The Validity and Utility of the BPI Interference Measures for Evaluating the Impact of Osteoarthritic Pain

Transcription

1 48 Journal of Pain and Symptom Management Vol. 31 No. 1 January 2006 Original Article The Validity and Utility of the BPI Interference Measures for Evaluating the Impact of Osteoarthritic Pain Valerie S. L. Williams, PhD, Meredith Y. Smith, PhD, and Sheri E. Fehnel, PhD RTI Health Solutions (V.S.L.W., S.E.F.), Research Triangle Park, North Carolina; and Purdue Pharma LP (M.Y.S.), Stamford, Connecticut, USA Abstract The psychometric properties of the Brief Pain Inventory (BPI), a widely used measure of pain and its impact on functioning, were assessed using data from two clinical trials of controlledrelease oxycodone in osteoarthritis (OA) patients. Specifically, the pain-related functional interference subscale and the sleep item from that subscale were examined. In Study 1 (n ¼ 133), night awakenings with pain was positively correlated with the BPI interference score and sleep item and both correlated negatively with quality of sleep. In Study 2 (n ¼ 107), pain experienced at night while in bed correlated higher with sleep interference than with the BPI interference subscale. Intraclass correlations denoted adequate test-retest reliability; moderate-to-large Guyatt s statistics provided evidence of responsiveness. These analyses address a gap in the literature regarding the psychometric properties of the BPI interference measures in noncancer pain patients, confirming their reliability, validity, and responsiveness as potential endpoints in trials of pain medications involving patients with OA. J Pain Symptom Manage 2006;31: Ó 2006 U.S. Cancer Pain Relief Committee. Published by Elsevier Inc. All rights reserved. Key Words Pain measurement, arthritis, osteoarthritis, Brief Pain Inventory (BPI), reliability, validity, responsiveness This research was supported under a contract with Purdue Pharma LP. Portions of this research were presented at the 23rd Annual Scientific Meeting of the American Pain Society in Vancouver, BC, Canada (May 6--9, 2004) and at the Annual Meeting of the International Society for Quality of Life Research in Prague, Czech Republic (November , 2003). Address reprint requests to: Valerie Williams, PhD, RTI Health Solutions, 3040 Cornwallis Road, P.O. Box 12194, Research Triangle Park, NC , USA. vwilliams@rti.org. Accepted for publication: June 7, Ó 2006 U.S. Cancer Pain Relief Committee Published by Elsevier Inc. All rights reserved. The Brief Pain Inventory (BPI) 1 is an established instrument for assessing the severity of cancer pain and its effect on patients functioning. The use and psychometric properties of the BPI for the measurement of cancer pain are documented extensively in the literature. 2,3 The primary impetus for the present investigation of the BPI interference scale and, specifically, the BPI sleep interference item, is the belief that sleep is an important outcome to assess in pain trials involving cancer or noncancer patients. To the extent that sleep is disturbed, there are pervasive negative effects on psychological and physical functioning /06/$--see front matter doi: /j.jpainsymman

2 Vol. 31 No. 1 January 2006 Validity and Utility of BPI Interference Measures 49 The BPI assesses two pain-related dimensionsdintensity and interference (or impact)d which have been found to be correlated but not redundant 3,7 and have led to a better understanding of the multidimensional nature of pain and its relation to patient functioning and health-related quality of life. It asks the patient about pain relief, pain quality, and perceived cause of pain. The BPI has been translated into many languages and widely used in different cultures and contexts. 7,8 Other notable advantages of the BPI are that it is generic rather than condition-specific, simple to administer and score, and is based on patient report. In particular, the BPI interference subscale is short (seven items) and can be a valuable tool in clinical trials where respondent burden is a significant consideration. Furthermore, the BPI questions and response options are easily understood by patients. 8 There is a growing consensus that chronic pain clinical trials should assess outcomes representing multiple domains beyond pain intensity itself. 9 Sleep is one such outcome to consider because it is affected by pain and impacts both psychological well-being and functioning in daily life. 4--6,10 However, there is a lack of brief, psychometrically validated measures designed to assess the impact of pain on sleep in patients with noncancer pain. Although originally developed for the study of cancer and cancer pain, the BPI has been used increasingly in studies of other disease states and analgesia. 3 The BPI has been used to assess pain and the impact of pain treatment in those with AIDS; the results related to the BPI interference subscale agreed with results related to diverse pain outcomes, such as physical symptom distress and performance status, pain management and relief ratings, and satisfaction with analgesic therapy. Galer et al. 16 used the BPI in their investigation of complex regional pain syndrome (CRPS) and found that results related to patient quality of life as evaluated by the BPI interference subscale agreed substantially with results related to other measures of neuropathic pain and CRPS symptoms. A study using the BPI for the assessment of pain-related quality of life associated with the progression of Fabry disease 17 showed improvement in BPI interference scores associated with improvement in renal, cardiac, and metabolic functioning. A large study by Tan et al. 18 involving mostly male military veterans with chronic nonmalignant pain showed the BPI interference scale to be significantly correlated with self-reported disability and responsive to pain treatment over time. The BPI has also been used as an outcome measure in studies of osteoarthritis (OA), specifically, low back pain OA 19,20 and temporomandibular joint OA. 21 In Thie et al., 21 the BPI interference subscale agreed with other measures of orofacial pain such as pain-free mouth opening and extra-oral masticatory muscle pain. This general finding provides limited support for the validity of the BPI with arthritis patients. Keller et al. 19 used the BPI in a convenience sample of patients with noncancer pain. They reported internal consistency reliabilities of 0.95 and 0.94 for the sevenitem interference scale in samples of arthritis and low back pain patients, respectively, while Tan et al. 18 reported a reliability of These reliabilities are similar to those reported in Jensen s 2 reviewdinternal consistencies ranged from 0.78 to 0.97 in cancer patients from various cultures. A limitation of both the Tan et al. 18 and the Keller et al. 19,20 investigations is that neither provides information about the test-retest reliability of the BPI. Similarly, the Thie et al. 21 study omits information about the stability of the BPI, as well as the instrument s validity and responsiveness. The foregoing research has set the stage for expanded claims of validity and wider use of the BPI interference subscale as a secondary endpoint in clinical trials of patients with persistent, moderate to severe pain of a nonmalignant nature. None of these studies, however, have sought to examine the sleep interference item specifically. As a result, there is a need for a brief, psychometrically robust measure of sleep quality. The purpose of this study was to assess three central psychometric characteristics of both the BPI interference scale and the sleep interference item in osteoarthritic patients participating in randomized clinical trialsdreliability (internal consistency and test-retest reliabilities), validity (construct and discriminant validities), and responsiveness (i.e., sensitivity to differences within and between the treatment and placebo groups).

3 50 Williams et al. Vol. 31 No. 1 January 2006 Methods Sample The present report summarizes the results of analyses of data available from two doubleblind, randomized, placebo-controlled multicenter clinical trials of controlled-release oxycodone in OA, in which the BPI was administered in its entirety. Eligible patients, 18 years of age or older, with a diagnosis of OA and a history of moderate to severe pain in the most affected joint or region for at least 1 month, were recruited by physicians at a number of different U.S. medical centers. The study protocols and informed consent forms were approved by an institutional review board at each of the study centers, and written informed consent was obtained from each patient before enrollment into either study. Where possible, analyses were conducted using data from both studies; however, the two databases were not combined, but analyzed separately. Study 1. Sixty-six patients 22 (intent-to-treat n ¼ 132) completed this 2 week study, with data collected at three time points (Days 0, 7, and 14). There were three study arms: 10 mg every 12 hours controlled-release (CR) oxycodone, 20 mg every 12 hours CR oxycodone, and placebo. Study 2. Forty patients 23 (intent-to-treat n ¼ 107) completed this 3 month study, with data collections at six time points (Days 0, 15, 30, 45, 60, and 90). Study 2 had two arms, 10 mg titrated every 12 hours CR oxycodone and placebo. Measures BPI. The BPI 1 is an established instrument for assessing the severity of cancer pain and its effect on patients functioning. The full 20-item BPI is easily administered and takes about minutes to complete, with the seven-item interference subscale requiring fewer than 5 minutes. The seven interference items assess general activity, mood, walking ability, normal work, relations with people, sleep, and enjoyment of life. All items use an 11-point response scale ranging from 0 ¼ does not interfere to 10 ¼ completely interferes. In Study 1, the BPI was administered at Days 0 (baseline), 7, and 14 (end of study), and in Study 2, the BPI was administered at baseline or Days 0 (baseline), 15, 30, 45, 60, and 90 (end of study). Activities and Lifestyle Questionnaire. The Activities and Lifestyle Questionnaire (ALQ), 24 administered in Study 1 at Days 0, 7, and 14, asks respondents to rate the difficulty of executing eight activities of daily living at this moment (e.g., dress yourself or get in and out of a car ) on a 4-point scale (1 ¼ without any difficulty to 4 ¼ unable to do ). The items of the ALQ were averaged to create a summary score. Number of Night Awakenings with Pain and Quality of Sleep. In Study 1, as part of the clinical visit evaluations at Days 0, 7, and 14, patients were asked to report the number of times they awakened during the previous night because of pain. Patients were also asked to rate their sleep quality for the previous night on a 5-point scale (1 ¼ very poor, 2 ¼ poor, 3 ¼ fair, 4 ¼ good, and 5 ¼ excellent ). Western Ontario and McMaster Universities Osteoarthritis Index. The Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC), 25 one of the most commonly used measures of physical functioning in rheumatoid and OA, was administered in Study 2. This measure contains 24 items asking respondents to rate their stiffness, pain, or difficulty experienced in the past 48 hours while performing a number of activities. Each WOMAC item was presented as a visual analog scale, requiring patients to place an X on a horizontal line, anchored at the left end by no pain and at the right by extreme pain. The WOMAC provides three subscale scores (a 5- item pain score, a 2-item stiffness score, and a 17-item physical functioning score) as well as a total score. One WOMAC item is particularly relevant to sleep interference due to pain: referring to the last 24 hours, How much pain do you have at night while in bed? Analytic Methods Cronbach s Alpha. The internal consistency of the BPI pain-related interference scale was

4 Vol. 31 No. 1 January 2006 Validity and Utility of BPI Interference Measures 51 evaluated by calculating Cronbach s alphas. 26 For research purposes, a coefficient alpha of 0.70 or greater is typically considered acceptable. 27 Intraclass Correlation Coefficient. Test-retest reliability was assessed by computing Intraclass Correlation Coefficients (ICC) for both measures of interest using data from subjects who were presumed to be in stable pain control: those patients who were in the highest-dose oxycodone treatment group at the end of Study 1. Specifically, the last two time points were used in Study 1: Day 7 to Day 14 (end of study) BPI data from the 20 mg every 12 hours CR oxycodone treatment group were analyzed (n ¼ 43). It is generally recommended for research purposes that intraclass correlations for multi-item scales be at least Guyatt s Responsiveness Statistic. To establish the ability of both BPI measures to detect change, a conservative Guyatt s responsiveness statistic 29,30 was calculated for both the BPI interference subscale and the single-item painrelated sleep interference score. The statistic is computed as the difference in average change scores between the placebo and treatment group divided by the standard deviation of change scores in the placebo group; the resulting value is an estimate of the effect of the treatment on pain-related interference. 30 Cohen 31 provides a general rule-of-thumb for the interpretation of such effect size estimates: effect sizes of about 0.20 represent small effects, those of about 0.50 represent moderate effects, and those around 0.80 or more represent large effects. Correlations and t-tests To evaluate the construct validity of the two BPI interference measures, we computed Pearson correlations to estimate the degree of association between the BPI measures and other variables of interest. Specifically, in Study 1 the BPI interference scale will correlate highly with the ALQ 24 and the BPI sleep item will correlate highly with patients quality of sleep and number of night awakenings with pain. In Study 2, the BPI interference scale will correlate highly with the subscales of the WOMAC, 25 a well-validated measure of pain intensity and physical functioning in rheumatoid arthritis and OA. Furthermore, the BPI sleep item will correlate highly with the WOMAC item pain at night while in bed. A set of known group analyses was conducted to provide evidence in support of the discriminant validity of the two BPI measures. Specifically, in both studies independent group t-tests were used to test subgroup differences at end of study in both pain-related interference measures between oxycodone treatment and placebo subgroups. Another set of t-tests assessed whether BPI interference subscale scores and sleep interference item scores would be greater in patients experiencing high pain in the past 24 hours compared to patients experiencing low or moderate pain. Patients were classified as high pain or low/moderate pain on the basis of the BPI worst pain question ( please rate your pain by circling one number that best describes your pain at its worst in the past 24 hours on a numeric scale ranging from 0 ¼ no pain to 10 ¼ pain as bad as you can imagine ), which was administered in both studies. In both studies, the pain classifications were based on cutoff scores recommended in the literature: the low/moderate pain group was defined as those patients who selected 0 to 7 on the 11-point pain scale, and the high pain group included those who selected 8 to 10. Results Descriptive Statistics In Study 1, 132 patients were included in the intent-to-treat population at baseline and 66 completed the 2 week study, while in Study 2, 107 patients were included in the intent-totreat population and 40 completed the full 90 day study. Table 1 summarizes the demographic characteristics of each sample. There were no BPI items with substantial missing data, nor were there floor or ceiling effects of concern. In Study 1, the means of the BPI interference items at baseline are ordered as follows: work (M ¼ 6.74, SD ¼ 2.5), walking (M ¼ 6.54, SD ¼ 2.7), enjoyment of life (M ¼ 6.20, SD ¼ 2.9), general activity (M ¼ 5.99, SD ¼ 2.6), sleep (M ¼ 5.43, SD ¼ 3.0), mood (M ¼ 5.29, SD ¼ 2.8), and relations with others (M ¼ 4.10, SD ¼ 2.9). The ordering was similar in Study 2: work

5 52 Williams et al. Vol. 31 No. 1 January 2006 Table 1 Descriptive Statistics (n and %) of Study 1 and Study 2 Participants Study 1 Study 2 Baseline (n) Randomized 44, 33.3% 51, 47.7% to placebo Females 98, 73.7% 78, 72.9% White 116, 87.2% 100, 93.5% Average age 62.4 (SD ¼ 11.8) 63.0 (SD ¼ 11.5) Less than 65 years of age 76, 57.1% 58, 54.2% Completed (n) Randomized 18, 27.3% 14, 35.0% to placebo Females 48, 72.7% 25, 62.5% White 59, 89.4% 39, 97.5% Average age 63.1 (SD ¼ 9.8) 62.4 (SD ¼ 12.0) Less than 65 years of age 35, 53.0% 23, 57.5% logical that pain-related sleep interference might be relatively unrelated to interference with walking. Also displayed in Table 2 are the baseline correlations between the scores of the seven BPI interference items and the total score for that scale. These correlations were, as anticipated, strong and positive, and the patterns for both Study 1 and Study 2 indicate that the sleep item obtained the lowest correlation with the BPI interference score (0.62 and 0.60, respectively), compared to the other six BPI items. These correlations are large enough in magnitude to assure an important connection between the two underlying constructsdthe impact of pain on a patient s ability to function normally and the extent to which pain interferes with a patient s sleep; however, they are not so large as to suggest redundancy. (M ¼ 6.71, SD ¼ 2.2), general activity (M ¼ 6.60, SD ¼ 2.2), walking (M ¼ 6.58, SD ¼ 2.4), enjoyment of life (M ¼ 6.34, SD ¼ 2.6), sleep (M ¼ 6.02, SD ¼ 3.0), mood (M ¼ 5.81, SD ¼ 2.5), and relations with others (M ¼ 4.02, SD ¼ 2.7). The ordering of these item means suggests that in both studies, OA pain interfered substantially with work, while relations with others were reportedly less hindered. To describe the relationships among items, interitem correlations were computed among the seven items of the BPI interference scale and are displayed in Table 2. As would be expected, all correlations were positive and statistically significant except between walking and sleep (Study 1 and Study 2) and between general activity and sleep (Study 2). It seems Reliability The reliability of the two BPI measures was assessed by examining both internal consistency and test-retest reliability, two primary aspects of reliability. The internal consistency of the BPI pain-related interference subscale was evaluated by calculating Cronbach s alphas. 26 Using baseline data from Study 1, alpha was computed as 0.89, and for Study 2, alpha was computed as Test-retest reliability was assessed by computing ICCs using data from patients on the highest-dose 20 mg every 12 hours CR oxycodone treatment group (n ¼ 43) at the end of Study 1, based on the assumption that they were in stable pain control. The 1 week (Day 7 to Day 14) intraclass correlations were 0.65 (95% confidence interval: ) for the BPI sleep Table 2 Correlations Among BPI Interference Measures at Baseline (Study 1 below the Main Diagonal and Study 2 above the Main Diagonal) BPI Measure General activity d 0.43 a 0.46 a 0.58 a 0.43 a a 0.68 a 2 Mood 0.60 a d 0.29 b 0.42 a 0.45 a 0.43 a 0.53 a 0.72 a 3 Walking ability 0.64 a 0.43 a d 0.60 a 0.36 a a 0.67 a 4 Normal work 0.78 a 0.63 a 0.65 a d 0.41 a 0.30 a 0.51 a 0.76 a 5 Relations 0.57 a 0.78 a 0.46 a 0.59 a d 0.32 a 0.47 a 0.71 a with people 6 Sleep 0.33 a 0.53 a a 0.50 a d 0.39 a 0.60 a 7 Enjoyment of life 0.61 a 0.62 a 0.49 a 0.62 a 0.61 a 0.43 a d 0.78 a 8 BPI interference scale 0.82 a 0.84 a 0.69 a 0.84 a 0.83 a 0.62 a 0.80 a d a P < b P < 0.01.

6 Vol. 31 No. 1 January 2006 Validity and Utility of BPI Interference Measures 53 item and 0.81 (95% confidence interval: ) for the multi-item BPI pain-related interference scale. Construct Validity In Study 1, construct validity was examined using three patient-reported variables related to patients sleep: (1) number of night awakenings with pain, (2) quality of sleep for the previous night rated on a 5-point scale (1 ¼ very poor, 2 ¼ poor, 3 ¼ fair, 4 ¼ good, and 5 ¼ excellent ), and (3) the ALQ. These variables were correlated with the two measures of the impact of pain in all patients at Days 7 and 14. It was expected that the BPI sleep item would correlate most highly with number of night awakenings with pain and quality of sleep, while the BPI painrelated interference scale would correlate most highly with the ALQ. In the case of number of night awakenings with pain, the correlations were positive and statistically significant, as expected. The correlation between the BPI interference scale and number of awakenings at Day 7 was 0.36 (P < ), and 0.47 (P < ) at Day 14. The BPI sleep item correlated 0.55 (P < ) with number of awakenings at Day 7, and 0.57 (P < ) at Day 14. As anticipated, the correlations were negative and statistically significant in the case of quality of sleep: At Day 7, the correlation between the BPI interference scale and sleep quality was ÿ0.53 (P < ), and at Day 14 the correlation was ÿ0.63 (P < ). The correlation between the BPI sleep item and sleep quality at Day 7 was ÿ0.65 (P < ), and ÿ0.72 (P < ) at Day 14. Study 1 ALQ scores were correlated with the BPI interference scale and the sleep interference item. As expected, these correlations were all positive and statistically significant (P < ): at baseline, the ALQ summary score correlated 0.55 with the BPI, and 0.34 with the BPI sleep item; at Day 7, the ALQ correlated 0.65 with the BPI, and 0.46 with the BPI sleep item; and at Day 14, the ALQ correlated 0.65 with the BPI, and 0.49 with the BPI sleep item. In each case, the correlation between the ALQ and BPI interference scores was consistently higher than the correlation between the ALQ and the single sleep interference item. This result is as expected given that the ALQ assesses activities of daily living similar in content to many of the BPI items. In Study 2, construct validity was examined using the WOMAC. Combining both study arms, baseline scores on the BPI interference scale and the sleep interference item were correlated with the four WOMAC scores and the pain at night while in bed item. It was hypothesized that these correlations would be positive; however, due to the small sample size, it was not necessarily expected that the correlations would achieve statistical significance. These correlations, displayed in Table 3, indicate a high degree of association between the BPI and WOMAC measures. Only the correlations between the BPI sleep item and the WO- MAC stiffness and WOMAC physical function subscales failed to obtain statistical significance. The correlations between the BPI sleep item and the WOMAC scores are consistently lower than the correlations between the BPI interference scale and the WOMAC scores. Whereas sleep disturbance is clearly related to pain (r ¼ 0.42, P < ), sleep disturbance is not as strongly related to bodily stiffness (r ¼ 0.12, P > 0.01). The strong correlations between the BPI interference scores and the WOMAC scores indicate that the two measures assess highly similar constructsdthe BPI assesses the effect of pain on patient functioning in the past 24 hours, and the WOMAC assesses the amount of pain experienced in the past 2 days. The correlation between the pain at night while in bed item and the BPI sleep item was 0.55 (P < ) and the correlation between this item and the BPI interference scale score was 0.42 (P < ). The fact that this item correlated more highly with the sleep interference item than with the BPI interference Table 3 Correlations Between the WOMAC Measures and the BPI Interference Measures at Baseline (Study 2) BPI Interference Scale BPI Sleep Item WOMAC total 0.62 a 0.29 b WOMAC pain 0.62 a 0.42 a WOMAC stiffness 0.39 a 0.12 WOMAC physical function 0.64 a 0.25 b WOMAC pain at night while in bed item 0.42 a 0.55 a a P < b P < 0.01.

7 54 Williams et al. Vol. 31 No. 1 January 2006 scale score indicates that pain experienced at night while in bed is more closely connected to the extent to which pain interferes with sleep than with other activities. Discriminant Validity The results of the first set of known-groups analyses show that in Study 1, neither the BPI interference measure (t 41 ¼ÿ0.71, P > 0.05) nor the BPI sleep interference item (t 23.6 ¼ÿ1.47, P > 0.05) differed significantly across placebo and oxycodone treatment groups, although differences were in the predicted direction. The placebo group reported slightly greaterdbut statistically nonsignificantdpain-related interference compared to the oxycodone group. The results were the same in Study 2: again, neither the BPI sleep interference item (t 38 ¼ÿ1.73, P > 0.05) nor the BPI interference score (t 38 ¼ÿ1.81, P > 0.05) differed significantly across treatment and placebo groups, although the means showed the expected patterndplacebo group patients reported only somewhat greater painrelated interference than oxycodone patients. It was further hypothesized that patients experiencing high pain in the past 24 hours would report higher pain-related interference on the two BPI measures compared to patients experiencing low/moderate pain. In Study 1, baseline BPI interference scale scores differed significantly (t 130 ¼ÿ5.66, P < ) between patients with low/moderate pain (M ¼ 4.68, SD ¼ 2.0, n ¼ 58) and high pain (M ¼ 6.60, SD ¼ 1.9, n ¼ 74). However, the BPI sleep interference item did not distinguish between the two groups (t 129 ¼ÿ1.57, P > 0.05), although the difference was in the predicted direction. In Study 2, the BPI interference measure (t 104 ¼ÿ2.73, P < 0.01) was found to differ significantly across patients with low/moderate pain (M ¼ 5.36, SD ¼ 1.7, n ¼ 35) and high pain (M ¼ 6.33, SD ¼ 1.7, n ¼ 71) at baseline. Average scores on the BPI sleep interference item also differed significantly (t 104 ¼ÿ2.02, P < 0.05) between low/moderate pain patients (M ¼ 5.17, SD ¼ 2.8) and high pain patients (M ¼ 6.39, SD ¼ 3.0). Responsiveness Using data from Study 1, change was defined as the 2 week difference between baseline and end of study, and also as the 1 week difference between baseline and Day 7. The 20 mg every 12 hours CR oxycodone group was compared with the placebo group, excluding the 10 mg every 12 hours CR oxycodone arm from these analyses. (Use of the largest contrast was appropriate for establishing the responsiveness of the BPI measures, given that it was not the purpose of the analyses to verify the efficacy of the medications.) For Study 2, change was defined as the 90 day difference between baseline and end of study. An additional analysis defined change as the 1 month difference between baseline and Day 30, by which time patients were expected to have achieved stable pain control. The computed Guyatt s responsiveness statistics are displayed in Fig. 1, with graphical representation of the changes across time for both groups. Both the BPI interference scale and the Sleep interference item demonstrate strong responsiveness, with Guyatt s statistics ranging from 0.46 in Study 1 (BPI interference score, baseline to Day 7) to 1.14 in Study 2 (BPI sleep interference item, baseline to Day 30). Discussion The present analyses examined the reliability, validity, and responsiveness of two measures assessing the impact of pain on functioning, the BPI interference scale, and the BPI sleep interference item. In terms of reliability, the internal consistencies for the seven-item interference subscale are excellentd0.89 (Study 1) and 0.82 (Study 2). These values compare favorably to the alpha of 0.91 reported in McDowell and Newell 3 and fall within the range of optimal alphas between 0.70 and 0.90 as outlined by Streiner and Norman. 27(p65) These alphas are also very similar to those reported by others within populations of noncancer pain patients. 18,19 Furthermore, the size of the present coefficients confirms that the items comprising the BPI interference subscale are sufficiently unidimensional to permit scoring of the items together as a single subscale. The test-retest reliability of the BPI interference scale satisfies the psychometric requirement commonly applied to multi-item scales, with an intraclass correlation of It is generally recommended for research purposes that intraclass correlations for multi-item scales be at least 0.70; however, the single-item

8 Vol. 31 No. 1 January 2006 Validity and Utility of BPI Interference Measures 55 Mean Change in BPI Sleep Item (SD) Mean Change in BPI Interference Scale (SD) Placebo Oxycodone 0.68 (1.85) 1.54 (2.77) Study (1.66) 2.66 (2.03) 0.72 (1.84) 2.28 (2.62) 1.02 (2.08) 3.0 (2.41) 0 Day 0 to Day 7 Day 0 to Day 14 Day 0 to Day 30 Day 0 to Day 90 Guyatt's Statistic: Placebo Oxycodone 0.78 (2.66) 2.46 (3.91) 1.41 (3.73) 3.4 (2.97) 0.44 (2.55) 3.34 (3.20) Study 2 Study 1 Study (2.04) 4.0 (3.09) 0 Day 0 to Day 7 Day 0 to Day 14 Day 0 to Day 30 Day 0 to Day 90 Guyatt's Statistic: Fig. 1. Mean change and Guyatt s statistics for the BPI interference scale (above) and the BPI sleep interference item (below). BPI sleep interference score does not achieve this criterion with an intraclass correlation of 0.65dthis is a level described by Shrout and Fleiss 32 as merely good. One reason is that it seems likely that there are, in fact, fluctuations in pain interference even among patients within the highest-dose treatment arm of these studies. At least within Study 1, the average BPI sleep item score at Day 7 was 1.35 (SD ¼ 1.94) and at Day 14, 2.22 (SD ¼ 1.98)dthis represents a mean difference from test to retest of about 0.44 of a standard deviation change. Second, because it is well known that reliability is a direct function of the number of items in a measure, 28 it is important to note that in the present context, test-retest reliability is being evaluated for a single-item measure of pain-related sleep interference. The accepted criterion for reliability cited above, 0.70, is applied to multi-item scales and as a result, a somewhat lower value (i.e., 0.65) does not appear unreasonable in this context. Our

9 56 Williams et al. Vol. 31 No. 1 January 2006 report of test-retest reliability is a unique contribution to the literature on the application of the BPI to noncancer pain. Neither study was designed specifically to validate the BPI measures of the impact of OA pain. However, several variables were present in both studies that permitted the evaluation of construct and discriminant validity. Using data available from Study 1, it was demonstrated that self-reported number of night awakenings with pain was positively and significantly correlated at Day 7 and Day 14 with both the BPI interference scale and the sleep interference item. Both BPI measures also correlated negatively and significantly with patient-reported quality of sleep at both time points. Additionally, the correlations of these two sleep-specific items were higher with the BPI sleep interference item than with the BPI interference measuredthis is as expected, considering the focus on sleep. These correlational results support the construct validity of the BPI sleep item when used in patients with OA-related pain. The strong correlations between the BPI interference score and both the WOMAC and ALQ reflect the hypothesized associations among the impact of pain on patient functioning and the amount of pain and stiffness experienced (WOMAC) and the performance of activities of daily living (ALQ). Again supporting our hypothesis, the single WOMAC item particularly relevant to sleep interference due to pain ( How much pain do you have at night while in bed? ) correlated more highly with the BPI sleep interference item than with the BPI interference score, indicating that pain experienced at night while in bed is more closely connected to the extent to which pain interferes with sleep than with other activities. Taken together, the direction, strength, and pattern of the correlations in both Study 1 and Study 2 provide strong and important evidence for the construct validity of the BPI sleep interference item and the BPI interference score as measures of the impact of osteoarthritic pain for use in clinical trials. Despite the small sample sizes that attenuated the statistical significance of comparisons between the treatment and placebo groups, Guyatt s statistics provide very strong evidence of the responsiveness of the two BPI measures of pain-related interference. This was true for both studies, with moderate and large effect size estimates ranging from 0.46 to 1.14, signifying that the treatment effects, as assessed by the two measures of the impact of pain, were considerably larger in the treatment groups compared to the placebo groups. Importantly, both the BPI sleep item and the BPI interference scale were sensitive enough to demonstrate these treatment advantages. The limitations of the present investigation are the small sample sizes and the fact that neither study was designed for psychometric validation purposes. The present analyses, however, were able to provide important evidence in support of the reliability, validity, and responsivity of the two BPI measures of the impact of OA pain. The process of instrument validation can be described as a gradual accumulation of evidence. 33 Validation is not an event whereby an instrument is deemed valid once and for alldrather, it is a periodic and systematic examination of a measure that helps to build a body of evidence in support of an instrument s validity and utility. The results of this study supplement the existing wealth of knowledge and experience with the BPI by confirming the utility and validity of the BPI interference subscale in noncancer pain patients and extending the inferences made by Keller et al. 19 and Tan et al. 18 This study also provides new evidence for the BPI sleep interference item as a promising endpoint in future trials of pain medications involving patients with OA. References 1. Cleeland CS. Measurement and prevalence of pain in cancer. Semin Oncol Nurs 1985;1: Jensen MP. The validity and reliability of pain measures in adults with cancer. J Pain 2003;4(1): McDowell I, Newell C. Measuring health: A guide to rating scales and questionnaires, 2nd ed. New York: Oxford University Press, Palermo TM, Kiska R. Subjective sleep disturbances in adolescents with chronic pain: relationship to daily functioning and quality of life. J Pain 2005;6(3): Menefee LA, Frank ED, Doghramji K, et al. Self-reported sleep quality and quality of life for individuals with chronic pain conditions. Clin J Pain 2000;16(4):

10 Vol. 31 No. 1 January 2006 Validity and Utility of BPI Interference Measures Morin CM, Gibson D, Wade J. Self-reported sleep and mood disturbance in chronic pain patients. Clin J Pain 1998;14: Cleeland CS, Nakamura Y, Mendoza TR, et al. Dimensions of the impact of cancer pain in a four country sample: new information from multidimensional scaling. Pain 1996;67: Caraceni A, Mendoza TR, Mencaglia E, et al. A validation study of an Italian version of the Brief Pain Inventory. Pain 1996;65: Turk DC, Dworkin RH, Allen RR, et al. Core outcome domains for chronic pain clinical trials: IM- MPACT recommendations. Pain 2003;106: Zelman DC, Gore M, Dukes E, Tai KS, Brandenburg N. Validation of a modified version of the Brief Pain Inventory for painful diabetic peripheral neuropathy. J Pain Symptom Manage 2005;29(4): Breitbart W, McDonald MV, Rosenfeld BD, et al. Pain in ambulatory AIDS patients. I: Pain characteristics and medical correlates. Pain 1996;68: Breitbart W, Passik S, McDonald MV, et al. Patient-related barriers to pain management in ambulatory AIDS patients. Pain 1998;76: Breitbart W, Rosenfeld BD, Passik SD, et al. The undertreatment of pain in ambulatory AIDS patients. Pain 1996;65: Breitbart W, Rosenfeld BD, Passik SD, et al. A comparison of pain report and adequacy of analgesic therapy in ambulatory AIDS patients with and without a history of substance abuse. Pain 1997;72: Newshan G, Lefkowitz M. Transdermal fentanyl for chronic pain in AIDS: a pilot study. J Pain Symptom Manage 2001;21: Galer BS, Henderson J, Perander J, Jensen MP. Course of symptoms and quality of life measurement in complex regional pain syndrome: a pilot survey. J Pain Symptom Manage 2000;20(4): Schiffman R, Kopp JB, Austin HA, et al. Enzyme replacement therapy in Fabry disease: a randomized controlled trial. JAMA 2001;285(21): Tan G, Jensen MP, Thornby JI, Shanti BF. Validation of the Brief Pain Inventory for chronic nonmalignant pain. J Pain 2004;5(2): Keller S, Bann CM, Dodd SL, et al. Validity of the Brief Pain Inventory for use in documenting the outcomes of patients with nonmalignant pain. Clin J Pain 2004;20(5): Keller S, Bann CM, Dodd SL. Power of the Brief Pain Inventory to document changes in arthritis and low back pain. Paper presented at the Drug Information Agency 8th Annual Symposium, 2001 March ; Hilton Head, SC. 21. Thie NMR, Prasad NG, Major PW. Evaluation of glucosamine sulfate compared to ibuprofen for the treatment of temporomandibular joint osteoarthritis: a randomized double blind controlled 3 month clinical trial. J Rheumatol 2001;28: Roth SH, Fleischmann RM, Burch FX, et al. Around-the-clock, controlled-release oxycodone therapy for osteoarthritis-related pain: placebocontrolled trial and long-term evaluation. Arch Intern Med 2000;160: Caldwell JR, Hale ME, Boyd RE, et al. Treatment of osteoarthritis pain with controlled release oxycodone or fixed combination oxycodone plus acetaminophen added to nonsteroidal antiinflamatory drugs: a double-blind, randomized, multicenter, placebo-controlled trial. J Rheumatol 1999;26: Fries JF, Spitz PW, Kraines RG, Holman HR. Measurement of pain outcome in arthritis. Arthritis Rheum 1980;23(2): Bellamy N, Buchanan WW, Goldsmith CH. Validation of the WOMAC: a health status instrument for measuring clinically important patient relevant outcomes to anti-rheumatic drug therapy in patients with osteoarthritis of the hip or knee. J Rheumatol 1988;15: Cronbach L. Coefficient alpha and the internal structure of tests. Psychometrika 1951; Streiner DL, Norman GR. Health measurement scales: A practical guide to their development and use, 2nd ed. New York: Oxford University Press, Nunnally JC, Bernstein IH. Psychometric theory, 3rd ed. New York: McGraw-Hill, Guyatt GH, Walter SD, Norman G. Measuring change over time: assessing the usefulness of evaluative instruments. J Chronic Dis 1987;40(2): Fehnel SE, Mcleod LD, Brandman J, et al. Responsiveness of the Acne-Specific Quality of Life Questionnaire (Acne-QoL) to treatment of acne vulgaris in placebo-controlled clinical trials. Qual Life Res 2002;11: Cohen J. Statistical power analysis for the behavioral sciences, 2nd ed. Hillsdale, NJ: Erlbaum, Shrout PE, Fleiss JL. Intraclass correlations: uses in assessing rater reliability. Psychol Bull 1979;86: American Educational Research Association, American Psychological Association, National Council on Measurement in Education. Standards for educational and psychological testing. Washington, DC: APA, 1999.