Introduction
Clinical prognostic estimates
How accurate are clinical predictions of survival?
Temporal and probabilistic predictions
Setting | Patients | Prediction | Accuracy defined as | Outcome (95% CI, if available) | |
---|---|---|---|---|---|
Temporal predictions | |||||
Amano 2016 | 16 palliative care units, 19 hospital palliative care teams, 23 home palliative care services Japan | N = 2036 Locally advanced or metastatic cancer | Specific temporal (for additional analysis also separated into categories < 7 days, 7–13 days, 14–27 days, 28–41 days, 42–55 days, 56–83 days, and 84 days+) | CPS lies between 0.67 and 1.33* times the AS, i.e. error was less than + 33% Spearman’s correlation coefficient between CPS and AS | - Accurate in 35% (33–37%); overestimated in 45% (43–47%) - Categories concordant 35% - Spearman’s correlation coefficient 0.69 reflecting good agreement |
Hui 2016 and Farinholt 2018 | Hospital, USA; secondary analysis of study of new prognostic marker | N = 222 Advanced cancer with palliative care consults | Specific temporal | C-index. AUC at 30 days and 100 days | - C-index 0.58 (0.47–0.68) - AUC at 30 days 0.58 (0.47–0.68) - AUC at 100 days 0.62 (0.54–0.70) |
Urahama 2018^ | Hospice, Japan | N = 101 Cancer referred from hospital to hospice for end of life care | Specific temporal | CPS lies between 0.67 and 1.33* times the AS, i.e. error was less than + 33% | - Accurate in 22.8%; overestimated in 67.3% |
Tavares 2018 | Hospital, Portugal | N = 38 Cancer known to the palliative care team (excluded haematological malignancy) | Specific temporal | If CPS and actual survival differed by less than 1 week | - Junior doctors accurate in 10.5% - Palliative care physicians accurate in 23.7% |
Razvi 2018 | Hospital, Canada | N = 172 Advanced cancer referred for palliative radiotherapy | Specific temporal (for additional analysis converted to categories ≤12, 13–26, 27–52, > 52 weeks) | Mean and median difference between CPS and AS in weeks | - Median difference overestimated by 14 weeks (IQ 1.3–36.4) - Categories concordant 41% |
Vasista 2019 | Secondary analysis of INTEGRATE trial of regorafenib in multiple countries (Australia, New Zealand, Canada, South Korea) | N = 152 Locally recurrent or metastatic gastric or oesophago-gastric cancer progressing after 1 or 2 lines of chemotherapy, ECOG status 0–1 | Specific temporal | CPS lies between 0.67 and 1.33* times the AS, i.e. error was less than + 33% C statistic | - Accurate in 29% - C-index 0.62 (0.57–0.68) |
Gramling 2019 | 2 hospitals, USA | N = 230 Metastatic cancer referred for palliative care consultation with a clear comfort only plan (excluded haematological malignancy) | Categorical temporal (< 24 h, 24 h to < 2 weeks, 2 weeks to < 3 months, 3 months to < 6 months, and > 6 months) | If category selected by physician was concordant with actual survival | - Accurate in 41%, overestimate in 50%# |
Simmons 2019 | 9 regional cancer centres, 7 palliative care units, UK | N = 463 Incurable cancer (excluded breast or prostate cancer with only bone metastasis) | Categorical temporal (≤14 days, 15–56 days, and ≥ 57 days) | AUC at 1 and 3 months | - AUC 1 month 0.71 (0.63–0.79) - AUC at 3 months 0.68 (0.62–0.73) |
Ermacora 2019 | 2 palliative care units, Italy | N = 334 Advanced cancer with no indications for oncology treatments, KPS 10–50% | Categorical temporal (1–2, 3–4, 5–6, 7–10, 11–12 weeks, or over 12 weeks) | AUC at 30 days separated by who made the prediction | - Nurse AUC 0.78 (0.72–0.82) - Physician 1 AUC 0.77 (0.77–0.81) - Physician 2 AUC 0.76 (0.71–0.81) |
Probabilistic predictions | |||||
Malhotra 2019 | Outpatient oncology, USA; secondary analysis of VOICES trial (communication and decision-making intervention) | N = 263 Stage IV non-haematological cancer or stage III plus oncologist would “not be surprised” if the patient died within 12 months | Probabilistic—chances of survival at 2 years with options of 0%, about 10%, about 25%, about 50%, about 75%, about 90%, or 100% | If difference between status at 2 years and prediction − 0.49 to 0.49+ AUC at 2 years | - Accurate predictions 62% - Pessimistic predictions 26% - AUC 0.81 (0.75–0.87) |
Surprise Question
Predicting imminent death
Are clinicians more accurate at predicting imminent death?
How do clinicians recognise the last days of life?
Performance scales and prognostic tools
Prognostic scale or tool | Prediction made | Factors included in the scale or tool | Comments |
---|---|---|---|
Palliative Performance Scale (PPS) | Patient assigned to a decile (from 100 to 0%). Each decile shows a distinct ordered survival difference with lower deciles associated with shorter survival. | Users find the “best fit” across the following domains to assign a decile •Ambulation •Activity and evidence of disease •Self-care •Intake (food and fluid) •Conscious level | - Does not rely on blood results or clinician predictions of survival. - Greatest accuracy found at the lower levels; a PPS of 10%, 20%, and 30% had a median survival of 2, 4, and 13 days respectively. |
Palliative Prognostic Score (PaP) | Cumulative total assigns patients to one of three groups with < 30%, 30–70%, or > 70% probability of surviving 30 days. | Each component assigns a score towards a cumulative total •Symptoms of dyspnoea •Symptoms of anorexia •Karnofsky performance status •Clinician predicted survival •White cell count •Lymphocyte % | - Clinician predicted survival is weighted heavily within the scoring system - Requires blood tests to be taken |
Delirium Palliative Prognostic Score (D-PaP) | Same factors as PaP with the addition of presence or absence of delirium | ||
Palliative Prognostic Index (PPI) | Score assigns patients to one of three groups of estimated survival of < 3 weeks, 3–6 weeks, or < 6 weeks | Each component assigns a score towards a cumulative total •Palliative performance scale •Oral intake •Presence of oedema •Presence of delirium •Symptoms of dyspnoea | - Does not rely on blood results or clinician predictions of survival - Contains PPS within the scoring system; it is unclear whether the PPI provides additional prognostic accuracy above PPS alone |
GPS (Glasgow Prognostic Score) or mGPS (modified GPS) | Score assigns patients to one of three groups: good, intermediate, and poor prognostic groups | Each component assigns a score towards a cumulative total •C-reactive protein •Albumin | - Requires blood tests to be taken - Allocation to prognostic group does not give indication of survival time; individual studies have to be found to understand median survival times for specific cancer groups, e.g. gastric cancer |
B12/CRP Index (BCI) | Score assigns patients to one of three groups with % probability of surviving 90 days | Total score derived from multiplying •Serum vitamin B12 level (pmol/L) and •Serum CRP level (mg/L) | - Requires blood tests to be taken - The only confirmatory study conducted found partial support for the BCI; they did not find statistically significant differences between all prognostic groups |
Prognosis in Palliative care study (PiPS - A) score | Provides a probability of surviving days (0–14 days), weeks (15–55 days), or months (> 55 days) | Prediction is based on a regression equation using the following variables •Diagnostic information •Sites of metastases •Presence or absence of key symptoms •Cognitive status •ECOG functional status | - Does not rely on blood results or clinician predictions of survival - Results can be calculated using an online calculator |
Prognosis in Palliative care study (PiPS - B) score | Similar factors as for PiPS-A but with addition of blood results | - Does not rely on clinician predictions of survival. - In one study was found to be better than a doctor’s or a nurse’s survival prediction |