Prognostication: How good or (bad) are we?
Dr Vincent Thai MBBS, MMed (Int Med) (S), MRCP (UK), C.C.F.P (C), ABPHM (USA) Director - Palliative Care Services (UAH site) Associate Clinical i l Prof - Division i i of Palliative Medicine i Department of Oncology Edmonton Zone Palliative Program
Usefulness of prognostication Helps to determine goals of care Patients t and families need to know Need to communicate with honesty and compassion Accurate prognostication is important for patients and their families in order to plan and prepare for the rest of the patients lives and to help with decision making. Easy to understand format Poor performance status, rapid declining function, wt loss, dyspnea, delirium, dysphagia Best approach is to combine clinical judgment with validated tools, but in a probablistic format the most appropriate care in the best possible setting. resource allocation and future planning
Background Various tools that can be used to enhance prognostication Debate whether bloodwork should be part of the prognostication Is there a difference in prognosticating cancer patients t vs non-cancer patients t How do we apply the tool directly to the patient? Difference between the discriminative value vs individual actual prediction
Literature review Clinician prediction of survival (CPS) has been studied since the early 1970 s and is shown to be affected by many types of biases (Groopman JE, 2007, Glare P, European J of cancer 2008) Preferably to prognosticate by Dr at arms length from the care (Christakis, Journal Digest, 2000) Physicians have been shown to predict survival in an overly optimistic fashion (Glare P, European J of cancer 2008, Heyse-Moore LH, Palliat Med 1987, Addington-Hall JM, Br J Cancer 1990, Evans C, Lancet 1985;1:1204-6)
Literature review Statistical indices only useful for group or population level, not individual level (Henderson, J Med Ethics, 2005) Clinician prediction survival remains a powerful prognostic indicator, though miscalibrated, is enhanced by experience but affected by biasness (Glare, Eur J of Cancer, 2008, Glare P, Systemic review, BMJ 2003) A systematic review by Chow et al ( Clin Oncol 2001; 13(3):209-18) showed that the use of prognostic tools can improve the accuracy of clinician survival prediction estimates CPS superior to Karnofsky Performance Status (Maltoni M, Eur J of Cancer 1994)
What tools to use? 5 clinically most used and studied tools are 1) Karnofsky Performance Scale (KPS) 2) Eastern Cooperative Oncology Group Performance Scale (ECOG), 3) Palliative Performance Score version 2 (PPS) 4) Palliative Prognostic Score (PaP) 5) Palliative Prognostic Index (PPI). KPS, PPS and ECOG are based on functional performance and thus are functional tools PaP and PPI are aggregates of several prognostic indicators (including functional status) and thus used as prognostic tools.
Comparison of the Prognostic Accuracy of Four Clinical Assessment Tools and Clinician Prediction of Survival of Terminally Ill Patients in Different Palliative Care Settings (n=674) This study attempts to provide further insight into these applications by a comparison of the four most commonly used tools and CPS. Key issues in developing of prognostic tools is the inception cohort which affects the generalisability of to different palliative populations p and even the location of the patients. (Chow E, Predictive model for survival in patients with advanced cancer JCO, 2008)
Problems with Tools Approach to prognostication not standardized Some tools have a temporal approach to prognostication and some express prognosis in terms of probability Unclear which tool is best suited to predict survival at different time points in the disease trajectory.
Disease Trajectories and Variable Prognosis
Results- CPS vs PaP vs PPI vs PPS vs ECOG CPS = highest specificity at all of the chosen survival cut-off time points (<3 weeks, <30 days, <6 weeks) In clinical i l terms this means that t by using CPS, clinicians will have the best chance of accurately predicting which patients with a life-limiting illness will survive the longest Factors contributing to the process that clinicians use to formulate a survival prediction have yet to be fully elucidated in future studies. Exploration of this interesting subject will have to include qualitative methodology
Results- CPS vs PaP vs PPI vs PPS vs ECOG) PaP was most accurate in predicting the chance of having a short survival within the poor prognostic groups (highest PPV) across all three cutoff time points (highest chance of detecting those patients who would live shortest in the group that we assigned a poor prognosis). PPI was the most accurate tool to predict the risk of short survival for all patients who died (highest sensitivity = has the highest chance of pointing out those patients in the total study cohort would live shortest. This accuracy was uniform across all three cutoff time points. Collapsing PPS into three distinct subgroups has demonstrated the ability of this tool to predict survival similar to the ability of the prognostic tool subgroups. The collapsed subgroups of PPS and CPS were comparable in survival prediction ability with the corresponding subgroups of PPI and PaP.
Results- CPS vs PaP vs PPI vs PPS vs ECOG) ECOG, with its broad groupings, proved to have the highest accuracy in predicting longer survival in the low risk group at all cutoff time points (highest NPV). High NPV = high chance of detecting the patients who would live longest in the group that we assigned a good prognosis. With the lowest sensitivity, ECOG is the least useful tool to predict short survival in the gravely ill. This result suggests that the use of the ECOG score in the oncology setting is most appropriate for early treatment decisions. The best correlations were found between CPS and PPI and PaP scores, followed by high correlations between PPI and PaP. Correlations between actual survival time and the prognostic scores or CPS were lower.
Practical Difficulties for Various Prognostic Tools The tools were close in their predictive ability at the chosen cut-off timepoints. All five tools performed best at the cut-off time point of less than 3 weeks. PaP, followed by CPS, had the best predictive accuracy at all three cut-off points and they were closely followed by the other tools. However in our clinical i l experience both PaPP and PPI pose challenges to clinicians. i i PaPP can only be used in settings where contributing predictive factors like WBC counts are available, which limits its application in community settings and situations where the patient s goals of care do not include laboratory testing. The two symptoms, dyspnea and anorexia, included in the PaP, are difficult to dichotomize as present or absent. Clinically the term dyspnea relates to a continuum from slight shortness of breath to severe air hunger. This is further complicated if a patient s t dyspnea is well-controlled with the use of oxygen and medications. Similarly loss of appetite is not a dichotomous concept. Some patients might only experience a slight, not complete, loss of appetite. Others loose their appetite, but force themselves to maintain a near normal intake. PPI has similar limitations for presence/absence of edema and dyspnea at rest.
Comparison of the Prognostic Accuracy of Four Clinical Assessment Tools and Clinician Prediction of Survival of Terminally Ill Patients t in Different Palliative Care Settings - Conclusion This prospective study has demonstrated that the studied tools performed well in survival prediction, with a similar level of accuracy. All tools also showed specific strengths and weaknesses.. In practice clinicians need a tool that is capable of identifying patients at both ends of the prognostic spectrum. In this study PaP, followed by PPS and CPS, had the best combination of PPV and NPV, reflecting that these three are capable of identifying patients with both poor and good prognosis. In our clinical practice PPS in combination with CPS are most widely used to guide decisions regarding appropriate medical treatment and care setting. We postulate that our widespread use and longstanding experience with PPS might be a contributing factor to the higher CPS accuracy demonstrated in our study. This strengthening of CPS could then also partly account for the improved accuracy that we found in PaP. Further prognostication studies are needed to eliminate the weaknesses of each tool and to examine the subjectivity of CPS.
Thoughts. Prognostic tools have to be very easy to use but yet sophisticated tall order Bloodwork may not be available Prognosis has to to be understood easily by both clinician and patient with family CPS is the easiest to use ie the clinician Experienced clinicians have better prognostication How can we understand CPS as a bigger picture and how accurate/ inaccurate it can be? How does it tie in with treatment options?
CLINICIAN PREDICTION SURVIVAL OF PALLIATIVE PATIENTS
Research Question With experienced palliative consult clinicians (both MD & RN), how accurate are the CPS? How does the CPS affect overall management of the palliative patient as part of the bigger picture? Difference in the accuracy between cancer and non-cancer patients?
Methods All palliative patients referred to the acute Palliative Consult at 2 major tertiary hospitals and referred to the Community Consult team, which include community hospitals, hospices and home. Prospective cohort Information collected over a 1 year period from 2009 to 2010. 3 months follow up ensued to account for all deaths that occurred up to the end of the study period Death dates provided independently from provincial i Cancer Registry and Zone palliative care program database Total n=1570 cancer patients, 312 non-cancer patients
CPS what are the cut offs? < 2 wks : may not be worthwhile to transfer to a palliative unit 3-6 wks : consider transfer to a hospice, maybe a pleurx catheter,? IV antibiotics (ab), 7-12 wks: transfer to hospice, IV ab, insertion of pleurx catheters, RT treatment, surgery for impending # >12 wks: hard to place in hospice, consider long term care, consider IV ab, insertion of catheters for symptom relief, RT treatment, surgery for impending #
Preliminary results
Table on patient characteristics
CPS for Cancer Patients - Survival Curve
CPS Cancer patients No.of deaths for CPS < 2 wks No. of deaths for CPS 3 to 6 wks 86.6% 45.2% 27.9 % 10.8%
CPS Cancer patients No.of deaths for CPS 7 to 12 wks No of deaths for CPS >12 weeks 47.4% 52.1% 23.5% 24.4% 45.2%
Conclusion 1/2 of the cancer patients will die before the prediction ¼ of patients will be in the prediction ¼ of the patients tend to exceed the prediction CPS has strong discrimination value For non-cancer patients, we are even more over-optimistic (ie more patients die before prediction) Probably, we tend to overtreat unless patient does not want the treatment
Acknowledgments CHIR thru Partnership in Health System Improvement grant University of Victoria Dr. Francis Lau Dr. Yoko Tarumi Drs. (Ingrid, Nosh) and RNs in the Zone Palliative Program Ju Yang Hue Quan Dr. Michael Downing Rachel Elston
THANKYOUVERYMUCH ANY QUESTIONS