Predictive models in palliative care


  • Presented at the Inside Track Conference “Predictive Modeling in Prostate Cancer,” organized by the European School of Oncology, Venice, Italy, April 17-19, 2008.


It is important to identify prognostic and predictive factors concerning both life expectancy and quality of life in palliative care patients to facilitate ethical, clinical, and organizational decisions, but also to use resources in the best possible way. The authors reviewed the literature to identify the major factors that can predict survival of patients with solid tumors. They found only a few prospective assessments of prognostic factors. Clinical prognostic/predictors of survival based on physician's and/or nurse's judgment, performance status, dyspnea at rest, anorexia, dysphagia, or delirium are all considered to be of primary importance. Despite several contrasting findings, it is generally agreed that the type and site of the primary tumor and metastasis, psychosocial factors, and quality of life should be considered secondary to the organic effects in the final stages of life. Leukocytosis, lymphocytopenia, and elevated C-reactive protein are all reported to have prognostic significance, and low serum albumin and high lactate dehydrogenase levels must also be taken into consideration. Cancer 2009;115(13 suppl):3128–34. © 2009 American Cancer Society.

There have been several studies of prognostic indicators of survival for advanced cancer patients in the last 15 years, but to our knowledge prospective studies are still few. The most frequent population sampling settings are home care programs, hospices, palliative care units, and hospital wards. We have examined the identification of prognostic factors for the survival of patients with advanced, terminal prostatic cancer.

The majority of the evidenced-based data on prognostic indicators of survival were recently summarized by the Steering Committee of the European Association for Palliative Care1 and pertain to patients with various types of cancer receiving palliative care, with a median survival ≤90 days. Specific prognostication for prostatic cancer patients is very limited, and drawn either from subgroup analysis of randomized trials or from general prospective evaluation of the prognosis of all-types advanced cancer. Our examination of the literature therefore focused not only on advanced and terminal prostatic cancer, but on patients in the terminal stage of any cancer.

According to Parkes, the terminal period of care (median survival <12 weeks) is the period during which there is evidence of progressive malignancy and therapy cannot realistically be expected to prolong life significantly.2 Saunders3 and other authors4 describe the beginning of the terminal period as the time when goals must be redefined, and it is appropriate to shift from treatments aimed at control of the tumor to treatments primarily intended for symptom control.

The identification of prognostic and predictive factors concerning both life expectancy and quality of life in palliative care patients is very important to facilitate ethical, clinical, and organizational decisions, but also to use resources as efficiently as possible. In some countries, such as the United States, Canada, and Italy, a life expectancy of ≤6 months is required for admission to government-funded hospices or regional palliative care programs5-7 and to receive healthcare and benefits (Medicare Hospice Benefit in the United States; “Attendance Allowance” in the United Kingdom). However, predictive models in palliative care should mainly concern the prognostication of life expectancy with the following objectives:

  • To inform patients, families, and caregivers about the prognosis, in an appropriate, individually tailored way within the patient-physician relationship to involve them in the global decision making;

  • To make clinical decisions concerning withdrawing curative-intent oncological therapies;

  • To make clinical decisions regarding the use of oncologic therapies such as chemotherapy/radiotherapy and hormonotherapy and/or bisphosphonates, with palliative intent;

  • To minimize the risks of under- or overtreatment;

  • To define the “appropriate time” for initiating a certain palliative option;

  • To establish the most appropriate therapy for symptom control in each individual patient;

  • To select the most appropriate setting of care (hospital, hospice, home care); and

  • To answer the patient's question: “How long do I have, physician”?8-10 with the aim of helping patients, families, and caregivers make personal decisions and plans, naming someone to take decisions, and sorting out their financial affairs.

Studies focusing on prognostication of survival in palliative care have taken the following into consideration: clinical predictors of survival, performance status, some physical symptoms, some biologic markers, some psychosocial and economic variables, and tumor type and stage.1, 11-18 Two prognostic scores, the Palliative Prognostic (PaP) score and the Palliative Prognostic Index (PPI), have been constructed and validated.16-21

Despite contrasting findings in the literature, it is generally agreed that the type and site of primary tumor and metastasis, psychosocial factors, and quality of life are secondary considerations in relation to the clinical presentation in the final stages of life.22-25

The clinical prognostic/predictors of survival (CPS) are defined as the clinical assessment of the length of survival. Clinical trials frequently use CPS as inclusion/exclusion criteria, but the evaluation is based only on the treating investigator's clinical experience. CPS can be defined as “the physician's ability to estimate the probable survival potentially remaining to the patient.” Studies assessing the accuracy of the physician's predictions are conflicting and contradictory; however, this judgment is often imprecise when used alone, and some authors consider that it tends to overestimate survival by a factor of 3 to 5,1, 15, 26-44 and is overly pessimistic in only approximately 20% of cases. CPS are more accurate for short-term than long-term survival (“horizon effect”).15, 26, 42, 43 Therefore, for greater accuracy the CPS need to be reassessed at planned, regular intervals. In the prospective study by Vigano et al,12 CPS were found to be independently and strongly associated with survival.

CPS are valid tools, but must be considered in conjunction with more widely used criteria such as biochemical findings and the prognostic index to make the prediction more accurate, rather than as a single criterion on which to base the choice of therapeutic interventions or healthcare programs for terminally ill cancer patients.

Performance Status

All the various indices (Karnofsky performance status [PS] [KPS], palliative PS, Eastern Cooperative Oncology Group PS) of activity and functional autonomy are considered prognostically significant.1, 34 However, PS can potentially provide reliable short forecasts only in 2 situations: first, when the score is mainly in the middle to low range22, 43-46 and second, if the score suddenly drops steeply. A good PS is not a firm basis for long-range predictions, but it is the best tool for predicting survival with treatment. As is the case with the physician's judgment, the prognostic utility of PS increases in combination with an assessment of other symptoms.25, 32, 44


Delirium, anorexia, cachexia syndrome, dyspnea, and cognitive failure can be of some importance in predicting survival.1, 15, 16, 44 Most of these symptoms are not specific to any single disease, and in fact are part of the terminal pathway for almost all cancer patients.25 Cognitive function is important, because delirium or cognitive impairment has a significant place in survival prognostication.20, 33, 47-49

In a total of 181 patients, Teunissen et al50 prospectively investigated: 1) the role of symptoms in advanced cancer as independent prognostic factors for death in hospitalized patients in oncology wards who were referred to the palliative care team; 2) clusters of proven predictive symptoms (present or absent) as a means with which to improve prognostication; 3) KPS; 4) sociodemographic data collected by the physicians and nurses; and 5) the length of survival (the time between the first consultation and death or last follow-up). The correlations between sociodemographic variables, diagnosis, KPS, and symptoms occurring in ≥10% of cases were analyzed to define the prognostic indicators for survival. The patients reported a median of 4 symptoms (range, 1-8 symptoms). Pain, fatigue, anorexia, and anxiety were the most frequently reported, but multivariate analysis demonstrated that only nausea, dysphagia, dyspnea, confusion, and absence of a depressed mood were independent prognostic factors for survival. Patients with 2, 3, or all 4 of these symptoms at the same time had a 2.7 to 9.0 times higher risk of dying than patients without any of the symptoms. The presence of 4 of these significant symptoms resulted in 83% mortality at 1 month and 100% mortality at 6 months, compared with 20% and 48%, respectively, for patients without these clusters.

Among the symptoms with prognostic significance, dyspnea is extremely important. It has been defined as an uncomfortable awareness of breathing; it is an unpleasant subjective sensation and cannot be defined by the physical abnormalities that accompany it. Dyspnea appears to be common in patients with advanced cancer, and is reported frequently during the last weeks of life. The prevalence of dyspnea ranges from 21% to 79%.25, 51-54 Although it is obviously observed often among patients with lung cancer or pulmonary metastases, it is also frequent in patients with no demonstrable tumor involvement in the lung. Table 1 summarizes the studies on dyspnea and other prognostic indicators of survival.32, 53, 55-59

Table 1. Dyspnea and Other Prognostic Indicators of Survival
StudyNo. of PatientsType of CancerParameters StudiedCorrelation With SurvivalNotes
  1. KPS indicates Karnofsky performance status; PS, performance status.

Ruben 1986531592VariousKPS, dyspnea, dysphagia, weight loss, anorexia, dry mouthDyspnea is associated with the greatest expected survival reductionThe symptoms had independent prognostic value
Ventafridda 199055120VariousSymptoms difficult to control while keeping the patient alert and awareDyspnea, delirium, pain, vomiting, correlated with reduction in survivalThe symptoms had independent prognostic value
Krech & Walsh 19915625Inoperable pancreatic cancer diagnosed 1 mo previouslyPS, dyspnea, anorexia, dysphagia, weight loss, dry mouthOnly with dyspnea and PSPatients with dyspnea had a median survival of 2 mo; patients without dyspnea had a median survival of 5 mo
Heyse-Moore 199157292VariousDyspnea, primary cancer, sex, age, presence/absence of lung metastasesDyspnea significantly affected survival; significant inverse correlation noted between severity of dyspnea and survivalDyspnea was more prevalent in patients with carcinoma of the bronchus and breast
Hardy 199458107VariousDifferent symptoms, PS, expected outcomes, type of cancerDyspnea, immobility, lung cancer, decubitus gave the highest relative risk of deathThe parameters had independent prognostic value
Maltoni 199532540VariousDyspnea and other symptoms, PS, hospitalizationDyspnea was statistically correlated with a poor prognosis (median, 32 d)PS and hospitalization were independent predictors of survival
Escalante 199659122VariousType of cancer and survivalLung cancer, 4 wk; breast cancer, 22 wk; other cancer, 27 wkLung cancer patients with dyspnea had much shorter survival than patients with other cancers

Dyspnea may reflect the consequences of cancer cachexia and progress toward the terminal syndrome. There is evidence that good symptom control, even by an experienced palliative care team, is achieved less frequently for dyspnea than for other symptoms such as pain or nausea.52 In addition, research and education are limited on the adequate assessment and management of dyspnea in cancer patients.

Several reports list dyspnea in a significant percentage of advanced cancer patients with no intrathoracic malignancies. The National Hospice Study25 found a frequency of 24% in patients with no known lung or heart disease. Cachexia occurs in >80% of patients with advanced cancer.60


Leukocytosis, lymphocytopenia, and elevated C-reactive protein are all reported to have prognostic significance.1, 12, 14, 21 Cook et al13 evaluated survival in patients with hormone-refractory advanced prostate cancer and bone metastases participating in a multicenter clinical trial of zoledronic acid; they correlated the outcome in relation to specific markers of osteoblastic and osteoclastic activity such as serum bone-specific alkaline phosphatase (BAP) and urinary N-telopeptide. These 2 markers were found to be significantly correlated with short overall survival (P < .0001) on univariate analysis, whereas multivariate analysis confirmed that only high levels of serum BAP were able to predict short-term prognosis.

Barbot et al61 assessed clinical, laboratory, and subjective (patient's preferences) prognostic factors in 177 hospitalized patients with advanced solid tumors (7.3% with prostatic cancer) and with an estimated survival of <6 months. The patients' preferences were collected by means of a questionnaire and concerned their choice among 4 attitudes, ranging from very curative to very palliative. Clinical symptoms and biochemistry confirmed those reported in the literature. In the multivariate Cox analysis, number of metastases, KPS, serum albumin, and lactate dehydrogenase were found to have an independent prognostic value. Patients preferring a very palliative attitude survived for less time, with a median survival of 22 days versus 54 days for patients with curative expectations and 126 days for those with very curative expectations.

Prognostic Scores

Various prognostic scores have been developed.1, 19, 21, 62 Examples are the PaP score38 and the PPI,63 the terminal cancer prognostic (TCP),64 the poor prognostic indicator of Bruera et al,33 and the Chuang et al prognostic score.65

The PaP score has been validated in several countries, settings, and stages of solid cancers.16, 21, 38, 66 Survival times were measured from the date of study enrollment, and the outcome was death from all causes. The score includes clinical32 and biologic variables16 acknowledged as having independent prognostic value: 1) symptoms such as dyspnea and anorexia (presence or absence); 2) KPS; 3) the CPS of survival, ranging from 1 to 2 weeks up to 12 weeks; 4) total white blood cell count; and 5) lymphocyte counts (percentages; classified as normal, high, or very high). A numerical partial score is given to each variable, based on the relative weight of the independent prognostic significance shown by each single category in the multivariate analysis. The sum of the single scores gives the total, which can range from 0 to 17.5 (all values maximally altered). The higher the score, the lower the likelihood of survival at 30 days.

The PaP scores divide patients into 3 groups: A: total score of 0 to 5.5 (≥70% probability of survival at 30 days); B: total score of 5.6 to 11.0 (30%-70% probability of survival at 30 days); and C: total score of 11.1 to 17.5 (<30% probability of survival at 30 days). It was subsequently noted that the presence or absence of delirium divided patients into 2 further prognostic subgroups,20 because delirium and the PaP score were independently predictive of survival in patients with advanced cancer. The PaP score was also used in hospital in-patients with advanced cancer who were treated as oncology cases, not as part of a palliative care program.

The prognostic variables of the PPI are: KPS, amount of oral intake, presence/absence of edema, dyspnea at rest, and delirium. A numeric partial score was given to each variable, and the sum of the single scores gives the total, which ranges from 0 to 15. Patients were stratified into 3 groups depending on their score (Group A: PPI ≤2.0; Group B: PPI of 2.0 to ≤4.0; and Group C: PPI >4.0). Group B survived significantly longer than Group C, and Group A survived significantly longer than either of the others. When a PPI of >6 was adopted as a cutoff, 3-week survival was predicted with 80% sensitivity and 85% specificity. When a PPI of >4 was used as a cutoff, 6-week survival was predicted with 80% sensitivity and 77% specificity. In conclusion, the PPI can acceptably predict whether a patient will survive for longer than 3 or 6 weeks This index does not include the CPS, but Morita et al19 reported that when the 2 were used together, the prognosis was more accurate.

Other prognostic index have been developed; The assessment of Bruera et al included activity, pain, nausea, depression, anxiety, anorexia, dry mouth, dyspnea, dysphagia, weight loss, and cognitive status.33 However, the small number of patients analyzed means further validation is needed; in other cases, scores have been developed in some particular ethnicity,33, 65 and this situation also calls for confirmation.


Prognostic evaluation is based on different criteria in patients with advanced stage cancer compared with patients with earlier stages of disease, when it depends mainly on the primary site and histologic grade of the tumor.

CPS, PS, dyspnea, anorexia, dysphagia, delirium, leukocytosis, and lymphocytopenia are all considered factors of primary importance in survival prognosis in patients with advanced solid cancers. However, further research is still needed to better identify the prognostic factors for survival in palliative care patients at any stage of disease, with the aim of improving their quality of life and symptom control and personalizing the care setting, as well as for improving communication between the patients and their caregivers.

Conflict of Interest Disclosures

Sponsored by ASTRA Zeneca and the European School of Oncology.