TO SHOW THE EFFECTIVENESS of their interventions

Transcription

1 70 Comparative Responsiveness of Locomotor Tests and Questionnaires Used to Follow Early Recovery After Total Knee Arthroplasty Eric Parent, MSc, Hélène Moffet, PhD ABSTRACT. Parent E, Moffet H. Comparative responsiveness of locomotor tests and questionnaires used to follow early recovery after total knee arthroplasty. Arch Phys Med Rehabil 2002;83: Objectives: To compare the responsiveness of 3 locomotor tests and 2 questionnaires in the early stage after a total knee arthroplasty (TKA) and to determine if the 4 responsiveness statistics ranked the measures similarly. Design: Longitudinal study. Setting: Rehabilitation institute. Participants: Twenty-five men and 40 women with knee osteoarthritis scheduled for a first TKA. Interventions: Not applicable. Main Outcome Measures: Six-minute gait distance, in-laboratory gait speed and stair ascent duration, Western Ontario and McMaster Universities (WOMAC) Osteoarthritis Index difficulty subscale, and Medical Outcomes Study Short-Form 36-Item Health Survey role physical and physical functioning subscale scores. Effect size, standardized response mean, paired t test, and relative efficiency statistics were computed for 3 time intervals: (1) before TKA to 2 months after TKA, (2) 2 to 4 months after TKA, and (3) before TKA to 4 months after TKA. Results: Responsiveness varied according to tests and intervals considered. For all intervals, the WOMAC difficulty subscale was the most responsive questionnaire and the 6-minute gait test was the most responsive locomotor test. Stair ascent duration was the least responsive measure. Of the responsiveness indices used, only effect size ranked the tests differently. Conclusions: The 6-minute gait test and the WOMAC difficulty subscale are recommended for outcome assessment during the early recovery period after TKA. Because interpretation guidelines are available and confidence intervals can be calculated for it, the standardized response mean is the most useful statistic. Key Words: Arthroplasty; Gait; Knee; Osteoarthritis; Quality of life; Questionnaires; Recovery of function; Rehabilitation by the American Congress of Rehabilitation Medicine and the American Academy of Physical Medicine and Rehabilitation From the Department of Rehabilitation, Laval University; and Center for Interdisciplinary Research in Rehabilitation and Social Integration, Quebec Rehabilitation Institute, Quebec City, Que, Canada. Accepted in revised form February 26, Supported by the Medical Research Council of Canada, Fonds de la Recherche en Santé du Québec, and the Institut de Recherche en Déficience Physique de Québec. Presented in part at the American Physical Therapy Association s Physical Therapy 99 Scientific Meeting, June 1999, Washington DC. Reprint requests to Hélène Moffet, PhD, Center for Interdisciplinary Research in Rehabilitation and Social Integration, Quebec Rehabilitation Institute, 525 Blvd Hamel, B-77, Quebec City, Que, G1M 2S8, Canada, helene.moffet@rea.ulaval.ca /02/ $35.00/0 doi: /apmr TO SHOW THE EFFECTIVENESS of their interventions to third-party payers, employers, and subjects, 1-3 health professionals have been asked to use clinical instruments with good measurement properties. Classically, validity and reliability are the metrologic properties that have been studied to the greatest extent However, to measure longitudinal changes, an additional property is needed. 5,11,12 This property, responsiveness, is either defined as: the capacity of a measure to detect (quantify) change in subjects functional status over time and to distinguish between subjects who change by differential amounts 13,14 or as an instrument s ability to detect clinically important changes, even if these changes are small. 3,11,15,16 In persons with severe osteoarthritis (OA), total knee arthroplasty (TKA) provides gradual pain reduction and progressive improvement in functional capacity and health-related quality of life (QOL) These changes have been noted over periods as long as 1 to 2 years after the surgery. 18,21,22 Thus, responsive measures of these attributes are required to document the longitudinal progress of subjects, the effectiveness of TKA, and its associated rehabilitation interventions during the overall recovery period. Recently, a panel of experts in OA research reached a consensus regarding which dimensions should be evaluated in trials of knee OA. The importance of evaluating physical function was emphasized and incorporating healthrelated QOL measurements in the core set of outcome measures to be used in such trials was suggested. 23 Among determinants of functional capacity, locomotor performance appeared to be very important. In a study 24 in which subjects with OA were asked to select the most important activity to mirror evolution in function, most chose difficulty with stair negotiation and walking as significantly impaired activities. Validated and reliable functional capacity questionnaires as well as measures of locomotor activities already exist. However, though the responsiveness of questionnaires of functional capacity has been documented, studies that simultaneously compare the responsiveness of functional capacity questionnaires, locomotor tests, and health-related QOL measures are scarce. 25 Furthermore, comparisons of such measures during the early follow-up period after TKA, a period in which the most important changes occur, are even rarer. 20 Although a need exists to establish the comparative responsiveness of functional outcome measures in subjects with TKA, there is no agreement on how responsiveness should be quantified. 9,11,16,26-30 The many statistical methods and research designs proposed to document responsiveness show this lack of agreement. 5-9,11,15,17,28,30-35 This deficiency can be explained by the fact that no criterion standard exists for the measurement of change in functional capacity. Considering the importance of evaluating responsiveness, the following recommendations have been proposed. First, to overcome the absence of a criterion standard of change, researchers have suggested that the best way to document responsiveness is to compare many measures according to many responsiveness statistics in many controlled experimental trials or longitudinal studies. 3,11,17,29,36-38 It has also been recommended that responsive-

2 RESPONSIVENESS OF OUTCOME MEASURES AFTER TKA, Parent 71 ness of outcome measures be compared in the situation in which they are intended to be used and with respect to time intervals, populations, and interventions involved. 39 The first objective of the present study was to compare the relative responsiveness of 3 locomotor tests, a disease-specific subscale and 2 generic questionnaire subscales in a group of persons before TKA and after (2 and 4mo postsurgery) by using 4 different responsiveness statistics. Our second objective was to determine if the 4 statistics used to quantify the relative responsiveness would lead to similar ranking of the outcome measures evaluated. METHODS Participants Subjects were recruited from the surgical waiting lists of 23 orthopedic surgeons working in the 5 main hospitals of the Quebec City metropolitan area, between March 1996 and October To be included, subjects were required: to have a diagnosis of primary OA of the knee, to be waiting for a first TKA, to live in the Quebec City metropolitan area, to be ambulatory with or without a walking aid, and to give informed consent. Subjects were excluded if they were planning a second surgery of the lower limbs during the first year after TKA, had associated conditions impeding their performance of locomotor tests, had undergone previous surgery of the lower limbs affecting their gait pattern, had neuromuscular or neurodegenerative diseases, had a knee infection after TKA or had other major complications (eg, loosening, emboli excluding thrombophlebitis), or had problems following the instructions of the study protocol. The present study was part of a randomized controlled trial in which half of the subjects participated in a new intensive functional rehabilitation program between the second and the fourth month after TKA. Because we included subjects from both control and experimental groups, we expected a variety of change scores during follow-up measurements. Such variety is ideal when evaluating responsiveness. Design and Protocol Subjects participated in 3 evaluations of approximately 2 hours each: before TKA surgery (PRE), 2 months postsurgery (POST1), and 4 months postsurgery (POST2). At each evaluation, the same procedure was repeated. The tests were administered as follows: 6-minute gait test, biomechanical analysis of gait and stair ascent, Western Ontario and McMaster Universities (WOMAC) Osteoarthritis Index, and Medical Outcomes Study Short-Form 36-Item Health Survey (SF-36). The subjects medical history and sociodemographic information were collected by a standardized telephone interview before TKA. The evaluations were conducted at the Quebec Rehabilitation Institute by 4 physical therapists trained as evaluators. One of the authors (EP) trained all evaluators according to a standard procedure. That training included observation of at least 3 evaluations and supervised practice with pilot subjects until adequate familiarity with the procedures was shown. At all times, 2 evaluators were present with the participants, ensuring continuous supervision in the application of the standard procedure. Construct of change. When assessing responsiveness, one must use a research design in which a real change is expected to occur. This logical expectation of change is referred to as a construct of change. 6 In the present study, responsiveness was measured by using various constructs of change and responsiveness statistics as previously suggested. 38 The first construct of change was based on natural recovery after TKA. As previously shown, 40 a functional deterioration is expected during the first time interval under study (PRE POST1) whereas an improvement should occur during the second time interval (POST1 POST2). However, this deterioration-improvement pattern makes it much more difficult to predict the outcome during the third interval (PRE POST2). A second construct of change based on a differential recovery pattern in subjects with and without an intensive rehabilitation was considered during the second and third time interval. Between POST1 and POST2 evaluations, an intensive functional rehabilitation program was offered to approximately half the subject sample. As a result, we expected a variety of change scores for the intervals including the period between POST1 and POST2 evaluations. Overall, such a design should enable one to test responsiveness to positive and negative changes. Because our purpose was not to test the effectiveness of the intensive functional rehabilitation program, we describe it only briefly here. The program includes 12 exercise sessions with a duration of approximately 90 minutes spread over a 6- to 8-week interval between the POST1 and POST2 evaluations. Each exercise session included 5 components: (1) a warm-up period consisting of general muscular and articular stretching exercises (duration, 5 10min); (2) specific lower limb strengthening exercises without weight bearing (isometric at various angles, isotonic with light resistance; 15min); (3) functional exercises with partial and total weight bearing and of varying complexity level with the purpose of improving locomotor performance (15 20min); (4) endurance exercises of walking and/or bicycling (5 20min); and (5) stretching exercises and analgesia if necessary (10 15min). Description of Outcome Measures Six-minute gait test. The 6-minute gait test measures the maximal distance covered by a subject walking at a selfselected speed during 6 minutes in a 50-meter long corridor. A walking aid was used if needed. No encouragement was given to the subject during the test. 41 Rest periods were allowed when needed. Construct and concurrent validity of the 6-minute gait test has been extensively shown in populations with cardiopulmonary problems In populations with OA, a good concurrent validity was also shown. Good correlation was found with respiratory measures, 52,53 knee extensor strength, 53 stride characteristics, 43 and timed up and go. 54 Good test-retest reliability was found even without any familiarization trial in a variety of populations. 45,48,53 Although this test has been used to measure the effectiveness of interventions for persons with OA, 43,52,55,56 its responsiveness has not been extensively studied. An improvement ranging from 30 to 60 meters or 15% to 18% was considered clinically significant in persons with pulmonary diseases. 41,48,57-59 In a study 60 including persons with TKA, the 6-minute gait test was found to be the most responsive of locomotor tests and the third most responsive of all variables tested. WOMAC difficulty subscale. The WOMAC is a diseasespecific questionnaire developed specifically for persons with OA of the hip and knee. By using visual analog scales (VASs), its 24 items probe 3 dimensions considered important by persons with OA: pain (5 items), stiffness (2 items), and functional difficulty (17 items). In the present study, 60 the valid and reliable French-Canadian version of WOMAC was used 61 and only the functional difficulty subscale was assessed. The difficulty score corresponds to the sum of each item score divided

3 72 RESPONSIVENESS OF OUTCOME MEASURES AFTER TKA, Parent by the total number of items (n 17). Scores range from 0 to 100, where 100 indicates the worst possible state. The psychometric properties of the WOMAC have been thoroughly studied. 62 Content validity was ensured by retaining only items with high relevance to persons with OA Furthermore, many studies have shown the construct validity of the questionnaire 22,26,39,65-71 as well as its good reliability. 26,39,65,67,70 In pharmacologic and surgical trials, the WOMAC showed similar responsiveness to its signal version, 24,72 the Lequesne and Doyle indices, 73 the Patient-Specific Index, 26,69 the Harris hip score, 26 the Health Assessment Questionnaire, 39 and a better responsiveness than the 50-feet walk time, range of motion measurements, and intermalleolar straddle measurements. 65,70 As expected for disease-specific questionnaires, the WOMAC was more responsive than the generic SF-36 questionnaire. 19,26,39,69,71 Furthermore, in a recent study 60 of persons who underwent surgery for TKA, the WOMAC was ranked as the most responsive outcome measure, followed by the Knee Society Clinical Rating Scale, the 6-minute gait test, the 30-second stair climb, the SF-36, and time trade-off. SF-36 physical function and role physical subscales. The SF-36 is a health-related QOL questionnaire designed for use in clinical practice, research, and general population survey. It is increasingly used in populations with OA. 14,18,19,25,26,39,71,74-79 This generic questionnaire is recognized for its simplicity, conciseness, and comprehensiveness. 80,81 It contains 36 items assessing 8 different health concepts: physical functioning (10 items); social functioning (2 items); role physical (4 items); bodily pain (2 items); mental health (5 items); (6) role emotional (3 items); (7) vitality (4 items); and (8) general health (6 items). Scores, as computed by the authors, 82,83 ranged from 0 to 100, with 0 indicating the least favorable health state. The SF-36 s content validity has been recognized. 80 In persons with OA, construct validity was shown in relation to many disease-specific corresponding subscales. 26,39,69,71,77,84,85 It is internationally recognized that the SF-36 questionnaire produces data of good quality and that consistency and itemdiscriminant validity are appropriate for a variety of diagnoses and for various levels of disease severity. 68,78,82,84,86-91 The discriminative capacity of SF-36 has also been shown along with good to excellent test-retest reliability. 39,68,84,86,87,92,93 With respect to responsiveness, some concerns were raised for role subscales that were prone to floor and ceiling effect, 39 and for physical functioning, social functioning, and bodily pain, for which a modest ceiling effect was also observed. 68,78,82,90,91 However, SF-36 showed the best responsiveness when compared with other generic measures of health status. 39,93 Bodily pain and physical functioning subscales were the best subscales for discriminating between changed and unchanged and treated and untreated persons with arthritis in pharmacologic trials. 74,76,77 On the other hand, responsiveness of the SF-36 was generally inferior to that of specific questionnaires, especially for dimensions related to mental components of health. 14,19,26,39,68,69,71,77,84 In the present study, only the responsiveness of physical functioning and role physical subscales of the French-Canadian translation 94 of the SF-36 was studied. Gait speed. Kinematic, kinetic, and spatiotemporal variables were measured with a 2-dimensional biomechanical gait analysis system. Among these variables, only gait speed was retained for analysis. Gait speed was calculated by dividing the horizontal displacement of a reflective marker placed over the greater trochanter of the hip by the duration of a gait cycle determined by footswitches. Subjects were asked to walk at their natural cadence along a 10-meter walkway. The mean gait speed of 3 gait cycles was used for subsequent analyses. Stair ascent cycle duration. Stair ascent was also assessed by means of a 2-dimensional biomechanical analysis system. Subjects were asked to climb a flight of 4 steps at their natural cadence using a reciprocal lower limb pattern. Stair ascent cycle duration was the variable studied; it was defined as the time measured between the first contact of the evaluated foot on the first step and the subsequent contact of the same foot on the third step. Duration was determined by means of footswitches taped under the heel, midfoot, and toe. No ambulatory aid was used during stair ascent and the use of a banister was avoided whenever possible. Psychometric properties of the biomechanical evaluations. Our biomechanical analysis system 95 has been described previously. For the determination of cycle duration and speed, intraclass correlation coefficients (ICCs) between.88 and.92 were obtained for test-retest reliability and between.97 and.98 for intertester reliability. 96 Other investigators also found very good reliability for intra- (ICC range,.90.98) and intersession measurements (ICC range,.81.94) of speed and cycle duration. 97 Although results from locomotion analysis are frequently used to follow the progress of subjects undergoing joint replacement, responsiveness of gait speed and stair ascent duration has not been formally evaluated. Data Analysis Descriptive statistics. For categoric variables, we calculated proportions and frequency counts, whereas for continuous variables, we computed means and standard deviations (SDs). We performed comparisons between performances at the time of the 3 evaluations by using an analysis of variance for repeated measures. Tukey s post hoc tests were used for pairwise comparisons, maintaining alpha at.05. Statistics for the assessment of responsiveness. For each outcome measure and the 3 time intervals, we calculated 4 responsiveness indices: effect size, 8 standardized response mean 7,31 (SRM), paired t statistic, 17,30 and relative efficiency. 5 Time intervals were defined as follows: PRE to POST1 first time interval; POST1 to POST2 second time interval; and PRE to POST2 third time interval. Effect size. Effect size was first introduced to estimate the magnitude of recorded change relative to the measure s variability. It enabled researchers to judge the clinical importance of statistically significant changes. With this method, 8 the mean change during a given interval is divided by the SD of the initial score. The value obtained can be used to compare different measures. It can also be compared with Cohen s benchmarks: effect size of.20 small,.50 moderate, and.80 large. 98 Standardized response mean. The SRM is another kind of effect size proposed in an effort to find optimal techniques to evaluate outcomes. 17 As such, it can also be interpreted by using Cohen s rule of thumb, presented earlier. To calculate the SRM, the mean change is divided by the SD of the change scores. It is also possible to estimate a SD and a confidence interval (CI) for the SRM. 17 Thus, differences between the responsiveness of the measures under study can be identified. For each time interval studied, comparison of the SRM value of the different outcome measures was performed by using 90% CIs. Paired t statistic. The paired t statistic, 30 which was derived by computing a paired t statistic based on within-patient score changes for each scale, is a simple method for evaluating responsiveness. This paired t test is calculated by dividing the mean change by the SD of change; the latter is divided by the

4 RESPONSIVENESS OF OUTCOME MEASURES AFTER TKA, Parent 73 Sociodemographic Characteristics Table 1: Subject Characteristics at Enrollment (N 65) Clinical Characteristics Age (yr) Operated limb Gender Right 28 (43.1) Left 37 (56.9) Men 25 (38.5) Women 40 (61.5) Walking aid None 46 (70.8) Height (cm) Cane (long distance) 10 (15.4) Men Cane (always) 9 (13.8) Women Both Comorbidity Cardiac 17 (26.2) Weight (kg) Vascular 21 (32.3) Men Pulmonary 21 (32.3) Women Diabetes 8 (12.3) Both Renal 1 (0.02) Neurologic 1 (0.02) Body mass index (kg/m 2 ) Cancer 2 (0.03) Men Women Time of symptom onset (mo) Both NOTE. Values are mean SD or n (percentage). square root of the number of subjects in the sample. The outcome measure associated with the largest absolute paired t statistic value is identified as the most sensitive and is thought to provide the best statistical power for a given sample size. Relative efficiency. This method is related to the paired t statistic method and was first introduced by Liang et al. 5 It consists of the squared ratio of the t statistics of the 2 measures being compared. For instance, if the t statistics obtained for a given time interval for the 6-minute distance and gait speed were 2.00 and 4.00, respectively, then the relative efficiency of gait speed would be 4 [(4/2) 2 4] and that of the 6-minute distance would be 1 [(2/2) 2 1]. This assumes that the 6-minute distance was used as the reference variable. In the present study, the 6-minute gait test was arbitrarily chosen as the reference measure against which other measures were compared. Relative efficiency values are easy to interpret because they are always positive and a higher value indicates higher responsiveness. Furthermore, if the relative efficiency of a given test is greater than 1, the measure being compared is more responsive than the reference measure (6-minute gait test). Liang 5 also used relative efficiency as an interval measure of responsiveness stating, for instance, that a measure with a relative efficiency value of 0.5 was half as responsive as the reference measure, and so on. RESULTS Sample Description Participants were recruited between March 1996 and October During this period, 767 patient files were screened from the 5 recruiting hospitals and a total of 147 persons were identified as eligible for the project. From this number, 65 agreed to participate in the project. Thus, the acceptance rate was 44%. The follow-up of the last participant was completed in February Only 1 participant did not complete the 4-month follow-up because he withdrew from the study after the POST1 evaluation. Eight subjects were not able to do the stair ascent test before TKA. This number increased to 10 subjects 2 months after TKA and decreased to 4 subjects 4 months after TKA. Subjects sociodemographic and clinical characteristics are shown in table 1. On average, the subjects length of hospital stay was days. Mean Performances and Scores Mean test performance at each of the 3 evaluations are shown in table 2. Normative values of healthy elderly people of similar age are also presented for comparison. During the first time interval, from PRE to POST1, a significant decrease in the distance walked in 6 minutes was noted. Concomitantly, we found significant improvements in both the WOMAC difficulty and SF-36 physical function subscales. Performances and scores in the other tests did not change significantly during this interval. In the second interval, from POST1 to POST2, a significant improvement was observed in all tests and questionnaires except stair ascent duration. Finally, in the third interval studied (PRE POST2) significant improvements were found in all questionnaire scores and the 6-minute gait test. No statistically significant differences were observed in gait speed and stair ascent duration. Responsiveness Results First interval (PRE POST1). Responsiveness indices for this time interval are shown in table 3. As shown, the WOMAC difficulty subscale was the most responsive by the 3 concordant responsiveness indices (SRM, paired t statistic, relative efficiency). Only this outcome measure had a large SRM value as determined by Cohen s rule of thumb. 98 It was followed by the 6-minute gait test, which had a moderate SRM value. On the other hand, the SF-36 physical function subscale, gait speed and stair ascent duration (small SRM values), and the SF-36 role physical subscale had very low SRM values. The effect size ordered 2 outcome measures differently; however, according to Cohen s rule, only the 6-minute gait test would be qualified differently. Figure 1A presents the 90% CI around the estimated SRM values for the first time interval. The WOMAC difficulty subscale (ranked no. 1) had a similar responsiveness to that of the 6-minute gait test, but was significantly more responsive

5 74 RESPONSIVENESS OF OUTCOME MEASURES AFTER TKA, Parent Table 2: Mean Performance During Locomotor Tests and Questionnaire Scores (N 65) Variables Normative Values* PRE POST1 POST2 Six-minute distance (m) Gait speed (m/s) Stair ascent duration (s) WOMAC difficulty (/100)** SF-36 physical function (/100) SF-36 role physical (/100) * From a sample of 21 healthy elderly (mean age, yr) except for gait speed, in which a sample of 18 healthy elderly (mean age, yr) was used. All subjects were evaluated with the same procedure. Mean 1 SD. No significant difference between PRE and POST1 evaluations (P.05, Tukey). No significant difference between PRE and POST2 evaluations (P.05, Tukey). No significant difference between POST1 and POST2 evaluations (P.05, Tukey). PRE, n 57; POST1, n 55; POST2, n 61. ** 100 indicates worst state. 100 indicates best QOL. than stair ascent duration (no. 5) and SF-36 role physical (no. 6), as shown by the nonoverlapping CI of these last 2 tests. Similarly, the 6-minute gait test (no. 2) was significantly more responsive than the SF-36 role physical (no. 6). No other significant difference in responsiveness was observed during this interval. Second interval (POST1 POST2). During this interval, the 6-minute gait test was ranked as the most responsive measure followed by gait speed (table 4). Both measures had large SRM values. The WOMAC difficulty subscale, SF-36 physical function, and role physical subscales were the next most responsive with moderate SRM values. Last, the stair ascent duration had very low responsiveness. During this interval, differences in outcome measure ranking by the effect size and the 3 concordant indices occurred more frequently than during the first interval. The most striking difference was the SF-36 role physical subscale, which was ranked second by effect size and fifth by the 3 other indices. For this interval, comparison of the measures responsiveness by using CIs lead to the following findings (fig 1B): the 6-minute gait test (no. 1) was significantly more responsive than all other tests (no. 2 no. 6) and gait speed (no. 2) was more responsive than stair ascent duration (no. 6). Third time interval (PRE POST2). The WOMAC difficulty subscale was ranked as the most responsive measure by the concordant indices and it was the only measure showing a large SRM value during this interval (table 5). The next most responsive measures were the SF-36 physical function score and the 6-minute gait test with moderate SRM values. The SF-36 role physical score and gait speed followed with small SRM values. Stair ascent duration was ranked last with a very low SRM value. During this time interval, measures ranked 3 and 4 by effect size were ranked 4 and 3, respectively, by the 3 other indices. When we compared the interpretation of effect size and SRM by using Cohen s rule, 2 differences were observed: (1) effect size value of SF-36 physical function was considered large whereas the SRM value was judged moderate, and (2) effect size value of the 6-minute distance was considered small whereas the SRM value was moderate. The 90% CI around the SRM values (fig 1C) showed that the WOMAC difficulty subscale (no. 1) was significantly more responsive than all other tests (no. 2 no. 6) and the SF-36 physical function subscale (no. 2) was significantly more responsive than the SF-36 role physical subscale (no. 4), gait speed (no. 5), and stair ascent duration (no. 6). Comparison of the outcome measure s responsiveness at the 3 time intervals. A summary of the ranking for the 3 intervals by the 3 concordant indices is shown in table 6. The Table 3: Responsiveness Statistics for the First Time Interval Outcome Measures SRM* Value TT* Value RE* Value ES Value WOMAC difficulty Six-minute distance SF-36 physical function Gait speed Stair ascent duration SF-36 role physical Abbreviations: ES, effect size; SRM, standardized response mean; TT, paired t statistic; RE, relative efficiency. * SRM, TT, and RE produced the same rankings of the outcome measures. Deterioration during this interval. Third and second rankings inverted compared with SRM, TT, and RE. Fig 1. SRM values and 90% CIfor all measures and the 3 time intervals (A) before to 2 months after TKA, (B) 2 to 4 months after TKA, and (C) before to 4 months after TKA. Abbreviations: SF-36 RP SF-36 role physical subscale; Lab SA, in-laboratory stair ascent duration; Lab Speed, in-laboratory gait speed; SF-36 PF, SF-36 physical function subscale; 6-MIN, 6-minute gait test; WOMAC Dif, WOMAC difficulty subscale.

6 RESPONSIVENESS OF OUTCOME MEASURES AFTER TKA, Parent 75 Table 4: Responsiveness Statistics for the Second Time Interval Outcome Measures SRM* Value TT* Value RE* Value ES Value Six-minute distance Gait speed WOMAC difficulty SF-36 physical function SF-36 role physical Stair ascent duration * SRM, TT, RE produced the same rankings of the outcome measures. Same ranking using effect size. Ranked second by effect size. WOMAC difficulty subscale and the 6-minute gait test are among the 3 most responsive measures for the 3 intervals evaluated. Of note, SRM values calculated for the WOMAC difficulty subscale and the 6-minute gait test were either moderate or large regardless of the interval considered. With respect to the responsiveness of the questionnaires, we observed that the WOMAC difficulty subscale was always more responsive than both SF-36 subscales. DISCUSSION The present study is the first to show the relative responsiveness of locomotor tests and questionnaires to monitor early recovery after TKA. Two main results emerged. First, responsiveness levels and ranking of the outcome measures varied according to the time intervals studied. As a result, a single most responsive measure for all intervals did not exist. A measure s responsiveness differed between intervals, indicating that its appropriateness for evaluating change varied depending on when it was used in the recovery continuum. Second, though responsiveness differed across time intervals, trends could be observed. The difficulty subscale of the WOMAC questionnaire and the distance covered during the 6-minute gait test were consistently ranked among the 3 most responsive measures regardless of the time interval considered. This finding suggests that these 2 outcomes are the most appropriate for monitoring early recovery after TKA, as well as for evaluating the outcome of surgery and rehabilitation programs. These measures had an adequate responsiveness during the entire period studied (before TKA to 4mo after). Conversely, it is clear that outcome measures such as stair ascent duration and SF-36 role physical subscale are not responsive and should not be used as main outcome measures during the early recovery period. The most responsive measures. The WOMAC difficulty score and 6-minute distance were the most responsive questionnaire and locomotor variable for the studied intervals. These 2 measures were ranked one after the other as the most responsive of the 6 measures under study. Although their SRM values varied, they were always considered moderate or large. Furthermore, the WOMAC difficulty score and 6-minute distance were the only variables to have a significantly better responsiveness than at least 1 other measure. These results agree with those of Kreibich et al 60 who studied subjects with TKA during 2 time intervals: 0 to 3 months and 0 to 6 months after TKA. They also found the WOMAC and the 6-minute gait test to be the most responsive functional capacity questionnaire and locomotor test, respectively. Because their sample size (N 71) was similar to ours, their larger paired t values are probably attributable to the longer time intervals studied. Many reasons can explain why the WOMAC difficulty subscale had such a good level of responsiveness in the present study and why it was also among the most responsive measures in pharmacologic and surgical studies of subjects with OA. 19,39,65,69-73 First, it was previously found 60,99,100 that questionnaires are better outcome measures for detecting changes in early recovery stages than locomotor tests, especially in subjects with severe disabilities. Compared with a locomotor test, a questionnaire such as the WOMAC probes a larger array of functional activities with different biomechanical and motor requirements. Thus, one can suppose that easier activities are recovered more rapidly in the early recovery stage after TKA than others, such as locomotion. Also, evaluations of difficulty level in performing activities may be influenced by the experienced level of pain. After TKA, an important pain relief occurred, especially in the first 2 months. This experience may have favored better responsiveness of the WOMAC in the first interval. For instance, in the present study, we observed that during the first time interval, the WOMAC was more responsive than the 6-minute test whereas the situation was reversed during the second time interval. During the first time interval, changes 3 times as large were observed in the WOMAC (41% improvement), compared with the 6-minute walking test (12% deterioration). Among the activities that have the largest change, difficulty getting out of bed and rising from a chair can be mentioned. The opposite result was observed during the second time interval. The 6-minute test was the most responsive measure, just before the WOMAC, even though both measures had a similar magnitude of change (25%: 6-minute; 31%: WOMAC). Our results show that locomotor capacity is recovered more slowly than other daily life activities. The results of Richards et al 99,100 in persons with stroke also support such findings. They observed that for subjects with stroke who walked very slowly, questionnaires were better at discriminating different functional levels, and that later, during follow-up, when gait speed had increased, locomotor tests were more responsive. Second, the WOMAC was more responsive than SF-36 subscales. This finding was expected because disease-specific questionnaires are more responsive than generic ones. A disease-specific questionnaire such as the WOMAC assesses difficulty in completing activities that are judged to be important by subjects with OA in the lower limbs. Because TKA is a surgical intervention intended to provide relief from OA complaints from the knee, it is likely that the WOMAC score would show larger change than a generic questionnaire score. Even though the physical functioning and role physical subscale of the SF-36 contain questions about physical health, they are not all related to lower limb activities. 4,12,26,71,101 Third, the structure of the WOMAC allowed detection of more subtle changes compared with the SF-36. The WOMAC Table 5: Responsiveness Statistics for the Third Time Interval Outcome Measures SRM* Value TT* Value RE* Value ES Value WOMAC difficulty SF-36 physical function Six-minute distance SF-36 role physical Gait speed Stair ascent duration * SRM, TT, RE produced the same rankings of the outcome measures. Rankings inverted by effect size.

7 76 RESPONSIVENESS OF OUTCOME MEASURES AFTER TKA, Parent Table 6: Summary of the Ranking for the 3 Time Intervals by the 3 Concordant Statistics Rank First Interval (PRE POST1) Second Interval (POST1 POST2) Third Interval (PRE POST2) 1 WOMAC difficulty* 6-minute distance* WOMAC difficulty* 2 6-minute distance Gait speed* SF-36 physical function 3SF-36 physical function WOMAC difficulty 6-minute distance 4 Gait speed SF-36 physical function SF-36 role physical 5 Stair ascent duration SF-36 role physical Gait speed 6 SF-36 role physical Stair ascent duration Stair ascent duration * Large SRM based on Cohen s rule. Moderate SRM. difficulty subscale contains questions with infinite response options (VAS), whereas both SF-36 subscales contain Likert scales. Theoretically, the continuous scales of the WOMAC should be more sensitive to small changes. 4,12,71,101 This hypothesis is supported by the fact that the Likert scaling version of the WOMAC is slightly less responsive than the VAS version. 65 The WOMAC difficulty subscale contains a larger number of questions (17 items) than the SF-36 physical function (10 items) and role physical (4 items) subscales. Consequently, the response continuum of the disease-specific WOMAC is even larger and can again, theoretically, detect more subtle changes in functional capacity. 4,101 Among the locomotor tests, the better responsiveness of the 6-minute gait test also warrants comments. First, compared with in-laboratory gait speed, the 6-minute gait test measured locomotor performance over a longer distance and duration. As a result, the 6-minute gait test can detect larger changes and have better responsiveness. This potential is corroborated by our present results. During the 3 time intervals studied, the SRM value of the 6-minute gait test was nearly twice that of in-laboratory gait speed. Furthermore, the 6-minute gait test can capture both speed and endurance ability changes whereas gait tests over short distances assess only speed ability. This ability was particularly relevant during the second time interval during which the responsiveness of the 6-minute and gait speed tests differed significantly and the magnitude of change in the 6-minute test (25.4%) was larger than that of in-laboratory gait speed (16.4%). This finding has an implication for the sample size needed to show treatment effectiveness. Because of the large variability in gait speed, a larger number of subjects would be needed to detect the same percentage of difference in gait speed compared with the 6-minute distance. For example, because the variability in walking distance is about 15%, a sample size of 16 subjects per group would allow detection of a 15% change in walking distance, at an 80% power level and.05 alpha level. In contrast, in the same condition twice the number of subjects (n 31 per group) would be needed for the gait speed test knowing that the variability is larger (SD, 21%). The least responsive measures. The poor responsiveness of both SF-36 subscales was discussed earlier. A smaller change was expected compared with the WOMAC and locomotor tests. Also, change in health-related QOL may be expected over a longer period than the ones studied. On the other hand, the poor responsiveness of stair ascent duration was unexpected. The higher biomechanical requirement of the stair ascent test may explain such results. 102 Studies describing the demand of different functional tasks have shown that stair ascent requires about 65% and gait about 25% of maximal activation level of the knee extensors. 103,104 Moreover, the muscle moment in the lower extremity extensor muscles (support moment) required during walking is about 33% of that required to ascend stairs In fact, in the present study, many subjects were unable to perform the stair ascent test (12.3% at pre-test, 15.4% at POST1, 6.1% at POST2). The remainder maintained a poor performance and, consequently, few changes occurred in the intervals studied. Thus, by definition, the stair ascent is affected by a floor effect. 77 Indeed, the stair ascent test cannot detect changes in subjects who have a poor performance or who cannot complete the test. The percentage of change observed during each interval was smaller than 1%. Therefore, this test was determined to be too difficult to follow the recovery continuum from the acute to the chronic stages. Great difficulty with stair ascent was reported for persons even 1 year after TKA. 108 At this point, a deficit of 51% in climbing time was measured whereas walking speed was reduced by only 17% compared with normal values. However, in other studies stair ascent was found to be a responsive test later in the post-tka recovery process and in subjects with near-normal performance. 60,104, The poor responsiveness of the variable chosen to represent stair ascent ability suggests that other variables may be more appropriate to follow the early locomotor recovery. The inlaboratory analysis of stair ascent generates a large number of spatiotemporal and biomechanical variables. In the present study, cycle duration was selected because it was assumed to be a good indicator of overall performance, as gait speed is for walking. However, cycle duration is quite short and might not be the most responsive functional indicator derived from the stair ascent evaluation. The mean change observed was very small ( 1% of initial measures) compared with the intersubjects variability of this change that was 7 times larger. It is possible that other variables such as the time required to ascend a longer flight of steps, or energy expenditure might be more responsive. Of note, a subgroup of subjects of the present sample were not able to climb a flight of 4 steps. This limitation will always be in disfavor of a good responsiveness of such activity in the early stage of recovery. Finally, in the present study, impairment variables such as kinematics and kinetics were not considered because these variables do not necessarily reflect the overall functional ability. Comparisons of the 4 responsiveness indices. The SRM appears to be the most useful and interpretable statistic in the present study. It can easily be interpreted by using Cohen s rule. 98 Furthermore, it is the only responsiveness statistic for which a CI can be calculated. 17 Thus, statistically significant differences in responsiveness between outcome measures can be identified. As Stratford et al 6 pointed out, SRM is not dependent on sample size; this is an advantage relative to the use of paired t statistic and relative efficiency if one plans to compare the results of different studies. Concerning consistency in ranking, similar findings to ours were observed in studies that used SRM, paired t statistic, and relative efficiency. In these studies, 5,6,17,26,37,68,112 all statistics generated the same rankings when used for samples of equal

8 RESPONSIVENESS OF OUTCOME MEASURES AFTER TKA, Parent 77 size. This finding is not surprising because a mathematic relationship exists between these 3 statistics. Calculation of effect size differs most from the calculation of the other statistics. The SD, which is used as the denominator for effect size, is based on initial values instead of change. Results among studies vary regarding ranking generated by effect size and the 3 other statistics: SRM, paired t statistic, or relative efficiency. In some studies, all 4 statistics ranked the tests similarly, 26,27,37 though in another study 112 different rankings were reported. Theoretically, both situations could occur. The relative responsiveness results found in the present study should apply to any representative sample evaluated with the same methodology over similar time intervals. Larger representative samples will result in larger absolute estimates of responsiveness and narrower CIs but not in different rankings. With huge samples, measures that have shown the best responsiveness in smaller samples would still perform the best. As a result, it appears important to estimate absolute responsiveness by using sample sizes that are practical for clinical trials. In the present study, the same practical sample size was used for all measures. At all evaluations, some subjects were not able to complete the stair ascent test. Thus, it is possible that the absolute responsiveness of stair ascent duration was slightly underestimated because of this sample size difference. Responsiveness could also have been estimated by using the SRM statistic and its CI in combination with a method based on an external criterion for change. 26,39,92,113 Studies that use both methods provide readers with 2 types of information: the ability of an outcome measure to identify statistically significant changes (SRM) and the ability to detect clinically important changes (external criteria). This last analysis was not performed in the present study because the question used as an external criterion, with only 3 response options (deteriorated, stable, improved), lacked sensitivity to clinically important change. 6 As a result, nearly all subjects rated themselves as improved regardless of the interval studied. Thus, this method was useless in discriminating which measures were the most and the least responsive to different levels of improvement. Limitations. Results of the present study may not generalize to every setting. For instance, we included only persons with OA who underwent surgery for a first TKA. Subjects with multiarticular involvement may not exhibit as much change and, as a result, lower levels of responsiveness may be found. This finding would not necessarily imply a different ranking of the outcome measures. However, an effort was made to improve generalizability by including many surgeons and hospitals. Moreover, the sample recruited was similar in age, height, and weight to other OA samples presented in the literature. 60 It would be interesting to see if our present results can be replicated in another city, socioeconomic context, or with persons who underwent surgery for other diagnoses than idiopathic OA. Because a measure s responsiveness varied between intervals in the present study, it would be necessary to evaluate the responsiveness of these outcome measures during other intervals of interest. For instance, physical therapists working in the acute postoperative phase would need information regarding the tests responsiveness during the first 10 days postsurgery. The good responsiveness observed for the WOMAC questionnaire and the 6-minute gait test when evaluating groups is probably an indication that these outcome measures can be appropriate for follow-up. A need exists to evaluate responsiveness to individual change of these outcome measures to provide clinicians with definitive recommendations for the selection of measurement tools. 114 CONCLUSION In the present study, the WOMAC difficulty score and the 6-minute gait distance were the most responsive outcome measures for evaluating early (first 4mo) recovery after TKA. Because these outcome measures are simple and were proven valid, reliable, and responsive, we recommend their use in monitoring persons with OA who have had a TKA, at least up to the fourth month after surgery. This study also showed that the responsiveness of the 6-minute gait distance was superior to that of in-laboratory gait speed and stair ascent duration. This latter measure did not show a sufficient responsiveness to be used as a main outcome measure during the first 4 months after TKA. Even if SF-36 is often less responsive than functional capacity measures, health-related QOL is considered an important outcome to measure; it permits health professionals to detect what effect comorbidities may have on various QOL dimensions and to compare results across diagnoses. 25,27,38,39,66,69,84, Furthermore, the SF-36 has the potential to measure side effects or treatment complications that may be unrelated to specific health conditions. 39 The SRM statistic was the most interesting responsiveness statistic because interpretation guidelines are available and CIs can be calculated. Future studies are needed to determine the responsiveness of the same outcome measures later in the recovery stage after TKA and to verify if our results apply to individual change. Acknowledgments: The authors thank the orthopedic surgeons for referring subjects to the project. Recruitment was performed in the following hospitals from the Centre Hospitalier Universitaire de Québec: Hôtel-Dieu de Québec, Saint-François D Assise and Centre Hospitalier de l Université Laval; and from the Centre Hospitalier Affilié Universitaire de Québec: Enfant-Jésus and St-Sacrement. References 1. Ottenbacher KJ, Johnson MB, Hojem M. The significance of clinical change and clinical change of significance: issues and methods. Am J Occup Ther 1988;42: van der Putten JJ, Hobart JC, Freeman JA, Thompson AJ. Measuring change in disability after inpatient rehabilitation: comparison of the responsiveness of the Barthel index and the Functional Independence Measure. J Neurol Neurosurg Psychiatry 1999;66: Deyo RA. Measuring functional outcomes in therapeutic trials for chronic disease. Control Clin Trials 1984;5: Liang MH. Evaluating measurement responsiveness. J Rheumatol 1995;22: Liang MH, Larson MG, Cullen KE, Schwartz JA. Comparative measurement efficiency and sensitivity of five health status instruments for arthritis research. Arthritis Rheum 1985;28: Stratford PW, Binkley FM, Riddle DL. Health status measures: strategies and analytic methods for assessing change scores. Phys Ther 1996;76: Meenan RF, Anderson JJ, Kazis LE, et al. Outcome assessment in clinical trials. Evidence for the sensitivity of a health status measure. Arthritis Rheum 1984;27: Kazis LE, Anderson JJ, Meenan RF. Effect sizes for interpreting changes in health status. Med Care 1989;27 Suppl 3: Deyo RA, Inui TS. Toward clinical applications of health status measures: sensitivity of scales to clinically important changes. Health Serv Res 1984;19: Tuley MR, Mulrow CD, McMahan CA. Estimating and testing an index of responsiveness and the relationship of the index to power. J Clin Epidemiol 1991;44: Deyo RA, Centor RM. Assessing the responsiveness of functional scales to clinical change: an analogy to diagnostic test performance. J Chronic Dis 1986;39: Krishner B, Guyatt G. A methodological framework for assessing health indices. J Chronic Dis 1985;38:27-36.