Background
Patients approaching the end of their lives, as well as their relatives and clinicians, value accurate prognostic information.1–5 This information is usually obtained by clinicians using their clinical intuition. However, clinician prediction of survival (CPS) is often inaccurate and overoptimistic.6 The need for more accurate methods of prognostication was one of the issues highlighted by the Neuberger report7 into the shortcomings of the implementation of the Liverpool Care Pathway.8
Systematic identification of patients approaching the ‘end of life’ is a key recommendation of the Department of Health and Social Care’s end-of-life care strategy.9 The Gold Standards Framework (GSF) service improvement programmes (widely used in general practice, nursing homes and, increasingly, acute hospitals) have produced proactive identification guidance to identify patients approaching the last year of life.10 However, many patients who would potentially benefit from inclusion in such programmes are currently unidentified by clinicians. More reliable prognostic estimates may facilitate the identification of such patients and may improve the prioritisation of patients who are referred to palliative care services. Improved prognostication would benefit patients and their carers by providing them with better-quality information to inform their choices about future care.3 Improved prognostication would also help clinicians to plan services and to ensure that patients are cared for in the most appropriate environment and with the most appropriate treatments. Prognostic scores could also facilitate comparison of services by more accurately describing the case mix of referrals.
The Prognosis in Palliative care Study (PiPS) predictive models of survival were previously developed by members of our research team to provide an objective aid to clinicians’ intuition.11 The original study prospectively recruited a cohort of 1018 patients with advanced cancer who were no longer undergoing disease-modifying treatment. This was a multicentre study involving 18 specialist palliative care services across England. Separate prognostic models were created for patients without or with available blood results [PiPS-A (Prognosis in Palliative care Study – All) and PiPS-B (Prognosis in Palliative care Study – Blood), respectively]. Logistic regression identified 11 core variables [i.e. pulse rate, general health status, mental test score, performance status, presence of anorexia, presence of any site of metastatic disease, presence of liver metastases, serum C-reactive protein (CRP) concentration, white blood cell count, platelet count and serum urea concentration] that were independently predictive of both 2-week and 2-month survival. Four variables had prognostic significance for 2-week survival only (i.e. dyspnoea, dysphagia, bone metastases and alanine transaminase concentration) and eight variables had prognostic significance for 2-month survival only (i.e. primary breast cancer, male genital cancer, tiredness, weight loss, lymphocyte count, neutrophil count, alkaline phosphatase concentration and albumin concentration). The receiver operating characteristic (ROC) area under the curve (AUC) for all models varied between 0.79 and 0.86.
Each PiPS model consists of two prognostic submodels [PiPS-A 14-day model (PiPS-A14) and PiPS-A 56-day model (PiPS-A56); PiPS-B 14-day model (PiPS-B14) and PiPS-B 56-day model (PiPS-B56)] that are combined using a ‘decision rule’. PiPS-A14 and PiPS-B14 predict whether a patient will die within days (< 14 days) or will live for ≥ 14 days. PiPS-A56 and PiPS-B56 predict whether a patient will live for ‘months+’ (≥ 56 days) or will die ≤ 55 days. The output of the two submodels are combined using the PiPS decision rule, which determines that, if the 14-day model predicts that the probability of survival is < 50%, then the patient is predicted to die within days (0–13 days); if the 56-day model predicts that the probability of survival is > 50%, then the patient is predicted to live for months+ (56 days or more); and, if the probability of 14-day survival is > 50% and the probability of 56-day survival is < 50%, then the patient is predicted to survive for weeks (14–55 days).
Thus, PiPS scores are able to predict whether a patient is likely to live for days (< 14 days), weeks (2–7 weeks) or months+ (≥ 2 months). These survival categories were chosen as they were deemed to have the greatest face validity among clinicians. Both PiPS-A and PiPS-B were shown to perform as well as CPS. The PiPS-B prognostic estimate was found to be significantly better than clinicians’ or nurses’ prognostic estimates but no better than a multidisciplinary agreed prognosis.
Following the publication of the PiPS prognostic tools, the National Institute for Health Research (NIHR) Health Technology Assessment (HTA) programme issued a commissioned call for studies to undertake ‘the validation of models of survival to improve prognostication in advanced cancer care to include the Prognosis in Palliative care Study (PiPS) predictor models’.
The commissioning brief12 from the NIHR was to evaluate the PiPS tools and other prognostic indices. Based on the results of systematic reviews,13,14 four other prognostic models were identified that might also be useful in clinical practice and that were in need of further evaluation. These were the Palliative Prognostic Index (PPI),15 the Palliative Performance Scale (PPS),16 the Palliative Prognostic (PaP)17 score and the Feliu Prognostic Nomogram (FPN).18
The PPI and the PPS can be calculated without the need for a blood test (like PiPS-A). The PPI model stratifies patients into three groups: survival for < 3 weeks, survival for < 6 weeks and survival for > 6 weeks.15 The PPI has shown a high level of accuracy in patients with short estimates of survival.19 The PPS is a measure of functional status and is one of the variables included in the PPI score. Although not specifically designed as a prognostic instrument, and therefore lacking some face validity as a stand-alone prognostic tool, the PPS has been found to have prognostic significance in patients with advanced disease.20,21 Using retrospective data from large observational studies, the PPS was found to distinguish between groups of patients with different probabilities of dying across a range of survival times.20
The PaP and the FPN require blood test results (like PiPS-B). The PaP classifies patients into three risk groups based on a 30-day survival probability of < 30%, 30–70% and > 70%.17 There is increasing evidence to support its validity in a variety of settings.22–26 The total PaP score was shown in one study to be more accurate than a simple CPS.27 One practical concern with the PaP score is that it relies on CPS. This can make the PaP challenging to use when clinicians are unsure about survival times or when an ‘objective’ estimate is required that is free from the influence of CPS. The FPN predicts survival at 15, 30 and 60 days.18 In one study, the FPN was found to be more accurate than the PaP18 and it does not rely on subjective CPS.
Although clinician estimates of survival have been shown to be inaccurate and overoptimistic, it is important that they are included as a comparator in any evaluation of prognostic scores because this is the method by which most clinicians currently form their opinion regarding likely survival. Our own work in this area has suggested that a multidisciplinary estimate of survival is more accurate than a nurse’s estimate of survival. Accordingly, the PiPS2 study compared the accuracy of PiPS-B with a clinician’s, a nurse’s or a multiprofessional estimate of survival.
The NIHR HTA commissioning brief12 also requested that an ‘assessment of the acceptability to patients and clinicians of the use of prognostic models’ should be included. Our research has tackled these questions using qualitative methods. We also assessed the acceptability of the models to the relatives/carers of patients. This is particularly relevant because in clinical practice it is often the relatives and carers who most wish to have access to accurate prognostic information.
As highlighted in the 2013 Neuberger report,7 a key research priority for the NHS is to determine the best ways to communicate uncertainty to patients and families about prognostic estimates. Previous research has shown that the majority of patients (61%) would want to know their prognosis if such information was available.28 Our qualitative substudy arm was designed to explore with patients and carers the type and extent of prognostic information they require and the best (and most sensitive) way to present this to them. The qualitative substudy also asked clinicians about the acceptability and practical utility of using prognostic indicators to support their subjective estimates and any facilitators of or barriers to their use.
