Prospective Validation of the Palliative Prognostic Index in Patients with Cancer
Article Outline
Abstract
The Palliative Prognostic Index (PPI) was devised and validated in patients with cancer in a hospice inpatient unit in Japan. The aim of this study was to test its accuracy in a different population, in a range of care settings and in those receiving palliative chemotherapy and radiotherapy. The information required to calculate the PPI was recorded for patients referred to a hospital-based consultancy palliative care service, a hospice home care service, and a hospice inpatient unit. One hundred ninety-four patients were included in the study, 43% of whom were receiving chemotherapy /or radiotherapy or both. Use of the PPI split patients into three subgroups based on PPI score. Group 1 corresponded to patients with PPI
≤4, median survival 68 days (95% confidence interval [CI] 52, 115 days). Group 2 corresponded to those with PPI>4 and ≤6, median survival 21 days (95% CI 13, 33), and Group 3 corresponded to patients with PPI>6, median survival five days (95% CI 3, 11). Using the PPI, survival of less than three weeks was predicted with a positive predictive value of 86% and negative predictive value of 76%. Survival of less than six weeks was predicted with a positive predictive value of 91% and negative predictive value of 64%. The PPI is quick and easy to use, and can be applied to patients with cancer, in hospital, in hospice, and at home. It may be used by general physicians to achieve prognostic accuracy comparable, if not superior, to that of physicians experienced in oncology and palliative care, and by oncology and palliative care specialists, to improve the accuracy of their survival predictions.
Key Words: Advanced cancer, prognostic factors, prognostic tool, palliative care
Introduction
Prediction of survival of patients who are terminally ill has a central role in decision making by the doctor regarding treatment, place of care, and timing of referral to palliative care. The provision of an accurate estimation of prognosis allows the patient to make a more informed decision regarding treatment options and facilitates their preparation for death and timely resolution of end-of-life issues. However, prognostication is not taught in medical schools, and doctors' survival predictions for their terminally ill patients are often wrong and usually optimistic. A prospective cohort study of 504 terminally ill patients and their 365 doctors found that only 20% of their predictions were accurate, whereas 63% were optimistic.1
To address this shortcoming, the Research Network of the European Association for Palliative Care (EAPC) established a working group charged with the task of providing evidence-based clinical recommendations concerning prognosis in patients with advanced cancer.2 Their examination of the literature identified performance status, symptoms of the anorexia–cachexia syndrome, dyspnea and delirium as the clinical signs and symptoms most closely associated with prognosis, and laboratory parameters (leukocytosis, lymphopenia, and high C-reactive protein) as being significantly associated with life expectancy. They found clinical predictions of survival to be associated with actual survival but limited by the strength of the association, with correlation coefficients varying from 0.2 to 0.65. They found evidence, however, that increased accuracy in prognostication is associated with experience in oncology and palliative care.
One of the six key recommendations of the working group was the systematic use by health workers of prognostic scores designed to divide patients into groups with significantly different survival times. They also recommended that a clinical prediction of survival (CPS) be used in partnership with attention to prognostic factors when assessing prognosis. The two prognostic scores or tools specifically considered in the report of the working group were the Palliative Prognostic Score (PaP) and the Palliative Prognostic Index (PPI). The PaP Score is based on Karnofsky Performance Status, the presence or absence of dyspnea and anorexia, white blood cell counts, and the clinician's prediction of survival, which is given significant weighting. It has been validated successfully, both in Italy and Australia, in hospital and hospice inpatients when used by experienced specialists in palliative medicine and oncology.3, 4, 5 Potential limitations of the PaP are the omission of delirium, the dependence on laboratory testing and the weighting given to the CPS. Researchers in a hospice setting in England who failed to validate the PaP Score highlighted the effect of inaccurate predictions by less experienced doctors on their results.6
The PPI was developed and successfully validated in hospice inpatients with advanced malignant disease by Morita et al. in Japan.7 They subsequently conducted a study, which demonstrated improved accuracy of physicians' survival predictions with use of the PPI.8 The PPI relies on assessment of performance status using the Palliative Performance Scale (PPS),9 oral intake, and the presence or absence of dyspnea, edema, and delirium, but does not require blood tests or incorporate a clinical prediction of survival (Table 1). The resulting score puts the patient into one of three groups, predicting survival of shorter than three weeks (PPI score greater than 6), shorter than six weeks (PPI score greater than 4), or more than six weeks (PPI score less than or equal to 4).
Table 1. Palliative Prognostic Index
| Performance Status/Symptoms | Partial Score |
|---|---|
| Palliative Performance Scale | |
| 10–20 | 4 |
| 30–50 | 2.5 |
| ≥60 | 0 |
| Oral intake | |
| Mouthfuls or less | 2.5 |
| Reduced but more than mouthfuls | 1 |
| Normal | 0 |
| Edema | |
| Present | 1 |
| Absent | 0 |
| Dyspnea at rest | |
| Present | 3.5 |
| Absent | 0 |
| Delirium | |
| Present | 4 |
| Absent | 0 |
Our study represents the first attempt to validate the PPI in a geographically and culturally different population, in patients receiving palliative radiotherapy and/or chemotherapy and those being cared for at home or in hospital.
Method
This study was carried out in a specialist palliative care service that included a consultancy service based in a large teaching hospital, a community-based hospice home care service and a six-bedded hospice inpatient unit. Approval for the study was granted by the Ethics and Medical Research Committee at St. Vincent's University Hospital. All patients with cancer, with the exception of those referred for assistance with symptom control who had potentially curable disease, referred to the service over a six-month period were included. Patients were assessed on one occasion only—at the time of first contact with the service. Demographics and information required to determine the PPI were recorded by the clinical nurse specialist or doctor who first assessed the patient. Experience in palliative care of the nurse specialists and doctors ranged from days to several years.
The five variables used to determine the PPI are as follows: oral intake, the presence or absence of edema, dyspnea at rest, delirium, and performance status, as measured by the PPS. The PPS measures physical performance and is measured in 10% decrements, from fully ambulatory and healthy (100%) to death (0%). The PPS score, oral intake, and presence of dyspnea at rest were recorded as reported by the patient. If this was not possible, they were determined by observation and by discussion with family or nursing staff. Patients receiving total parenteral nutrition or feeding via enterostomies were recorded as having a “normal” oral intake. Delirium was diagnosed according to the criteria outlined in the Diagnostic and Statistical Manual of Mental Disorders, 4th edition. Delirium was judged to be absent if considered to be caused by a single medication, as per the protocol of the original development and validation study of the PPI.
Information regarding date of death was obtained from each of the three facilities or from death notices and subsequently the actual survival, in days, from enrollment was calculated.
Application of the cutoff points of PPI 4 and PPI 6 splits the sample into three groups based on PPI score. Kaplan–Meier survival curves were constructed for each of the three groups and a Cox proportional hazards regression was used to examine the relationship between survival and PPI as a continuous covariate. Positive predictive value (PPV) and negative predictive value (NPV) of predictions of survival of less than three weeks and less than six weeks were calculated. Statistical analysis was conducted using R version 2.4.0.10
Results
A total of 201 patients were included in the study. Seven were lost to follow-up as a result of being discharged from, or voluntarily ceasing contact with, the services. Of the remaining 194 patients, 143 (73.7%) made first contact with the hospital-based team, 50 (25.8%) with the hospice home care team, and only one patient enrolled on admission to the hospice inpatient unit. This reflects the fact that almost all of the patients admitted to the inpatient unit were referred from the two other services and, therefore, already had been included in the study. A significant proportion of patients (43%) were receiving palliative radiotherapy and/or chemotherapy (Table 2). A minority (13%) scored 10%–20% on the PPS, consistent with being bedbound (Table 3). At the time of analysis, actual survival data were available for 151 (78%) patients who had died. Censored survival times were recorded at this time for the 43 (22%) remaining patients whose survival from enrollment to analysis ranged from 44 to 211 days.
Table 2. Patient Demographics
| Patient Demographics | n (%) |
|---|---|
| Total number of patients | 194 |
| Male | 100 |
| Mean age | 69.9 years |
| Having chemotherapy | 60 |
| Having radiotherapy | 36 |
| Both | 13 |
| Bronchial carcinoma | 53 (27) |
| Colorectal | 26 (14) |
| Breast | 20 (10) |
| Hematological | 15 (8) |
| Prostate/bladder/kidney | 16 (8) |
| Pancreas/hepatobiliary | 16 (8) |
| Gynecological | 12 (6) |
| Upper GI | 10 (5) |
| Other | 26 (14) |
Table 3. Patients' Performance Status and Clinical Symptoms
| Performance Status/Symptoms | n (%) |
|---|---|
| Palliative Performance Scale | |
| 10–20 | 25 (13) |
| 30–50 | 86 (44) |
| ≥60 | 83 (43) |
| Oral intake | |
| Mouthfuls or less | 32 (16.5) |
| Reduced but more than mouthfuls | 94 (48.5) |
| Normal | 68 (35) |
| Edema | |
| Present | 62 (32) |
| Absent | 132 (68) |
| Dyspnea at rest | |
| Present | 39 (20) |
| Absent | 155 (80) |
| Delirium | |
| Present | 18 (9) |
| Absent | 176 (91) |
Direct comparison of patient characteristics in this study with those in the study by Morita et al. shows that there are a number of demographic variances between the patients included in this study and those patients in whom the tool was originally developed and validated (Table 4).
Table 4. Comparison of Patient Characteristics with Those of Population in Which PPI Was Developed
| Characteristic | Morita et al.7 | Stone et al. |
|---|---|---|
| Mean age | 67 years | 70 years |
| Place of care | Hospice 100% | Hospital 73.7% |
| Home 25.8% | ||
| Hospice 0.5% | ||
| Palliative Performance Scale | 19% PPS≥60 | 43% PPS≥60 |
| Antineoplastic therapy | 0% | 43% |
| Cancer diagnosis | Lung 25% | Lung 27% |
| Upper GI 23% | Upper GI 5% | |
| Prevalence of delirium | 23% | 9% |
The PPI was used to split patients into three subgroups as follows: Group 1 corresponds to patients with PPI less than or equal to 4, Group 2 corresponds to those with PPI greater than 4 and less than or equal to 6, and Group 3 corresponds to patients with PPI greater than 6. A Kaplan–Meier curve was constructed for each of the groups. The median survival for Groups 1, 2, and 3 were 68, 21, and 5 days, respectively. The 95% confidence intervals (CI) for these are summarized in Table 5, and demonstrate the substantial statistically significant difference in the median survival times. The actual Kaplan–Meier survival curves for the three groups are shown in Fig. 1. A Cox proportional hazards regression was used to examine the relationship between survival and PPI as a continuous covariate; the Hazard Ratio associated with a one unit increase in PPI score is 1.36 (95% CI 1.29, 1.43), P<0.001.
Table 5. Characteristics of the Three Groups as Defined by PPI
| Group | n | Events | Median Survival in Days | 95% Confidence Interval for Median | |
|---|---|---|---|---|---|
| Lower Bound | Upper Bound | ||||
| 1 | 117 | 76 | 68 | 52 | 115 |
| 2 | 27 | 25 | 21 | 13 | 33 |
| 3 | 50 | 50 | 5 | 3 | 11 |
Fig. 1
Kaplan-Meier survival curves for each PPI group.
Using the PPI, survival of less than three weeks was predicted with a PPV of 86% and negative predictive value NPV of 76% (sensitivity 56%, specificity 94%). Survival of less than six weeks was predicted with a PPV of 91% and NPV of 64% (sensitivity 63%, specificity 92%) (Table 6).
Table 6. Accuracy of Predictions Using the PPI
| Positive Predictive Value (%) | Negative Predictive Value (%) | Sensitivity (%) | Specificity (%) | |||||
|---|---|---|---|---|---|---|---|---|
| Stone et al. | Morita et al.7 | Stone et al. | Morita et al.7 | Stone et al. | Morita et al.7 | Stone et al. | Morita et al.7 | |
| PPI>4 | 91 | 83 | 64 | 71 | 63 | 79 | 92 | 77 |
| PPI>6 | 86 | 80 | 76 | 87 | 56 | 83 | 94 | 85 |
Discussion
In this study, the first to independently validate the PPI, we have confirmed the ability of the PPI to divide a population of patients with advanced cancer into three groups, each with a significantly different survival profile. In addition, we have shown that the scope of validity of the PPI extends beyond the characteristics of the population in which it was originally developed, which comprised solely of patients in an inpatient hospice setting.
We identified a number of differences between our patient population and the population in which the tool was originally developed. In this study, 73.7% of patients' place of care was hospital, 25.8% home and 0.5% hospice at the time of enrollment in the study, whereas all patients included in the initial validation study were hospice inpatients. Cancer of the upper gastrointestinal tract accounted for 5% of patients in this study, compared with 23% of patients in the original study, reflecting recognized variances in cancer demographics between Japan and Western Europe. The mean age in this study was 70 years vs. 67 years in the original study. The incidence of recorded symptoms was comparable, with the exception of delirium (30% vs. 9%). Research into the incidence of delirium in patients with advanced cancer has shown it to be in the order of 26%–44% of admissions to hospitals and hospices.11 The comparatively low incidence of delirium and high proportion of people with good performance status in this study is likely to reflect the fact that many patients were referred to palliative care services earlier in their disease trajectory and that patients receiving antineoplastic therapy were included.
In this study, predictions of survival of less than three weeks using a PPI greater than 6 and of less than 6 weeks using a PPI greater than 4 were highly specific, with specificity and PPV greater than in the original study. The sensitivity and NPV, however, were less than in the original study. The high specificity and only moderate sensitivity in this study suggests that although the PPI has a high level of accuracy in those patients it identifies as having a short prognosis, it will not identify all patients with a short prognosis.
There is some evidence that experience in oncology or palliative care is associated with increased prognostic accuracy.2 In a study of accuracy of survival estimates, two physicians experienced in the care of patients with advanced cancer predicted four-week survival with a PPV of 73%–86% and an NPV of 60%–64%.12 The fact that the PPI predicts three-week survival with a PPV of 80%–86% and an NPV of 76%–87% suggests that it may be used by medical and nursing staff not experienced in oncology or palliative care to attain accuracy comparable, if not superior to that of physicians with significant experience. In addition, use of the PPI has been shown to improve the clinical predictions of survival by doctors experienced in palliative care.8 This finding of increased accuracy using a combination of a prognostic tool and the CPS has previously been demonstrated using the SUPPORT prognostic model.13
The EAPC working group on prognosis recommended the use of prognostic tools in combination with the CPS to estimate survival of patients with advanced cancer. The PPI is quick and easy to use and does not require blood testing. It incorporates all of the four clinical signs and symptoms shown to be most significantly associated with prognosis in patients with advanced cancer and does not require specialist knowledge to ensure accuracy. It may be used by general physicians and clinical nurse specialists to achieve prognostic accuracy comparable or superior to that of specialists in oncology and palliative care, and by specialists in oncology and palliative care to improve the accuracy of their predictions of survival. We have shown that it is valid for use in patients with advanced cancer who are receiving antineoplastic therapies in addition to those in the terminal phase. It has now been validated in patients with endstage cancer in a number of settings: hospital, hospice, and home.
In the recommendations concerning prognosis in patients with advanced cancer, the working group of the EAPC recommended use of the PaP Score over the PPI, as at that time it had been more comprehensively validated. The findings of our study add significantly to the knowledge base in this area of prognostication in cancer patients. We propose that the PPI has a number of potential advantages over the PaP Score in terms of ease of use, reliability, and breadth of validity. Further comparative work is required to establish whether the EAPC recommendations in respect of choice of prognostic scale are still valid.
References
- . Extent and determinants of error in doctors' prognoses in terminally ill patients: a prospective cohort study. Br Med J. 2000;320:469–473
- Prognostic factors in advanced cancer patients: evidence-based clinical recommendations – a study by the Steering Committee of the European Association for Palliative Care. J Clin Oncol. 2005;23:6240–6248
- A new Palliative Prognostic Score: a first step for the staging of terminally ill cancer patients. J Pain Symptom Manage. 1999;17:231–239
- . Independent prospective validation of the PaP Score in terminally ill cancer patients referred to a hospital-based palliative medicine consultative service. J Pain Symptom Manage. 2001;22:891–898
- . Diagnostic accuracy of the Palliative Prognostic Score in hospitalized patients with advanced cancer. J Clin Oncol. 2004;22:4823–4828
- . The accuracy of the clinical prediction of survival: a comparison of doctors' and nurses' estimations and the failure to validate the Palliative Prognostic Score (PaP). [abstract] Palliat Med. 2004;18:155
- . The Palliative Prognostic Index: a scoring system for survival prediction of terminally ill cancer patients. Support Care Cancer. 1999;7:128–133
- . Improved accuracy of physicians' survival prediction for terminally ill cancer patients using the Palliative Prognostic Index. Palliat Med. 2001;15:419–424
- . Palliative Performance Scale (PPS): a new tool. J Palliat Care. 1996;12:5–11
- . R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing; 2006;http://www.r-project.org
- . Delirium in advanced cancer patients. Palliat Med. 2004;18:184–194
- . Estimate of survival of patients admitted to a palliative care unit: a prospective study. J Pain Symptom Manage. 1992;7:82–86
- The SUPPORT prognostic model: objective estimates of survival for seriously ill hospitalised adults. Ann Intern Med. 1995;122:191–203
PII: S0885-3924(07)00798-1
doi:10.1016/j.jpainsymman.2007.07.006
© 2008 U.S. Cancer Pain Relief Committee. Published by Elsevier Inc. All rights reserved.
