A prospective multicenter assessment of the Edmonton staging system for cancer pain

348 Journal of Pain and STmptomM a ~ t

Vol 10 No. 5 lul7 1995

Or/g/ha/Art/de

A Prospective Multicenter Assessment of the Edmonton Staging System for Cancer Pain Eduardo Bruera, MD, Teresa Schoeller, MD, Roberto Wenk, MD, Tara MacEachern, RN, BScN, Silvana Marcelino, RN, J o h n Hanson, MSc, and Maria Suarez-Almazor, MD, PhD Edmonton General Hospital (E.B.), Cross C.mw.erlnstitute (E.B., T.M., fH.), Edmonton, Arena, C~na~" CentroDe Emutos E Pesquisas Onagogi~ (T.S., S.M.), Ftorianopotis, Bro.zi~"Argentinian Palliative Care Program (R. W.), San Nicolas, Argentina; and Healthcare Quality and Outcomes Resmrch Center (M.S-A.), University of Alberta, Edmonton, Alberta, Canada.

Abstract Two hundred and seventy-seven patients were admitted to this prospective multicenter study in order to assess the accuravy of a staging system for cancer pain. The staging system (SS) was completed by a trained physician during the initial constdtation. This system included the assessment of pain mechanism (PM, neuropathic versus nonneuropathi¢), pain charactaistic (PC, continuous versus incidental), previous opioid dose (OD), cognitive function (CF), psychological distress (PD), tolerance (T), past history of alcohol or drugs (A). During day 21, a final assessnmU of pain control was made. Agreement for staging was observed in 96% of mses for investigato~ 1 and 2 (kappa 0.76, P < 0.001), and in 84% of cases between iavestigators 1 and 3 (kappa 0.723, P < 0.001). 0 f 2 7 6 evaluable patients, 86/92 Stage I (good ~ ) patients achieved good PC (93%) versus 102/184 Stage II and Ill (poor prognosis) patients (55%, P < 0.001). Sensitivity and s p e c i ~ of the system werefound to be 0. 93 and 0.46, respectively. Univariate correlation found significant correlation baween pain control and all variables except CE. In logistic regression, CF and OD showed no significant correlation. We, therefore, propose a more simple SS of f we categor~ (PM, PC, PD, T, and A) and two stages (good and poor prognosis). We conclude that the SS is highly accurate in predicting patients with good prognosis, but patients with "poor prognosis" can still achieve good pain control in more than 50% of cases. J Pain Symptom Manage 1995;10:348-355. / ~ Won/s Canceg, pain, staging, analgesia

Ina,oduet/on The recognition of poor prognostic features has led to the development of sta~in[ systems Addressrepin~ mfu~ts to: Eduardo Bruera, MD, Palliative Care Program, Edmonton General Hospital, 11111 Jasper Avenue, Edmonton, Alberta, Canada TSK 0L4. Arreptedforpublicatio~'January 13, 1995. © U.S. Cancer Pain Relief Commiuee, 1995 Published by Elsevier, New York, New York

for different primary tumors. 1'2 Although these systems required frequent changes as knowledge about the biology of cancer developed, s'4 they allowed researchers to speak a common language and practitioners to apply treatments in a logical and predictable fashion. The precise definition of patient characteristics in clinical research trials results in accurate interpretation of data, successful 0885-$g24/95/$9.50 SSDI 0885-$924(95)00052-Z

Vot I0 No. 5 ~u~ 1995

Edmonton Stagin~ S~stem

application of therapies, and subsequent formulation of more-advanced clinical research studies. Unfortunately, there is no clinical staging system for cancer pain. Although the International Association for the Study of Pain has included neoplastic pain in its taxonomy, the most-important prognostic factors in cancer pain patients are not included because it is not the purpose of the taxonomy to establish prognosis, s The consequence of the absence of a comm o n staging system for cancer pain is that the results obtained by different groups with different therapeutic techniques are frequently impossible to compare. Most publications describe patients as having "pain due to cancer," although this statement is probably as simplistic and difficult to assess and interpret as "carcinoma of the breast." It is possible that large discrepancies in the efficacy of a given treatment between groups can simply result from different characteristics in the populations u n d e r study. Although pain is a subjective sensation and more difficult to assess than characteristics of the tumor, a n u m b e r of definable features are well known to influence its response to different treatments. A simple staging system based on these features that could be used by clinicians at the bedside might help the process of referral to Pain Services and the process of c o m p a r i n g results o f t r e a t m e n t strategies between different centers. This system might also help in the characterization of patient populations in clinical trials. We have previously described a simple system for the staging of pain in patients with cancer. 6 The purpose of this study was to assess the accuracy of the Edmonton Staging System for cancer pain in a multicenter prospective study.

Pat/ents and Methods Two h u n d r e d a n d seventy-seven patients with cancer pain were admitted to this noncontrolled, prospective study in three different centers. Patient characteristics from each of the three centers are summarized in Table 1. For the purpose of this study, cancer pain was

349

Tab/e 1 Patient Characteristics Number Age (years) Gender (female/male)

276 59-+14.22 147/103

Uor

Genitourinary Lung Gastrointestinal Breast Head and neck Hematological Unknown Sarcoma

Other

64 54 51 48 29 11 11 4 4

Treatment

Mean daily equivalent dose of parenteral morphine day 1 (mg) Mean daily equivalent dose of parenteral morphine day 21 (mg) Strong opioids (no. of patients) a Weak opioids (no/. of patients) b

130-+496 177-+522 229 47

Route (day 1)

Oral Parenteral

150 126

Route (day 21)

Oral 188 Parenteral 88 Strong opioids:morphine,hydromorphone,dow-release morphine, levorphanol,and meperidine. Weakopioids:propoxyphene,acetamlnophen+ codeine, acetaminophen + oxycodone,and aspirin + oxycodone. defined as pain directly related to the presence of a malignant tumor. Patients were admitted to the study consecutively when seen by one of the authors in the outpatient clinic or as inpatients for the first time. All patients were assessed and treated by the authors. After the initial consultation, this a t t e n d i n g physician c o m p l e t e d the staging form. O n e week after the initial consultation, all patients were contacted personally or by telephone for the assessment of pain intensity a n d dose of opioids required. Three weeks after the initial consultation, patients were seen by one of the investigators for final assessm e n t of pain control. The intensity of pain was determined by the patient using visual analogue scales (VAS, 0 = no pain; 100 = worst possible pain), and the treatment was decided by an experienced palliative care physician after a complete multidimensional assessment of pain. 7,s After the ini-

350

Bruera et aL

tiai assessment, patients were treated with a nonopioid analgesic for mild pain (low intensity, no previous opioid treatmen0; a so-called "'weak" opioid, usually codeine for moderate pain (usually less than 75 on a VAS of 0--100); and a so-called "strong" opioid, usually morphine or hydromorphone, for severe pain (usually greater than 75 on a VAS of 0--100, or previous exposure and no response to mild opioids). Adjuvant analgesic drugs were used following a common protocol in all cases. Amitriptyline (first-line) or mexiletine (secondline) was used for neuropathic pain. Carbamazepine (first-line) followed by phenytoin (second-line) or baclofen (second-line) were used for lancinating neuropathic pain. Senna plus docusate sodium were used as laxatives, supplemented as necessary by enemas. Metoclopramide (first-line) or d e x a m e t h a s o n e (second-line) was used for chronic emesis. No benzodiazepines were used for hypnosis. Haloperidol was used for agitated delirium. Dose and routes of administration (oral followed by subcutaneous) of opioids and adjuvants were the same for the three groups. All patients were treated following recommended procedures for the assessment and treatment of cancer pain. 7-a° In all cases, opioid analgesics were employed regularly or infused continuously. All patients had access to "rescue doses," (extra doses to be taken as needed, each dose being approximately 10% of the daily opioid dose). Adjuvant drugs were considered in all cases as well as radiotherapy, chemotherapy, anesthesiological or neurosurgical procedures, acupuncture, or transcutaneous nerve stimulation. All investigators underwent training at the Palliative Care Unit of the Edmonton General Hospital, and the analgesic procedures in use at the Unit were considered standard treatment. These consisted of at least once-a-day assessment of pain intensity, nausea, and sensation of well-being using VAS, as well as daily assessment of constipation and need for extra analgesic dose. In all cases, analgesics were administered regularly, and the regular daily dose was increased to include the n u m b e r of milligrams of rescue doses received during the previous day.

VoL I0 No. 5July 1995

Clinical Staging System After the first consultation, the treating physician completed a form that considered seven features in the patient's pain. A. Mechanism ofpain. It has been suggested that patients in whom nerve damage can be demonstrated as the mechanism for pain are less likely to i m p r o v e a f t e r p h a r m a cologic s'll-ls and nonpharmacologic treatments a4 than those who have nonneuropathic pain. Another category in this group includes those patients in whom a combination of neuropathic and nonneuropathic pain can be demonstrated. The patient is placed into one of the following groups: A1. V/seem/pa/~. Patients in this category have pain due to visceral involvement (for example, liver metastases): This pain is usually described as not well localized, "aching," or "dull." Occasionally, the patient can have a "cramp" characteristic (bladder, biliary, or urinary spasm). A2. Bone or soft tissue. This pain is usually described as an "ache" on the affected bone or soft tissue area, aggravated by pressure or movement, and usually is well localized. A3. Neuropathic pain. Pain is located in the region where the nerve or nerve route has been damaged, associated or not with m o t o r or sensory deficit, a u t o n o m i c changes, paresthesias, or paroxysmal episodes of pain; the pain usually has a "burning" or an "electrical" character. A4. M/xed. This category applies when components of both neuropathic and nonneuropathic pain can be identified. A3 must be present for the diagnosis of mixed pain. A5. Unknown. This category applies if, after clinical history, physical examination, and imaging techniques, the mechanism of pain remains uncertain. Although it is obvious that categories A1, A2, and A3 include several types of pain, it is not so obvious at the present time that subtypes of pain within each of these categories show different responses to treatments.

Vol. 10 No. 5 July 1995

Edmonton Sta~ing S~stem

B. Pain charaamistics. Intensity, duration, location, or radiation have not been correlated with different responses to treatment. One characteristic that has been associated with poor response to analgesics is onset with some maneuver or so-called incidental pain. 15'16 B1. Nonineidemal pain. Any other pattern of pain. B2. Inc/denta/pa/n. Pain is aggravated suddenly as a result of movements, swallowing, defecation, or urination. Pain control is usually excellent if the patient remains immobile or refrains from performing the pain-causing maneuver.

C. Prev/ou, op/o/d eeposure. Although there are several types of opioid analgesics, equivalent doses can be estimated using standard tables. 17 There is no evidence that one opioid is significan@ better than another for one particular type of pain. The previous dose of opioid that a patient received, however, has major prognostic implications particularly in reference to dose selection. The previous dose is a continuous variable, and, therefore, any break point must be artificial. We have chosen three doses. C1. Less than 60 rag of equivalent oral moq~ne/ day. This dose level would include patients receiving no opioid or most of the so
351

orientation, and simple mental tasks (counthag and adding), the patient is judged to have normal cognitive function. D2. Impaired cognitive functio~ The patient has dear alteration in some of the mental functions to a degree that makes the assessment of pain treatment unreliable. E. Psycholog/cal d/m~. The degree of psychological distress may be a major prognostic factor. Major depression, anxiety, or hostility can make the treatment more difficult. ~-~4 The assessment was again made at the bedside by the treating physician as part of the assessment of the pain syndrome. El. Pat/ents without major psycho~g/ca/d/stm,~ E2. Major psychological distress. Patients with somatization alone or somatization accompanied by symptoms such as depression, anxiety, hostility or neuroticism severe enough to jeopardize the success of the analgesic treatment. E Tdemnee. The need for an increase in the dose of opioid may be due to true pharmacologic tolerance (related to changes at a receptor level) or to increased pain. In any case, the rate of required dose escalation has prognostic implications. The rate of increase in dose can be calculated according to the following formula: Percentage daily increase of initial dose of equivalent oral morphine = initial dose / (final dose - initial dose) × 100 / Number of days of treatment This is a continuons variable, but for practical reasons, patients can be divided into two groups: F1. Increase of less than 5 % of initial dose/day. These patients can usually be managed well with dose adjustment on a weekly or bimonthly basis. There are patients in whom tolerance or other causes of dose escalation is not a serious problem in management. F2. Increase of 5 % or greater of initial dose~day. These patients need dose increases several times a week. In these patients, tolerance can be a serious management problem. All features of the staging system except F were completed after the inidal consultation. F (tolerance) was completed at least 1 week after the initial assessment.

352

Bruera et at

Tab/e 2 Clinical Stalling System Stage I: Good prognosis A1 (visceral) C1 (<60 mg morphine/day) B1 (nonincidental) D1 (normal cognition) E1 (no somatization) F1 (no tolerance) G1 (no alcohol or drugs) A2 (bone or soft tissue) C2 (60-300 mg morphine/day) Stage II: Intermediate prognosis (any patient who is not stage I or III) A4 (mixed, if not stage III) C$ (>300 mg morphine/day, if not stage III) D2 (cognitive failure, if not stage lII) A5 (unknown) Stage III: Poor prognosis A3 (neuropathic, any B-C-D-E-F-G) B2 (incidental, any A-C-D-E-F-G) E2 (somatization, any A-B-C,-D-F-G) F2 (tolerance, any A-B-C-D-E-F) G2 (alcohol or drugs, any A-B-C-D-E-F)

VoL 10 No. 5 Jul7 1995

pain and its treatment may be very difficult. In these patients, some treatments, such as opioids or psychoactive drugs, may aggravate a confusional syndrome. 9

Stage III: Poor prognosis. Patients less likely to respond well to any analgesic treatment. This stage includes any patient from groups A3, B2, E2, F2, or G2, n o matter what the o t h e r group classes are for that patient. Assessment of Pain Control D u r i n g the final a s s e s s m e n t at day 21, patients completed a VAS (0 = no pain, 100 = worst pain) and an ordinal scale (0 = no pain; 3 = severe pain). In addition, a complete medication review was made. After this review, the treating physician and the patient r e a c h e d consensus about o n e o f these two categories:

Good pain control. No pain most of the time and need for two or fewer rescue doses per day. Poor pain control. Mild to moderate-to-severe

G. Past h/story. A past history of addictive personality with demonstrated addiction to alcohol or drugs may predispose patients to opioid addiction, even if these drugs are given for the treatment of cancer pain. s The assessments of both pain and response to different treatments in these patients become extremely difficult.

pain most of the time and need for three or more rescue doses per day. The type of overall assessment was made by a treating physician without knowledge of the patient's stage. The method of overall assessment by an experienced physician and consensus with the patient was preferred to a certain VAS value because it better reflected the nature of routine practice.

G1. Negat/ve history for a/coho/ism or drug add/tO.on.

Statistical Analysis

G2. Positive historyfor alcoholian or drug addiction.

All of the investigators u n d e r w e n t training in E d m o n t o n for at least 3 weeks. During this period, a blinded cross-assessment o f 50 consecutive patients took place between investigator 1 a n d i n v e s t i g a t o r 2. A g r e e m e n t was observed in 96% o f cases (95% confidence interval, 55-97; kappa = 0.76; P < 0.001). Fifty consecutive b l i n d e d cross-assessments also took place between investigator 1 and investigator 3. A g r e e m e n t was observed in 84% of cases (95% confidence interval, 51-93; kappa 0.723; P < 0.001). Due to logistic problems, cross-assessments between investigator 2 and 3 could not be p e r f o r m e d . Sensitivity, specificity, and positive and negative predictive values were d e t e r m i n e d for each of the clinical stages o f pain. Chi-squared test was employed for the comparison o f proportions. Logistic regression was used to determine the correlation o f each of the independ e n t factors with o u t c o m e and the odds ratio

Construction of Clinical Stages All t h e p r o g n o s t i c f a c t o r s p r e v i o u s l y described were summarized in three prognostic stages (Table 2).

Stage I: Good prognosis. Patients m o r e likely to respond well to any analgesic treatment. This stage included patients from groups A1, A2, B1, C1, C2, D1, El, F1, and G1. Stage II: Intermediate prognosis. This stage includes patients from groups A4, AS, C3, or D2, whenever the remaining group classes d o not belong to Stage III. Patients in this group are those who do not fit Stages I or III. A4 and A5 are groups of unpredictable outcome. C3 includes patients less likely to respond to most treatments than do C1 and C2, but in whom a g o o d r e s p o n s e can still be a c h i e v e d . D2 i n d u d e s patients in whom the assessment o f

Vol. 10 No. 5 ~ul~ 1995

Edmonton Sta~in~ System

353

Tab/e 3 F'mai Pain Control According to Initial Stage Pain control

Good

Poor

Total

Stage I 86 True6 False92 Stages II and III 102 False+ 82 True + 184 Total 188 88 276 Sensitivity,TP/TP+FN = 0.93; specificity,TN/TN+FP = 0.46; +predictivevalue,0.93; -predictivevalue,0.44. for each of the variables. Data were analyzed according to the Statistical Package for the Social Sciences. 25

Negative predictive value, defined as the ratio of patients who achieved poor pain control and were Stage II or III (N = 82) to the total n u m b e r of patients who were Stage II or III (N = 184), was 0.44. Positive predictive value, defined as the n u m b e r of patients who achieved good pain control and were Stage I (N-- 86) to the total n u m b e r of individuals who were Stage I (N = 92), was 0.93. Table 4 summarizes the univariate correlations and logistic regression between each prognostic factor and pain control. The table shows that both C (previous dose) and D (cognitive function) did not show i n d e p e n d e n t correlation with pain control. The remaining five parameters showed strong i n d e p e n d e n t correlation. The relative risk for p o o r pain control in those patients who had the poor prognostic factor ranged between 2 and 5.86. To simplify the staging system, we removed the two prognostic factors with no independ e n t correlation. This new staging system is summarized in Table 5. Table 6 summarizes the results after application of the new staging system to our study population. Overall results, specificity, and sensitivity were not significantly different with this newer and simpler system.

Resu/ts Results in the 276 evaluable patients are summarized in Tables 3 and 4. Ninety-three percent of Stage I patients (86/92) achieved good pain control versus 55% of Stage II and III patients (102/184) (P < 0.001). Only 12 patients were considered Stage II. Because of the small n u m b e r of patients and because it had been decided that the categories previously included under Stage II would be part of the poor prognostic feature in the new system, these patients have been analyzed with Stage III patients in this report. Screening test parameters were determined for the staging system, taking the " t r u e " value for Stage I to be the achievement of good pain control and for Stage II or III to be the achievement of poor pain control. Sensitivity was d e f i n e d as the ratio o f patients who achieved poor pain control and were Stage II and III (N = 82) to these plus those who achieved poor pain control and were Stage I (N = 6). This value was 0.93. Specifidty was defined as the ratio of patients who achieved good pain control and were Stage I (N = 86) to these patients plus those who achieved good pain control and were Stage II and III (N= 102). This value was 0.46.

D/scuss/on In this prospective study, we assessed the effectiveness of a staging system in predicting the outcome of patients with cancer pain. We Tabk 4

Correlation Between Different Prognostic Factors and Pain Control

Logimc A (mechanism) B (incidental) C (previous dose) D (cognition) E (somatization) F (tolerance) G (alcohol or drugs)

Univariate

Regression

Relative Risk

<0.001 <0.001 0.05 0.72 0.009 <0.001 <0.001

<0.001 0.03 0.3 0.14 0.005 <0.001 <0.001

4.86 2.09 1.51 0.45 4.01 3.58 5.96

354

Bruera et a£

Tab/e 5

New S t , l ~ Symm Stage 1: Good prognosis A1 (visceral) B1 (nonincidental) E1 (no somatization) F1 (no tolerance) G1 (no alcohol or drugs) A2 (bone or soft tissue) Stage 2: Poor prognosis A3 (neuropathic, any DE-F-G) A4 (mixed, any DE-F-G) A5 (unknown, any F-E-F-G) B2 (incidental, any A-E-F-G) E2 (somatization, any A-B-F-G) F'2 (tolerance, any A-B-E-G) (;2 (alcohol or drugs, any A-B-E-F) C and D are eliminated.

chose a 3-week term for the duration of the study because it allowed for adequate titration of opioid analgesics and adjuvant analgesic drugs, and for consideration and administration of other treatments, such as nerve blocks, cordotomies or radiotherapy. A longer follow-up would result in fewer evaluable patients because of death, transfer, or confusion. It may also be complicated by a possible change in the nature of the original pain syndrome. The staging system could be applied after a standard consultation and did not extend the consultation more than 5-10 rain. All prognostic factors except for F (tolerance) could be determined during the initial assessment. Tob erance was determined in all cases within 7 days after the initial visit. Our data suggest that this system is accurate in predicting the outcome of patients with good prognosis. However, patients with more difficult pain syndromes can still achieve good pain control in almost one-half of the cases. These results are similar to those of other screening instruments such as the b o n e scan. ~6 The accuracy of this staging system could he improved by two main mechanisms: 1. More-sophisticated diagnostic instruments could be applied for the assessment of psy-

VoL lO No. 5 Jul7 1995

chological distress, tolerance, or addictive personalities. These more-sophisticated assessment techniques might allow for more discriminating power and, therefore, better prognostic accuracy. However, these techniques might significantly prolong the staging system and make it less likely to be utilized both in the staging of patients for treatment and in clinical research. 2 Another potential improvement in accuracy could occur by further reducing measurement error. Our data, however, suggest that, when the assessment is made by an experienced physician, a high level of agreement (96% and 84%, respectively) is observed. Further improvement is unlikely to have significantly impacted on the prediction of outcome. 2. The identification of new poor prognostic factors could improve the accuracy of a staging system without necessarily increasing significantly its complexity or duration. Several authors have described neuropathic pain (A), incidental pain (B), total suffering (E), and addictive personality (G) as factors contributing to intractable pain syndromes. 9-11.14-16 They are all included in Stage III of our system. To identify new prognostic factors, we will need to perform epidemiological trials in large numbers of patients. The results of these trials will likely change our staging of cancer pain in a similar way as the identification of prognostic factors such as hormone receptors modified the staging of breast cancer. 4 We believe that this staging system needs independent testing and validation, and that new data may require several changes. Our data, however, suggest that this is a simple and reliable instrument that might allow for better planning in the care of these patients and for b e t t e r d e s c r i p t i o n o f the characteristics patients admitted to different clinical trials.

Tab/e 6 Reauim When Applying the New System to the Population Pain control

Good

Poor

Total

Stage 1 (good) Stage 2 (poor) Total

98 True 90 False + 188

10 False78 True + 88

108 168 276

Sensitivity, 0.88; specificity,0.52.

VoL 10 No. 5 Jul~ 1995

Edmonton Staging System

R0r 1. American Joint Committee for Cancer Staging and End Result Reporting. Manual for staging of cancer. Chicago: American Joint Committee, 1977. 2. Paterson AHG. Clinical staging and its prognostic significance. In: Stall B, ed. Pointers to cancer prognosis. Dordrecht: Martinus Nijhoff, 1988:3748.

3. Begent R, Bagshawe K. Management of highrisk choriocarcinoma. Semin Oncol 1982;9:198203. 4. Valagussa P, Bonadonna G, Veronesi V. Patterns of relapse and survival following radical mastectomy. Cancer 1978;41:1170-1178. 5. ASP Subcommittee on Taxonomy. Classification of chronic pain. Pain 1986;(suppl 3):S1-$225. 6. Bruera E, Macmillan IL Hanson J, MacDonald RN. The Edmonton staging system for cancer pain: preliminary report. Pain 1989; 37:203-209. 7. Cancer pain relief and palliative care. Report of a WHO Expert Committee. World Health Organization Technical Report Series 804; Geneva: World Health Organization, 1990. 8. Foley IL The treatment of cancer pain. N Engl

J Med 1985;313:84-95. 9. Mackiewicz R, Vouckoms A, Martin J. Drug therapy of neuropathic pain. Clin J Pain 1985;1:3949. 10. Tasker R, Tsuda T, Hanrylyshin P. Clinical neurophysiological investigations of deafferentation pain. In: Bonica J, Lindblom V, Iggo A, eds. Advances in pain research and therapy, vol 5. New York: Raven, 1981:713-738.

355

14. Swedlow M, Cundill J. Intra-ventricular drugs used in the treatment of neuropathic pain: a comparison. Anaesthesia 1981;36:1129-1132. 15. Portenoy RK, Hagen NA. Breakthrough pain definition, prevalence and characteristics. Pain 1990;41:273-281. 16. Bruera E, Fainsinger R, MacEachern T, et al. The use of methylphenidate in patients with incident cancer pain receiving regular opiates: a preliminary report. Pain 1992;50:75-77. 17. Cancer pain: a monograph on the management of cancer pain. Health and Welfare Canada: Minister of Supply and Services Canada, Ottawa H42-2/5 1984E. 18. Bruera E, Chadwick S, Weinlick A, MacDonald RN. Delirium and severe sedation in patients with terminal cancer. Cancer Treat Rep 1987;71:787788. 19. Massie MJ, Holland J, Glass E. Delirium in terminally ill cancer patients. Am J Psychiatry 1983; 140:1048-1050. 20. Bruera E, Miller L, McCallionJ, et al. Cognitive failure in patients with terminal cancer: a prospective study. J Pain Symptom Manage 1992;7:192-195. 21. Bruera E, Fainsinger RL, Miller MJ, et al. The assessment of pain intensity in patients with cognitive failure: a preliminary report. J Pain Symptom Manage 1992;7:267-270. 22. SprockJ, Braff DL, Saccuzzo DP, et ai. The relationship of depression and though disorder in pain patients. BrJ Med Psychol 1983;56:351-360. 23. Cleeland CS. The impact of pain on the patient with cancer. Cancer 1984;$4:2635-2641.

11. Arner G, Meyerson BA. Lack of analgesic effect of opioids on neuropathic and idiopathic forms of pain. Pain 1988;33:11-23.

24. Dout RL, Cleeland CS. The prevalence and severity of pain in cancer. Cancer 1982;50:19131918.

12. Portenoy RIL Foley KM, Inturrisi CE. The nature of opioid responsiveness and its implications for neuropathic pain: new hypothesis derived from studies of opioid infusions. Pain 1990;43:273-86.

25. Statistical Package for the Social Sciences. Chicago: McGraw-Hill, 1986.

13. Max MB. Towards physiologically based treatment of patients with neuropathic pain. Pain 1990; 42:131-133.

26. Maucb PM, Drew MA. Treatment of metastatic cancer to bone. In: De Vita VT, Hellman S, Rosenberg SA, ed. Cancer: principles and practice of oncology. New York: JB Lippincott, 1985:21322141.