Vol. 22 No. 5 November 2001 Journal of Pain and Symptom Management 891 Original Article Independent Prospective Validation of the PaP Score in Terminally Ill Patients Referred to a Hospital-Based Palliative Medicine Consultation Service Paul Glare, FRACP and Kiran Virik, FRACP Department of Palliative Care, Royal Prince Alfred Hospital, Sydney, New South Wales, Australia Abstract The aim of this prospective study was to validate the Palliative Prognostic (PaP) Score in a population of hospitalized patients in Australia in order to determine its applicability in a different setting to that in which it was originally developed. Individual PaP scores were calculated for 100 terminally-ill patients consecutively referred to a palliative medicine consultation service based in a university teaching hospital. The PaP score was able to subdivide this heterogeneous patient population into three groups, the differences being highly statistically significant. Median survivals for the three groups were, respectively, 60 days (95% confidence interval 41 89 days), 34 days (25 40), and 8 days (2 11). The percentage survival at 30 days for the three groups was 66%, 54%, and 5% respectively. These data suggest that the PaP scoring system is a reasonably robust method for prognostication in advanced cancer that appears to be independent of the setting. The short survival of the third group in this study, which is consistent with the presence of a subset of gravely ill patients within the hospital setting who are referred to specialist palliative care services very late in the course of their illness, raises important issues for the care and treatment of these individuals. J Pain Symptom Manage 2001;22:891 898 U.S. Cancer Pain Relief Committee, 2001. Key Words Prognosis, prognostic score, palliative care, survival, advanced cancer Address reprint requests to: Kiran Virik, FRACP, Department of Palliative Care, Royal Prince Alfred Hospital, Missenden Road, Camperdown 2050, Sydney, New South Wales, Australia. Accepted for publication: January 14, 2001. Introduction The importance of accurate prognoses in the care of patients with advanced cancer and other eventually fatal illnesses is being increasingly recognized. 1,2 Concurrent with this recognition is the acknowledgment that doctors and other health care professionals are not very accurate when they rely solely on clinical judgement to make their prognostications. 3 5 To try to improve prognostic accuracy, a number of methods have been developed. These range from using simple clinical measures like performance status 6 to applying complex mathematical formulae that are not suitable for routine use. 7 Pirovano et al. recently published details of a prognostic scoring system termed the Palliative Prognostic (PaP) score, which classifies patients with very advanced cancer into homogeneous risk groups for survival based on various clinical and laboratory parameters (namely, presence or absence of certain symptoms; performance status; clinician s prediction of survival [CPS]; white blood cell counts). 8 U.S. Cancer Pain Relief Committee, 2001 0885-3924/01/$ see front matter Published by Elsevier, New York, New York PII S0885-3924(01)00341-4
892 Glare and Virik Vol. 22 No. 5 November 2001 This method was subsequently validated in 451 patients entered into the hospice programs of fourteen Italian palliative care centers, illustrating its usefulness in clinical practice. 9 Consistent with the authors recommendation, trials of the PaP score in other settings and countries are now needed. The aim of this study was to independently validate the method in the acute care setting in a different country (Australia). Methods This prospective study was conducted in a single Australian center over a 4-month period. During this time, all patients referred to the palliative medicine consultative service based in a university teaching hospital who were seen by one of the authors were considered to be eligible for the study. Oncological services within the hospital cater to a broad spectrum of tumors at the specialist, tumor-specific level. Patients were subsequently excluded if they were not terminally ill (i.e., they did not have an ultimately fatal illness that was progressive and at an advanced stage). In patients with advanced disease, the PaP Score was determined on the day of first contact with the palliative medicine consultant during that admission. This score comprises 4 clinical and 2 laboratory parameters that are amenable to evaluation in hospitalized patients: 1) presence/absence of dyspnea; 2) presence/ absence of anorexia; 3) Karnofsky Performance Status (KPS); 4) Clinical Prediction of Survival (CPS); 5) white blood cell count (WBC); and 6) lymphocyte count. The CPS contains 5 categories dividing survival into periods under 12 weeks and one category for survival over 12 weeks. The first two parameters are ascertained by direct questioning of the patient, KPS and CPS are clinical estimates performed by the physician, and the last two parameters are available from the full blood count which is routinely performed in most hospitalized patients. A partial score is given for each of the six parameters, and the sum of these is the total score. The total score is then used to classify individual patients into high, intermediate, and low probabilities of surviving the next 30 days (Table 1). To maintain uniformity, high WBC and low lymphocyte counts were defined as in the original study (i.e., high WBC: 8500 11,000 cells/mm, 3 very high WBC: 11,000 cells/mm; 3 low lymphocyte percentage: 12.0 19.9%; very low lymphocyte percentage: 12.0%). All the clinical parameters of the PaP score, including the CPS, were scored by the same physician (possessing specialist knowledge and experience in the field). The blood counts were all analyzed by the same laboratory. Survival curves for the three prognostic risk groups were constructed using the Kaplan Meier method, and the log-rank test was calculated. All analyses were carried out using SPIDA software (1992, Statistical Computing Laboratory, Macquarie University, Sydney, Australia). Results Of the 104 subjects eligible, 4 were excluded as they were not judged to be terminally ill. Subject characteristics are reported in Tables 2 and 3. The median age was 66.5 years (range 16 92). Sex distribution revealed more males (M:F 1.4:1). A cancer diagnosis was present in 91 patients, of which the commonest subtype among the wide spectrum of primary tumors was lung cancer (15%). A non-cancer di- Table 1 PaP Score Partial score Dyspnea No 0 Yes 1 Anorexia No 0 Yes 1.5 Karnofsky performance status 30 0 10 20 2.5 Clinician s estimate of survival (weeks) 12 0 11 12 2 7 10 2.5 5 6 4.5 3 4 6 1 2 8.5 Total white cell count ( 10 9 /l) 8.5 0 8.6 11.0 0.5 11 1.5 Lymphocyte percentage (of total WCC) 20 40% 0 12 19.9% 1 12% 2.5 (Normal range: 20 40%) Risk groups Total score A (30 day survival probability 70%) 0 5.5 B (30 day survival probability 30 70%) 5.6 11 C (30 day survival probability 30%) 11.5 17.5
Vol. 22 No. 5 November 2001 Australian Prospective PaP Score Validation 893 Table 2 Patient Demographics Age: median 66.5 (range 16 92) Sex: M 59, F 41 Diagnosis 1. Cancer 91 (%) Lung 14 (15%) Hematological 9 (10%) Gynecological 8 (9%) Renal 8 (9%) Pancreas/Hepatobiliary 8 (9%) Colorectal 8 (9%) ACUP 6 (7%) Breast 5 (5.5%) Brain 5 (5.5%) Prostate 4 (4%) Melanoma 4 (4%) Upper GI 4 (4%) H&N 3 (3%) Miscellaneous 5 (5.5%) 2. Non-cancer (pathophysiological diagnosis) 9 (%) Multi-system failure (sepsis post-op, vasculitis, PVD a ) 3 (33%) AIDS 2 (22%) Heart failure (end-stage CCF b ) 1 (11%) Kidney failure (ESRF c due to hypertension) 1 (11%) Liver failure (Chronic liver disease complicated by hepato-renal syndrome) 1 (11%) Severe hypoxic brain damage 1 (11%) Median length of hospital stay: 11 days (IQR 6 19 days) Median time in hospital prior to PaP score: 3 days (IQR 1 8 days) Median time in hospital post PaP score: 6.5 days (IQR 3 12 days) Outcome of admission Home 54 (54%) Hospice transfer 18 (18%) Died 28 (28%) a Peripheral vascular disease. b Congestive cardiac failure. c End stage renal failure. agnosis was present in 9 patients. Patients were hospitalized for a median of three days prior to the PaP score being determined and left the hospital a median of 6.5 days later. Approximately half (54%) of the patients were discharged home while approximately a quarter (28%) of admissions culminated in death. Close to a quarter of the patients were clearly terminal, as evidenced by a KPS of 10 20 in 23% of patients. Anorexia and dyspnea were common (present in approximately half of the cases). As defined by the PaP scoring criteria, the haematological parameters were frequently abnormal. The median WBC count was 10.1 10 9 /l [interquartile range (IQR) 6.5 14.9] and the median lymphocyte count as a percentage of the total WBC count was 8.6% (IQR 5.0 17.2%). The survival status at 30 days post-prognostication is known for all but one subject, who is lost to follow-up (upon discharge from hospital, left to return home overseas and was censored at the day of discharge [Day 5]). At the time of analysis (1 August 2000), 85 of the remaining 99 (86%) evaluable patients had died and the remaining 14 survivors were censored on this day for the purpose of analysis. The follow-up times of these censored patients ranged from 138 to 249 days. The overall survival curve for the group is illustrated in Figure 1, with an estimated median survival of 30 days (95% CI: 24 40 days). This suggests that patients with quite advanced disease constituted a considerable portion of the sample. The existence of population heterogeneity within the sample is evidenced by 26% of patients surviving less than 2 weeks and 28% surviving more than 2 months. To validate the PaP Score method, the scoring procedure shown in Table 1 was applied. This resulted in: 42 subjects being categorized into Group A (30-day survival probability 70%), 37 into Group B (30-day survival probability intermediate), and 21 into Group C (30-day survival probability 30%). Kaplan Meier survival curves for the three risk groups are shown in Figure 2. Estimated median survival and rel-
894 Glare and Virik Vol. 22 No. 5 November 2001 Table 3 Main Clinical and Biochemical Characteristics of 100 Terminally Ill Hospital Patients 1. KPS Scores a n 10 20 23 30 77 30 40 38 50 34 unrated ( 20) 5 2. Clinical estimate of survival in weeks (CPS) n 12 32 11 12 6 7 10 25 5 6 13 3 4 12 1 2 13 3. Symptoms b n Dyspnea 46 Anorexia 57 4. Total White cell count ( 10 9 /l) n 8.5 41 8.6 11.0 13 11 46 5. Lymphocyte percentage n 20% 17 12 19.9% 17 12% 66 a KPS 10 moribund; fatal processes progressing rapidly; KPS 20 very sick; hospitalization necessary; active supportive treatment necessary; KPS 30 severely disabled; hospitalization is indicated, although death not imminent. b Excludes 10 patients who were too sick or confused to report their symptoms but were assumed to have both anorexia and dyspnea for the purpose of the PaP score. ative 95% CIs for the three groups were as follows: 60 days (41 89 days) for Group A (11 censored), 34 days (25 40 days) for Group B (3 censored), and 8 days (2 11 days) for Group C. There was a highly statistically significant difference in the survival rates between the three groups (log rank 74.87, P 0.0000001). The 30-day survival probability for each group, respectively, was 66%, 54% and 5%. The PaP Score risk groups and survival characteristics of the 9 patients without cancer is shown in Table 4 (one result censored). Of the nine non-cancer patients, 4 were in Group A and the remaining 5 in Group C, none being categorized with an intermediate 30-day survival probability (Group B). The estimated median survival (149.5 days) of the non-cancer Group A patients was considerably longer and the median survival of Group C (5 days) shorter than that of the group as a whole. The median survival of Group C may in reality be shorter than that seen as one of the 5 patients was receiving cardio-respiratory support in intensive care and thus the 11 day survival was artificially achieved. The 30-day survival probability in this subgroup of patients for Groups A and C, respectively, was 75% and 0%. The relative accuracy of the CPS is shown in Figures 3 and 4. CPS and actual survival are grouped according to the classification categories used in the PaP system: 1 2, 3 4, 5 6, 7 10, 11 12 and 12 weeks. A bias towards being overly optimistic was apparent. However, 45% of cases were categorized correctly and almost 70% were within one category of being classified correctly. Discussion The PaP scoring system demonstrated a predictive value for estimating survival in a sample of terminally-ill patients referred for palliative care, who were taken from both a different country and setting to that in which the system was developed. The concordance in the ability of these different data sets to differentiate groups of patients suggests that the PaP Score for prognostication in terminally ill patients possesses wide applicability, irrespective of the setting. This study concurs with the validation study by Maltoni et al. 9 in demonstrating the ability of the scoring system to divide a heterogeneous population into three groups with very different survival characteristics. This allows for a more tailored approach to the distinct therapeutic and care needs inherent in each group. Notably, patients who were referred to the palliative care service in the acute care setting were indeed terminal, as reflected by the similar survival characteristics to hospice and home care groups. Despite the different setting, the survival characteristics were very similar to the previous training and testing sets, 8,9 with the median estimated survival being approximately one month and almost a quarter of the patients being at each extreme of the survival times (less than 2 weeks and more than 2 months). Similar survival patterns were identified in patients enrolled in U.S. hospice programs by Christakis and Escarce, 10 suggesting a level of international uniformity in current palliative care referral patterns. The overall median survival found after enrollment in hospice programs (36 days), 10 in the training (32 days), 8 and testing set (33 days) 9 of the PaP scoring system falls within the 95% confidence interval
Vol. 22 No. 5 November 2001 Australian Prospective PaP Score Validation 895 Fig. 1. Overall survival of the 100 patients. (24 40 days) of our estimate of 30 days, thus corroborating this concept. Despite the similarities in the findings, there were important clinical differences between the patients in this study and those in the testing and training studies. One of the main differences was that the patients in this study were hospitalized, thus had a poorer overall state of health (Tables 2 and 3) compared to those previously studied: KPS scores were lower, and symptoms and hematological abnormalities were commoner. Consequently, although Groups A and B in this study had similar survivals to the previous samples, Group C had a worse survival. This suggests the presence within the hospital of a group of gravely ill patients (albeit a small number in this study) who will only be in contact with hospital-based palliative care services for a very brief period prior to death and in whom accurate prognostication is of clear merit with regards to decision-making by medical staff together with the patients and their families. The lower overall median survival found in this study as compared with those referred to previously may be due to the presence of the sicker patients in the acute care setting (Group C) exerting more influence in the overall estimate, although the effect of small sample size on our estimate must be borne in mind. While life expectancy is only one of many factors to influence clinical decision-making, the importance of accurate prognostication in estimating life expectancy in the acute care setting should not be underestimated. The more diverse options for life-prolonging treatment available to those that are hospitalized, compared to those who are being cared for at home, requires a judicious approach to guard against inappropriate use. In particular, if patients can be reliably identified who will die in the next few days irrespective of what is done to them, then burdensome, ineffective overtreatment may be avoidable. Likewise, undertreatment due to therapeutic nihilism in those with a good chance of survival may also be prevented. Such information may be particularly useful in ethical dilemmas such as writing Do Not Resuscitate orders. The PaP score depends heavily on the clinical prediction of survival (CPS). Given the limitations of CPS, 3 5 this is arguably a weakness of the PaP model. Significant variation in the CPS
896 Glare and Virik Vol. 22 No. 5 November 2001 Fig. 2. Survival of the three groups identified by the PaP Score. doubtless exists across prognosticators and is a function of experience and knowledge in the care of the terminally ill. However, a high degree of accuracy for CPS was obtained in this study, despite the known and previously documented bias to overestimation being observed. Despite its limitations, Pirovano et al. showed that the CPS does provide independent prognostic information. 8 This is presumably due to the ability of physicians to discriminate between patients with respect to their probability of survival, and this plausibly accrues from the integration of other clinical data which contribute to survival determination such as natural history of disease, rate of progression, response to treatment, comorbidities, and psychological issues. This ability to discriminate survival probabilities between patients appears to be distinct from the ability to accurately predict survival. Until more objective methods for Table 4 Survival Characteristics of the Nine Non-Cancer Patients PaP Score Risk Group A (30-day survival probability 70%) Pathophysiological diagnosis Observed survival (days) 1. Vasculitis 24 2. PVD 137 3. AIDS 162 4. AIDS 191 (censored) 75% (3/4) survived 30 days Overall observed median survival for group 149.5 days (IQR 52 184) PaP Score Risk Group C (30 day survival probability 30%) Pathophysiological diagnosis Observed survival (days) 1. CCF 1 2. hypoxic brain damage 2 3. decompensated CLD a 5 4. sepsis (multisystem organ failure) b 11 5. ESRF 15 0% (0/5) survived 30 days Overall observed median survival for group 5 days (IQR 1.5 13) a Chronic liver disease. b Receiving cardio-respiratory support in intensive care.
Vol. 22 No. 5 November 2001 Australian Prospective PaP Score Validation 897 Figure 3. Graphical representation of the clinical prediction of survival and actual survival. determining these factors are developed, these results acknowledge the continued place of CPS within the PaP scoring system. Despite the inherent subjectivity involved in the CPS, the results from this study support an overall consistency in the results of the CPS when used by physicians experienced in caring for terminally-ill patients. A prognostic scoring system reliant wholly on measured physiological or biochemical parameters is not appropriate for all terminally-ill patients (especially the subset who are gravely ill and in whom invasive tests are inappropriate) and thus the art and science of prognostication appear to marry well in the PaP scoring system. The applicability of the PaP score to a different setting lends itself to further exploration of its function as a prognostic measuring tool in other patient populations. Nine percent of the subjects in this sample did not have cancer. The PaP Score was able to divide this heterogeneous patient group into 2 distinct subgroups, each being more homogeneous for survival. The lack of patients with an intermediate probability (30 70%) of surviving 30 days (i.e., Group B) suggests that there are two clear temporal patterns of referral of non-cancer patients comparatively early or quite late in the course of the disease. The probability of survival at 30 days was accurately predicted in all but one of these (categorized as Group A but whose survival was only 24 days). The true median survival of non-cancer patients in Group C is most likely less then 5 days as in this small sample, because one of the 5 patients was receiving life-support in intensive care. This suggests that the median survival of these gravely ill patients (from the time of referral) is less than that of cancer patients, but a larger sample size is needed in order to reach any definitive conclusion. Prognostication in non-cancer patients with end-stage illnesses who seem to be approaching death is a well known problem, 2 and the validity of the PaP Score in patients with diseases such as heart failure, motor neuron disease, AIDS, and dementia merits further investigation. The alternative models of end-of life care proposed by the Institute of Medicine in its report Approaching Death 11 serve to raise awareness of the need to incorporate palliative care principles earlier in the course of ultimately fatal illnesses. It is anticipated that accurate prognostication will continue to be an important tool to augment patient care and health services planning within this framework. However, the PaP scoring system is unproven in such a population of earlier referrals and its utility in this setting remains to be explored. Figure 4. Difference between the clinical prediction of survival and actual survival according to classification categories of weeks as used in Figure 3. Each specified interval in weeks represents a category, for example, 1 2 weeks, 5 6 weeks. The difference between estimated and actual survival is therefore two survival categories. References 1. Christakis NA. Death foretold: prophecy and prognosis in medical care. Chicago: University of Chicago, 2000. 2. Von Gunten CF, Twaddle ML. Terminal care for noncancer patients. Clin Geriatric Med 1996;12: 349 358. 3. Parkes CM. Accuracy of predictions of survival in later stages of cancer. Br Med J 1972;2:29 31.
898 Glare and Virik Vol. 22 No. 5 November 2001 4. Vigano A, Dorgan M, Bruera E, Suarez-Almazor ME. The relative accuracy of the clinical estimation of the duration of life for patients with end of life cancer. Cancer 1999;86:170 176. 5. Christakis NA, Lamont EB. Extent and determinants of error in doctors prognoses in terminally ill patients: prospective cohort study. Br Med J 2000; 320:469 473. 6. Yates JW, Chalmer B, McKegney FB. Evaluation of patients with advanced cancer using the Karnofsky Performance Status. Cancer 1980;45:2220 2224. 7. Knaus WA, Harrell FE Jr, Lynn J, et al. The SUP- PORT prognostic model: objective estimates of survival for seriously ill hospitalised patients. Ann Intern Med 1995;122:191 203. 8. Pirovano M, Maltoni M, Nanni O, et al. A new palliative prognostic score: a first step in the staging of terminally ill cancer patients. J Pain Symptom Manage 1999;17:231 239. 9. Maltoni M, Nanni O, Pirovano M, et al. Successful validation of the palliative prognostic score in terminally ill cancer patients. J Pain Symptom Manage 1999;17:240 247. 10. Christakis NA, Escarse JJ. Survival of Medicare patients after enrollment in hospice programs. New Engl J Med 1996;335:172 178. 11. Committee on Care at the End of Life, Institute of Medicine; Field MJ, Cassel CK (eds). Approaching death: improving care at the end of life. Washington, DC: National Academy Press, 1997.