk of dependence in activities of daily living, hospitalizations, and mortality. 5,8,9 Many older adults with musculoskeletal conditions seek outpatient physical therapy services. A large, nationally representative survey of individuals over 18 years of age referred by primary care providers for outpatient physical therapy showed 30% were aged 60 years or older, 10 and the majority of these older adults were receiving physical therapy for musculoskeletal conditions. 10,11 There is research evidence that physical therapy interventions can benefit many common musculoskeletal conditions in older adults. [12][13][14][15][16][17] Little research, however, is available describing the outcomes of usual physical therapy care for older adults with musculoskeletal conditions. 18 Many older adults receiving outpatient physical therapy services are insured through Medicare part B coverage. The Centers for Medicare & Medicaid Services (CMS) reported that 8.5% of all Medicare beneficiaries received outpatient physical therapy services during 2006, equating to 3.9 million individual patients and more than $3 billion in expenditures. 19 One national survey estimated that 15% of all patients who received outpatient physical therapy were insured through Medicare. 10 The utilization of outpatient physical therapy by Medicare beneficiaries continues to expand, despite legislative changes and the imposition of financial limitations (ie, therapy caps). 19 -21 The continued growth of Medicare has led to calls for reform of the current fee-for-service structure, with a move toward alignment of payment strategies with highquality and efficient care. [22][23][24] Moving toward this goal requires information on both the process of care and clinical outcomes of services provided. The purposes of this study were: (1) to describe the baseline characteristics, clinical outcomes, and physical therapy utilization of Medicare beneficiaries receiving outpatient physical therapy for musculoskeletal conditions in communitybased, non-hospital clinics of a single integrated health care system in Utah; (2) to compare baseline characteristics, clinical outcomes, and utilization among Medicare beneficiaries based on the body region affected by the musculoskeletal condition; and (3) to examine factors predictive of clinical outcomes and utilization for Medicare beneficiaries as a whole and based on the body region affected.

Method Data Source
Data for this study were collected from 8 outpatient physical therapy clinics of Intermountain Healthcare Inc (IHC), a private, nonprofit, integrated health care delivery system. Each clinic is located in the region around Salt Lake City, Utah, and is a community outpatient facility that is not hospital-based. All physical therapists working in the participating clinics were salaried employees of IHC. Since 2002, each of these clinics has been tracking clinical outcomes for all patients receiving physical therapy. Data obtained from each new patient are entered into an Internet-based electronic database. At each visit, a region-specific disability outcome questionnaire and a numeric pain rating scale (NPRS) are collected from the patient, and the data are entered into the database. Additional elements entered into the database by the physical therapist performing the initial evaluation include the patient's sex and date of birth, date of onset of the patient's current symptoms or date of surgery, the body region involved, and the patient's diagnostic subgroup. An NPRS, which asked the patient to "rate your current level of pain" on a 0 to 10 scale anchored with the phrases "no pain" and "worst imaginable pain," was used for all patients regardless of the nature or location of the chief complaint. The NPRS has been used frequently in research for people with a variety of musculoskeletal conditions. Although there is some variation in the literature, the minimum clinically important difference (MCID) on the NPRS has been estimated at or slightly less than 2 points of change for patients with various musculoskeletal conditions. [25][26][27] We used 2 points of change as the MCID for change on the NPRS for all patients in this study as a somewhat conservative estimate.
The region-specific disability questionnaire used for each patient was based on the body region of the patient's chief complaint at the time of the initial physical therapy session. A modified version of the Oswestry Disability Questionnaire (OSW) was used for patients with a chief complaint related to the lumbar or thoracic spine. 28 The OSW is a 10-item scale, with the total score expressed on a scale of 0 to 100 points. Higher scores indicate greater disability. The MCID for this version of the OSW has been estimated as 6 points. 28 The Neck Disability Index (NDI) was the questionnaire used for patients with a chief complaint of neck pain. 29 The NDI is similar in structure to the OSW, with the total score expressed on a scale of 0 to 100 points. Higher scores are indicative of greater disability. The MCID for the NDI has been estimated as 7 points. 26 27 The Lower Extremity Function Scale (LEFS) was used for conditions affecting the lower extremity other than the knee (ie, hip, foot and ankle). 33 The LEFS has 20 items, and the total score ranges from 0 to 80, with lower scores indicating greater disability. The MCID for the LEFS is estimated as 9 points. 33,34

Study Sample
The sample for this study included episodes of care provided to patients with musculoskeletal conditions in a participating clinic between January 1, 2006, and December 31, 2008, in the clinical outcomes electronic database. An episode of care was defined as time from the date of the initial evaluation to the last visit. If no visits occurred for more than 45 days, we presumed the continuity of the care would likely be disrupted, and the episode of care was considered complete. Care provided for a musculoskeletal condition was identified by selecting patients for whom a body region-specific disability questionnaire was used. Additional criteria for inclusion in the analysis were: age 65 years or older on the date of the initial physical therapy session, at least 2 physical therapy visits

The Bottom Line
What do we already know about this topic?
Musculoskeletal conditions are common among older adults insured by Medicare. Many Medicare enrollees with musculoskeletal conditions receive physical therapy, yet little information is available on the types of physical therapy provided or the outcomes of physical therapy for these patients.

What new information does this study offer?
Low back pain was the most common musculoskeletal condition among the Medicare enrollees in this study, followed by shoulder and knee conditions. The majority of patients improved with physical therapy treatment. Factors associated with better outcomes of physical therapy included greater initial pain, greater initial disability, and attending more physical therapy sessions.
If you're a patient, what might these findings mean for you?
The study results suggest that physical therapy is helpful for older adults with musculoskeletal conditions.

Outpatient Physical Therapy for Medicare Beneficiaries With Musculoskeletal Conditions
included in the episode of care, and Medicare insurance provider for the episode of care. Records of patients meeting the inclusion criteria were linked to the billing database using the enterprise master patient index number. The primary insurance provider to which the episode of care was billed and the number of physical therapy sessions during the episode of care were extracted from the billing database. The insurance provider was categorized as Medicare, Medicaid, private, workers' compensation, or "other" (eg, self-pay, charity care).

Patient and Care Characteristics
For each included episode of care, the length of stay was recorded as the number of days between the initial and final physical therapy visits. The body region involved with the patient's musculoskeletal condition was recorded based on the data indicating the body region of the patient's chief complaint that were entered by the physical therapist into the electronic database. Body region was categorized as lumbar spine, cervical/thoracic spine, hip, knee, foot/ankle, shoulder, or elbow/wrist/hand. Within each body region, the diagnostic subgroup, as recorded in the electronic database, also was extracted. For patients with musculoskeletal conditions involving the spine (lumbar or cervical/thoracic), the diagnostic subgroup was recorded as: (1) postsurgical, (2) nonsurgical acute (onset of current symptoms Ͻ90 days), (3) nonsurgical chronic (onset of current symptoms Ն90 days), or (4) nonsurgical radicular. For patients with musculoskeletal conditions involving peripheral body regions (hip, knee, foot/ ankle, shoulder, or elbow/wrist/ hand), diagnostic subgroups were recorded as: (1) surgical (joint arthroplasty, ligament repair, and so on, depending upon the region), (2) nonsurgical soft tissue condition (eg, tendinitis, bursitis), (3) nonsurgical degenerative joint condition, or (4) other region-specific nonsurgical condition.

Clinic and Physical Therapist Characteristics
The proportion of episodes of care provided to older adult Medicare beneficiaries was recorded for each participating physical therapy clinic by dividing the number of episodes of care provided to individuals aged 65 years or older at the time of the initial visit who were Medicare beneficiaries by the total number of episodes of care provided during the study period. A similar procedure was used to calculate the proportion of older adult Medicare beneficiaries for each physical therapist included in the study sample.

Clinical Outcomes and Utilization
Clinical outcomes of care for included patients were based on the change in the NPRS score and region-specific disability score from the initial physical therapy session to the final session. Change in pain was calculated for each patient by subtracting the final NPRS score from the initial score for the patient. The points of change on the regionspecific disability questionnaire were used to categorize the episode of care for each patient as "improved" if the improvement from the initial to the final session met or exceeded the MCID for the particular questionnaire used with the patient. If the patient did not improve by an amount at least equal to the MCID value, the episode of care was categorized as "not improved." Patients whose initial disability score was less than the MCID value were categorized as improved if the discharge disability score indicated no disability (ie, a score of 0 for the DASH, NDI and OSW, a score of 100 for the KOS, or a score of 80 for the LEFS). Utilization outcomes included the number of physical therapy sessions attended during the episode of care. The duration of care was calculated as the number of days from the initial physical therapy session to the final session of the episode of care.

Data Analysis
Data analysis was performed using SPSS statistical software (version 17.0*).
Description of baseline characteristics, clinical outcomes, and utilization. Descriptive statistics were calculated for all Medicare beneficiaries, including means and standard deviations for normally distributed variables, medians with interquartile range for variables with skewed distributions (length of stay and duration of current symptoms), and frequency counts with percentages for categorical variables. Descriptive characteristics were calculated based on the body region involved (eg, lumbar spine, hip, knee). In addition, we evaluated diagnostic subgroupings within each body region and calculated descriptive statistics for any subgroup with at least 50 episodes of care to provide more-specific information for the most commonly encountered conditions.
Comparison of baseline characteristics, clinical outcomes, and utilization by body region. Baseline characteristics, physical therapy utilization, and clinical outcomes were compared among episodes of care based on body region using chisquare and one-way analysis of variance procedures for categorical and continuous baseline characteristics, respectively. Variables with skewed distributions were compared using Kruskal-Wallis tests. Comparison of baseline NPRS scores among body regions was performed using linear mixed model analysis in order to risk-adjust scores for available demographic data, including age, sex, and duration of symptoms, and for the effect of clustering of patients by physical therapist and within clinics using mixed model analysis. 35 Age, sex, and duration of symptoms were included as fixed-effect covariates. Because of the degree of skewness, duration of symptoms was transformed to an ordinal variable as acute (0 -29 days), subacute (30 -89 days), or chronic (90 or more days). The physical therapist responsible for the episode of care, nested within the physical therapy clinic, was included as a random factor with a variance components covariance structure. Comparison of riskadjusted change in NPRS scores among different body regions was performed in a similar manner, with addition of baseline NPRS score as a fixed effect covariate. Comparisons of disability scores among different body regions were not performed because different questionnaires were used. The proportion of episodes of care resulting in improvement were compared using multivariate logistic regression, controlling for age, sex, categorized duration of symptoms, the physical therapist and physical therapy clinic, and the proportion of Medicare beneficiaries at the physical therapist and physical therapy clinic level. A significance level of .05 was used for all comparisons.

Examination of predictors of clinical outcomes and utilization.
We examined predictors of clinical outcomes and physical therapy utilization using regression models. Separate models were developed for the most common body regions (lumbar spine, cervical/thoracic spine, shoulder, and knee). Regression models were not developed for body regions with fewer than 200 episodes of care (hip, foot/ankle, and elbow/wrist/ hand) due to concerns of overfitting the models based on small sample sizes. Patient characteristics (age, sex, symptom duration, postsurgical status, initial pain score), physical therapist and clinic characteristics (proportion of Medicare beneficiaries), and physical therapy utilization variables (number of visits, length of stay) were examined as potential predictors of clinical outcomes. An ordinary least squares model was constructed with change in pain (ie, change in NPRS score) as the dependent variable. An additional model was developed with the number of physical therapy visits as the dependent variable. The utilization variables were removed as potential predictors from this model. Assumptions of homoskedasticity and linearity between the independent and dependent variables were examined by plotting the residuals versus the predicted values. Normal probability plots of residuals were evaluated. Variance inflation factor (VIF) statistics were used to evaluate collinearity. A logistic regression model was developed using improved outcome as the dependent variable. Model fit was tested using the Hosmer-Lemeshow test and collinearity was examined with the VIF statistics. Linearity of the continuous independent variables was examined using the Box-Tidwell test.

Results
During the period of this study, a total of 14,060 episodes of care were provided for patients with musculoskeletal conditions in participating clinics, of which 2,396 (17.0%) were provided for individuals aged 65 years or older at the time of the initial visit. Among episodes of care provided for those aged 65 years or older, 332 were provided to individuals with private insurance (13.9%), 23 were provided to those receiving workers' compensation (1.0%), 5 (0.21%) were provided to Medicaid beneficiaries, 13 (0.54%) were provided to patients who had other coverage (self-pay or charity care), leaving 2,023 (84.4%) episodes of care provided to Medicare beneficiaries. Of the episodes of care provided to Medicare benefi-ciaries, 20 (0.99%) were not for a musculoskeletal condition, and 163 (8.1%) involved a single visit, leaving 1,840 episodes of care involving at least 2 visits. Patients attending a single visit had a mean age of 75.1 years (SDϭ6.5), and 72.3% were female. With respect to the body region involved, patients attending a single visit were less likely to have a cervical spine condition (Pϭ.002) and more likely to have a wrist/hand condition (PϽ.001). The 1,840 episodes of care with at least 2 visits represented 1,685 unique patients; 103 patients had 2 separate episodes of care during the study period, 18 patients had 3 episodes of care, 4 patients had 4 episodes of care, and 1 patient had 5 episodes of care.
The proportion of all episodes of care provided to Medicare beneficiaries within each participating physical therapy clinic ranged from 9.9% to 22.0%, with a mean of 14.9% (SDϭ4.8%). A total of 37 different physical therapists provided treatment for patients included in the study sample. The average number of Medicare beneficiaries treated by each therapist was 49.5, with a range from 1 to 205 patients. The proportion of Medicare beneficiaries managed by each physical therapist in his or her case load ranged from 1.4% to 24.3% (meanϭ9.9%, SDϭ5.9%).

Description of Baseline Characteristics, Clinical Outcomes, and Utilization
The study sample (nϭ1,840 episodes of care) had a mean age of 74.2 years (SDϭ6.3), and 1,201 (65.3%) were female. The most common body regions involved were the lumbar spine, followed by shoulder and knee. Patients were seen for physical therapy, on average, for 6.8 visits (SDϭ4.7) over a median of 27 days. Mean risk-adjusted change in NPRS scores for all episodes of care was 1.9 points (95% confidence inter-valϭ1.7, 2.0). The baseline disability Outpatient Physical Therapy for Medicare Beneficiaries With Musculoskeletal Conditions score was less than the MCID value for 54 episodes of care (2.9%). Thirty of these 54 episodes of care involved the DASH questionnaire: 24 for conditions affecting the shoulder (6.1% of all shoulder episodes of care) and 6 for conditions affecting the elbow/ wrist/hand (6.3% of all elbow/wrist/ hand episodes of care). The remaining episodes of care with baseline disability scores below the MCID value were 12 (1.9%) provided for lumbar spine conditions using the OSW, 4 (1.6%) provided for cervical/ thoracic spine conditions using the NDI, 2 (0.7%) provided for knee conditions using the KOS, and 5 (3.6%) provided for hip conditions and 1 (2.2%) provided for foot/ankle conditions using the LEFS. Overall, 63.9% of the episodes of care resulted in improvement, based on achieving at least an MCID level of change in disability across the episode of care (Tab. 1).

Comparison of Baseline Characteristics, Clinical Outcomes, and Utilization by Body Region
Comparison of baseline characteristics by body region involved showed patients with shoulder and lumbar spine conditions were more likely to be male than patients with cervical/ thoracic spine, hip, or foot/ankle conditions (PϽ.05) (Tab. 2). Lumbar and cervical/thoracic spine conditions were less likely to be postsurgical than conditions affecting any other body region (PϽ.001). Episodes of care provided for knee, shoulder, and elbow/wrist/hand conditions had shorter symptom durations than those provided for hip, foot/ankle, and lumbar and cervical/thoracic spine conditions (PϽ.05). Lumbar spine conditions had the highest initial pain ratings, significantly higher than shoulder conditions (Pϭ.002) and elbow/ wrist/hand conditions (Pϭ.002) (Tab. 2). Mean region-specific disability scores are reported for the intended disability questionnaire. There were instances of administration of an incorrect region-specific disability questionnaire. The LEFS was administered to 4 patients with lumbar spine conditions and to 18 patients with knee conditions. The OSW was used for 22 patients with thoracic region symptoms and 11 patients with hip conditions. Comparison of clinical outcomes among Medicare beneficiaries by body region showed the smallest

Examination of Predictors of Clinical Outcomes and Utilization
Regression models examining predictors of physical therapy clinical outcomes and utilization are reported in Tables 4, 5, and 6. Table  4 presents the predictors of change in pain across the episode of care.
Assumptions of the models were met, and no data transformations were undertaken. The contribution of individual predictors is expressed by the regression coefficient. A positive coefficient indicates the amount of change in pain was greater when more of the predictor variable was present, whereas a negative coefficient indicates the opposite relationship. The magnitude of the coefficient indicates the strength of the association between the predictor variable and the dependent variable and is interpreted as indicating the amount of increase or decrease in the dependent variable (change in pain) as a result of a 1-unit change in the predictor variable, with other predictors held constant. The models developed for different body regions explained 26% to 43% of the variability in the change in pain occurring with physical therapy. A higher initial pain rating was a predictor of change in pain for all regions examined. Additional predictors by body region are outlined in Table 4.
Logistic regression models examining predictors of an improved outcome are presented in Table 5. The predictor variables of the proportion of Medicare beneficiaries at the clinic and physical therapists were dichotomized for these analyses due to evidence of nonlinearity. A median split was used to create "high" and "low" categories for these variables. Adjusted odds ratios (aORs) are used to express the impact of significant predictor variables. The aOR indicates the effect of a 1-unit change in a predictor on the odds of an improved outcome, with other variables in the model held constant. The total explained variance of the models for predicting an improved outcome by body regions was low (Nagelkerke R 2 ϭ0.084-0.38) (Tab. 5). A higher initial regionspecific disability score was predictive of an improved outcome for all body regions, and a greater number of visits was predictive of an improved outcome for all body regions except the lumbar spine. Additional predictors by body region are outlined in Table 5.
Models examining predictors of physical therapy utilization for lumbar and cervical/thoracic spine conditions explained a very small amount of the variance in utilization (adjusted R 2 ϭ0.051 and 0.019, respectively), but a greater amount of variance for episodes of care provided for shoulder and knee conditions (adjusted R 2 ϭ0.36 and 0.22, respectively) (Tab. 6). A postsurgical condition and a higher baseline disability score predicted higher utilization for all body regions except cervical/thoracic spine. Additional predictors by body region are outlined in Table 6.

Discussion
This study described the characteristics and clinical outcomes, as well as factors predictive of outcomes and utilization, of episodes of outpatient physical therapy care provided to Medicare beneficiaries receiving outpatient physical therapy for musculoskeletal conditions in one health care system. The results showed that Medicare beneficiaries accounted for a consequential proportion of the caseload in the outpatient physical therapy clinics included in the study. Slightly more than half of the Medicare beneficiaries in this sample had conditions affecting the lumbar spine or shoulder. Although there were differences based on the body region involved, a majority of the patients improved with physical Outpatient Physical Therapy for Medicare Beneficiaries With Musculoskeletal Conditions therapy, as evidenced by achieving at least a minimum threshold of clinical improvement on region-specific disability questionnaires. Predictors of change in pain or an improved outcome differed by body region involved. Attending physical therapy following surgery and a higher initial level of pain predicted a greater number of physical therapy visits for most body regions. The results of this study suggest areas to target to improve physical therapy care, such as focusing on the most prevalent clinical conditions and identifying more-effective strategies for the treatment of Medicare beneficiaries. Determining optimal physical therapy utilization for Medicare beneficiaries also should be a target for further research.
There is a growing focus in health care generally, and by CMS specifically, on measuring the quality of health care services and designing strategies to incentivize more-efficient, higher-quality care. 22 Improvements in health care quality can be facilitated through standardized measurement procedures and information on care processes and clinical outcomes from representative practice environments. 36,37 The CMS has recognized the need for standardized measurement, calling for the development and implementation of quality measurement focused on the outcomes of care and resource utilization for conditions with a high impact on public health, such as joint conditions and arthritis. 38 In 2007, the CMS established a research project titled "Developing Outpatient Therapy Payment Alternatives" for the purpose of addressing the need for standardized measurements of quality specific to outpatient therapy services and developing strategies to link these measurements to reimbursement strategies. 24 Very little information has been published to date examining the outcomes of care and the factors influencing outcomes of physical therapy, and the available data have not focused specifically on Medicare beneficiaries. Data published on physical therapy for Medicare beneficiaries has focused largely on utilization with respect to various changes in payment regulations. 20,21,39 The goal of this project was to take advantage of a standardized measurement system to examine utilization, clinical outcomes, and factors influencing outcomes for physical therapy provided to Medicare beneficiaries with musculoskeletal conditions within one health care system. Variation in the utilization of health care services among Medicare beneficiaries is a potential threat to overall quality of care, particularly when variation does not appear to be attributable to illness severity or disease characteristics. 53,54 Andersen and Newman 55 modeled health care utilization as a function of 3 principal components: predisposing characteristics (eg, patients' attitudes, sociodemographic factors), enabling characteristics (eg, insurance coverage, access to providers), and need characteristics (eg, illness severity, disease characteristics). The most important predictors of higher physical therapy utilization in our sample were treatment for a postsurgical condition and higher initial pain ratings for patients with lumbar, shoulder, or knee conditions. These 3 factors represent need characteristics and would appear to be appropriate reasons for variation in utilization. Resnik et al 42 identified disability score at intake and number of prior surgeries as predictors of the number of physical therapy visits for patients of all ages with low back pain. Freburger and Holmes 56 examined predictors of physical therapy utilization in a nationwide sample of community-dwelling older adults and found the strongest predictors were primarily enabling characteristics (higher income level, living in a metropolitan community, greater local supply of physical therapists, not being in a managed care plan).
We were unable to model most of these characteristics because of either a lack of data or a lack of sufficient geographic variability in our sample. Additional research on the determinants of variation in utilization of physical therapy for older adults may help to identify areas of unwarranted disparity in utilization, resulting in overutilization or underutilization of physical therapy.
The results of this study need to be considered in light of several limitations. This study was conducted in only one geographic region within a single health care delivery system. Regional differences in health care, along with the wide degree of variability in intensity of physical therapy utilization in different regions of the United States, 56 may limit the generalizability of the results of this study. There is a need for further research to evaluate these issues in a more nationally representative sample. We were able to risk-adjust our outcome variables for several important factors (age, sex, symptom duration); however, we lacked data on other potential confounding variables such as comorbid health conditions, patient income or education level, and so on. The addition of these factors may have altered our comparisons and prediction models. Although we were able to include data on more than 90% of the patients receiving at least 2 physical therapy visits during the time frame of this study, the elimination of patients who received only 1 visit may have created a selection bias in our sample. Our clinical outcome data relied on patient-report questionnaires.

Conclusions
This study provides information, including utilization and clinical outcomes of care, for Medicare beneficiaries with musculoskeletal conditions who received outpatient physical therapy in communitybased, non-hospital clinics within one integrated health care system. We found that slightly more than half of all episodes of care were provided for musculoskeletal conditions affecting the lumbar spine or shoulder and that the least amount of clinical improvement tended to occur for spinal conditions. We identified several factors that were associated with greater clinical improvement among Medicare beneficiaries, including a greater number of physical therapy visits and treatment in a clinic with a greater proportion of older adults in its case mix. The results of this study provide information on utilization and outcomes of physical therapy for Medicare beneficiaries with musculoskeletal conditions and identify potential target areas for quality improvement.