Development of a Cancer Pain Prognostic Scale

Transcription

1 366 Journal of Pain and Symptom Management Vol. 24 No. 4 October 2002 Original Article Development of a Cancer Pain Prognostic Scale Shirley S. Hwang, RN, MS, AOCN, Victor T. Chang, MD, Diane L. Fairclough, DrPH, and Basil Kasimis, MD Section of Hematology/Oncology (S.S.H., V.T.C., B.K.), and Patient Care Services (S.S.H.), VA New Jersey Health Care System, East Orange, New Jersey; UMDNJ/School of Nursing (S.S.H.), and UMDNJ/ New Jersey Medical School (V.T.C., B.K.), Newark, New Jersey; and Center for Research Methodology and Biometry (D.L.F.), AMC Cancer Research Center, Denver, Colorado, USA Abstract The purpose of this study was to develop a Cancer Pain Prognostic Scale (CPPS) which could predict the likelihood of pain relief within 2 weeks for cancer patients with moderate to severe pain. Seventy-four (74) consecutive patients who presented with cancer-related pain were managed in accordance with the guidelines for pain management developed by the United States Agency for Health Care Policy and Research (AHCPR). Patients were followed weekly using the Brief Pain Inventory (BPI), and medications were recorded weekly for 3 weeks. Baseline scores from the Functional Assessment of Cancer Therapy (FACT-G), Mental Health Inventory (MHI), Karnofsky Performance Status (KPS), and Memorial Symptom Assessment Scale Short Form (MSAS-SF) at initial interview served as explanatory variables in a logistic regression model. Pain relief 80% at the end of weeks 1 and 2 were used as outcomes in this model. From this analysis, we developed a predictive formula, the CPPS, which includes the worst pain severity, FACT-G emotional well being, daily opioid dose, and pain characteristics. The rule yields a numerical score that ranges from Higher scores correspond to a higher probability of good pain relief. The CPPS has the potential to rapidly identify patients with poor pain prognosis. It can be used as a research tool to characterize pain in cancer patients. J Pain Symptom Manage 2002;24: U.S. Cancer Pain Relief Committee, Key Words Cancer, pain, prognosis, veterans, pain relief, predictive rule, MSAS-SF, FACT-G, Brief Pain Inventory Preliminary results presented at the American Pain Society 17th Annual Scientific Meeting, 5 8 November 1998 at San Diego, CA, and at the 34th Annual Meeting of the American Society for Clinical Oncology, May 1998 at Los Angeles, CA. The views expressed herein do not necessarily reflect the views of the Department of Veterans Affairs or of the U.S. Government. Address reprint requests to: Shirley S. Hwang, RN, AOCN, MS, Section Hematology/Oncology (111), VA New Jersey Health Care System at East Orange, 385 Tremont Avenue, East Orange, NJ 07018, USA. Accepted for publication: November 17, Introduction Cancer-related pain is one of the most prevalent and distressing symptoms in populations with cancer. It comprises a heterogeneous group of more than 100 pain syndromes with varied underlying pathophysiologies. 1,2 Although cancer pain can be successfully controlled by adhering to World Health Organization (WHO) guidelines, 3 it remains undertreated. The Eastern Cooperative Oncology Group study found that nearly 50% of cancer pain patients did not re- U.S. Cancer Pain Relief Committee, /02/$ see front matter Published by Elsevier, New York, New York PII S (02)

2 Vol. 24 No. 4 October 2002 Development of a Cancer Pain Prognostic Scale 367 ceive adequate analgesic therapy. 4 A major cause of undertreatment is poor assessment. One reason for poor assessment may be that patients are not asked about pain; another is the lack of an easy, simple bedside instrument to predict pain relief. Alternatively, lack of recognition of difficult pain syndromes may lead to poor pain control, and delay referral to pain specialists. Patients with advanced cancer may often experience pain syndromes which may require specialized diagnostic and therapeutic approaches. 5 Five to twenty percent of cancer pain patients require invasive modalities to achieve analgesia. 6,7 Currently, these patients are identified by their failure to respond to conventional treatment, as specified by the WHO analgesic pain ladder or the recommendations of the Agency for Health Care Policy and Research (AHCPR). 8 Other contributing factors may also be important. Patients who experience difficulties with coping, anxiety, and psychological distress may also have greater difficulty with pain control. The lack of an institutional, or individual, plan to act upon the information obtained may affect pain management. Recently, this point was made to explain why bedside charting of pain levels of inpatients failed to affect pain control. 9 Of the many current instruments for assessing cancer pain, only the WHO analgesic ladder has specific recommendations for pain management, 3 and none have prognostic capabilities. The development of an accurate pain staging system has only received attention recently. Currently, the only validated predictive scale is the Edmonton Pain Staging System developed by Bruera et al. 10 This system was validated in hospice patients by using candidate variables to assess pain prognosis from the following five dimensions: pain mechanism, psychological distress, pain characteristics, tolerance, and history of substance abuse. A 21-day period was selected to allow adequate time for dose titration and implementation of other treatments. Two prognostic categories were identified: 1) good prognosis category, which includes patients with visceral pain, somatic pain, non-incidental pain, absence of somatization, absence of tolerance and absence of substance abuse; and 2) poor prognosis category, which includes patients with neuropathic pain, mixed pain etiology, incidental pain, substance abuse, somatization, and tolerance. Our aim was to apply the concept of a simple multidimensional predictive instrument suggested by Bruera et al. to assess pain in the setting of a Hematology/Oncology outpatient and inpatient population. In this article, we report the development of Cancer Pain Prognostic Scale. Methods Theoretical Model The patient s perception of pain is a multidimensional construct consisting of physiologic, sensory, affective, cognitive, behavioral, and sociocultural dimensions. 11,12 In this study, a biopsychosocial model of factors affecting pain was used to assess the patient and served as a source of predictor variables for pain relief. Items related to pain assessment were divided into four dimensions: pain characteristics, physical and psychological symptom distress, quality of life (QOL), and personal characteristics. The primary outcome was adequate pain relief, defined as pain relief 80% at each follow-up week. Validated instruments were used to assess each of these dimensions, and variables derived from items in these instruments are summarized in Table 1. We applied the concept of clinical prediction rules, as defined by Laupacis et al., 13 in the development of a prognostic scale. A clinical prediction rule is a decision-making tool for clinicians that includes 3 or more variables obtained from history, physical examination, or simple diagnostic tests and that either provides the probability of outcomes or suggests a diagnostic or therapeutic action. 14 It can assist clinicians in quantifying data and make more accurate estimates of the likelihood of an event and its progress. Three standards in the development of clinical prediction rules 15 outcome definition, blind assessment of outcomes, and blind assessment of predictors were followed in this study. Patient Selection In this prospective, longitudinal study, patient recruitment began in January 1996 and ended in March 1997 at the VA New Jersey Health Care System (VANJHCS) at East Orange. The VANJHCS is the sole tertiary care teaching hospital that provides Hematology/ Oncology services for veterans residing in New

3 368 Hwang et al. Vol. 24 No. 4 October 2002 Table 1 Definition of Independent Variables Within Each Dimension (Based on the Initial Day 1 Assessment) Category Instruments/Measurements Variables 1. Pain Characteristics Brief Pain Inventory (BPI) Initial worst pain severity Initial pain relief BPI Pain Interference Score 24-hour opioid dose Number of pain sites Pain Syndromes Assessment Bone pain Neuropathic pain Nociceptive pain Breakthrough pain 2. Personal Characteristics Karnofsky Performance Status Demographic data Mini Mental Status Exam 3. Symptom Distress Memorial Symptom Assessment Scale Short Form (MSAS-SF) Mental Health Inventory (MHI) 4. Quality of Life Functional Assessment of Cancer Therapy (FACT-G) Outcome variable: Pain relief measured by BPI 80% at week 1, week 2. Mixed pain KPS Education History of substance abuse Presence of care givers at home Primary cancer diagnosis History of psychiatric disorder Age Marital status MMSE score Physical symptom distress (PHYS) Psychological symptom distress (PSYCH) Global distress Index (GDI) Number of symptoms (NS) Anxiety Psychological distress Depression Positive affect Psychological well-being Physical well-being Functional well-being Emotional well-being SUMQOL Jersey. The study was approved by the VANJHCS Institutional Review Board, and all patients signed informed consent before participating. Seventy-four (74) consecutive patients with poorly controlled cancer-related pain were recruited from the outpatient Hematology/Oncology clinic and patients admitted to the Hematology/Oncology service. Patients with worst pain 4 out of 10 were asked to participate, as these patients can experience significant interference with function. 16 Cancer-related pain was defined as pain caused by cancer or by treatment of cancer. Moribund or delirious patients who were not able to fill out the questionnaires were excluded. Assessments and Instruments Initial assessment included age, sex, primary site, extent of disease, care-giver status, marital status, and Karnofsky Performance Status (KPS). 17 Each patient was asked to complete four instruments: the Brief Pain Inventory (BPI), 18 Functional Assessment Cancer Therapy (FACT-G), 19 Memorial Symptom Assessment Short Form (MSAS-SF), 20 and Mental Health Inventory (MHI). 21 At the time of study entry, the Mini-Mental Status Exam (MMSE) 22 was assessed upon study entry to determine the presence and degree of cognitive impairment. An initial pain diagnosis was recorded. 23 Standard guidelines from the AHCPR Cancer Pain Guidelines for pain assessment, analgesic interventions, and management of neuropathic and bone pain were followed. 8 Preferences for pharmacologic management were implemented as described in Table 13 of the Guidelines 8 and by Cherny et al. 24 Our management differed in that establishing a pain diagnosis took place at the same time analgesics were titrated for pain relief. Patients and family members received the standard AHCPR patient guidelines upon study participation. Fol-

4 Vol. 24 No. 4 October 2002 Development of a Cancer Pain Prognostic Scale 369 low-up assessment was done at 1-week intervals for 3 weeks. At each follow-up visit, the patients were assessed for change in medications, side effects, and pain variables according to the BPI. The primary care providers in the Hematology/Oncology section (SSH, VTC) were also the pain clinicians for the patients. Interviews with the various instruments were usually conducted by an interviewer, but were at times also conducted by the care providers when circumstances dictated, such as inpatients first seen after working hours. Other than information from the BPI and side effect profiles, the results of the other assessment instruments were not available to the clinicians. The KPS is an 11-point rating scale ranging from (0 dead, 100 normal function) to assess patients functional level related to cancer and its treatment. The FACT-G (version 3) is a validated, 28-item general patientrated measure of QOL for cancer patients with any tumor type. Each item is scored from 0 4 and anchored from not at all to very much. There are 5 subscales: Functional Well Being (FWB) (7 items), Physical Well Being (PWB) (7 items), Social/Family Well Being (SFWB) (7 items), Relationship with MD (RMD) (2 items), and Emotional Well Being (EWB) (5 items), with total QOL scores ranging from The FACT-G has been used widely in clinical trials, it is easy to complete, and has demonstrated sensitivity according to performance status and extent of disease. The MSAS-SF is a validated patient-rated instrument that includes patient assessment for symptom frequency or distress for 32 highly prevalent physical and psychological symptoms. Each symptom was scored from 0 4 ranging from no symptom to very much. MSAS- SF subscales include the Global Distress Index (GDI) (4 psychological symptoms: feeling sad, worrying, feeling irritable, and feeling nervous; and 6 physical symptoms: lack of energy, pain, lack of appetite, feeling drowsy, constipation, dry mouth). The physical symptom distress score (PHYS) includes 12 prevalent symptoms: lack of energy, pain, lack of appetite, feeling drowsy, constipation, dry mouth, nausea, vomiting, change in taste, weight loss, feeling bloated, and dizziness. The psychological symptom distress score (PSYCH) includes 6 prevalent psychological symptoms: worrying, feeling sad, feeling nervous, difficulty sleeping, feeling irritable, and difficulty concentrating. The number of symptoms (NS) is derived from screening for the presence of 32 symptoms at each interview. The MSAS-SF and FACT-G instruments have been validated at our institution and reference values determined for our patients. 25 The RAND Mental Health Inventory (MHI) is a validated instrument to measure anxiety, distress, and depression and is used as part of the initial psychological evaluation. The Brief Pain Inventory Short Form is a validated and widely used instrument for patientrated pain, with a numerical 0 10 scale, anchored from no pain at all to as bad as you can imagine. Patients rate their worst, least, average and immediate pain severity, pain relief ranging from 0 100%, and functional interference caused by pain in the areas of daily activity, mood, walking, sleeping, movement, enjoyment of life, and relationship with others. The sum of answers to the interference questions, the total pain interference score, ranges from Statistical Analysis Pain Characteristics (Table 2). Wilcoxon signed rank test was used to assess the difference between each week in the following variables: worst pain severity, average pain severity, pain interference scores, pain relief and morphine equivalent daily dose. Predictors of Pain Relief (Table 3). To determine predictors of pain relief at week 1, 2, and 3 within each dimension, screening tests and exploratory analyses were performed to identify potentially important predictors. Logistic variables were defined from continuous variables measured in the FACT-G, MSAS-SF, MHI, and pain interference score by dichotomizing each variable around its median value. Univariate and multivariate logistic regression models were performed within each of the dimensions. The multivariate models used a stepwise selection procedure. Predictors of pain relief within each dimension were then combined in multivariate logistic analyses to develop multidimensional models of pain relief at 1 and 2 weeks (Table 3). The resulting predictive rule was tested to determine sensitivity, specificity, and positive and negative predictive values based upon a predictive probability 0.5.

5 370 Hwang et al. Vol. 24 No. 4 October 2002 Table 2 Summary of Pain Characteristics at Each Visit 2A: Entire Population Day 1 (n 74) Week 1 (n 66) Week 2 (n 57) Week 3 (n 53) Median Range Median Range Median Range Median Range Worst pain Average pain Pain relief (%) Interference MEDD (mg) B: Limited to Patients Who Remained on Study (n 53) Day 1 Week 1 Week 2 Week 3 Median Range Median Range Median Range Median Range P value a Worst pain Average pain Pain relief (%) Interference MEDD (mg) a P value by Wilcoxon signed-rank test between day 1 and week 1. Worst pain and average pain No significant difference between week 1 and week 2; Pain Relief No significant difference between week 1 and week 2, and between week 2 and week 3; Pain Interference and MEDD (morphine oral equivalent daily dose) No significant difference between week 1 and week 2, week 1 and week 3, and week 2 and week 3. Development of CPSS. A scale for the perception of pain relief ( 80%) at 1 and 2 weeks was developed using multivariate logistic analyses in two steps. In the first step, a stepwise procedure was used to select the strongest predictors of relief after 1 week ( 0.01) from characteristics that were predictive of relief in univariate analyses ( 0.05). In the second step, the stepwise selection procedure was repeated ( 0.05), with the predictors from the first step forced into the model (Table 4). Because the purpose of this exploratory analysis was to develop an easily usable scale, the actual coefficients from the logistic regression analysis were used as a guide for determining the weights of the scale. The objective was to keep the scale simple, using easily remembered integers that could be used in clinical practice without requiring a calculator. Areas under the Receiver Operated Characteristic (ROC) curves and the sensitivity, specificity, and positive and negative predictive values based upon different cutoff points of CPPS scores were generated to determine the predictive abilities of the CPPS. 26 Results Demographics and Pain Characteristics Eighty-nine (89) patients were encountered with severe pain. Three patients were too ill and one patient refused to participate. Of the 85 patients who gave consent to participate, 4 were found not to have cancer-related pain and removed from analysis. This left 81 patients, of whom 7 were excluded because of inability to complete the questionnaires. Seventy-four patients were available for analysis. Median age was 63 years (range 40 82); the median education level was 12th grade (range 6 18), and the median MMSE score was 27 (range 18 30). There were 39 (53%) inpatients and 35 (47%) outpatients; 45 (61%) patients had a caregiver at home. Among these 74 patients, 31 (42%) were seen for the first time and 43 (58%) were patients known to the service with a new acute pain problem. Eleven (15%) patients had a history of drug abuse and 48 (65%) patients had a history of alcohol abuse. The pain syndromes and pain diagnoses were completed by two of investigators (SSH, VTC). The median number of pain sites was one, with a range from 1 5. Forty (54.8%) patients had one pain diagnosis, 34 (46%) patients presented with two pain diagnoses. The pain diagnoses included nociceptive pain (50 patients, 67%), bone pain (25 patients, 34%), neuropathic pain (43 patients, 58%), breakthrough pain (52 patients, 70%), and mixed pain (30 patients, 40%). Primary cancer sites were lung (24 patients, 32%), prostate (16 patients, 22%), head and neck (9 patients, 12%),

6 Vol. 24 No. 4 October 2002 Development of a Cancer Pain Prognostic Scale 371 Pain Relief 80% at week 1 Table 3 Stepwise Multivariate Logistic Regression Analysis Area under receiver operated characteristic curve (ROC) 0.83 Sensitivity 78.38% Specificity 73.08% Positive prediction value 80.56% Negative prediction value 70.37% Independent Predictors Unit a Odds ratio P value 95% C.I. BPI Worst pain severity FACT-G Emotional well-being 0, MHI Anxiety 0, Pain Relief 80% at week 2 Area under receiver operated characteristic curve (ROC) 0.85 Sensitivity 86.11% Specificity 63.16% Positive prediction value 81.58% Negative prediction value 70.59% Independent Predictors Unit a Odds ratio P value 95% C.I. Initial opioid dose FACT-G Emotional Well-Being 0, Nociceptive pain 0, Alcohol abuse 0, There are 14 variables in the model: worst pain severity (BPI), mixed pain, neuropathic pain, nociceptive pain, initial opioid dose, number of pain sites, history of alcohol abuse, emotional well-being (FACT-G), physical well-being (FACT-G), psychological distress (MHI), depression (MHI), anxiety (MHI), psychological well-being (MHI), and global distress index (MSAS-SF). The predictive probability cutoff is 0.5. a Logistic variables were defined from continuous variables by dichotomizing each variable around its median value, 0 was assigned to the value less than median and 1 was assigned to the value equal to or greater to median. worst pain severity: 0 to 10. FACT Emotional well-being (EWB): median 17, 0 EWB 17, 1 EWB 17. MHI Anxiety: median 27, 0 anxiety score 27, 1 anxiety score 27. Initial opioid dose: median 60 mg, 0 60 mg/day, 1 60 mg/day. Nociceptive pain: 0 no nociceptive pain, 1 positive nociceptive pain. Alcohol: 0 no history of alcohol abuse, 1 positive history of alcohol abuse. colorectal (6 patients, 8%), lymphoma (3 patients, 4%) and other (16 patients, 22%); primary pain syndromes were bone pain (20 patients, 27%), local disease invasion (14 patients, 19%), brachial plexopathy (6 patients, 8%), cranial neuralgias (5 patients, 7%), epidural compression (5 patients, 7%), hepatic distension syndrome (5 patients, 7%), lumbosacral plexopathy (5 patients, 7%), chest wall syndrome (3 patients, 4%), acute pain related to radiotherapy (2 patients, 3%), and other (9 patients, 12%). Changes in the pain characteristics (pain severity, pain relief, pain interference) over time are summarized in Table 2. There was significant attrition during the 3-week period, with 10% loss of patients every week to follow-up (74, 66, 57, and 53 patients respectively at day 1, week 1, week 2, and week 3 interview). The most common reason was patient death (12 instances), followed by physical deterioration with inability to complete the follow-up forms (13 instances), refusal to answer questions (4 instances), incompletely filled forms (3 instances), and miscellaneous (4 instances). Most of the attrition was the result of progressive disease. The results limited to those remaining in the study are summarized in Table 2B. These results can be summarized as follows. First, most patients achieved pain relief 80% by week 1 and this remained steady until week 3. On day 1, median pain relief was 40% with interquartile range of 0 70%. By week 1, median pain relief was 80% with an interquartile range of %, and remained at 80% for weeks 2 and 3. The number of patients with pain relief 80% over time ranged from 13/74 (18%) on day 1, to 38/66 (58%) on week 1, to 38/57 (67%) on week 2, and to 40/53 (75%) on week 3. This improvement was still present after limiting the analysis to individuals who remained on the study. Second, pain severity decreased significantly from day 1 to week 1, with

7 372 Hwang et al. Vol. 24 No. 4 October 2002 worst pain decreasing from 9 6 out of 10, and average pain decreasing from 6 4 out of 10. Improvement continued in the remaining 2 weeks for a total of 4 point drop in both worst and average pain severity categories. Again, the results were unchanged after limiting the analysis to individuals who remained on the study. Third, the median morphine equivalent oral daily dose was doubled from 60 mg 120 mg by week 1 and remained steady afterwards. Besides increasing opioid doses, other interventions included opioid switching, addition of adjuvant analgesics, referral for radiation therapy, and admission for pain management. The median number of interventions at the first week was 2 with a range of 1 5. Details are summarized in another report. 27 Fourth, significant and continuous improvement in pain interference scores was seen. Predictors of Pain Relief Four dimensions were evaluated at initial assessment: pain characteristics, personal characteristics, QOL, and symptom distress. The independent variables in each dimension are listed in Table 1. Each variable was examined to see which variable correlated most closely with the outcome of pain relief 80% at each week. The significant predictors of pain relief at each week were different, and none of the candidate variables predicted pain relief at week 3. A multivariate logistic regression was then performed within each dimension to identify independent predictors at weeks 1 and 2. Ten variables were predictors of pain relief at week 1: worst pain severity, number of pain sites, nociceptive pain, anxiety, depression, psychological distress, psychological well-being, global distress index, physical well-being, and emotional well-being. Seven out of 10 variables were related to symptom distress and QOL. At week 2, 10 variables were identified: initial opioid dose, mixed pain, neuropathic pain, nociceptive pain, worst pain severity, initial pain relief, initial pain interference score, depression, emotional well-being, and history of alcohol abuse. Most of the variables (7 out of 10) were related to pain characteristics. The worst pain severity, nociceptive pain, depression, and emotional well-being were the only variables identified in both weeks. The final stepwise multivariate logistic regression model was formulated with 14 variables based upon the week 1 and week 2 results. These were: worst pain severity (BPI), mixed pain, neuropathic pain, nociceptive pain, initial opioid dose, number of pain sites, history of alcohol abuse, emotional well-being (FACT-G), physical well-being (FACT-G), psychological distress (MHI), depression (MHI), anxiety (MHI), psychological well-being (MHI), and GDI (MSAS-SF). The results of a stepwise multivariate logistic regression analysis, with area under the ROC, sensitivity, and specificity, are summarized on Table 3. Independent predictors for pain relief at week 1 included initial worst pain severity, emotional well-being, and anxiety (odds ratio 1.69, 7.73, 0.23 with P value 0.004, 0.004, respectively). The independent predictors for week 2 included initial opioid dose, emotional well-being, nociceptive pain, and history of alcohol abuse (odds ratio 0.23, 4.94, 0.15, 0.47 with P value 0.04, 0.03, 0.02, 0.02 respectively). Both models showed high sensitivity (78.38% for week 1 and 86.11% for week 2) and moderate specificity (73.08% for week 1 and 63.16% for week 2) with a probability cutoff 0.5. The area under the ROC for weeks 1 and 2 were.83 and.85 respectively. These models suggest that pain relief is a dynamic process as different independent predictors were identified at different time points of the outcome evaluation. In summary, patients without nociceptive pain, with higher worst pain scores, currently managed on less intensive pain medication, with better emotional well-being, less anxiety, and no history of alcohol abuse were more likely to perceive better pain relief when the AHCPR pain management protocol is followed. Development of the Cancer Pain Prognostic Scale (CPPS) After identifying predictors of pain relief for week 1 and week 2, we then developed the CPPS by combining the information obtained from both models. In the first step, initial rating of worst pain and a score 17 on the emotional well-being subscale of the FACT-G were the strongest predictors of relief after 1 week ( 0.01) (Table 4). The observed coefficients of 0.48 and 2.11 suggested the weights of 1 and 4 respectively. In the second step, this intermediate scale (worst pain severity (BPI) + 4 [FACT-G emotional well-being 17]) was included as one of the potential explanatory vari-

8 Vol. 24 No. 4 October 2002 Development of a Cancer Pain Prognostic Scale 373 Table 4 Development of the CPPS Scale Step 1: Pain Relief 80% at week 1 Predictors (from week 0) Coefficient P value BPI Worst Pain Severity FACT-G Emotional Well-Being Intermediate scale Worst Pain Severity (BPI) 4 [FACT-G Emotional Well-Being 17] Step 2: Pain Relief 80% at week 2 Predictors (from week 0) Coefficient P value Intermediate scale a Daily opioid 60 mg morphine PO Presence of mixed pain CPPS Scale Intermediate Scale 4 [Daily opioid dose 60 mg morphine orally] 4 [Presence of mixed pain] 3 Fig. 1. Cancer Pain Prognostic Scale week 1 estimated probability for pain relief 80%. ables for relief at two weeks. The intermediate scale, initial lower opioid dose, and the absence of a mixed pain syndrome were significant predictors of pain relief after 2 weeks ( 0.05). Again, using the observed logistic regression coefficients of 0.40, 1.76, and 1.66 as guides, the weights for the CPSS scale were determined: CPPS 3 + worst pain severity (BPI) + 4 [FACT-G emotional well-being 17] 4 [daily opioid dose 60 mg morphine orally] 4 [presence of mixed pain]. Interestingly, the mixed pain syndrome was identified as a significant predictor by adding the intermediate scale from the first step as one of the explanatory variables. Possible values of the scale range from 0 17 with higher scores corresponding to a higher probability of pain relief one and two weeks after assessment. The estimated probability of pain relief after 1 and 2 weeks on the pain management protocol as a function of the CPPS are presented in Figures 1 and 2. The predictive statistics (sensitivity, specificity, positive predictive value, and negative predictive value) of CPPS by using different CPPS score as cutoff points for pain relief 80% at weeks 1 and 2 were calculated. The results indicate that the CPPS scores can be grouped into three groups high CPPS scores (13 17), intermittent CPPS scores (7 12), and low CPPS scores (1 6) based on the sensitivity and specificity (Table 5). For high CPPS group, the sensitivity ranged from with specificity ranging from For intermittent CPPS group, the sensitivity ranged from with the specificity ranging from , and for the low CPPS group, the sensitivity ranged from with the specificity varying from It was noted that the predictors of relief at three weeks were quite different than the predictors at two weeks. It was decided that it would be unlikely to be able to develop a scale that was useful for prediction of both short and long term pain relief, therefore scale development stopped at this point. Comparison of CPPS s Predictive Power to the Edmonton Pain Staging System We calculated prognostic scores for our patients with the formula and obtained a median value of 10, interquartile range of 8 12, and a range from The percentage of patients in each quartile with good pain control (80% pain relief at 1 week, 2 weeks, and 3 weeks) was calculated. According to the Edmonton Pain Fig. 2. Cancer Pain Prognostic Scale week 2 estimated probability for pain relief 80%.

9 374 Hwang et al. Vol. 24 No. 4 October 2002 Table 5 Sensitivity and Specificity of CPPS Scores Score Sensitivity Specificity Week 1 Week 2 Week 1 Week Staging System, 7 of our patients would have had a good prognosis, and 67 would have had a poor prognosis. The differences between the Edmonton Pain Staging System and CPPS predictions at each week of follow-up are summarized on Table 6. At three weeks, both predictive systems perform poorly. At 1 2 weeks, the CPPS offers a wider range of possible outcomes. Discussion In this study, we prospectively followed 74 patients with significant pain encountered in a medical oncology setting, and managed them in accordance with published guidelines with pharmacologic, psychological, and radiation therapy interventions. Patients seen in a referral pain clinic or hospice might have a different spectrum of problems or concerns, and studying a hematology/oncology population may minimize selection bias. 28 Based upon these outcomes, we have developed a cancer pain prognostic scale. Pain relief has been the primary outcome, although definitions of pain relief have varied in previous studies of cancer pain The choice of 80% pain relief as good pain relief was based on the observation that most of our patients reported 80% pain relief. Previous predictors of good pain relief in patients seen in cancer clinics have focused on the responsiveness of the underlying disease to cancer therapy for good long-term pain relief 30 and the presence of metastatic disease. 32 Currently, the only validated predictive scale is the Edmonton Pain Staging System developed by Bruera et al. 10 This system was validated in hospice patients. A difficulty with this system is low specificity. In the original study, more than 50% of patients with poor prognosis stage could still achieve good pain control by 3 weeks. In another survey of Belgian patients, all patients were classified as poor prognosis. 33 One reason may be an incomplete range of prognostic factors. Possible prognostic factors not included by Brueraet al. include extent of disease, physical distress, performance status, psychological factors, number and severity of other symptoms, prior analgesic therapy, patient barriers to pain management, age, and opioid responsiveness of pain. 34 Another reason may be that with multiple interventions and changes in the course of disease, 3 weeks may be a long time for patients and therefore insensitive to gradations of treatable pain. Our analysis suggests that many variables may no longer be relevant at 3 weeks evaluation. As pain is a multidimensional phenomenon, we developed a multidimensional model for pain relief to include a wider range of possible prognostic factors. We hypothesized that the four dimensions of the model (pain characteristics, the patient s social situation, overall symptom distress, and general overall QOL) may all be important in the success of pain therapy. The resulting model provides some interesting ideas regarding pain management by domain. For pain characteristics, neuropathic pain was not a significant independent predictor variable in the weekly model. One reason may have been the high incidence of epidural compression, which is considered a form of neuropathic pain. Another reason may have been that at higher opioid doses, neuropathic pain is responsive to opioids (31). In this study, pain diagnoses were categorized as nociceptive, bone pain, neuropathic pain, breakthrough pain, and mixed (Table 1). Bone pain was seen in 25 patients (34%) and is a subtype of nociceptive pain. Breakthrough pain describes a temporal pattern and was seen in 52 patients (70%). The breakthrough pain was neuropathic (15 patients, 20%), nociceptive (13 patients, 18%) or mixed pain syndrome (24 pa-

10 Vol. 24 No. 4 October 2002 Development of a Cancer Pain Prognostic Scale 375 Table 6 Comparison of Edmonton Pain Staging System and Cancer Pain Prognostic Scale (CPPS) 1. Differences Between the CPPS and the Edmonton Staging System Edmonton Pain Staging System Cancer Pain Prognostic Scale Population Hospice patients Medical Oncology patients Outcomes Good/Bad prognosis Graded range from 0 to 17 Time of assessment Three weeks One week Criterion MD/patient consensus Patient rated pain relief 80% Variables Differences Between Prediction Results for Pain Relief 80% at each Follow-up Week Edmonton Pain Staging System Cancer Pain Prognostic Scale Week 1 Week 2 Week 3 Week 1 Week 2 Week 3 n % n % n % n % n % n % All population 40/ / / / / /53 76 Edmonton System CPPS score Poor prognosis 35/ / / / / / / / /10 70 Good prognosis 5/7 71 5/6 83 5/ / / / / / /18 72 tients, 46%). 35 This categorization may have influenced the CPPS results, as nociceptive pain is described as a predictor for slow pain relief. A second finding is the importance of the opioid dose as a predictor variable. Patients on low doses of opioids are likely to benefit from dose escalation whereas patients in pain on higher doses are likely to have a significant pain problem and require more attention. For personal characteristics, previous use of alcohol or illicit drugs was not an adverse factor. In this population, older men with prostate cancer and bone pain were the largest group of patients who had used alcohol in the past. General symptom distress parameters were not predictive of pain relief, suggesting that patients can focus specifically on pain regardless of other possibly distracting symptoms. Quality of life evaluation suggested that the dimension of emotional well-being is an important predictor, and highlights the potential contribution of psychological interventions in the management of pain. A new finding is the possibility that pain relief depends upon different variables at each time of assessment. Worst pain severity is the most important independent predictor at 1 week follow-up. The predictive value of higher worst pain severity can be explained if we consider higher pain scores as an indicator of undertreatment. Following the AHCPR guidelines should lead to better and fast pain relief in these patients. Worst pain severity is also a predictive factor for responsiveness of patients with painful bone metastases to radiation. 36 Patients emotional states measured by emotional well-being and anxiety also were important variables at immediate follow-up. This suggests that the patient s emotional state may contribute significantly in the perception of pain relief at follow-up. At week 2 follow-up, emotional well-being, initial opioid dose, and pain syndrome were more important predictors. The reason for the importance of different variables at different time points is not known. The variables used to derive the CPPS were generated from variables identified by dimensions important for pain management. As the percentage of patients achieving pain relief, and the predictors for pain relief, changed with time, it was necessary to combine the results of two logistic regressions at weeks 1 and 2 to select variables for the final CPPS. The intermediate scale (worst pain severity + 4 [FACT-G emotional well-being 17]), which can be considered as the predictive scale for pain relief at week 1, was used in the derivation of the second logistic model. Our initial results suggest that such an approach may be feasible

11 376 Hwang et al. Vol. 24 No. 4 October 2002 in developing predictive rules for dynamic cancer pain outcomes. One interesting consequence of this approach is that a variable, which was not present in the models for week 1 or week 2, mixed pain, was selected as a predictive variable. Perhaps the most interesting feature of the CPPS is its ability to provide a continuous distribution of probabilities that a patient can achieve good pain relief, as illustrated in Figures 1 and 2. Also of interest is that the CPPS scores provided similar sensitivity and specificity ratings for both weeks. We categorized two cutoff points of CPPS for clinical utilization after carefully examining the predictive statistics by using each CPPS score as probability cutoff point (Table 6). In this regard, the CPPS extends the potential uses of a pain prognostic scale. A high CPPS score suggests that patients are more likely to achieve pain relief 80% in either week, which consists of CPPS scores cutoff 13 with high sensitivity ( 0.75) and low specificity ( 0.1). The low probability group consists of CPPS scores 6. These scores have extremely low sensitivity ( 10%) and high specificity ( 70%) values, which indicates that these lower CPPS scores may help identify a low probability group of patients who may have poor pain relief after conventional pain management strategies. The intermediate probability group consists of patients with CPPS scores ranging from 7 12 with sensitivity approximately from 13 55% and specificity from 12 58% for both weeks. Ultimately, with a CPPS score and an associated probability, a clinician could combine this information with other clinical information to decide whether to proceed with standard care or to pursue more intensive care. Such an instrument could be useful for nurses and housestaff who may not have much experience with assessment and treatment of cancer pain. Not all cancer pain patients are alike. Each institution could determine, based upon its population, thresholds for initiating pain consultations and aggressive management. Similar ranges could also be specified and studied for quality of care purposes. A possible clinical application of the scale is illustrated in Figure 3. Our results suggest that development of clinical predictive rules for cancer pain management may result in simpler and more informative assessment instruments. The majority of patients were able to complete these questionnaires, did not find them burdensome, and remarked that they felt the questionnaires helped with communication with health professionals. It took approximately 20 minutes to complete the initial package and 10 minutes for the follow-up packages. There are some limitations in our study. First, the study was conducted at a VA Medical Center and the results may not be generalizable to all advanced cancer patients. Second, the study included a small number of patients with poor risk factors, such as substance abuse problems, mental illnesses, severe bone pain, impaired mental status, and tolerance to opioid medications. Third, the small sample size also made it difficult to study other possible predictive factors, such as the site of pain and primary cancer site. Fourth, the selection of different variables and weights may depend upon the frequencies of pain syndromes encountered. Fifth, in this study, we used pain relief 80% as outcome variable to develop the CPPS. However, the CPPS results could have been different if other outcome items, such as worst pain severity of 4 or less or a decrease in pain severity by 2 points, were used. Further studies can effectively address these limitations. We are in the process of gathering additional patient data. Hopefully, as development of this tool continues, a better tool will Fig. 3. Possible application of the CPPS.

12 Vol. 24 No. 4 October 2002 Development of a Cancer Pain Prognostic Scale 377 emerge to provide appropriate direction for pain treatment interventions. References 1. Cherny NI. Cancer pain: principles of assessment and syndromes. In: Berger A, et al, eds. principles and practice of supportive oncology. Philadelphia: Lippincott-Raven Publishers, 1998: Grond S, Zech D, Diefenbach C, et al. Assessment of cancer pain: a prospective evaluation in 2266 cancer patients referred to a pain service. Pain 1996;64: Jadad AR, Browman GP. The WHO analgesic ladder for cancer pain management. Stepping up the quality of its evaluation. JAMA 1995;275: Cleeland CS, Gonin R, Hatfield AK, et al. Pain and its treatment in outpatients with metastatic cancer. N Engl J Med 1994;330: Hanks G, Portenoy RK, MacDonald N, et al. Difficult pain problems. In: Doyle D, Hanks GWC, Mac- Donald N, et al., eds. Oxford textbook of palliative medicine, 2nd edition. New York: Oxford University Press, 1998: Malone BT, Beye R, Walker J. Management of pain in the terminally ill by administration of epidural narcotics. Cancer 1985;55: Gildenberg, PL. Considerations in the selection of patients for surgical treatment caused by malignancy. In: Arbit E, ed. management of cancer related pain. Mt Kisco, NY: Fuutura, 1993: Jacox A, Carr DB, Payne R, et al. Management of cancer pain. Clinical Practice Guideline No. 9. AHCPR Publication No Rockville, MD. Agency for Health Care Policy and Research, US Department of Health and Human Services, Public Health Service, March, Kravitz RL, Delafield JP, Hays RD, et al. Bedside charting of pain levels in hospitalized patients with cancer: a randomized controlled trial. J Pain Symptom Manage 1996;11: Bruera E, Schoeller T, Wenk R, et al. A prospective multicenter assessment of the Edmonton Staging System for cancer pain. J Pain Symptom Manage 1995;10: Ahles TA, Blanchard EB, Ruckdeschel JC. The multidimensional nature of cancer-related pain. Pain 1983;17: McGuire DB. Comprehensive and multidimensional assessment and measurement of pain. J Pain Symptom Manage 1992;7: Laupacis A, Sekar N, Stiell IG. Clinical prediction rules. A review and suggested modification of methodological standards. JAMA 1997;277: Barry HC, Ebell MH. Test characteristics and decision rules. Endocrinol Metab Clin North Am 1997; 26: Wasson JH, Sox HC, Neff RK, et al. Clinical prediction rules: applications and methodological standards. N Engl J Med 1985;313: Serlin RC, Mendoza TR, Nakamura Y, et al. When is cancer pain mild, moderate or severe? Grading pain severity by its interference with function. Pain 1995;61: Karnofsky DA, Burchenal JH. The clinical evaluation of chemotherapeutic agents in cancer. In: Macleod CM, ed. Evaluation of chemotherapeutic agents. New York: Columbia University Press, 1949: Daut RL, Cleeland CS, Flanery RC. Development of the Wisconsin Brief Pain Questionnaire to assess pain in cancer and other disease. Pain 1983; 17: Cella DF, Tulsky DS, Gray G, et al. The Functional Assessment of Cancer Therapy Scale: development and validation of the general measure. J Clin Oncol 1993;11: Chang VT, Hwang SS, Feuerman M, et al. The Memorial Symptom Assessment Scale Short Form. Validity and reliability. Cancer 2000;89: Avies AR, Sherbourne CD, Peterson JR, et al: Scoring manual: adult health status and patient satisfaction measures used in RAND s Health Insurance Experiment, A Rand Note N HHS. 22. Crum RM, Anthony JC, Bassett SS, et al. Population-based norms for the Mini-Mental State Examination by age and educational level. JAMA 1993;269: Portenoy RK. Cancer pain: pathophysiology and syndromes. Lancet 1992;339: Cherny NI, Portenoy RK. Cancer pain management. Current strategy. Cancer 1993;72: Chang VT, Hwang SS, Feuerman M, et al. Symptom and quality of life survey of oncology patients at a VA Medical Center. Cancer 2000;88: Hanley JA, McNeil BJ. A method of comparing the areas under receiver operating characteristic curves derived from the same cases. Radiology 1983; 148: Chang VT, Hwang SS, Kasimis B. Longitudinal documentation of cancer pain management outcomes: a pilot study at a VA medical center. J Pain Symptom Manage (in press). 28. Crombie IK, Davies HTO. Selection bias in pain research [editorial]. Pain 1998;74: Mercadante S, Maddaloni S, Roccella S, et al. Predictive factors in advanced cancer pain treated only by analgesics. Pain 1992;50: Moulin DE, Foley KM. A review of a hospitalbased pain service. In: Foley KM, Bonica JJ, Ventafridda V, eds. Proceedings of the 2nd International Congress on Cancer Pain. Amsterdam: Elsevier, 1990:

13 378 Hwang et al. Vol. 24 No. 4 October Banning A, Sjogren P, Henriksen H. Treatment outcome in a multidisciplinary cancer pain clinic. Pain 1991;47: Cohen RS, Ferrer-Brechner T, Pavlov A. Prospective evaluation of treatment outcome in patients referred to a cancer pain center. Clin J Pain 1985;1: Lossignol DA. Pitfalls in the use of opiates in the treatment of cancer pain. Support Care Cancer 1993;1: Max MB, Portenoy RK. Pain research: designing clinical trials in palliative care. In: Doyle D, Hanks GWC, MacDonald N, eds. Oxford textbook of palliative medicine. New York: Oxford University Press, 1993: Hwang SS, Chang VT, Kasimis B. Cancer breakthrough pain characteristics and responses to treatment at a VA Medical Center. Pain (in press). 36. Rutten EH, Crul BJ, van der Toorn PP, et al. Pain characteristics help to predict the analgesic efficacy of radiotherapy for the treatment of cancer pain. Pain 1997;69: