SEARCH

SEARCH BY CITATION

Keywords:

  • Cancer;
  • Prognosis;
  • Electronic Medical Record;
  • Palliative Care;
  • Advance Care Planning;
  • Hospitalized

Abstract

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. FUNDING SOURCES
  8. REFERENCES

BACKGROUND

This study sought to develop a predictive model for 30-day mortality in hospitalized cancer patients, by using admission information available through the electronic medical record.

METHODS

Observational cohort study of 3062 patients admitted to the oncology service from August 1, 2008, to July 31, 2009. Matched numbers of patients were in the derivation and validation cohorts (1531 patients). Data were obtained on day 1 of admission and included demographic information, vital signs, and laboratory data. Survival data were obtained from the Social Security Death Index.

RESULTS

The 30-day mortality rate of the derivation and validation samples were 9.5% and 9.7% respectively. Significant predictive variables in the multivariate analysis included age (P < .0001), assistance with activities of daily living (ADLs; P = .022), admission type (elective/emergency) (P = .059), oxygen use (P < .0001), and vital signs abnormalities including pulse oximetry (P = .0004), temperature (P = .017), and heart rate (P = .0002). A logistic regression model was developed to predict death within 30 days: Score = 18.2897 + 0.6013*(admit type) + 0.4518*(ADL) + 0.0325*(admit age) − 0.1458*(temperature) + 0.019*(heart rate) − 0.0983*(pulse oximetry) − 0.0123 (systolic blood pressure) + 0.8615*(O2 use). The largest sum of sensitivity (63%) and specificity (78%) was at −2.09 (area under the curve = −0.789). A total of 25.32% (100 of 395) of patients with a score above −2.09 died, whereas 4.31% (49 of 1136) of patients below −2.09 died. Sensitivity and positive predictive value in the derivation and validation samples compared favorably.

CONCLUSIONS

Clinical factors available via the electronic medical record within 24 hours of hospital admission can be used to identify cancer patients at risk for 30-day mortality. These patients would benefit from discussion of preferences for care at the end of life. Cancer 2013;119:2074–2080. © 2013 American Cancer Society.


INTRODUCTION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. FUNDING SOURCES
  8. REFERENCES

The majority of cancer patients, when asked about end-of-life care preferences, prefer to die at home rather than in a hospital.[1] In spite of this, approximately 30% of these patients die in the hospital, and 9% receive an intensive care unit level of care.[2] Cancer patients who die in the hospital may experience a substantially greater physical and psychological burden.[7] At the same time, families report significant distress associated with in-hospital death and an increased risk of developing posttraumatic stress and prolonged grief.[10, 11] Advance care planning is often limited to care at the close of life with discussion of hospice, location of death, advance directives, and health care surrogacy only being done when there are no further anticancer treatments available, patients are hospitalized, and/or patients are developing refractory, difficult-to-control symptoms.[12, 13] Improvement in end-of-life care and avoidance of undesired hospital deaths in patients who have advanced cancer might be achieved if patients with an increased risk of short-term mortality could be identified.

Physicians' ability to prognosticate is also flawed. On average, physicians tend to overestimate life expectancy by a factor of 3.[14] Although significant clinical symptoms and poor performance status can portend a poor prognosis (eg, dysphagia, weight loss); prediction of survival becomes less accurate and more difficult in patients with a good performance status.[18, 19]

Although there are prognostic instruments to help clinicians, the tools currently available have significant shortcomings. These include difficulty in clinical application and lack of validity in the inpatient hospital setting. For example, validated instruments including the Palliative Prognostic Score (PaP), the Palliative Prognostic Index (PPI), and more recently the Japan Palliative Oncology Study Prognostic Index, and Prognosis in Palliative Care Study require face-to-face clinical assessment to determine key prognostic factors including performance status, extent of disease and symptom burden.[20] The PaP, and Japan Palliative Oncology Study Prognostic Index each necessitates a clinical prediction of survival which, as pointed out above, has inherent flaws. The external validity of these measures may also be questionable, because most were not derived or rigorously applied in an inpatient hospital setting with a diverse population of patients.[25, 26] In fact, most of these studies were done on patients who were already determined to be terminally ill, making these tools less relevant for patients who are still receiving anticancer therapy.[22, 27] Tools such as the PaP and PPI were also developed at a time when the electronic medical record (EMR) could not be appropriately leveraged to aid in clinical decision-making. Chiang et al attempted to overcome this limitation using computer-assisted technology, but still focused on cancer patients who were already determined to be terminally ill and/or enrolled in hospice.[30] Finally, even the most vetted prognostic tools available do not meet the most rigorous criteria of high discriminatory ability (C statistic > 0.90), independent validation, and transportability.[31]

Given the disconnect between the expressed wishes of patients regarding where they wish to die and what actually happens, the lack of early discussion around advance care planning, along with the shortcomings of available prognostic measures, we sought to develop a prognostic tool that predicts 30-day mortality among hospitalized cancer patients, including patients still receiving active cancer therapy. The tool was constructed using data only from the EMR with the goal that every cancer patient can be screened within 24 hours of hospitalization.[32, 33] The rapid identification of patients with increased 30-day mortality allows health care providers to address the goals of care of these patients, identify patient and family needs, and provide care that is consistent with these goals.[34]

MATERIALS AND METHODS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. FUNDING SOURCES
  8. REFERENCES

Study Population

This is an observational cohort study that included all adult patients who were admitted to Northwestern Memorial Hospital, Chicago, Illinois, to the hematology/oncology service between August 1, 2008, and July 31, 2009. Patients admitted through malignant hematology, oncology, and bone marrow transplant were included. The final cohort consisted of 3188 patient encounters, with 3062 included in the final analysis. Repeat admissions were counted as separate incidents. The study was approved by the Institutional Review Board at Northwestern University and Northwestern Memorial Hospital.

Data Collection

Variables were considered for inclusion in the model if they had been associated with mortality within 30 days in the literature or were judged likely to be associated with mortality within 30 days by an expert panel including specialists in medical oncology, quality, and palliative care. Variables also had to be readily available in the EMR within 24 hours of admission. Variables meeting either of the first 2 criteria but which were not available in the EMR were not considered, because the objective of our study was to develop a tool based on electronic data only. Fifteen candidate variables included patient age, sex, admission type (elective versus emergent), number of previous admissions in the 3 months preceding the index hospitalization (≤ 1/> 1), scheduled current admission (yes/no), involuntary weight loss (yes/no), functional status as defined by difficulties with activities of daily living (ADLs; yes/no), temperature, heart rate, respiratory rate, oxygenation (yes/no), pulse oximetry, blood pressure, white blood cell (WBC) count, and bilirubin. Age, temperature, heart rate, respiratory rate, pulse oximetry, systolic blood pressure, and WBC count were treated as continuous variables. The remaining variables were treated as dichotomous. Because bilirubin was not routinely drawn on all cancer patients on day 1, it was missing for one-third of all patients, and thus was excluded from the final prediction models.

Data were collected from the Northwestern Medicine Enterprise Data Warehouse (NMEDW), a single integrated repository of all medical data sources on the campus used for facilitation of operations, patient care, and research reporting. A data abstraction instrument was developed by consensus from the research team (K.J.R., J.W.S., M.S., E.S., and S.W.) as well as our quality improvement team. The NMEDW was reviewed for all patients who met the study's inclusion criteria and for whom the data were complete. All relevant patient data were collected and stripped of identifying information by a quality specialist. The Social Security Death Index was used to determine 30-day mortality after hospital discharge.

Statistical Methods

The study included 3188 patient encounters. Using the full dataset, the 14 candidate variables were entered into a multiple logistic regression model with outcome (died versus survived within 30 days of discharge) as the dependent variable. This model resulted in 8 significant variables (P < .05): patient age, functional status as defined by difficulties with ADLs (yes/no), admission type, temperature, heart rate, oxygen use (yes/no), pulse oximetry, and systolic blood pressure. The logistic regression model with only these 8 variables resulted in a final sample size of 3062. This overall sample was randomly split into a derivation sample (n = 1531) and a validation sample (n = 1531). Continuous candidate variables were compared for the 2 groups (died versus survived) using an independent 2-sample t test. Dichotomous variables were compared using Fisher's exact test. For the 8 variables which were significant in the multivariate logistic model, parameter estimates, including standard errors, P values, and odds ratios (ORs), with 95% confidence intervals (CIs), are also reported. Within the derivation sample, discrimination of the model was assessed by the area under the receiver operating characteristic (ROC) curve (c = 0.791). The Hosmer-Lemeshow chi-square statistic was used to assess the model's calibration. Different thresholds corresponding to fixed specificities of 75%, 80%, 85%, 90%, and 95% were used to classify patients at high risk of dying. To test the model for robustness, these thresholds were evaluated in the derivation sample using sensitivity and positive predictive tests. To validate the prediction model, the derivation model was applied to the validation sample and the same training sample thresholds from the derivation model were evaluated in the validation sample using sensitivity and positive predictive tests. Positive and negative likelihood ratios are reported.

RESULTS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. FUNDING SOURCES
  8. REFERENCES

Patient Characteristics

The derivation sample had a death rate of 9.5%, whereas the validation sample death rate was 9.7%. A total of 294 patients of 3062 (9.60%) died within 30 days based on data from the Social Security Death Index.

Frequencies and descriptive summary statistics for both groups 1 and 2 (died versus survived) are reported in Table 1. Patients who died were more likely to be older (P < .0001). There were no differences between the 2 groups with regard to sex (P = .066). With regard to hospital admission data, we found that 28% of admissions were elective or routine among patients who survived versus 9% in patients who died (P < .0001). Equal percentages of patients in both groups had prior or scheduled admissions. Unintended weight loss was similar between groups. Functional status was different between the 2 groups, with 52% of patients who died having difficulty with ADLs as opposed to only 27% of those who survived (P < .0001). Patients who died were more likely to have vital sign abnormalities with regard to heart rate (tachycardia), blood pressure (hypotension), and pulse oximetry (hypoxia) (P < .05). They were also more likely to be on oxygen (41% versus 13%) (P < .0001). Patients who died had a higher WBC count (P = .04) compared with those who survived past 30 days.

Table 1. Baseline Characteristics of Study Participants in Derivation Samplea
CharacteristicDied (n = 145)Survived (n = 1386)Pa
  1. Abbreviations: ADL, activities of daily life; cont. var., continuous variable; SD, standard deviation; WBC, white blood cell count.

  2. n = 1531

  3. a

    Independent sample t test for continuous variables, represented by mean (SD); Fisher's exact test for dichotomous variables, represented by n (%).

  4. b

    Not included in prognostic model due to missing data.

Age at admission, mean (SD)66.37 (15.9)56.31 (15.9)<.0001
Sex (male), n (%)64 (44.1%)727 (52.5%).066
Admission type (elective), n (%)13 (9.0%)388 (28.0%)<.0001
Previous admission in the last 3 months (>1 month), n (%)40 (27.6%)381 (27.5%).999
Preadmit/current unscheduled (yes), n (%)36 (24.8%)226 (16.3%).014
Involuntary weight loss (yes), n (%)46 (32.2%)414 (30.2%).633
Functional status (ADL) (yes), n (%)76 (52.4%)381 (27.5%)<.0001
Temperature (cont. var.), mean (SD)97.63 (2.1)97.81 (1.5)0.329
Heart rate (cont. var.), mean (SD)97.54 (20.8)90.72 (18.5)<.0001
Systolic blood pressure (cont. var.), mean (SD)118.90 (25.4)124.2 (22.5).016
Pulse oximetry (cont. var.), mean (SD)95.37 (3.9)97.13 (2.5)<.0001
Oxygen use (yes), n (%)60 (41.4%)186 (13.4%)<.0001
Respiratory rate (cont. var.), mean (SD)19.10 (4.5)18.47 (3.4).106
WBC (cont. var.), mean (SD), n = 131912.82 (17.9)9.58 (16.6).043
Total bilirubin (≥ 2.0), n (%), n = 1053b21 (22.8%)98 (10.2%).0008
Total bilirubin (cont. var.), mean (SD), n = 1053b2.65 (5.8)1.29 (2.6).028

The variables independently associated with death within 30 days are presented in Table 2. These variables included increasing age (OR = 1.033; 95% CI = 1.02-1.05), functional status (difficulty with ADLs) (OR = 1.57; 95% CI = 1.07-2.31), and vital sign abnormalities including temperature, heart rate, systolic blood pressure, pulse oximetry, and oxygen use (OR = 2.37; 95% CI = 1.58-3.55). Emergent admission type was kept in the model at P = .059 (OR = 2.82; 95% CI = 0.98-3.41), because it was significant in the initial model with all candidate variables. The final logistic model was moderately discriminative (C statistic = .79). The Hosmer-Lemeshow goodness-of-fit test was not significant with P = .27.

Table 2. Multivariate Analysis of Derivation Model (Died Versus Survived)a
DescriptorEstimateStandard ErrorPOdds Ratio (OR)95% Confidence Interval for OR
  1. Abbreviation: ADL, activity of daily life.

  2. Intercept = 18.290 (standard error = 6.790)

  3. n = 1531, c = 0.791, Hosmer-Lemeshow chi-square = 9.97, P = .27

Age at admission0.0330.007<.00011.0331.02-1.05
Functional status (reference: No ADL)0.4520.198.02221.5711.07-2.31
Temperature−0.1460.061.01690.8640.77-0.97
Heart rate0.0190.005.00021.0191.01-1.03
Systolic blood pressure−0.0120.004.00240.9880.98-1.00
Pulse oximetry−0.0980.028.00040.9060.86-0.96
Oxygen use (reference: No)0.8620.207<.00012.3671.58-3.55
Admission type (reference: Elective)0.6010.319.05911.8250.98-3.41

We used this logistic regression model to obtain a score which predicted death within 30 days:

Score = 18.2897 + 0.6013*(admit type) + 0.4518*(ADL) + 0.0325*(admit age) − 0.1458*(temperature) + 0.019*(heart rate) − 0.0983*(pulse oximetry) − 0.0123 (systolic blood pressure) + 0.8615*(O2 on admission). The predicted probability of death is given by: exp(score)/[1 + exp(score)] where exp is the exponential function.

The median (range) of the scores was −2.88 (−5.36 to 1.11) for patients who survived and −1.73 (−4.15 to 2.39) for patients who died. Although the premodel prevalence of dying was 9.5%, the positive predictive value of this model was 22.9%. The ROC curve is given in Figure 1.

image

Figure 1. Receiver operating characteristic (ROC) curve is shown for the predictive 30-day mortality derivation model.

Download figure to PowerPoint

Based on an ROC analysis, the largest sum of sensitivity plus specificity was at a threshold of −2.09, where sensitivity was 63% and specificity was 78%. We altered our threshold to increase specificity and to minimize false positives. These results are presented in Table 3. As one proceeds down the rows of Table 3, fewer patients are classified as high risk. At a threshold of −1.0426, the sensitivity is 30%, the specificity is 95%, and the positive predictive value is 39.3%.

Table 3. Derivation Sample Model Performance by Varying Thresholda
ThresholdNo. Above ThresholdNo. of False PositivesNo. of False NegativesNo. of True PositivesNo. of True NegativesPositive Predictive ValueSensitivitySpecificity
  1. Total sample size = 1531.

−2.19924393455194104121.4%64.8%75.1%
−1.99843652775788110924.1%60.7%80.0%
−1.77542792057174118126.5%51.0%85.2%
−1.49312051377768124933.2%46.9%90.1%
−1.04261126810144131839.3%30.3%95.1%

In Table 4, the model was applied to the validation sample, with the sensitivity and positive predictive value evaluated at the same training thresholds. There was no statistical evidence of over-fit as demonstrated by sensitivity estimations in the validation sample. Using the threshold of −2.09, sensitivity was 67% and specificity was 79%. Above this threshold, 25.3% (100 of 395) of patients died within 30 days, whereas below this threshold, 4.3% (49 of 1136) of patients died. The likelihood ratio of a positive test (probability that score >−2.09 in deaths/probability that score >−2.09 in survivors) is 3.13. The likelihood ratio of a negative test (probability that score <−2.09 in deaths/probability that score <−2.09 in survivors) is 0.41. By varying across the same set of thresholds as in the test sample, sensitivity and positive predictive value on this independent data set compared favorably with those of the training sample, being at or above those of the training sample.

Table 4. Validation Sample Model Performance by Varying Thresholda
ThresholdNo. Above ThresholdNo. of False PositivesNo. of False NegativesNo. of True PositivesNo. of True NegativesPositive Predictive ValueSensi-tivitySpeci-ficityLR+LR-
  1. Total sample size = 1531.

  2. LR+ = Likelihood ratio positive = Sensitivity/(100% − Specificity)

  3. LR− = Likelihood ratio negative = (100% − Sensitivity)/Specificity

−2.199243032847102105423.7%68.5%76.3%2.890.41
−2.0939529549100108725.3%67.1%78.6%3.130.42
−1.99843612675594111526.0%63.1%80.7%3.270.46
−1.77542892046485117829.4%57.0%85.2%3.850.50
−1.49312251507475123233.3%50.3%89.1%4.610.56
−1.0426118669752131644.1%34.9%95.2%7.270.68

DISCUSSION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. FUNDING SOURCES
  8. REFERENCES

In this study, we were able to identify clinical and administrative factors readily obtainable through the EMR that were associated with an increased risk of 30-day mortality among cancer patients admitted to a hematology/oncology or bone marrow transplant service. Clinical factors at time of hospital admission associated with an increased risk of death included older age and impairment in functional status. Variables obtained on hospital admission that were significant included vital sign abnormalities (temperature, heart rate, systolic blood pressure, pulse oximetry), and oxygen use. Finally, admission type (elective versus emergent) was also associated with 30-day mortality. Our findings support the feasibility of using the EMR data to identify cancer patients on day 1 of admission who are at a higher risk of 30-day mortality.

Similar to prior studies, we found that certain clinical factors such as performance status, oxygen use, and patient age are associated with a higher risk of patient death.[21, 22, 35] Interestingly, some laboratory parameters that would be expected to be associated with mortality, such as WBC count, were not significant in the multivariate model.[21, 38] We differed from prior studies in that we excluded factors that we believed would require face-to-face clinical assessment, such as extent of disease, intake, clinical signs (ie, ascites) or symptoms, and clinical prediction of survival.[20, 28, 37, 39, 40] These factors would limit the utility of this tool as an automated screen.

Two key factors distinguish our study from prior research in this area: 1) a focus on hospitalized cancer patients who have not yet progressed to intensive care, and 2) use of the EMR exclusively to glean data for the predictive model.

There is a dearth of prognostication models developed for the inpatient setting prior to admission into intensive care. Most of the well-known models that have been validated for patients with cancer were initially developed in the home-hospice or palliative care setting. This includes well-known prognostic tools such as the PaP, the PPI, as well as the Palliative Performance Scale.[21, 22, 28, 39, 40] Consequently, these scoring systems are better predictors in patients with a shorter prognosis, and are less well-suited for patients with the potential for a longer prognosis. A number of prognostic scoring systems/prediction models, such as the Mortality Prediction Model, Acute Physiology and Chronic Health Evaluation (APACHE) II, or the Simplified Acute Physiology Score (SAPS) II,[38, 41] are used in cancer patients who are already admitted into intensive care units Although these models are useful for the medical team, they are of limited value to patients and families. Patients who have progressed to intensive care are frequently unable to contribute substantively to a discussion of their goals and values. Thus, identifying patients who are at risk of mortality in the near term while they are still able to clearly communicate their end-of-life wishes is imperative.

Amarasingham et al developed a real-time electronic predictive model based on hospital admission data to predict 30-day mortality or readmission for patients with heart failure.[48] This was also done by Tierney et al in an outpatient cardiac patient population.[32] Our model is the first of its type we are aware of for cancer patients. Similar to Amarasingham et al, we show that a simple equation using baseline admission data can identify cancer patients at higher risk for 30-day mortality. This type of model would facilitate large-scale screening of cancer patients on day 1 of admission and allow physicians to identify patients who may benefit from further assessment of their prognosis via clinical evaluation. Patients who screen positive and are determined to have a short prognosis based on clinical judgment would benefit from further goals of care conversations and advance care planning.[17, 19]

Our study has several limitations. The study was conducted at a single tertiary care institution, which limits generalizability to other hospitals and settings. Our data relies on the EMR with inherent variability in accuracy, availability of data (eg, missing data), and standard assessments that may not conform with our clinical assessments (ie, nursing determination of functional status via ADLs versus ECOG or Karnofsky performance status). Our assumption would be that as the EMR evolves and becomes more clinically salient, tools such as these can also become more accurate and relevant. Our findings are based on a retrospective cohort analysis, and future studies need to prospectively validate these results. Also, the clinical utility of this tool that includes patient, caregiver, and system-related outcomes needs to be established. In an effort to address some of these limitations, particularly the exclusion of potentially important variables due to missing data, we plan to refine this instrument in future analyses that use larger and more complete data sets.

In this study, we establish the feasibility of using the EMR to identify cancer patients on hospital admission who are at increased risk for 30-day mortality by fitting the model on a training sample and validating it on an independent test sample. This approach has the potential to rapidly and continuously screen all cancer patients admitted to the medical/surgical units. This is particularly important near the end of life, when patient, family, and medical team discussions are needed to address the preferred location of death, goals of symptom management, use of hospice, and hospital readmission. The application of such an approach can help clinicians readily identify patients with a high risk of short-term mortality to ensure prognosis is integral to development of the care plan. This will avoid potentially harmful interventions, and ensure incorporation of patient goals and values into the patient's plan of care.

FUNDING SOURCES

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. FUNDING SOURCES
  8. REFERENCES

No specific funding was disclosed.

CONFLICT OF INTEREST DISCLOSURE

The authors made no disclosure.

REFERENCES

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. FUNDING SOURCES
  8. REFERENCES
  • 1
    Bruera E, Russell N, Sweeney C, Fisch M, Palmer JL. Place of death and its predictors for local patients registered at a comprehensive cancer center. J Clin Oncol. 2002;20:2127-2133.
  • 2
    Higginson IJ, Sen-Gupta GJ. Place of care in advanced cancer: a qualitative systematic literature review of patient preferences. J Palliat Med. 2000;3:287-300.
  • 3
    Higginson IJ, Astin P, Dolan S. Where do cancer patients die? Ten-year trends in the place of death of cancer patients in England. Palliat Med. 1998;12:353-363.
  • 4
    Townsend J, Frank AO, Fermont D, et al. Terminal cancer care and patients' preference for place of death: a prospective study. BMJ. 1990;301:415-417.
  • 5
    Earle CC, Neville BA, Landrum MB, Ayanian JZ, Block SD, Weeks JC. Trends in the aggressiveness of cancer care near the end of life. J Clin Oncol. 2004;22:315-321.
  • 6
    Sharma G, Freeman J, Zhang D, Goodwin JS. Trends in end-of-life ICU use among older adults with advanced lung cancer. Chest. 2008;133:72-78.
  • 7
    Wright AA, Keating NL, Balboni TA, Matulonis UA, Block SD, Prigerson HG. Place of death: correlations with quality of life of patients with cancer and predictors of bereaved caregivers' mental health. J Clin Oncol. 2010;28:4457-4464.
  • 8
    Nelson JE, Meier DE, Oei EJ, et al. Self-reported symptom experience of critically ill cancer patients receiving intensive care. Crit Care Med. 2001;29:277-282.
  • 9
    Teno JM, Clarridge BR, Casey V, et al. Family perspectives on end-of-life care at the last place of care. JAMA. 2004;291:88-93.
  • 10
    Wright AA, Zhang B, Ray A, et al. Associations between end-of-life discussions, patient mental health, medical care near death, and caregiver bereavement adjustment. JAMA. 2008;300:1665-1673.
  • 11
    Billings JA, Kolton E. Family satisfaction and bereavement care following death in the hospital. J Palliat Med. 1999;2:33-49.
  • 12
    Daugherty CK, Hlubocky FJ. What are terminally ill cancer patients told about their expected deaths? A study of cancer physicians' self-reports of prognosis disclosure. J Clin Oncol. 2008;26:5988-5993.
  • 13
    Keating NL, Landrum MB, Rogers SO Jr, et al. Physician factors associated with discussions about end-of-life care. Cancer. 2010;116:998-1006.
  • 14
    Lamont EB, Christakis NA. Prognostic disclosure to patients with cancer near the end of life. Ann Intern Med. 2001;134:1096-1105.
  • 15
    Christakis NA, Lamont EB. Extent and determinants of error in doctors' prognoses in terminally ill patients: prospective cohort study. BMJ. 2000;320:469-472.
  • 16
    Parkes CM. Accuracy of predictions of survival in later stages of cancer. Br Med J. 1972;2:29-31.
  • 17
    Vigano A, Dorgan M, Buckingham J, Bruera E, Suarez-Almazor ME. Survival prediction in terminal cancer patients: a systematic review of the medical literature. Palliat Med. 2000;14:363-374.
  • 18
    Bruera E, Miller MJ, Kuehn N, MacEachern T, Hanson J. Estimate of survival of patients admitted to a palliative care unit: a prospective study. J Pain Symptom Manage. 1992;7:82-86.
  • 19
    Glare P, Virik K, Jones M, et al. A systematic review of physicians' survival predictions in terminally ill cancer patients. BMJ. 2003;327:195-198.
  • 20
    Hyodo I, Morita T, Adachi I, Shima Y, Yoshizawa A, Hiraga K. Development of a predicting tool for survival of terminally ill cancer patients. Jpn J Clin Oncol. 2010;40:442-448.
  • 21
    Pirovano M, Maltoni M, Nanni O, et al. A new palliative prognostic score: a first step for the staging of terminally ill cancer patients. Italian Multicenter and Study Group on Palliative Care. J Pain Symptom Manage. 1999;17:231-239.
  • 22
    Morita T, Tsunoda J, Inoue S, Chihara S. The Palliative Prognostic Index: a scoring system for survival prediction of terminally ill cancer patients. Support Care Cancer. 1999;7:128-133.
  • 23
    Gwilliam B, Keeley V, Todd C, et al. Development of prognosis in palliative care study (PiPS) predictor models to improve prognostication in advanced cancer: prospective cohort study. BMJ. 2011;343:d4920.
  • 24
    Gripp S, Moeller S, Bölke E, et al. Survival prediction in terminally ill cancer patients by clinical estimates, laboratory tests, and self-rated anxiety and depression. J Clin Oncol. 2007;25:3313-3320.
  • 25
    Tassinari D, Montanari L, Maltoni M, et al. The palliative prognostic score and survival in patients with advanced solid tumors receiving chemotherapy. Support Care Cancer. 2008;16:359-370.
  • 26
    Maltoni M, Nanni O, Pirovano M, et al. Successful validation of the palliative prognostic score in terminally ill cancer patients. Italian Multicenter Study Group on Palliative Care. J Pain Symptom Manage. 1999;17:240-247.
  • 27
    Maltoni M, Pirovano M, Scarpi E, et al. Prediction of survival of patients terminally ill with cancer. Results of an Italian prospective multicentric study. Cancer. 1995;75:2613-2622.
  • 28
    Yun YH, Heo DS, Heo BY, Yoo TW, Bae JM, Ahn SH. Development of terminal cancer prognostic score as an index in terminally ill cancer patients. Oncol Rep. 2001;8:795-800.
  • 29
    Stone P, Kelly L, Head R, White S. Development and validation of a prognostic scale for use in patients with advanced cancer. Palliat Med. 2008;22:711-717.
  • 30
    Chiang JK, Cheng YH, Koo M, Kao YH, Chen CY. A computer-assisted model for predicting probability of dying within 7 days of hospice admission in patients with terminal cancer. Jpn J Clin Oncol. 2010;40:449-455.
  • 31
    Yourman LC, Lee SJ, Schonberg MA, Widera EW, Smith AK. Prognostic indices for older adults: a systematic review. JAMA. 2012;307:182-192.
  • 32
    Tierney WM, Takesue BY, Vargo DL, Zhou XH. Using electronic medical records to predict mortality in primary care patients with heart disease: prognostic power and pathophysiologic implications. J Gen Intern Med. 1996;11:83-91.
  • 33
    Evans C, McCarthy M. Prognostic uncertainty in terminal care: can the Karnofsky index help? Lancet. 1985;1:1204-1206.
  • 34
    Morrison RS, Meier DE. Clinical practice. Palliative care. N Engl J Med. 2004;350:2582-2590.
  • 35
    Chow E, Harth T, Hruby G, Finkelstein J, Wu J, Danjoux C. How accurate are physicians' clinical predictions of survival and the available prognostic tools in estimating survival times in terminally ill cancer patients? A systematic review. Clin Oncol (R Coll Radiol). 2001;13:209-218.
  • 36
    Maltoni M, Caraceni A, Brunelli C, et al; Steering Committee of the European Association for Palliative Care. 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.
  • 37
    Chow E, Fung K, Panzarella T, Bezjak A, Danjoux C, Tannock I. A predictive model for survival in metastatic cancer patients attending an outpatient palliative radiotherapy clinic. Int J Radiat Oncol Biol Phys. 2002;53:1291-1302.
  • 38
    Staudinger T, Stoiser B, Müllner M, et al. Outcome and prognostic factors in critically ill cancer patients admitted to the intensive care unit. Crit Care Med. 2000;28:1322-1328.
  • 39
    Chuang RB, Hu WY, Chiu TY, Chen CY. Prediction of survival in terminal cancer patients in Taiwan: constructing a prognostic scale. J Pain Symptom Manage. 2004;28:115-122.
  • 40
    Suarez-Almazor ME, Newman C, Hanson J, Bruera E. Attitudes of terminally ill cancer patients about euthanasia and assisted suicide: predominance of psychosocial determinants and beliefs over symptom distress and subsequent survival. J Clin Oncol. 2002;20:2134-2141.
  • 41
    Kress JP, Christenson J, Pohlman AS, Linkin DR, Hall JB. Outcomes of critically ill cancer patients in a university hospital setting. Am J Respir Crit Care Med. 1999;160:1957-1961.
  • 42
    Groeger JS, Lemeshow S, Price K, et al. Multicenter outcome study of cancer patients admitted to the intensive care unit: a probability of mortality model. J Clin Oncol. 1998;16:761-770.
  • 43
    Guiguet M, Blot F, Escudier B, Antoun S, Leclercq B, Nitenberg G. Severity-of-illness scores for neutropenic cancer patients in an intensive care unit: Which is the best predictor? Do multiple assessment times improve the predictive value? Crit Care Med. 1998;26:488-493.
  • 44
    Sculier JP, Paesmans M, Markiewicz E, Berghmans T. Scoring systems in cancer patients admitted for an acute complication in a medical intensive care unit. Crit Care Med. 2000;28:2786-2792.
  • 45
    Benoit DD, Vandewoude KH, Decruyenaere JM, Hoste EA, Colardyn FA. Outcome and early prognostic indicators in patients with a hematologic malignancy admitted to the intensive care unit for a life-threatening complication. Crit Care Med. 2003;31:104-112.
  • 46
    Soares M, Fontes F, Dantas J, et al. Performance of six severity-of-illness scores in cancer patients requiring admission to the intensive care unit: a prospective observational study. Crit Care. 2004;8:R194-R203.
  • 47
    Groeger JS, White P Jr, Nierman DM, et al. Outcome for cancer patients requiring mechanical ventilation. J Clin Oncol. 1999;17:991-997.
  • 48
    Amarasingham R, Moore BJ, Tabak YP, et al. An automated model to identify heart failure patients at risk for 30-day readmission or death using electronic medical record data. Med Care. 2010;48:981-988.