Abstract
Background Physical therapy for hip osteoarthritis (OA) has shown short-term effects but limited long-term benefit. There has been limited research, with inconsistent results, in identifying prognostic factors associated with a positive response to physical therapy.
Objectives The purpose of this study was to identify potential predictors of response to physical therapy (exercise therapy [ET] with or without adjunctive manual therapy [MT]) for hip OA based on baseline patient-specific and clinical characteristics.
Design A prognostic study was conducted.
Methods Secondary analysis of data from a multicenter randomized controlled trial (RCT) (N=131) that evaluated the effectiveness of ET and ET+MT for hip OA was undertaken. Treatment response was defined using OMERACT/OARSI responder criteria. Ten baseline measures were used as predictor variables. Regression analyses were undertaken to identify predictors of outcome. Discriminative ability (sensitivity, specificity, and likelihood ratios) of significant variables was calculated.
Results The RCT results showed no significant difference in most outcomes between ET and ET+MT at 9 and 18 weeks posttreatment. Forty-six patients were classified as responders at 9 weeks, and 36 patients were classified as responders at 18 weeks. Four baseline variables were predictive of a positive outcome at 9 weeks: male sex, pain with activity (<6/10), Western Ontario and McMaster Universities Osteoarthritis Index physical function subscale score (<34/68), and psychological health (Hospital Anxiety and Depression Scale score <9/42). No predictor variables were identified at the 18-week follow-up. Prognostic accuracy was fair for all 4 variables (sensitivity=0.5–0.58, specificity=0.57–0.72, likelihood ratios=1.25–1.77), indicating fair discriminative ability at predicting treatment response.
Limitations The short-term follow-up limits the interpretation of results, and the low number of identified responders may have resulted in possible overfitting of the predictor model.
Conclusions The authors were unable to identify baseline variables in patients with hip OA that indicate those most likely to respond to treatment due to low discriminative ability. Further validation studies are needed to definitively define the best predictors of response to physical therapy in people with hip OA.
It is well recognized that osteoarthritis (OA) is a heterogeneous condition,
1,2 with variation in clinical presentation and response to treatment. Results of randomized controlled trials (RCTs) of nonpharmacological-based interventions have demonstrated short-term benefit,
3–5 but there is limited evidence of a long-term effect.
6 The achievement of optimal outcomes of interventions in RCTs is challenging, as it is unlikely that one type of intervention will be effective for all people.
7 Identification of patients who respond optimally to certain interventions is important because certain characteristics may result in varied outcomes for different patients. Baseline characteristics in an RCT can be evaluated as potential predictors of treatment outcome.
8 A Cochrane review of exercise for OA of the hip recommended that possible predictors of treatment responsiveness should be included in future RCTs of hip OA.
9
Four known studies
10–13 have attempted to predict outcome of physical
therapy–based interventions such as exercise
therapy (ET), manual
therapy (MT), and education for hip OA, with mixed results. The variables of female sex,
11 absence of depressive symptoms,
11 past use of complementary medicine,
11 low comorbidity,
11 unilateral hip pain,
10 age ≤58 years,
10 pain severity ≥6/10, symptom duration less than 1 year,
10 40-m self-paced walk test measurement of ≤25.9 seconds,
10 mild or moderate radiological severity,
12 and low use of nonsteroidal anti-inflammatory drugs (NSAIDs)
13 have been identified as postive predictors of outcome.
However, a number of variations among the studies in terms of the population types included, interventions delivered, comparison groups, and outcomes recorded limit the interpretation of the findings. Two studies
11,13 included people with knee OA; therefore, predictors may not be specific to hip OA. Two studies used prospective cohort designs,
11,13 and 2 studies were RCTs.
10,13 Variation in the types of interventions, such as ET or MT alone
10,12 or in combination,
10 ET and drug
therapy,
13 and inpatient rehabilitation,
11 make comparisons difficult across the studies. Definitions of “responders” also varied among studies, with the OMERACT/OARSI responder criteria
14 used by some studies
10,13 and the minimal clinically important difference of the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC)
11 and a Likert-type transition scale of health status
11 also used. Due to these differences, there has been little consistency in agreeing on predictors of response to rehabilitation-based interventions for hip OA.
One of these studies
10 used data from an RCT that evaluated ET and MT in isolation and in combination, specifically for hip OA. French et al
15 conducted the Exercise and Manual Physio
therapy Arthritis Research Trial (EMPART), which evaluated the effect of ET with and without adjunctive MT. The results showed no significant difference in the primary outcome of WOMAC physical function subscale score between the ET group (n=66) and the ET+MT group (n=65) at 9 weeks (mean difference=0.09; 95% confidence interval [CI]=−4.41, 5.25) or at 18 weeks (mean difference=0.42; 95% CI=−3.98, 6.83), or in other outcomes, except patient satisfaction with outcome, which was higher in the ET+MT group (
P=.02). Due to similarities between Wright and colleagues' RCT and ours, we conducted analyses to determine whether similar predictors could be identified. The aim of this secondary analysis was to identify potential predictors of response to physical
therapy–based interventions (ET with or without adjunctive MT) for hip OA based on baseline patient-specific and clinical characteristics.
Method
Secondary analysis of patients recruited to the EMPART trial (
ClinicalTrials.Govregistration number:
NCT00709566) was undertaken to establish what potential baseline variables and characteristics could predict response to treatment. The methods and results of the primary trial analyses are reported elsewhere.
15 In brief, participants with a diagnosis of OA of the hip according to American College of Rheumatology criteria
1 and aged 40 to 80 years were recruited from physical
therapy waiting lists in 4 acute care teaching hospitals in an urban area. Exclusion criteria included previous hip arthroplasty, congenital or adolescent hip disease, clinical signs of lumbar spine disease, physical
therapy in the previous 6 months for hip symptoms, pregnancy, hip fracture, contraindications to ET,
16inflammatory arthritis, on waiting list for hip joint replacement within the next 7 months, intra-articular hip corticosteroid injection in the previous 30 days, or insufficient knowledge of the English language to complete questionnaires.
Eligible participants initially were randomly allocated into 1 of 3 groups: an ET group, and ET+MT group, and a waiting list control group. Details of the interventions (8-week duration) are described elsewhere.
15 Control group participants remained on the waiting list until reassessment at 9 weeks, after which they were re-randomized into ET or ET+MT groups. Outcomes were completed at 9 and 18 weeks posttreatment.
Responder Criteria
The OMERACT/OARSI criteria include a composite index based on pain, function, and patient global assessment as determined by an international task force.
14They are currently the most recognized criteria used to determine response to pharmacological,
17–19 nonpharmacological,
20,21 or surgical
22 interventions for knee or hip OA and for this reason were chosen as the responder criteria for this study. Although the responder criteria were initially developed to evaluate the efficacy of pharmacological interventions for OA, they have been validated for use in nonpharmacological interventions.
23 These criteria are defined as improvement of ≥50% and absolute change of 20 points in pain or function or as at least 2 of the following: (1) improvement of ≥20% and absolute change in function of ≥10 points, (2) improvement of ≥20% and absolute change in pain of ≥10 points, and (3) improvement of 20% as defined by a global rating of change (GRC) score. For the purposes of this analysis, the latter criteria were used, with pain severity with activity (measured with a numeric rating scale [NRS]) and WOMAC physical function subscale used as the pain and function measures, respectively. As the GRC score was based on a 7-point Likert scale in the EMPART study,
15 a positive increase of 2 increments on the scale represented 20% improvement.
Predictor Variables
Baseline measures were used as predictors based on identified predictors of positive response to surgical, pharmacological, or nonpharmacological treatment of hip OA as well as expectations based on the authors' clinical experience. The variables of age,
24,25 sex,
25 body mass index, symptom duration,
10 number of comorbidities,
26 baseline pain with activity
10 (measured with an NRS), baseline physical function (measured with the WOMAC physical function subscale), baseline mood
25 (measured using the Hospital Anxiety and Depression Scale [HADS]), baseline range of motion (ROM),
12 and treatment adherence
27 were chosen.
Data Analysis
Data were analyzed using Stata version 12 (StataCorp LP, College Station, Texas). Analysis was based on the final group allocation of all 131 participants into the ET group (n=66) and the ET+MT group (n=65). Participants were dichotomized into responders and nonresponders at both 9 and 18 weeks based on the OMERACT/OARSI criteria. A total of 10 predictors were first tested for collinearity by calculating bivariate correlation using the Pearson product moment correlation for continuous variables and Pearson chi-square tests for categorical variables. Variables with a correlation value of >.5 for continuous variables or
P value of <.2 for categorical variables were considered to demonstrate collinearity and were not used together in the regression model. Univariate logistic regression was undertaken for each of the 10 variables. Criteria for entry into the model and for removal were
P<.05 and
P>.1, respectively. Variables that fulfilled these criteria were entered into the multivariate regression model in a forward stepwise manner. Variables that did not fulfill the criteria but had been identified in previous predictor studies, such as age,
10 symptom duration,
10 and number of comorbidities,
11 also were added to the model. All multivariate analyses were adjusted for the initial 3-group allocation, which included the waiting list control group.
Logistic quantile regression was used for relevant variables to further quantify the contribution of baseline severity to treatment response. This method allows further exploration of continuous outcomes by banding them into quartile categories to quantify the impact of baseline severity of the predictor on treatment response.
28 For continuous variables with a significant univariate relationship, sensitivity and specificity values were calculated for possible cutoff points and plotted as receiver operating characteristic (ROC) curves,
29 plotting the sensitivity (true positive) on the y axis against 1 − specificity (false negative) on the x axis. Sensitivity, specificity, and likelihood ratios were calculated for the predictor variables that were retained in the final model for the full cohort and the waiting list control group prior to receipt of treatment.
Sensitivity is defined as the proportion of people with the condition whom the test identifies as positive, and
specificity refers to the proportion of people who do not have the condition that the test identifies as negative. The closer that values are to 1, the better the baseline measure is at distinguishing between those who will and will not respond to treatment. Lower sensitivity values are more likely to give a false-negative result, and lower specificity values are more likely to give false-positive results.
Results
Of the 131 participants with OA of the hip who were recruited between May 2008 and February 2010, 123 (93.8%) attended the 9-week follow-up, and 112 (85.5%) attended the 18-week follow-up. The results of the main RCT showed no significant difference in clinical outcomes between the 2 final groups of ET and ET+MT. There was a significant difference in satisfaction with outcome, with higher satisfaction reported by the ET+MT group.
15 Using the OMERACT/OARSI criteria, a total of 46 participants (35.1%) were classified as responders at the 9-week follow-up, and 36 participants (27.5%) were considered responders at the 18-week follow-up. The number of responders in the waiting list control group (n=43) prior to reallocation to one of the intervention groups was 4 (9.3%). Baseline characteristics of all participants, as defined by responder status, are shown in
Table 1. Significant differences occurred between the 2 groups for WOMAC physical function subscale responder status at 9 weeks (
P=.002).
Table 1.
Baseline Variables by OMERACT/OARSI Criteria Response to Treatmenta
Of the 10 predictors that were evaluated for collinearity, a significant linear relationship was reported between baseline pain as measured with the NRS and baseline pain as measured with the physical function subscale of the WOMAC. These predictor variables were not used together in the multivariate regression model.
Univariate analysis of responder criteria at 9 weeks showed that baseline pain with activity, WOMAC physical function subscale score, HADS score, and male sex were all significantly associated with a positive response to treatment. At 18 weeks, none of the baseline variables were significantly associated with a favorable response to the intervention (
Tab. 2). Due to the high collinearity (
r=.76) between baseline pain and WOMAC physical function score, only the latter variable was retained for the multivariate regression model. When these variables were entered into a multiple regression model, all variables (male sex, baseline WOMAC physical function score, and HADS score) remained significant. Quantile regression of WOMAC physical function score and pain (
Tab. 3) showed that the odds of a positive response were influenced by severity, whereby at higher scores (fourth quartile), the odds of a positive response were reduced compared with the first quartile.
Table 2.
Univariate Logistic Regression for Possible Predictor Variables at 9 and 18 Weeksa
Table 3.
Quantile Regression for Western Ontario and McMaster Universities Physical Function Subscale (WOMAC PF) and Pain Severity at 9 Weeksa
The final model identified male sex, lower WOMAC physical function subscale score, and lower HADS score as predictive of a positive response to treatment at 9 weeks (
Tab. 4). Summary estimates of sensitivity and specificity and their corresponding 95% CI values were calculated for those variables that predicted response to treatment on univariate analysis (
Tab. 5). The 4 predictor variables demonstrated moderate discriminative ability, with higher estimates of specificity than sensitivity, indicating that the predictor variables are more useful at ruling in those who would respond to treatment. Of the 4 final predictor variables, participants' sex demonstrated the highest likelihood ratio value, followed by HADS scores. These results mean that men are 1.77 times more likely than women to have a positive treatment response, and those with HADS scores <9 are 1.6 times more likely to have a positive treatment response than those with scores ≥9. The 95% CI values for these variables were wide, indicating some lack of precision of the result. The 95% CI values for the WOMAC and pain severity variable are even less robust, as the CIs fell below 1.
Table 4.
Multivariate Regression Model for Response to Treatment at 9 Weeksa
Table 5.
Diagnostic Accuracy Statistics for Predicting Response to Treatment at 9 Weeks for All Participants Who Received Exercise Therapy or ExerciseTherapy + Manual Therapy (N=131)a
The sensitivity estimates for the control group (
Tab. 5) were broadly similar, but the lower specificity estimates indicate poorer discriminative ability of these variables at ruling in those who would have a positive response to treatment. The likelihood ratios were not statistically significant.
Table 6, which shows the diagnostic accuracy statistics for the control group prior to re-randomization, highlights that no variables predicted response in those participants who received no treatment at this time point.
Table 6.
Diagnostic Accuracy Statistics for Predicting Response to Treatment at 9 Weeks in the Control Group (n=43)
Discussion
Although the results of this secondary analysis showed that the odds of a positive response to physical
therapy treatment at 9 weeks was associated with some baseline variables, due to their fair discriminative ability, none of the variables predicted which people with hip OA are more or less likely to respond to treatment. Variables associated with a positive response to treatment identified in previous studies were not found in this study. The most significant predictor was sex, where male participants were more than twice as likely to respond to treatment compared with female participants. The odds of a positive treatment response were reduced by 3% and 9% for baseline WOMAC physical function and HADS scores, respectively. Although all predictors demonstrated statistical significance, the 95% CI values were wide and approached 1, which demonstrates uncertainty in the results. These results were reinforced by the diagnostic accuracy statistics. The likelihood ratios' proximity to 1 provides weak evidence that these variables are predictive of outcome.
30
None of the baseline variables predicted outcome at 18 weeks. This finding might have been due to the lower proportion of responders at this time and indicates the short-lived nature of the effect of the interventions delivered in the main RCT. Further exploration of pain severity and WOMAC physical function subscale scores using quantile logistic regression showed that response to treatment reduced as baseline severity increased, but was only significant at the fourth quartile, which indicated the highest level of severity for both outcomes. It could be argued that the NRS would be preferable to the WOMAC for inclusion in the multivariate model due to its ease of administration in a clinical setting. The choice to retain the WOMAC physical function subscale over pain severity as measured with the NRS was based on the World Health Organization's recognition of the WOMAC as a condition-specific measure of choice for lower limb OA
31 and the Initiative on Methods, Measurement, and Pain Assessment in Clinical Trials (IMMPACT)
32 and its inclusion in the OMERACT responder criteria.
14
The identification of baseline variables that can predict response to treatment is clinically useful, as it can allow researchers to develop targeted interventions based on baseline characteristics of the population of interest and assist clinicians in making decisions regarding which intervention suits patients on an individual level. Osteoarthritis is defined as being a heterogeneous condition
33 with variation in clinical presentation, so subgrouping of patients based on certain characteristics would appear to be appropriate to enhance treatment outcomes. This outcome has been demonstrated successfully in other chronic musculoskeletal conditions such as LBP, where interventions previously shown to have little effect in the general chronic LBP population were more effective when given to selected LBP subgroups.
34,35 As improved outcomes of exercise interventions for hip OA have to date been short-lived and of small effect,
36 there is a need to identify which patients respond to specific interventions.
These results are somewhat incongruent with the findings of other prognostic studies of rehabilitation interventions. Although Hoeksma et al
12 used hip function, pain, and ROM as responder variables in their study, predictors of improvement occurred only in hip ROM based on baseline radiological severity. They used a median split method to dichotomize their predictor variables, which may have resulted in a loss of information and statistical power.
37 Predictors of favorable response to inpatient rehabilitation for hip or knee OA included female sex, absence of depressive symptoms, past use of complementary medicine, and low comorbidity.
11 Although low levels of anxiety and depression were predictive of a positive response in our study, comorbidity was not predictive of such a response, and male participants were more likely than female participants to respond positively. Weigl and colleagues' cohort study
11 did not include a control group, which may limit interpretation of the results because the treatment effect cannot be quantified and RCT methods are recommended for determining response to treatment.
7
In the study by Wright and colleagues,
10 unilateral hip pain, age ≤58 years, pain ≥6/10 as measured with the NRS, symptom duration less than 1 year, and a 40-m self-paced walk test measurement ≤25.9 seconds were predictive of positive response to physical
therapy treatment. Having 3 or more of these variables increased posttest probability of success to 99%.
10 Age and symptom duration were not predictive of outcome in our analyses. Although pain severity was identified as a predictor by Wright et al,
10 they determined a cutoff of >6/10 was a positive predictor of outcome, whereas, in our study, a cutoff of ≥6/10 was a negative predictor of treatment response. We also demonstrated, using logistic quantile regression, that those individuals at the highest levels of functional severity were less likely to have a positive response to treatment. The study by Wright et al
10 was most like ours, as it included people only with hip OA, used MT and ET interventions in a secondary care facility, and used similar responder criteria. As with our RCT, the results of the main RCT on which their secondary analysis was based also showed that a combination of MT and ET added no additional benefits compared with an exercise-only approach,
38 so it is difficult to ascertain why the results of the 2 studies were incongruous. Studies of hip replacement surgery that included predictors such as sex,
39 age,
25 waiting times for surgery,
25 baseline physical function,
39 and comorbidities
39 also have demonstrated a lack of consistency in the identification of treatment outcome.
Using ROC analysis, a cutoff score of 34 on the WOMAC physical function subscale was derived, indicating that those who had a baseline score of ≥34 would have a poor response to treatment. Cutoffs were similarly devised for pain severity (NRS >6) and for combined anxiety and depression HADS outcome (baseline score of >9) and were indicative of nonresponse to treatment. The sensitivity analysis undertaken for the control group suggests that these predictors are specific to the interventions and are not just random findings. However, the sensitivity and specificity values in this analysis could be considered as having only fair discriminative ability for the predictor variables. Caution should be applied to the interpretation of these baseline measures in attempting to predict who would respond favorably to ET with or without adjunctive MT.
Although pain severity was assessed and considered as a predictor variable in this study, further exploration of the underlying pain physiology associated with OA is needed. Osteoarthritis is historically considered to be a disease associated with nociceptive pain due to joint damage, but there is emerging evidence that central sensitization may be involved in the processing of pain associated with OA.
23,40,41Consequently, the pain experience may be modulated by psychological and social factors and can add to the complexity of both presentation and management.
42
The lack of consistency across different studies in identifying predictors of outcome for physical therapy interventions for the hip and the findings of this study that none of the predictors demonstrated a strong effect in predicting outcome highlight the need for more derivation and validation studies with large sample sizes to ensure all relevant baseline variables can be considered. To date, the prediction studies have been conducted in secondary care facilities. Exploration of the relevance of disease severity and duration in relation to treatment response is needed. Replication in a primary care setting with larger sample sizes would be needed for patients with milder clinical presentations and would allow for exploration of more predictor variables.
Although more validation studies are needed to determine prognostic variables for response to pharmacological, nonpharmacological, and surgical interventions for hip OA, the results of this study may suggest that treatment should include interventions to address psychological symptoms such as anxiety or depression. There have been few RCTs investigating such approaches in hip OA. A behavioral graded approach incorporated into an exercise program produced comparable outcomes with usual physical
therapy,
43 and the active involvement of the patient improved long-term adherence.
44
Strengths and Limitations of the Study
A relatively large sample was used in this study compared with similar studies, which allowed for a greater number of predictor outcomes to be tested. We also used a recognized reference standard for outcome, used in previous research,
10,13,23 which allows for comparison with other studies. However, the relatively low number of identified responders in relation to the number of predictor variables may have resulted in overfitting of the model. The inclusion of the WOMAC physical function subscale and pain severity as the outcome variables, as components of the OMERACT criteria, and as predictor variables may have resulted in some confounding. There was considerable heterogeneity within the sample, as people with all levels of severity of hip OA were included in the trial, which may have compromised the predictive capacity of the baseline variables. It also may have affected the external validity of these results, as patients were recruited from secondary care centers. This study evaluated follow-up only up to 18 weeks postrandomization. The implication of response over a longer period of time is of particular relevance in a chronic disease such as OA.
Nenhum comentário:
Postar um comentário