| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Research Reports |
CA Kennedy, BScPT, MSc, is Research Associate, Institute for Work and Health, Toronto, Ontario, Canada; Research Associate, Martin Family Arthritis Care and Research Centre, Mobility Program Clinical Research Unit, St Michael’s Hospital, Toronto; and Clinical Lecturer, Department of Physical Therapy, University of Toronto
M Manno, MSc, is Biostatistician, Cancer Care Ontario, Toronto; Analyst, Institute for Work and Health; and Lecturer, Department of Public Health Sciences, University of Toronto
S Hogg-Johnson, PhD, is Senior Biostatistician, Institute for Work and Health, and Assistant Professor, Graduate Department of Community Health, University of Toronto
T Haines, MD, DOHS, is Associate Professor, Department of Clinical Epidemiology and Biostatistics and Program in Occupational Health and Environmental Medicine, McMaster University, Hamilton, Ontario, Canada
L Hurley is Project Coordinator, SCRIPT Project, Sponsored by The University Health Network, Toronto. At the time of the study, she was Manager, Quality Management Program, College of Physiotherapists of Ontario, Toronto
D McKenzie is Director of Clinical Services, Work Able Centres Inc, Toronto. At the time of the study, she was Research Coordinator, Institute for Work and Health
DE Beaton, BScOT, PhD, is Scientist, Institute for Work and Health; Scientist and Director, Mobility Program Clinical Research Unit, Martin Family Arthritis Care and Research Centre; Assistant Professor, Department of Occupational Therapy, Graduate Departments of Rehabilitation Science and Health Policy, Management, and Evaluation, University of Toronto.
Dr Hogg-Johnson, Dr Haines, Ms McKenzie, and Dr Beaton provided concept/idea/research design. Ms Kennedy, Dr Haines, and Dr Beaton provided writing. Ms Kennedy, Ms Hurley, Ms McKenzie, and Dr Beaton provided data collection and project management. Ms Kennedy, Mr Manno, Dr Hogg-Johnson, Dr Haines, and Dr Beaton provided data analysis. Ms Hurley and Dr Beaton provided fund procurement, facilities/equipment, and institutional liaisons. Ms Hurley provided subjects. Ms McKenzie provided clerical/secretarial support. Dr Hogg-Johnson, Dr Haines, and Ms McKenzie provided consultation (including review of manuscript before submission)
(dbeaton{at}iwh.on.ca) Address all correspondence to Dr Beaton at Institute for Work and Health, 481 University Ave, Ste 800, Toronto, Ontario, Canada M5G 2E9
Submitted July 19, 2005;
Accepted January 9, 2006
 | Abstract |
|---|
Key Words: Linear models • Prognosis • Shoulder • Shoulder pain • Soft tissue injuries
| Introduction |
|---|
 TopAbstract Introduction MethodResultsDiscussionConclusionReferences |
|---|
Shoulder disorders also are common in the workplace. Workers’ Compensation Board statistics in the province of Ontario document high rates of upper-limb disorders, with shoulders representing 6.1% (n=5,786) of all lost time claims in 2002.3 One surveillance study of largely computer-based workers at a large newspaper in Ontario revealed that 14% of the work force experienced upper-limb symptoms of moderate or worse severity at least once per month or for longer than 1 week over the past year.4 According to information from a mandatory annual practice survey of all registered physical therapists in Ontario, shoulder disorders were the second most common condition treated by physical therapists in 2002 (Marla Nayer, Director, Quality Programs, College of Physiotherapists of Ontario; personal communication; 2003). Shoulder disorders, therefore, represent a sizable effect on the population, the work force, and physical therapist practices.
Studies report unfavorable outcomes in many patients with new episodes of shoulder problems in primary care. Only 21% of patients reported complete recovery at 6 months, and only 49% of patients reported complete recovery at 18 months.5 Similarly, van der Windt et al6 reported that 41% of patients in their sample had persistent or recurrent shoulder complaints after 1 year.
Clinicians often are faced with challenging questions from patients, insurance or compensation providers, administrators, and policy makers about the outcome of soft tissue disorders of the shoulder. Frequent questions include "What will the extent of recovery be?" and "Are there any factors that could delay recovery?" The literature on prognosis and potential prognostic factors in shoulder disorders, however, is limited. Prognostic factors were assessed in 16 studies5–20 that were identified in the systematic review by Kuijpers et al,21 and 9 additional shoulder prognostic studies22–30 that were identified by the first author (CAK).31 The 9 additional studies have examined other prognostic factors, such as menopause,25 night pain,29 side affected,25 shoulder pain interrupting sleep,27 muscle wasting,29 drop arm test,27 painful arc,29 radiological findings,23, 26, 29 and pain on movement.29 Excluding the variable "shoulder pain interrupting sleep," none of these factors were found to be significantly associated with the outcome studied.
The systematic review by Kuijpers et al21 provides a best-evidence synthesis from 6 of 16 identified studies. This systematic review reported strong evidence that high pain intensity predicts poorer outcome in primary care populations6, 8 and that middle age (45–54 years) predicts a poorer outcome in people seeking occupational medicine.9, 10 The review also reported moderate evidence that long duration of complaints and high disability at baseline predict a poorer outcome in primary care.5, 8 The remaining identified studies provided only weak or inconclusive evidence.21
Many of the prognostic factors studied in the literature are not in the control of the physical therapist but can be helpful in better predicting the duration or outcome of an episode of care. This information may provide patients with adequate knowledge about their expected clinical course. This prognostic information is necessary for clinicians to distinguish between those patients who are likely to have a favorable outcome versus those who are at risk for more chronic pain or disability. If the prognostic findings can be replicated and evaluated as clinical prediction rules, then they can help guide clinical decision making.32 In turn, this improved clinical decision making by physical therapists can facilitate better communication with insurance and compensation providers regarding their patients’ expected outcome.
Studies of prognosis vary widely in terms of the outcome they are trying to predict. Although outcomes in the literature on soft tissue disorders of the shoulder have focused on pain and range of motion (ROM), patient self-reports of physical function and health are increasingly recognized as important measures in the evaluation of health outcomes33 and, we would suggest, in prognosis as well. A disability outcome was used in only 4 shoulder prognostic studies5, 17, 24, 25 of the 25 studies reviewed. Most shoulder prognostic studies have used the following outcomes: symptom,6, 8–10, 12, 13, 19, 20, 23, 28 ROM,16, 18, 22 or an aggregate scoring system (ie, some combination of pain, ROM, and function).7, 14, 15, 26, 29, 30
A systematic review of shoulder disability questionnaires identified and evaluated the clinimetric properties of 16 different questionnaires.34 Overall, the Disabilities of the Arm, Shoulder, and Hand (DASH) measure received the strongest ratings for its clinimetric properties.34 The DASH measure has been validated for shoulder conditions.35–39 The DASH reflects the effect of the disorder in terms of physical function and symptoms, which are the primary reasons patients seek care for musculoskeletal disorders.
Two approaches could be taken using health status as an outcome. First, a study of prognosis could focus on change in disability from baseline to follow-up, answering the question "What factors predict larger versus smaller change?" Second, it could focus on the followup health status alone, answering the question "What factors predict which patients will have higher levels of disability at discharge from rehabilitation?" Either approach provides clinically useful information, but they differ conceptually. Work by Jacobson and colleagues40, 41 suggests that treatment induces a change that moves someone outside the range of a dysfunctional population or within the range of the functional population. Factors predicting a level of disability at discharge from treatment will not tell us if someone has changed to get to that final state, and predictors of change in disability will not tell us where someone is at the end of treatment. If these different outcome formats lead to different prognostic factors, then this would suggest that the consumer of the literature must decide not only what type of outcome to look for, but also which format of that outcome is most useful to them.
The purpose of this study was to determine prognostic factors affecting response in soft tissue disorders of the shoulder in patients receiving physical therapy. This study focuses on 2 concepts of clinical response—change in and level of disability after treatment—using the DASH outcome measure.
| Method |
|---|
|
TopAbstractIntroductionMethodResultsDiscussionConclusionReferences |
|---|
Inclusion and Exclusion Criteria
Inclusion criteria.
All patients seen by the selected physical therapists for the treatment of soft tissue shoulder complaints were included in the study. Shoulder complaints were defined as any condition of pain or discomfort, including instances where there had been surgical treatment of the soft tissue shoulder disorder (eg, rotator cuff repair).
Exclusion criteria.
Patients were excluded from the study if they: (1) had fractures or dislocations associated with soft tissue pain, (2) received physical therapy for only one visit (eg, referred for equipment or single education session), or (3) were unable to read and write English and thus could not complete the questionnaire package independently.
Each participating therapist maintained a log to record the required information for all consecutive patients with soft tissue shoulder disorders seen during the 4-month recruitment period. This log was designed for the following reasons: (1) to ensure that consecutive patients were chosen for the study, (2) to document reasons for the exclusion of patients, and (3) to describe basic demographic characteristics (age, sex, and duration of symptoms) of all patients meeting the inclusion and exclusion criteria, including those who declined to participate. Signed consent was obtained by the treating physical therapist at the clinic. The rights of the patients were protected.
Data Collection
Training related to the study protocol (eg, recruitment process, inclusion and exclusion criteria) and the data collection process (eg, review of the questionnaires and recruitment logs) was provided to all physical therapists participating in the study. The training was provided in small group sessions led by the investigators, and the physical therapists could ask specific questions about the study material. For each patient eligible for the study, the therapist completed a questionnaire at initial assessment ("baseline") and at discharge. For study purposes, we defined discharge as the date when the discharge questionnaire was completed; this occurred either when the patient was discharged from physical therapy treatment or after a 12-week period of treatment (for the purposes of the study), whichever happened first. The 12-week window was selected because it is a typical time for the healing of a soft tissue injury42–44 and was judged by the investigators and study team as a reasonable point at which many therapists would finish treatment.
Dependent variables.
Patient response to therapy was assessed over a period of 12 weeks or less, using the DASH to measure disability at baseline and again at discharge from physical therapy. The DASH is a 30-item self-completed questionnaire designed to measure physical function (at the level of disability as defined by Verbrugge and Jette45) and symptoms in people with any or multiple disorders of the upper limb.35, 36, 46 Each item of the DASH has 5 response options. Summative scores range from 0 (no disability/symptoms) to 100 (greater disability/symptoms). A high DASH score indicates more disability. Change in DASH was calculated by subtracting the baseline DASH score from the discharge (or 12-week) DASH score. Therefore, a negative value for change in DASH indicates improvement. Previous studies have demonstrated that the DASH has strong construct validity,38 reliability (internal consistency and test-retest reliability: intraclass correlation coefficient [ICC]=.96),38, 47 and responsiveness (SRMs [standardized response means] in patients with shoulder conditions=0.81–1.44).38, 39
Final state and change in state were used to define clinical response. Therefore, the dependent outcome measure used for the 2 regression models were: (1) DASH score at discharge and (2) change in DASH score between baseline and discharge.
Candidate prognostic variables.
Information was collected from both the patient and the physical therapist on a wide range of descriptive and prognostic variables. The content of the questionnaires included variables chosen based on review of the literature on prognosis in musculoskeletal disorders,21, 31 key clinical findings identified through a survey of 12 physical therapist specialists (includes experienced clinicians, academics, and researchers from Ontario), and additional prognostic information that the research team was interested in exploring. Two investigators (CAK, DEB) selected or created prognostic variables, based on availability of supporting evidence in the musculoskeletal literature. Three variables on clinical findings at baseline (muscle wasting, restriction of ROM, and muscle strength) were chosen based on clinical rationale. In addition, we were interested in exploring variables related to both patient expectations and therapist predictions for recovery.
Wherever possible, standardized instruments with demonstrated measurement properties were selected for each variable. The variables were organized by domains: demographic, disorder-related and disability measures, medication use, clinical findings, and expectations for recovery. Table 1 contains a detailed description of all candidate prognostic variables considered for the regression models. In the model for change in DASH, we decided not to include the baseline DASH score as a candidate variable because this measure is a component of the dependent outcome.
|
Disorder-related.
Disorder-related information was provided by the patients. Recurrence of the shoulder problem was dichotomized to those patients who had a recurrent episode within the past 6 months and those who had a recurrent episode more than 6 months ago or who had not experienced a previous episode. We felt that the group with a recurrent episode within the past 6 months identified a worse condition because they were potentially dealing with a recurrence of the same problem. Patients described the nature of onset of their shoulder problem as either gradual or sudden. Duration of the current problem was partitioned into 3 phases of soft tissue recovery: less than 4 weeks, 4 to 12 weeks, longer than 12 weeks. Previous literature has consistently demonstrated that these phases represent transitional states in the recovery of soft tissue injuries from acute to subacute to chronic phases respectively.42–44 The Numerical Pain Rating Scale (NPRS) was used to assess each patient’s pain intensity at baseline. The NPRS asks the patient to circle a number from 10 ("no pain") to 100 ("pain as bad as can be") that best describes their average pain level over the past week. The NPRS has been shown to correlate with the visual analog scale (VAS) for pain,49 has good same-day test-retest reliability,50 and is considered easier for some patients to complete than the VAS.49–51 Operative (or surgical) conditions were defined as patients who had shoulder surgery within the past 6 months. We felt that those having surgery in the previous 6 months identified a unique group compared with either a nonsurgical group or those who had surgery more than 6 months previously.
Disability measures.
Measures of disability collected at baseline were: whether patients continued to work, whether a workers’ compensation claim had been filed, and a patient’s global rating of his or her shoulder problem. In the work domain, current working status that was affected by the shoulder condition and workers’ compensation claims for the current shoulder problem were of particular interest. In addition, the acute version of the 36-Item Short-Form Health Survey (SF-36) was used to assess the overall health of each client.52 We chose to use 2 aggregate summary scores, the Physical Component Score (PCS) and the Mental Component Score (MCS).
Medication use.
Patients reported current medication use (over-the-counter [OTC] and prescription) at baseline. We used a 5-point ordinal scale, ranging from no days to all days.
Clinical findings.
Physical therapists completed standardized forms of their clinical assessment. Three variables on clinical findings at baseline (muscle wasting, restriction of ROM, and muscle strength) were chosen based on the clinical rationale that these variables were the most relevant clinical measures in soft tissue disorders of the shoulder. Therapists recorded mild, moderate, or severe findings with respect to broad categories of: muscle wasting (supraspinatus, infraspinatus, deltoid), decreased muscle strength (supraspinatus, infraspinatus, serratus anterior, etc), and decreased active and passive ROM. We decided to collapse each of these variables (muscle wasting, muscle strength, and ROM) into "some degree" (mild, moderate, severe) versus "no" because of the potential for less reliability between observers (81 participating therapists) for these clinical findings.
Expectation for recovery.
We measured patients’ and therapists’ expectations for recovery using questions adapted from previous literature.53–55 At initial assessment, patients also were asked to predict how long it would take them to recover and how long it would take them to return to usual activities. Similarly, at initial assessment, therapists were asked to predict the patient’s functional activity level (with or without restrictions) at discharge and the time frame for recovery. Given the important role that recovery expectations have previously demonstrated in the literature,54, 55 we were interested in exploring these variables. For patient and therapist prediction for time to return to usual activity, a cut point of 4 weeks was chosen as a clinically relevant threshold for those who will get better quickly (acute) versus those who will take longer to recover (subacute or chronic). This threshold was based on work by Frank et al,42 who contend that understanding what predicts quick versus longer-term course is likely to be most helpful.
Data Management and Analysis
Therapists returned completed questionnaire packages (including both physical therapist and patient questionnaires) by courier to the study center. All data were entered into Access* databases with customized data entry screens designed to minimize data entry errors. Data were transferred into SAS version 6.08.
The purpose of the analysis was to build a prognostic model that could be used to determine the strength of each candidate variable in terms of its ability to predict clinical response (either DASH at discharge or change in DASH) on its own and when controlling for other variables. A strategy for investigating prognostic factors and developing prognostic models was determined a priori using approaches suggested in the literature.56, 57 Multiple linear regression techniques were used, based on approaches reported in the literature.58
Model building.
The analysis began with descriptive statistics on each outcome and candidate prognostic factor. We paid attention to the distribution of responses and suitability for a regression analysis.58 Categories were collapsed if distributions were too low in any one category.
A statistical strategy for variable selection was performed. Univariate models were built to obtain beta coefficients and a P value for each variable. Using a more conservative approach suggested by Hosmer and Lemeshow,56 variables that were associated with the outcome at a level of P<.25 progressed to the next phase of modeling. Collinearity among the remaining independent variables were assessed and if the correlation (Pearson for continuous variables or polychoric for dichotomous or ordinal variables) was greater than .75, then only one of the predictor variables was selected.58 Cramer’s V (a 
2-based measure) was used to assess the association between the nominal variables. The independent variable selected was the one with the highest correlation with the dependent variable or, in the case of similar associations, the one with the greatest clinical sensibility as a prognostic factor.
Following this selection process, the remaining variables (those significant at P<.25 and not collinear) were considered for the final regression model. These variables were entered into the model and, using backward manual elimination methods, the variable with the least significant P value was removed from the model. This process was repeated until all variables remaining in the final model had a beta coefficient with a P value less than .05. The same process of model building was repeated for each of the 2 outcome variables (DASH at discharge and change in DASH).
Regression diagnostics.
The distribution of each of the dependent outcome variables, DASH at discharge and change in DASH, was examined for normality. Goodness-of-fit indexes were assessed for each of the multiple regression models:
| Results |
|---|
|
TopAbstractIntroductionMethodResultsDiscussionConclusionReferences |
|---|
Study logs revealed that 534 patients met the inclusion criteria, but 154 were subsequently excluded based on a priori exclusion criteria. Therapists excluded an additional 23 patients based on their clinical judgment that the patient could not provide valid data. Following data collection, 5 patients were excluded from the database because of ineligibility. Nine cases were included in the study but were not documented in the tracking logs. Therefore, the final study sample was 361 clients.
We compared demographic variables describing participants and those considered nonparticipants (exclusion by eligibility criteria or the patient refused to participate) and did not find any statistically significant differences between the 2 groups. There were no sex and age differences between the participants group and nonparticipants group (P
.4). Although nonparticipants had their symptoms for a longer time before starting therapy than participants (381 versus 229 days), the difference was marginally significant (unpaired t test, P=.07).
The DASH outcome measures—at baseline, at discharge (to a maximum of 12 weeks), and change in DASH (DASHdischarge–DASHbaseline)—are presented in Table 2. Figure 1 illustrates a fairly normal distribution of DASH scores at baseline (starting physical therapy) with a mean DASH score of 40.1/100. At discharge, the distribution had shifted toward less disability, showing overall improvement with a mean of 17.9/100 (Fig. 2). The distribution of change in DASH scores (mean=–22.2/100) is illustrated in Figure 3.
|
|
|
|
Normality
The distribution of each of the dependent outcome variables (DASH at discharge and change in DASH) were examined for normality. DASH at discharge was right (positively) skewed (Fig. 2). Therefore, we chose to transform DASH at discharge to the dependent outcome log (1 + DASH at discharge), which displayed a fairly normal distribution (Fig. 4). Change in DASH was only slightly left (negatively) skewed, with a good distribution; therefore, we felt it was acceptable not to transform this outcome and rely on goodness-of-fit and regression diagnostics to identify any issues.
|
|
|
|
|
|
2-fold), age (by 4-fold), and PCS (by 8-fold).
Regression diagnostics.
The residuals of the models were randomly distributed across the continuous independent predictors. One exception to this was in the log (1 + DASH at discharge) model for the predictor variable "therapist prediction for return to usual activity," where the therapist predicted that only 4 patients would "not return to usual activity." We felt these were potentially patients of influence and ran further regression diagnostics. The regression diagnostic, Cook’s distance (di), was assessed in these 4 individuals. Cook’s distance measures the influence of an observation and how much the regression coefficients are changed by deleting the particular observation in question.58 Kleinbaum et al58 suggest that an observation with a di >1 may deserve closer scrutiny, and if the model is correct, then the expected di is <1. In these 4 individuals, Cook’s di were all less than 1, suggesting these were not patients of influence.58 In addition, the residuals of the models were generally equally distributed across each category for each of the categorical independent predictors. The model residuals for the predicted value of log (1 + DASH at discharge) and change in DASH were randomly distributed, which therefore suggested that all basic assumptions were held and that there were no problems with heteroscedasticity of the variance.
| Discussion |
|---|
|
TopAbstractIntroductionMethodResultsDiscussionConclusionReferences |
|---|
Although there was some overlap between the 2 models for the disability outcome, the different formats of outcome led to different prognostic factors. This would suggest that the consumer of the literature must decide not only what type of outcome to look for, but also which format of that outcome is most useful to them. Factors predicting a level of disability at discharge from treatment will not tell us if someone has changed to get to that final state and predictors of change in disability will not tell us where someone is at the end of treatment.
In both prediction models presented in this study, younger age (by decade) was associated with each of the outcomes (less disability at final state, more improvement in terms of change in DASH score). Strong evidence from shoulder prognostic studies of people seeking occupational medicine9, 10 have found that middle age (45–54 years) was associated with poorer outcome. Age was not identified as a significant factor in several other studies.5, 6, 8, 14–18, 23–25, 29
Shorter duration of shoulder symptoms was associated with greater improvement in terms of change in DASH score. Several studies5, 8, 11, 14, 15, 17, 28 have shown that this factor (shorter duration of shoulder symptoms) is important in predicting a more favorable outcome in terms of final status. However, many studies have not found the duration of shoulder symptoms to be significantly associated with various outcomes (including ROM16, 18, 22, 25; symptom improvement23; patient satisfaction25, 27; disability25; and aggregate score based on pain, motion, strength, and function29).
The current study found that a workers’ compensation claim was associated with the outcome higher DASH scores (more disability) at discharge. These findings are heightened by the fact that only 8.3% of the total sample reported a workers’ compensation claim related to their shoulder disorder; and despite this, the variable remained significant in this model. Similarly, other shoulder prognostic studies have found a workers’ compensation claim25 associated with dissatisfaction with their outcome or being on sick leave associated with a worse Neer shoulder score (an aggregate score including pain, function, ROM, and radiological findings)7 or taking more sick leave.11 Two other studies, one in a cohort of patients with rotator cuff tears receiving conservative care27 and another in a cohort of workers following carpal tunnel release,62 did not find an association between an insurance claim and the outcome studied. The evidence in the literature related to the impact of workers’ compensation status in people with soft tissue shoulder disorders is limited. However, in the carpal tunnel syndrome literature, longer-term followup studies by Katz and colleagues62, 63 found that factors other than simply having a workers’ compensation claim—such as longer duration of work absence or litigation issues—may be responsible for a delayed return to work that often has been assumed to be due to the claim itself. These more extensive studies involving workers and predictors of outcome do not exist in the shoulder literature and suggest a need for further study.
Physical therapist prediction for patient to return to activity was significant in the model for DASH at discharge. To the best of our knowledge, this is the first time this factor has been explored in any of the shoulder prognostic studies reviewed. Further study to identify more specific factors considered by clinicians in making a prediction of a patient’s degree of recovery would be helpful. Qualitative methods would likely be most helpful in addressing this question. Qualitative methods would help us identify the areas that we do not understand with our quantitative instruments. Furthermore, these methods are useful in this type of situation to get a rich understanding of a decision-making process.
Patients’ expectation for recovery has been studied in an injured worker population (including back and upper- and lower-limb disorders) and found to be significantly associated with time receiving wage replacement benefits.55 A systematic review of the evidence for a relationship between patients’ recovery expectations and health outcomes found that positive expectations were associated with better health outcomes in 15 of 16 studies.64 Despite this strong support in previous literature, 2 variables that were related to patients’ expectation for recovery (prediction for recovery and estimate of time to return to usual activity) were included in our analysis but were not significant predictors in either of the 2 multivariable regression models. In particular, we used the same wording and categorization as Hogg-Johnson and Cole55 for the variable client prediction for recovery. However, differences in their selection criteria (compensated occupational soft tissue injuries of back and upper and lower limb) and the dependent outcome assessed (duration on wage replacement benefits) may explain these differences.
In addition, only a modest correlation (polychoric correlation=0.36) was found between therapist and patient expectations for the variable "prediction for recovery" and the measure of association between the therapist and client "prediction for time to return to usual activity" were not significant (Cramer’s V, P=.12). Therefore, these findings suggest that the therapist and patient expectations were not cancelling each other out or that clinicians and patients decide on expectations using different frameworks.
Better physical health (PCS) at baseline was a significant predictor in the model for change in DASH. Given that most of our cohort showed improvement in disability from baseline to discharge from physical therapy and indicated that they were better on other indicators of change that were not used in this analysis,65 patients with better physical health had smaller change scores on the DASH and still considered that important. This also has been documented in other studies and interpreted as suggesting that change can still be meaningful, though smaller in magnitude, for those who begin at higher levels of baseline health compared with those who were sicker at baseline.66–68
In the current study, higher pain intensity at baseline was associated with more improvement in terms of change in DASH score. These findings are consistent with another study of patients presenting with acute shoulder pain in secondary care in which a multivariate analysis found that more severe pain at baseline was associated with greater improvement in pain.17 Two studies found that those with higher baseline severity of symptoms were more likely to have long-standing, persistent shoulder symptoms.6, 8 However, 3 studies did not find an association between pain intensity and satisfaction with outcome25, 27 or with sick leave.11
Those patients who had undergone shoulder surgery in the past 6 months showed more improvement in terms of change in DASH score. This finding makes clinical sense because we would expect those who have had recent surgery to be starting therapy in a more painful and disabled state. In this case, there would be more room for improvement in the first 12 weeks. Many of the shoulder prognostic studies did not include this variable because they were studying either surgical or nonsurgical cohorts.
In the current study, female sex was associated with a higher DASH score at discharge from physical therapy. Wirth et al30 had similar results in a cohort of patients with nonsurgical management of rotator cuff tears whereas males were associated with a better outcome reflected by a higher UCLA Shoulder Rating Scale score. In contrast, several studies did not find sex to be significantly associated with various outcomes (including persistent shoulder pain or complaints; shoulder pain; an aggregate score based on pain, function, ROM, strength, and patient satisfaction; ROM; and self-report measures of pain and disability).6, 8, 10, 14–17, 19, 24
One interesting finding in this study was that the baseline clinical findings from the physical examination (restriction in ROM and muscle weakness) were significant at the univariate stage and fell out of the model at the multivariable stage. Modest correlations (range=0.2–0.3) were found between these clinical findings and other independent variables such as client global rating of shoulder problem and baseline DASH score. A stronger correlation of 0.4 was found between muscle weakness and operative status. These findings suggest that these clinical data may be competing for predictive variance in the multivariate model. Overall, physical findings may be important in diagnostic classification,69 but this study did not support their utility over patient self-report measures in predicting outcome. We recognize that measurement noise or lack of sensitivity of the clinical measures used in this study also may have been responsible for the absence of these findings. Therefore, a limitation of this study is the relative weakness of the clinical measures used. With these clinical measures the greatest source of error is between observers, so we chose more gross, but valid, measures that may be less precise. We chose to err on including any deficit as a positive finding rather than depending on interobserver differences between mild and moderate. Very few people were rated in the severe category. We made this decision as a group but we felt it was the best one, given the variety of therapists and the distribution of findings. In our review of more than 25 studies, only 11 studies5, 7, 8, 11, 14, 15, 22, 25, 27, 29, 30 considered clinical findings as predictors and only 4 studies5, 11, 29, 30 found them to be significant predictors.
This study has several methodological strengths that distinguish it from previous studies in the literature. First, it includes a cohort of consecutive patients seeking physical therapy treatment for shoulder disorders. This cohort is most likely to represent the population of patients receiving physical therapy for shoulder problems and therefore more likely to give generalizable results. Cases were identified by physical therapists who were provided with an operational definition of cases, including exclusion criteria. These criteria were operationalized in a broad manner, again allowing for generalizable results. All clients entered the study at a common point (at the beginning of physical therapy treatment) and were followed until discharge (to a maximum of 12 weeks). These findings may not be generalizable to all persons with shoulder pain in Ontario, but rather those who seek treatment. Furthermore, this primary set of data focuses on physical therapist practice and patients rather than surgical or physician-based practices. This focus is indeed rare, because none of the 25 shoulder prognostic studies in our review were based solely on physical therapist practice.
Second, the data included a broad range of clinical variables: demographic, disorder-related and disability measures, medication use, clinical findings, and expectations for recovery. Wherever possible, standardized instruments with demonstrated measurement properties were selected for each variable. Variable reduction methods and model-building strategies suggested in the literature were used to explore the presence and strength of association between a limited number of variables collected in this study.4, 56, 70, 71
In response to ongoing uncertainty, we did not include baseline DASH in the change in DASH model because the baseline DASH score was taken into consideration in the formulation of the dependent variable (change in DASH). One of the major issues of concern is the violation of the assumption of independence of outcome and predictors that underlies regression analysis.58 Indeed, we found a Pearson correlation of –0.6 between baseline DASH and change in DASH. Thus, there is evidence to suggest that the magnitude of the change was dependent on the baseline level of disability. The direction of this change would be consistent with regression to the mean but could also reflect a ceiling effect at discharge or it could actually be a clinically relevant difference in response depending on the disability at baseline. Regression to the mean and ceiling (good health) effects at follow-up for those starting with less disability could be important limitations in the use of change scores as the format of the dependent outcome. Using discharge scores as the outcome, and controlling for baseline scores, is one alternative that theoretically removes the covariance due to baseline scores.
Third, in the current study, physical therapist and patient outcome assessments were done at treatment baseline and discharge (to a maximum of 12 weeks). This was considered a reasonable period of time for patients to have completed physical therapy. Useable follow-up data on the DASH outcome were available from 281 of 361 patients (77.8%) and 272 of 361 patients (75.3%) for DASH at discharge and change in DASH, respectively. One could criticize this study for limiting follow-up to the duration of physical therapy treatment (to a maximum of 12 weeks). Although this time frame captures a very important period in clinical recovery, it does not identify the clinical course and factors affecting response over a longer-term follow-up.
Fourth, the main outcome measure (DASH questionnaire) has demonstrated strong measurement properties in shoulder conditions.34–39 This outcome measure focuses on the clinical outcomes that matter most to patients: disability and symptoms.
Finally, with respect to analysis, this study demonstrates appropriate statistical methods. An analytic plan was developed a priori to address the research questions. The statistical tests chosen were valid for the outcome considered: specifically, continuous outcomes of DASH score at discharge and change in DASH score using multiple linear regression methods. Because the purpose was to develop a predictive model, variables were chosen based on the musculoskeletal literature relevant to prognosis, univariate analysis, and then using backward manual elimination methods. Multivariable methods were employed, thereby allowing for adjustment. The final models for DASH at discharge and change in DASH explained 35.6% and 22.5%, respectively, of the variability in the outcomes. Our review of the shoulder prognostic literature did not identify comparable studies using multiple linear regression methods. Therefore, we reviewed the low back pain prognostic literature for comparable analyses. Although our R2 values seem low, they are comparable to the existing low back prognostic literature, where Symonds et al72 had 32.2% and Dionne et al73 had 30% explained by their models. The highest R2 that we found was 51% of the variability explained in a prognostic study of diskogenic low back pain where manual labor, physician diagnosis of disk lesion, and prescribed days off were the main variables.74 As with nonspecific low back pain, however, we are dealing with a much less dramatic and definable condition in soft tissue shoulder disorders. It often is not possible to discern the soft tissue structure that is responsible for the patient’s complaint and, therefore, we would expect a lower R2 value.
We also ran models for 2 different forms of "disability," (ie, DASH at discharge and change in DASH). The results demonstrated 2 sets of predictors, that overlapped by only 1 variable: age (by decade). The other predictors differed between models. Occasionally this could be accounted for by chance but we believe that this is not a spurious finding. This could be validated by replication of the findings in a second sample. This highlights a new issue for reviews of prognosis. DASH at discharge is a concept that looks only at the final state of an individual, whether there was a lot of change to get there or not. Change in DASH focuses only on the amount of change and predictors are those sensitive to smaller versus larger amounts of change. They do not reveal where the person actually is at discharge. This means that when reviewing studies of prognosis it is no longer adequate to speak of predictors of "disability"; rather, one must also define how they conceptualized disability: the degree of change in disability or the final level of disability. The conceptualization will make a difference in the final set of predictors.
Furthermore, we do not know which outcome (the magnitude of the change or the final level of disability) is the most important to the patient. A qualitative study, which explored the concept of recovery in a group of people with upper-limb musculoskeletal disorders, would suggest that it could be the amount of change in disability, the final level of disability, or that an individual gets to a level where he or she can cope with the disability (either cognitively or behaviorally through adaptations).75 Other work supports another view, a view where treatment should induce a change in state greater than measurement error, and the final state would be in a normal or functional range of scores.41 These all suggest the need to consider how the outcome (disability) was used in a study of prognosis and its potential implications.
In our review of the shoulder prognostic literature, disability measures were often dichotomized, which adds yet another approach that could be taken and possibly lead to another set of predictors. Rather than being labeled "inconsistent findings" across studies of prognosis, our work would suggest that this variability in predictors could be the result of different formats of the outcome. Each approach to formatting the disability outcome has its own value, and the choice between models might best be made by selecting the one closest to the user’s clinical question.
A potential limitation of this study is that we have restricted the description of the outcome to the final status measure or the change over time. This may mask important information about the heterogeneity in the clinical course and outcome for different subgroups. Therefore, we undertook another research project that combined information on initial and final state with information on speed of recovery, which may offer a more useful description of clinical recovery for the clinician and researcher.76 Subsequently, this form of outcome may provide a more clinically relevant predictive model.
A further limitation of the study was that we did not include specific interventions because they would occupy a large amount of statistical power and our cohort design was not set up to evaluate treatment effectiveness. In addition, physical therapy intervention could not be controlled in this study, which involved 81 physical therapists. This cohort study aimed for more generalizable results—identifying factors at baseline that affect response in shoulder patients attending physical therapy. In the current physical therapist practice, there are a variety of interventions for soft tissue disorders of the shoulder; and rather than "disregarding" this, we have allowed physical therapy interventions to vary and considered only other treatments (such as prescription and nonprescription medications). However, further research studies could examine the interaction between the physical therapy interventions and examination findings to produce a more optimal outcome.
| Conclusion |
|---|
|
TopAbstractIntroductionMethodResultsDiscussionConclusionReferences |
|---|
| Footnotes |
|---|
* Microsoft Corp, One Microsoft Way, Redmond, WA 98052-6399. 
SAS Institute Inc, 100 SAS Campus Dr, Cary, NC 27513-2414.
| References |
|---|
|
TopAbstractIntroductionMethodResultsDiscussionConclusionReferences |
|---|
