Functional status (FS)

[ research report ] DENNIS L. HART, PT, PhD 1 MARK W. WERNEKE, PT, MS, Dip MDT 2 DANIEL DEUTSCHER, PT, PhD 3 STEVEN Z. GEORGE, PT, PhD 4 PAUL W. STRATFORD, PT, MS 5 Journal of Orthopaedic & Sports Physical Therapy Effect of Fear-Avoidance Beliefs of Physical Activities on a Model That Predicts Risk-Adjusted Functional Status Outcomes in Patients Treated for a Lumbar Spine Dysfunction TTSTUDY DESIGN: Retrospective analysis of a prospective, longitudinal cohort study of 30 858 patients being treated for a lumbar spine dysfunction in outpatient physical therapy. TTOBJECTIVES: To determine effect of adding a single-item screening variable classifying patients with elevated versus not-elevated scores of fearavoidance beliefs of physical activities at intake, on a model predicting risk-adjusted functional status (FS) outcomes. TTBACKGROUND: Outcomes must be risk-adjusted before making meaningful interpretations. Elevated fear-avoidance beliefs scores have been predictive of poor outcomes. But the importance of elevated fear-avoidance scores in a multivariable model predicting FS outcomes needs further study. TTMETHODS: Using retrospective analyses, predictive ability (R 2 ) of multivariable linear regression models of discharge FS with and without classification by elevated versus not-elevated fearavoidance scores were compared, while controlling for intake FS, age, symptom acuity, surgical history, gender, number of comorbidities, and payer. Percent variance controlled and beta coefficients (95% confidence intervals) of each variable in both models were compared. A split-half design was used for model cross-validation. Predictive ratios (predicted FS, divided by actual discharge FS) were assessed. TTRESULTS: Adding fear-avoidance beliefs classification to the discharge FS model improved (P<.001) model predictive ability but only slightly (R 2 without, and with, fear-avoidance classification, 0.2997 and 0.3010, respectively). Variables impacted models similarly (95% confidence intervals not different). Fear-avoidance classification added 0.2% data variance control to the existing model. Cross-validation was supported. Predictive ratios were 1.09 and 1.10, without and with fearavoidance, respectively. TTCONCLUSION: Although screening for elevated fear-avoidance beliefs of physical activities significantly improves the FS outcomes predictive model, the amount of additional meaningful interpretation of FS outcomes was minimal. Exploration of other clinically relevant variables designed to improve outcomes prediction is warranted. TTLEVEL OF EVIDENCE: Prognosis, level 2c. J Orthop Sports Phys Ther 2011;41(5):336-345, Epub 6 April 2011. doi:10.2519/jospt.2011.3534 TTKEY WORDS: computerized adaptive testing, outpatient rehabilitation, patient demographics, prediction models Functional status (FS) outcomes of patients with lumbar spine dysfunctions managed in outpatient rehabilitation clinics are affected by many variables, including patient demographic variables. 35 Before clinicians can interpret associations between treatments and outcomes in a meaningful way, FS outcomes must be statistically risk-adjusted to control the effect of variables that may or may not be related to clinical care. 8,23,33,35,52 Although statistical risk adjustment models can appear complicated, 53,54 results of these models tend to be clinically logical. For example, it is clinically logical that patients who are older or patients whose symptoms are more chronic should report less FS change compared to patients who are younger or have more acute symptoms, and, indeed, results support this inverse relationship (ie, as patients age, or symptoms become chronic, FS 1 Director of Consulting and Research, Focus On Therapeutic Outcomes, Inc, White Stone, VA. 2 Physical Therapist, Spine Rehabilitation at CentraState Medical Center, Freehold, NJ. 3 Director of Research and Development, Physical Therapy Service, Maccabi Healthcare Services, Tel Aviv, Israel. 4 Associate Professor, Department of Physical Therapy, Center for Pain Research and Behavioral Health, Brooks Center for Rehabilitation Studies, University of Florida, Gainesville, FL. 5 Professor, School of Rehabilitation Science and Associate Member Department of Clinical Epidemiology and Biostatistics, McMaster University, Ontario, Canada. This project was approved by The Institutional Review Board for the Protection of Human Subjects from Focus On Therapeutic Outcomes, Inc. Dr Hart acknowledges that he is an employee of and investor in Focus On Therapeutic Outcomes, Inc (FOTO), the database management company that manages the data analyzed in the manuscript. Analyses of data like the analyses presented in the manuscript are part of Dr Hart s regular daily work activities. Mr Werneke, Dr George, Dr Deutscher, and Mr Stratford have no financial relationship with FOTO. Mr Werneke works for CentraState Medical Center, which uses FOTO software to collect and manage outcomes data on their patients on a routine basis. Dr Deutscher works for Maccabi Healthcare Services, which uses FOTO software to collect and manage outcomes data on their patients on a routine basis. Address correspondence to Dr Dennis L. Hart, Director of Consulting and Research, Focus On Therapeutic Outcomes, Inc, 551 Yopps Cove Road, White Stone, VA 22578. E-mail: dsailhart@verizon.net 336 may 2011 volume 41 number 5 journal of orthopaedic & sports physical therapy

Journal of Orthopaedic & Sports Physical Therapy outcomes worsen). 8,25,30,34,35,50,52 Therefore, multivariable statistical models tend to control the effect of risk adjustment variables in clinically logical ways. To date, no consensus exists as to the optimal risk adjustment method for FS outcomes. 5 One of the more common methods of risk adjusting FS outcomes is a multiple linear regression model, which controls categorical and continuous variables simultaneously. 8,23,35,52 The challenge is to select the most important variables to include in the model to ensure the best prediction of the FS outcomes. Previous models have included variables like intake FS, 8,25,30,34,35,50,52 age, 8,25,30,34,35,50,52 symptom acuity, 8,25,30,34,35,50,52 exercise history, 8,52 payer, 8,25,30,34,35,50,52 surgical history, 1,2,5,8-11 number of comorbid conditions, 8,23,35,51 gender, 8,25,30,34,35,50,52 type of referring physician, 8,52 depression, 8,35 income, 48 educational level, 35 and ethnicity. 35 Because the burden of collecting clinical data is significant, not all variables have been included in any one model. The literature has suggested that psychosocial factors play an important role in the development of chronic disability in patients with low back pain. 16,39,49 In an effort to make FS outcomes prediction models more clinically relevant and to improve model predictive ability, our team investigated the inclusion of a measure of fear-avoidance beliefs into an existing risk adjustment model of discharge FS. We focused on the theory of fear and activity avoidance developed by Waddell et al, 68 as assessed using a single screening item from the Fear-Avoidance Beliefs Questionnaire (FABQ). 29 Fearavoidance beliefs 68 are embodied in the fear-avoidance model of musculoskeletal pain. 36 The model posits that an individual s response to an episode of pain falls along a continuum, ranging from avoidance (maladaptive) to confrontation (adaptive), and provides one explanation of why a minority of patients with acute low back pain syndromes may develop chronic disability. 37,65-67 The FABQ is a patient self-reported questionnaire with 2 scales: 1 assessing fear-avoidance beliefs regarding work activities and 1 assessing fear-avoidance beliefs regarding physical activities. 68 Because many people who seek outpatient therapy for lumbar spine dysfunctions are not working, we only used the physical activities subscale (FABQ-PA) for the current analyses. There is evidence that identifying patients with elevated fearavoidance beliefs and managing them accordingly may reduce fear and improve outcomes. 3-5,10-20,36,40,41,56,68 The purpose of this study was to determine the effect of adding a single-item screening variable that classified patients with elevated versus not elevated fearavoidance beliefs of physical activities on a model that predicts risk-adjusted FS outcomes. METHODS Study Design and Setting We performed a retrospective analysis of data from a prospective, longitudinal cohort study of 49 376 patients being treated for a lumbar spine dysfunction in 434 outpatient physical therapy clinics in 39 states of the United States. Each clinic was participating with Focus On Therapeutic Outcomes, Inc (FOTO, Knoxville, TN), an international medical rehabilitation database management company. 58,59 This project was a retrospective study, and no treatment was modified; therefore, no patient consent forms were required. The Institutional Review Board for the Protection of Human Subjects from Focus On Therapeutic Outcomes, Inc approved the study. Patients The sample consisted of patients with lumbar spine dysfunctions (TABLE 1). Patients were selected for analyses from the FOTO database if they had completed the lumbar computerized adaptive testing (CAT) FS survey at intake, between April 2007 and May 2010, and had complete data for all risk adjustment variables. All patients had intake FS measures, of which 62% (30 858) had discharge FS measures. 7 Differences between those with complete (intake and discharge FS measures) versus those with intake FS measures alone were assessed using chi-square tests of independence if data were categorical or t tests if data were continuous (TABLE 1). Patients with intake and discharge FS measures compared to patients with just intake FS measures were older (P<.01), had more surgery (P =.02), had less chronic symptoms (P<.01), were more likely to receive benefits from indemnity, Medicare Part A, Medicare Part B, and workers compensation plans (P<.01), reported higher intake FS (P<.01), received more visits (P<.01), and were treated over a longer treatment duration (P<.01). Groups were not different by number of comorbid conditions (P =.79), gender (P =.43), or classification by fear-avoidance at intake (P =.11). Data Collection All data, FS measures, and risk adjustment variables were collected using Patient Inquiry computer software developed by FOTO. Data were collected according to methods that have been described. 7,8,25,30,69 Briefly, patients seeking rehabilitation entered demographic, fear-avoidance, and FS data using a computer in the clinic, prior to initial evaluation and intervention. Clinicians entered demographic information at intake and discharge via the computer. Patient-entered demographic risk adjustment data included age (years, continuous), symptom acuity (calendar days between date of dysfunction onset to date of initial evaluation, grouped as 0 to 7 days, greater than 7 days to 14 days, greater than 14 days to 21 days, greater than 21 days to 90 days, greater than 90 days to 6 months, or greater than 6 months), surgical history (none, 1 or more), condition complexity (quartile of a total of 29 comorbid conditions), and gender (male, female). Clinicians entered type of payer (indemnity, litigation, Medicaid, Medicare journal of orthopaedic & sports physical therapy volume 41 number 5 may 2011 337

[ research report ] Journal of Orthopaedic & Sports Physical Therapy Part A, Medicare Part B, patient private pay, health maintenance organization [HMO], preferred provider organization [PPO], workers compensation, or other). Intake FS functioned as the covariate. Each variable (age, 8,25,30,34,35,50,52 symptom acuity, 8,25,30,34,35,50,52 surgical history, 8,25,30,34,35,50,52 condition complexity, 21,25-28,30 gender, 8 type of payer, 8,23,48 and intake FS 8,25,30,50,52 ) has been shown to influence discharge FS in similar data sets and was used in the risk-adjusted prediction models for the current analyses. Lumbar CAT and FS Measures The lumbar CAT has been developed and simulated, 25 and is in routine clinical use in the United States 76,77 and Israel. 7,8 Briefly, items in the lumbar CAT item bank represent functional activities, like bending or stooping. There are 6 response choices for items from the Back Pain Functional Scale 57 and 3 response choices for physical functioning items. 22,31,71 Unidimensionality, local independence, and internal consistency reliability (α =.92) of the 25-item bank have been supported. 25 Calibrated items formed an item bank with a clinically logical hierarchical structure. 25 The lumbar CAT was developed using data from patients with lumbar dysfunctions, which makes the lumbar CAT a body-part-specific or condition-specific CAT. The CAT was developed following common CAT logic. 60 The adaptive test starts by administering the most informative 44 item. Patient FS estimates with associated standard errors (SE) are calculated. 38 The CAT stops when 1 or 2 stopping rules related to the precision of the measure estimate are satisfied: (1) SE for the provisional level of FS is less than 4 out of 100 FS units (functional status estimates were scaled 0 to 100 using a linear transformation with higher FS measures representing higher functioning), and (2) change in provisional level of FS estimates for the last 3 items is 1 or less out of 100. The SE of smaller than 4 represents less than 0.36 standard deviations of the scale range. 25 As an example, TABLE 1 Differences Between Patients With Intake- Only Versus Intake and Discharge Data* Characteristic Intake Only (n = 18 518) Intake and Discharge (n = 30 858) Age, y 51 17 (18, 100, 54) 55 17 (18, 100, 56) 18 to <45, % 37 28 45 to <65, % 40 39 65 to <75, % 13 19 75, % 10 14 Missing, % 0 0 Symptom acuity, % 0 to 7 d 6 6 >7 to 14 d 7 8 >14 to 21 d 8 8 >21 to 90 d 24 24 >90 d to 6 mo 12 13 >6 mo 44 41 Surgical history, % None 83 82 1 or more 17 18 Comorbid conditions, % None 14 14 1 15 15 2 or 3 29 29 4 or more 42 42 Table continues on page 339. the CAT would stop if, after administering the 3rd, 4th, and 5th items, the FS estimates were 46, 47, and 46, which would imply that the change in FS estimates was 1 or less out of 100, thus there was no need to ask any more items to improve measure precision. FS, as assessed using the lumbar CAT items, which have been mapped to the ICF, represents the activity dimension of the World Health Organization s International Classification of Functioning, Disability and Health. 79 FS measure data were labeled intake when the patient completed the CAT prior to initial evaluation (ie, at admission) and discharge when the patient completed the CAT at the time of discharge. There are data to support that FS measures estimated using the lumbar CAT have adequate known group construct validity, 25 are precise, 25,30 sensitive to change, 30 responsive, 30 and usable, 8,77 and have clinically meaningful interpretations. 69 Screening for Fear-Avoidance Beliefs of Physical Activities Our team developed a single-item screening method that can be used to classify patients by elevated versus not-elevated fear-avoidance beliefs of physical activities. 29 Using item response theory methods, 44,61 we analyzed items from the original FABQ, answered by a large sample of patients from outpatient therapy clinics, and identified a screening process that adequately classified patients with elevated or not-elevated fear-avoidance beliefs of physical activities. One item (ie, I should not do physical activities which [might] make my pain worse ) that provided maximum information 44 at the median fear-avoidance scale was selected as the screening item to dichoto- 338 may 2011 volume 41 number 5 journal of orthopaedic & sports physical therapy

Journal of Orthopaedic & Sports Physical Therapy TABLE 1 mize patients by elevated (response unsure or higher using 5 response choices) versus not-elevated fear. The diagnostic accuracy of the screening item was supported (sensitivity, 0.82; specificity, 0.98; likelihood ratio [LR], 34.88; negative LR, 0.18). 29 All patients completed the fear-avoidance screening process as part of their intake survey. Data Analysis To assess the predictive ability of the variables in our model, data were regressed using a multiple linear regression model employing ordinary least-squares estimation. Discharge FS was the dependent Differences Between Patients With Intake- Only Versus Intake and Discharge Data* (continued) Characteristic Intake Only (n = 18 518) Intake and Discharge (n = 30 858) Gender, % Male 40 40 Female 60 60 Payer, % PPO 33 35 Indemnity 2 3 Litigation 1 1 Medicaid 8 3 Medicare Part A 3 5 Medicare Part B 17 22 Patient private pay 1 1 HMO 14 11 Workers compensation 8 9 Other 13 10 Fear-avoidance of physical activities, % Not elevated 59 59 Elevated 41 41 Intake FS 49 14 (3, 100, 49) 49 14 (3, 100, 50) Treatment visits 8 6 (2, 94, 6) 11 7 (2, 93, 9) Missing 11 1 Duration of care, d 32 31 (2, 351, 23) 44 33 (2, 364, 35) Missing 12 1 Abbreviations: FS, functional status; HMO, health maintenance organization; PPO, preferred provider organization. *Values are percents for categorical data and mean SD (minimum, maximum, median) for continuous data. t statistic was significant at P<.01. Chi-square statistic was significant at P<.01. variable. Intake FS was the covariate. All demographic risk adjustment variables were in the model. One level of each categorical variable was selected as the reference level, while other levels of each variable were transformed to dummy variables. Two models were run, 1 with and 1 without the FABQ-PA variable. The R 2 statistic of each model was used to assess the predictive ability of the model. Significance of each variable to the model was assessed using a t statistic. We set α to.05, and 95% confidence intervals (CIs) were estimated for each variable beta. Partial correlation coefficients for each variable were estimated to assess the strength each variable had in the predictive model (ie, amount of data variance controlled by each variable, as assessed by the partial correlation coefficient squared). We cross-validated the model-building procedures, with and without the FABQ-PA variable, by randomly separating patients into 2 equal samples (split-half validation). This produced 2 samples: the developmental sample (used to develop the model) and the testing sample (used to test the stability of the independent variable coefficients of the model and test predictions using beta coefficients from the developmental sample). 55 Data from both samples were fit to the same multivariable regression model as above. Beta coefficients, with 95% CIs, were estimated and compared, to determine whether the beta coefficients were stable between the samples, which could then be used to examine the predictive validity (ie, cross-validation) of the model. Finally, as another approach to the assessment of the predictive validity of the model, we estimated predictive ratios. 23,32 To generate predictive ratios, the developmental sample was used to estimate beta coefficients for the independent variables. Beta coefficients were used to predict discharge FS in the testing sample. Predictive ratios, one with and one without the FABQ-PA variable, were estimated using the testing sample by dividing predicted discharge FS by actual discharge FS. If the predicted discharge FS was close to the actual discharge FS, the predictive validity of the regression model would be supported. 32 RESULTS Models (TABLE 2) produced R2 values approximating 0.30. Beta coefficients should be interpreted according to the following example: compared to having no comorbid conditions, discharge FS on average will be 4.16 FS units less when the patient reports 4 or more comorbid conditions, while control- journal of orthopaedic & sports physical therapy volume 41 number 5 may 2011 339

[ research report ] TABLE 2 Predictive Models of Discharge Functional Status With and Without Fear-Avoidance Beliefs of Physical Activities Without FABQ-PA* With FABQ-PA Journal of Orthopaedic & Sports Physical Therapy Variable Beta (95% CI) t Beta (95% CI) t FS intake 0.52 (0.50, 0.53) 85.76 0.51 (0.50, 0.52) 83.33 Gender, male (reference) Female 1.33 ( 1.65, 1.01) 8.20 1.35 ( 1.67, 1.03) 8.33 Acuity, 0 to 7 d (reference) >7 to 14 d 2.81 ( 3.65, 1.98) 6.62 2.77 ( 3.60, 1.94) 6.52 >14 to 21 d 4.50 ( 5.28, 3.63) 10.62 4.42 ( 5.24, 3.60) 10.54 >21 to 90 d 6.71 ( 7.42, 6.00) 18.53 6.70 ( 7.40, 5.99) 18.52 >90 d to 6 mo 9.09 ( 9.87, 8.31) 22.84 9.09 ( 9.86, 8.31) 22.83 >6 mo 11.27 ( 11.96, 10.58) 32.00 11.26 ( 11.95, 10.57) 31.99 Age 0.07 ( 0.09, 0.06) 11.87 0.07 ( 0.09, 0.06) 11.81 Surgery, none (reference) 1 or more 3.24 ( 3.65, 2.83) 15.45 3.23 ( 3.64, 2.82) 15.40 Comorbidities, none (reference) 1 1.54 ( 2.11, 0.96) 5.27 1.57 ( 2.14, 1.00) 5.40 2 or 3 2.20 ( 2.70, 1.70) 8.61 2.27 ( 2.77, 1.77) 8.89 4 4.16 ( 4.66, 3.67) 16.42 4.24 ( 4.73, 3.74) 16.71 Payer, PPO (reference) Indemnity 0.98 ( 1.90, 0.06) 2.09 0.98 ( 1.90, 0.05) 2.08 Litigation 6.48 ( 8.96, 4.01) 5.13 6.35 ( 8.83, 3.88) 5.04 Medicaid 4.93 ( 5.83, 4.03) 10.78 4.91 ( 5.80, 4.01) 10.73 Medicare Part A 0.61 ( 0.18, 1.40) 1.51 0.72 ( 0.07, 1.51) 1.79 Medicare Part B 1.27 ( 1.77, 0.77) 5.01 1.25 ( 1.75, 0.75) 4.92 Patient private pay 2.12 ( 3.92, 0.31) 2.30 2.09 ( 3.89, 0.29) 2.28 HMO 0.23 ( 0.76, 0.30) 0.85 0.24 ( 0.77, 0.29) 0.87 WC 5.50 ( 6.08, 4.91) 18.50 5.40 ( 5.98, 4.82) 18.18 Other 1.56 ( 2.10, 1.01) 5.61 1.54 ( 2.08, 1.00) 5.56 FABQ-PA, low (reference) Elevated...... 1.24 ( 1.55, 0.92) 7.63 Constant 56.19 (55.08, 57.30) 99.01 57.09 (55.96, 58.23) 98.57 Abbreviations: FABQ-PA, Fear-Avoidance Beliefs Questionnaire physical activities subscale classification; FS, functional status; HMO, health maintenance organization; PPO, preferred provider organization; WC, workers compensation. *Model F 21,30836 = 628.35; R 2 = 0.2997; n = 30 858. Model F 22,30835 = 603.55; R 2 = 0.3010; n = 30 858. All t statistics were significant at P<.01, except for these variables. ling for the effects of the other variables in the model. Beta coefficient estimates suggested the risk adjustment variables performed as expected. For example, discharge FS decreased as symptoms became more chronic, age increased, number of comorbid conditions increased, and patients received benefits from workers compensation, litigation, Medicare Part B, or Medicaid. When the FABQ-PA variable was added to the model, the R 2 value increased negligibly from 0.2997 to 0.3010. Compared to patients whose fear was not elevated, those with elevated fear on average had 1.24 fewer units of discharge FS, while controlling for the other variables in the model. The addition of a classification (elevated versus not elevated) by fear-avoidance beliefs of physical activities to the model had no effect on beta coefficient estimates (95% CIs for coefficients per variable were not different between models). Strength of each variable in the predictive models is displayed in TABLE 3 and suggests that the most important risk adjustment variables, in decreasing importance, were intake FS, symptom acuity, number of comorbid conditions, surgical history, payer, age, FABQ-PA classification, and gender, with FABQ-PA classification controlling 0.22% of the data variance in this model. 340 may 2011 volume 41 number 5 journal of orthopaedic & sports physical therapy

Journal of Orthopaedic & Sports Physical Therapy There were no differences (P<.05) between beta coefficients for any combination of developmental and testing samples, with or without the FABQ-PA variable, suggesting stability in the predictive models. Predictive ratios (mean SD, 1.09 0.40 without FABQ-PA, versus 1.10 0.41 with FABQ-PA) were not different with (95% CI: 1.09, 1.10) versus without (95% CI: 1.10, 1.11) FABQ-PA. The average SD differences (predicted minus actual) in discharge FS, with and without the FABQ-PA variable, were 1.66 13.8 (95% CI: 1.44, 1.88) and 2.26 13.8 (95% CI: 2.04, 2.48), respectively. Data suggest that the predictive models tend to predict discharge FS well, but with values slightly higher than the actual values. DISCUSSION The primary purpose of this study was to assess the effect of adding a single-item screening variable that classified patients with elevated versus not elevated fear-avoidance beliefs of physical activities (FABQ-PA) on an existing model that predicted risk-adjusted FS outcomes. The results suggest that the FABQ-PA variable improves the predictive ability of the model but only slightly, thus predicting discharge FS within 10% of actual discharge FS, on average. Although the predictive ability of the models and the statistically significant addition of fear-avoidance beliefs of physical activities classification to the models were encouraging, there is much room for improvement. Although FABQ-PA classification added little to a predictive model of discharge FS, it is clinically logical to add a variable that quantifies the level of fearavoidance for patients with lumbar spine dysfunctions. As summarized by George et al, 19 when using the complete FABQ- PA, there is evidence indicating that elevated fear-avoidance beliefs can be predictive of poor outcomes for patients with lumbar spine dysfunctions. 12,15,56 In addition, George and Stryker 18 reported TABLE 3 that when dichotomizing patients of cervical, upper extremity, lumbar or lower extremity musculoskeletal pain using the psychometric methods described by Hart et al, 29 FABQ-PA was associated with functional outcomes, although patients with elevated FABQ-PA were associated with more functional change compared to those with not-elevated FABQ-PA. In our multivariable model, fear-avoidance added little to our understanding of FS outcomes, so we conducted a univariate analysis using the same data set. In a 1-way analysis of covariance (ANCOVA), with discharge FS as the dependent variable, intake FS as the covariate, and the FABQ-PA variable (elevated versus not elevated) as the main factor, fear classification significantly affected discharge FS (F 1,30855 = 44.4; P<.001; adjusted leastsquares mean, 65.0 with not-elevated fear versus 63.9 with elevated fear). Hence, the FABQ-PA classification variable affected discharge FS in almost identical ways in the univariate and multivariable models. But when the FABQ-PA classification variable was added to the multivariable model, fear added a statistically significant, albeit minimal, improvement to the power of that model. Possibly more importantly, evidence has been reported to suggest that there are treatment strategies therapists can use to improve elevated fear-avoidance beliefs that may, in turn, improve functional outcomes. 3,15,19,40,42,63,64 These Importance of Risk Adjustment Variables in the Discharge FS Predictive Models* Variable Without FABQ-PA With FABQ-PA FS intake 20.39 19.39 Symptom acuity 5.26 5.29 Comorbid conditions 1.02 1.06 Surgery 0.79 0.78 Payer 0.42 0.41 Age 0.17 0.34 FABQ-PA... 0.22 Gender 0.17 0.18 Abbreviations: FABQ-PA, Fear-Avoidance Beliefs Questionnaire physical activities subscale classification; FS, functional status. *Values are percent variance controlled for each variable in the multivariable model. studies did not use the same outcomes measure as those estimated in the current study. However, because fear-avoidance beliefs, as assessed at intake, may decrease (ie, improve) or in some cases increase (ie, worsen) during therapy, the variability in fear-avoidance scores during therapy may lower the predictive ability of the FABQ-PA variable compared to the other patient variables that do not change during therapy (age, symptom acuity, surgical history). Therefore, identification of elevated fear-avoidance beliefs at intake seems clinically important because treatment strategies may be applied to patients with elevated fear for the purpose of decreasing fear and increasing FS change. Change in fear during therapy might prove a stronger predictor of FS outcomes compared to fear level at intake, as has been suggested. 62 Regardless of the FS outcomes measure, 15,18,25 assessing the level of fear-avoidance beliefs at initial evaluation for patients with lumbar spine dysfunctions seems warranted, particularly if evaluation results are used to develop targeted interventions that are used to assist in patient management for improved functional outcomes. Therefore, we recommend that clinicians use methods like the one described here to identify patients with elevated fearavoidance beliefs and decide whether to use targeted interventions for those with elevated fear. Because the FABQ- PA classification variable added little to journal of orthopaedic & sports physical therapy volume 41 number 5 may 2011 341

[ research report ] Journal of Orthopaedic & Sports Physical Therapy the predictive model, one can argue that the FABQ-PA classification variable may not be important enough to be used for predicting FS outcomes, given the other variables in the model. Further research is warranted to clarify this issue. Although evaluating patients for elevated fear-avoidance beliefs appears clinically relevant for one patient at a time in the clinic, we are only beginning to understand how a fear-avoidance beliefs variable interacts with other important variables that affect FS outcomes. For example, Werneke et al 78 used a similar multivariable regression model and discharge FS measure to analyze a smaller (n = 207) sample of patients with lumbar dysfunctions. They reported that fearavoidance beliefs of physical activities did not affect discharge FS when the model contained intake FS, having chronic symptoms, receiving benefits from a preferred provider organization (PPO), and being classified as a person with noncentralizing symptoms (R 2 = 0.42). In another analysis, Werneke et al 77 applied a dual-level method of classifying patients based on pain pattern classification, 72-75 combined with level of fear-avoidance beliefs of physical activities. 77 Using a similar multivariable regression model and discharge FS measure analyzing the same (n = 207) sample of patients with lumbar dysfunctions, Werneke et al 77 reported that being classified as having noncentralizing symptoms and having elevated fear-avoidance beliefs was associated with 14 units less of discharge FS compared to being classified as having centralizing symptoms and having low fear-avoidance beliefs of physical activities, while controlling for intake FS, chronic symptoms, and receiving PPO benefits (R 2 = 0.42). In both of those studies, elevated fear-avoidance was identified by median split of the original Waddell et al 68 summative score method. In the current study, we assessed FS using a lumbar CAT, 25,30,69 and fearavoidance beliefs of physical activities 29 was assessed using a single-item screening method designed to classify patients by elevated versus not-elevated levels of fear. Both FS estimates and FABQ-PA classification are based on item response theory (IRT) methods. 44,61 Literature supports the psychometric and practical strength of the FS scores estimated using the IRT-based lumbar CAT administration, 8,25,30,70,76-78 but additional prospective study is needed to validate whether a single-item screening scale represents the construct of fear-avoidance beliefs of physical activities well. IRT methods are complex, but the advantages are clinically logical and practical. First, IRT methods produce linear measures, which tend to improve measure sensitivity to change, as compared to measures estimated using summative scored surveys that employ categorical item responses. 24,45 Second, IRT methods facilitate identification of where along the construct continuum the scale is most discriminating (ie, where the cut-score for elevated versus not elevated should be). 29,44 Third, IRT methods facilitate identification of the item in a scale that produces maximum information at the cut-score for a construct, given the ability of the patient, 29 which was instrumental in identifying the single screening item and the response choice in the FABQ-PA screening tool that produces a sensitive (0.82) and specific (0.98) classification of elevated fear. 29 Compared to the original summative scoring of the FABQ scales, 68 improved scale linearity, decreased measurement error, increased measure sensitivity to change, and more precise ways of identifying cut-scores and items for elevated versus not elevated fear available using IRT methods are promising. Fourth, use of CATs and single-item screening scales improves efficiency of data collection, which facilitates simultaneous assessment of multiple biopsychosocial constructs, with limited respondent burden. As more constructs are assessed, hopefully, more variables will be available to improve the predictive ability of models. More importantly, in the current study, we did not have a variable for classifying patients according to movement signs and symptoms. Data support use of treatment-based classification systems to enhance patient outcomes, 1,2,10,43 but clinical classification data have not been adequately represented in prior risk-adjusted models used to explain FS outcomes. 8,52,54 Data from our research group suggest that the predictive ability of models improves when variables for classifying patients in clinically logical ways are included. 76 Compared to the R 2 value for our current model with the FABQ-PA variable (R 2 = 0.30), the R 2 value reported by Werneke et al 76 when a classification variable was added to the same model (R 2 = 0.50) represents approximately a 66% gain in predictive ability. Hence, support is growing to include clinical classification variables in FS predictive models. Unfortunately, in a large, proprietary database like the one analyzed in the current study, not all clinicians are trained in patient response classification methods, 6,46 so classification data were unavailable. In addition, when classification data are used to direct treatment, treatment confounds outcomes assessment, which affects regression models. 9 Another concern to consider is that predictive models are dependent on the dependent and independent variables used. Previous authors who have reported fear-avoidance beliefs as predictive have used different measures of FS. 15,18 Whether there is a specificity based on outcome measure used is unknown. Also, simply adding independent variables to the regression model should increase model predictive ability, so researchers and clinicians should strive to include the most clinically relevant variables in their models. As an example, in a large (n = 24 276) sample of patients with lumbar spine dysfunctions, Resnik and Hart 52 included the following independent variables: age, intake FS, gender, symptom acuity, number of surgeries, payer, exercise history, and employment status. Gender was the only variable that did not significantly affect discharge FS, so consideration should be given to in- 342 may 2011 volume 41 number 5 journal of orthopaedic & sports physical therapy

Journal of Orthopaedic & Sports Physical Therapy cluding exercise history and employment status in discharge FS predictive models. In another large (n = 16 281) lumbar data set, Resnik et al 54 included type of referring physician (primary care, orthopaedic surgeon, etc) and primary diagnosis from ICD-9-CM codes, which affected discharge FS and warrant consideration in future predictive models. Fear-avoidance beliefs of work activities, assessed either as originally described by Waddell et al 68 or using a single-item screening variable, 29 were not included in our predictive models by design. Some may consider this a weakness of the current study, as fear of work activities is associated with FS outcomes in other studies 4,11,12,14,16,36,68 and can be screened efficiently. 29 However, the effect of type of payer on fear-avoidance beliefs of work items has not been studied. For example, if patients receiving benefits from workers compensation answer fear-avoidance beliefs items differently compared to patients receiving benefits from HMOs, which by definition is differential item functioning, 29,47 the presence of differential item functioning by payer would invalidate use of the single-item screen for fear-avoidance of work in models containing patients receiving benefits from different payer types. We would prefer to know if fear-avoidance beliefs of work activity items display differential item functioning by payer, before including fear of work activities in our models. For future research, we also recommend assessing long-term outcomes, as compared to our short-term outcomes assessment in this study. Long-term studies with large data sets that include many diverse risk adjustment variables should facilitate predictive model development. In addition, as larger, more complex data sets are analyzed, more sophisticated statistical techniques 50,51 may assist the examination of complex associations between different psychosocial measures (eg, George et al s psychometric study of fear-avoidance models) 19 and other clinically appropriate variables. Finally, although clinicians had access to patient-specific reports that detailed FS measures, and whether the patient had elevated or not-elevated fear-avoidance beliefs of physical activities each time the patient entered data into the clinical software, we had no data quantifying whether clinicians read the reports or used the results to modify treatments, or data on the treatments used. In addition, we did not study the effect of different degrees (eg, baccalaureate, master, doctorate), certifications (eg, Orthopedic Clinical Specialist), or clinical experience (eg, years of experience or advanced training), because these data were not part of the research question or design of this study. Future prospective designs are needed to continue the investigation into whether targeted interventions for patients with elevated fear-avoidance beliefs, as well as clinician degrees, experience, or training affect predictive validity of the models. CONCLUSION We have demonstrated that discharge FS can be modeled with introductory predictive ability using several common risk adjustment variables, which will require prospective analyses to confirm results. Adding a single-item screening variable for classifying patients by elevated versus not-elevated fear-avoidance beliefs of physical activities provided minimal additional meaningful interpretation of FS outcomes. Exploration of other clinically relevant variables designed to improve outcomes prediction is warranted, such as change of fear level during therapy or patient classification variables. t KEY POINTS FINDINGS: Results suggest that a singleitem variable for fear minimally improves prediction of FS outcomes. IMPLICATION: Clinicians should consider screening patients for elevated fearavoidance beliefs of physical activities to improve patient management, but other variables, like intake FS, symptom acuity, number of comorbid conditions, surgical history, payer, and age appear more important when predicting FS outcomes. CAUTION: Our predictive models used a single-item screening process for identifying elevated fear-avoidance beliefs. Exploration of other clinically relevant variables, such as a way to classify patients or tracking change in psychosocial constructs during therapy, should be examined for the purpose of improving risk-adjusted outcomes prediction. REFERENCES 1. Brennan GP, Fritz JM, Hunter SJ, Thackeray A, Delitto A, Erhard RE. Identifying subgroups of patients with acute/subacute nonspecific low back pain: results of a randomized clinical trial. Spine (Phila Pa 1976). 2006;31:623-631. http:// dx.doi.org/10.1097/01.brs.0000202807.72292.a8 2. Browder DA, Childs JD, Cleland JA, Fritz JM. Effectiveness of an extension-oriented treatment approach in a subgroup of subjects with low back pain: a randomized clinical trial. Phys Ther. 2007;87:1608-1618; discussion 1577-1609. http://dx.doi.org/10.2522/ptj.20060297 3. Burton AK, Waddell G, Tillotson KM, Summerton N. Information and advice to patients with back pain can have a positive effect. A randomized controlled trial of a novel educational booklet in primary care. Spine (Phila Pa 1976). 1999;24:2484-2491. 4. Cleland JA, Fritz JM, Brennan GP. Predictive validity of initial fear avoidance beliefs in patients with low back pain receiving physical therapy: is the FABQ a useful screening tool for identifying patients at risk for a poor recovery? Eur Spine J. 2008;17:70-79. http://dx.doi.org/10.1007/ s00586-007-0511-y 5. Crombez G, Vlaeyen JW, Heuts PH, Lysens R. Pain-related fear is more disabling than pain itself: evidence on the role of painrelated fear in chronic back pain disability. Pain. 1999;80:329-339. 6. Delitto A, Erhard RE, Bowling RW. A treatmentbased classification approach to low back syndrome: identifying and staging patients for conservative treatment. Phys Ther. 1995;75:470-485; discussion 485-479. 7. Deutscher D, Hart DL, Dickstein R, Horn SD, Gutvirtz M. Implementing an integrated electronic outcomes and electronic health record process to create a foundation for clinical practice improvement. Phys Ther. 2008;88:270-285. http:// dx.doi.org/10.2522/ptj.20060280 8. Deutscher D, Horn SD, Dickstein R, et al. Associations between treatment processes, patient characteristics, and outcomes in outpatient journal of orthopaedic & sports physical therapy volume 41 number 5 may 2011 343

[ research report ] Journal of Orthopaedic & Sports Physical Therapy physical therapy practice. Arch Phys Med Rehabil. 2009;90:1349-1363. http://dx.doi. org/10.1016/j.apmr.2009.02.005 9. Edmond SL, Werneke MW, Hart DL. Association between centralization, depression, somatization, and disability among patients with nonspecific low back pain. J Orthop Sports Phys Ther. 2010;40:801-810. http://dx.doi.org/10.2519/ jospt.2010.3334 10. Fritz JM, Delitto A, Erhard RE. Comparison of classification-based physical therapy with therapy based on clinical practice guidelines for patients with acute low back pain: a randomized clinical trial. Spine (Phila Pa 1976). 2003;28:1363-1371; discussion 1372. http:// dx.doi.org/10.1097/01.brs.0000067115.61673. FF 11. Fritz JM, George SZ. Identifying psychosocial variables in patients with acute work-related low back pain: the importance of fear-avoidance beliefs. Phys Ther. 2002;82:973-983. 12. Fritz JM, George SZ, Delitto A. The role of fearavoidance beliefs in acute low back pain: relationships with current and future disability and work status. Pain. 2001;94:7-15. 13. George SZ. Fear: a factor to consider in musculoskeletal rehabilitation. J Orthop Sports Phys Ther. 2006;36:264-266. http://dx.doi. org/10.2519/jospt.2006.0106 14. George SZ, Bialosky JE, Donald DA. The centralization phenomenon and fear-avoidance beliefs as prognostic factors for acute low back pain: a preliminary investigation involving patients classified for specific exercise. J Orthop Sports Phys Ther. 2005;35:580-588. http://dx.doi. org/10.2519/jospt.2005.2073 15. George SZ, Fritz JM, Bialosky JE, Donald DA. The effect of a fear-avoidance-based physical therapy intervention for patients with acute low back pain: results of a randomized clinical trial. Spine (Phila Pa 1976). 2003;28:2551-2560. http:// dx.doi.org/10.1097/01.brs.0000096677.84605. A2 16. George SZ, Fritz JM, Childs JD. Investigation of elevated fear-avoidance beliefs for patients with low back pain: a secondary analysis involving patients enrolled in physical therapy clinical trials. J Orthop Sports Phys Ther. 2008;38:50-58. http://dx.doi.org/10.2519/jospt.2008.2647 17. George SZ, Fritz JM, McNeil DW. Fear-avoidance beliefs as measured by the fear-avoidance beliefs questionnaire: change in fear-avoidance beliefs questionnaire is predictive of change in self-report of disability and pain intensity for patients with acute low back pain. Clin J Pain. 2006;22:197-203. 18. George SZ, Stryker S. Fear-avoidance beliefs and clinical outcomes for patients seeking outpatient physical therapy for musculoskeletal pain conditions. J Orthop Sports Phys Ther. 2011;41:249-259. 19. George SZ, Valencia C, Beneciuk JM. A psychometric investigation of fear-avoidance model measures in patients with chronic low back pain. J Orthop Sports Phys Ther. 2010;40:197-205. http://dx.doi.org/10.2519/jospt.2010.3298 20. George SZ, Zeppieri G. Physical therapy utilization of graded exposure for patients with low back pain. J Orthop Sports Phys Ther. 2009;39:496-505. http://dx.doi.org/10.2519/ jospt.2009.2983 21. Groll DL, To T, Bombardier C, Wright JG. The development of a comorbidity index with physical function as the outcome. J Clin Epidemiol. 2005;58:595-602. http://dx.doi.org/10.1016/j. jclinepi.2004.10.018 22. Hart DL. Assessment of unidimensionality of physical functioning in patients receiving therapy in acute, orthopedic outpatient centers. J Outcome Meas. 2000;4:413-430. 23. Hart DL, Connolly JB. Pay-for-Performance for Physical Therapy and Occupational Therapy: Medicare Part B Services Final Report. Grant #18-P-93066/9-01: Health & Human Services/ Centers for Medicare & Medicaid Services. Knoxville, TN: Focus On Therapeutic Outcomes, Inc; 2006. 24. Hart DL, Deutscher D, Werneke MW, Holder J, Wang YC. Implementing computerized adaptive tests in routine clinical practice: experience implementing CATs. J Appl Meas. 11:288-303. 25. Hart DL, Mioduski JE, Werneke MW, Stratford PW. Simulated computerized adaptive test for patients with lumbar spine impairments was efficient and produced valid measures of function. J Clin Epidemiol. 2006;59:947-956. http:// dx.doi.org/10.1016/j.jclinepi.2005.10.017 26. Hart DL, Wang YC, Stratford PW, Mioduski JE. Computerized adaptive test for patients with foot or ankle impairments produced valid and responsive measures of function. Qual Life Res. 2008;17:1081-1091. http://dx.doi.org/10.1007/ s11136-008-9381-y 27. Hart DL, Wang YC, Stratford PW, Mioduski JE. A computerized adaptive test for patients with hip impairments produced valid and responsive measures of function. Arch Phys Med Rehabil. 2008;89:2129-2139. http://dx.doi.org/10.1016/j. apmr.2008.04.026 28. Hart DL, Wang YC, Stratford PW, Mioduski JE. Computerized adaptive test for patients with knee impairments produced valid and responsive measures of function. J Clin Epidemiol. 2008;61:1113-1124. http://dx.doi.org/10.1016/j. jclinepi.2008.01.005 29. Hart DL, Werneke MW, George SZ, et al. Screening for elevated levels of fear-avoidance beliefs regarding work or physical activities in people receiving outpatient therapy. Phys Ther. 2009;89:770-785. http://dx.doi.org/10.2522/ ptj.20080227 30. Hart DL, Werneke MW, Wang YC, Stratford PW, Mioduski JE. Computerized adaptive test for patients with lumbar spine impairments produced valid and responsive measures of function. Spine (Phila Pa 1976). 35:2157-2164. http:// dx.doi.org/10.1097/brs.0b013e3181cbc17f 31. Hart DL, Wright BD. Development of an index of physical functional health status in rehabilitation. Arch Phys Med Rehabil. 2002;83:655-665. 32. Hwang W, Ireys HT, Anderson GF. Comparison of risk adjusters for medicaid-enrolled children with and without chronic health conditions. Ambul Pediatr. 2001;1:217-224. 33. Iezzoni LI. The risks of risk adjustment. JAMA. 1997;278:1600-1607. 34. Jette AM, Haley SM, Tao W, et al. Prospective evaluation of the AM-PAC-CAT in outpatient rehabilitation settings. Phys Ther. 2007;87:385-398. http://dx.doi.org/10.2522/ptj.20060121 35. Jette DU, Jette AM. Physical therapy and health outcomes in patients with spinal impairments. Phys Ther. 1996;76:930-941; discussion 942-935. 36. Leeuw M, Goossens ME, Linton SJ, Crombez G, Boersma K, Vlaeyen JW. The fear-avoidance model of musculoskeletal pain: current state of scientific evidence. J Behav Med. 2007;30:77-94. http://dx.doi.org/10.1007/ s10865-006-9085-0 37. Lethem J, Slade PD, Troup JD, Bentley G. Outline of a Fear-Avoidance Model of exaggerated pain perception--i. Behav Res Ther. 1983;21:401-408. 38. Linacre JM. Estimating measures with known polytomous item difficulties. Rasch Measure Trans. 1998;12:638. 39. Linton SJ. A review of psychological risk factors in back and neck pain. Spine (Phila Pa 1976). 2000;25:1148-1156. 40. Linton SJ, Boersma K, Jansson M, Svard L, Botvalde M. The effects of cognitive-behavioral and physical therapy preventive interventions on pain-related sick leave: a randomized controlled trial. Clin J Pain. 2005;21:109-119. 41. Linton SJ, Buer N, Vlaeyen JW, Hellsing A. Are fear-avoidance beliefs related to the inception of an episode of back pain? A prospective study. Psychol Health. 2000;14:1051-1059. 42. Linton SJ, Ryberg M. A cognitive-behavioral group intervention as prevention for persistent neck and back pain in a non-patient population: a randomized controlled trial. Pain. 2001;90:83-90. 43. Long A, Donelson R, Fung T. Does it matter which exercise? A randomized control trial of exercise for low back pain. Spine (Phila Pa 1976). 2004;29:2593-2602. 44. Lord FM. Applications of Item Response Theory to Practical Testing Problems. Hillsdale, NJ: L. Erlbaum Associates; 1980. 45. McHorney CA, Haley SM, Ware JE, Jr. Evaluation of the MOS SF-36 Physical Functioning Scale (PF-10): II. Comparison of relative precision using Likert and Rasch scoring methods. J Clin Epidemiol. 1997;50:451-461. 46. McKenzie R, May S. The Lumbar Spine: Mechanical Diagnosis and Therapy. 2nd ed. Waikanae, New Zealand: Spinal Publication, Ltd; 2003. 47. Millsap RE, Everson HT. Methodology review: statistical approaches for assessing measurement bias. Appl Psych Meas. 1993;17:297-334. 48. Nof L, Rone-Adams S, Hart DL. Relation between payer source and functional outcomes, visits, and treatment duration in U.S. patients with lumbar dysfunction. Internet J Allied Health 344 may 2011 volume 41 number 5 journal of orthopaedic & sports physical therapy