Prognostic models of survival in patients with advanced incurable cancer: the PiPS2 observational study

Patients with capacity to consent

Eligible patients were approached by a member of the clinical team about participation in the study. If eligible patients were not approached by a team member then the reason for failure to do so was recorded on the screening log.

Potential participants who had been approached by a member of the clinical team were asked if they were willing to speak to a member of the research team and were handed a patient information sheet (PIS). A member of the research team then discussed the study with the patient and sought their consent to participate. For community patients, this discussion sometimes occurred over the telephone. Written informed consent was usually obtained at least 24 hours after the PIS had been handed out but could occur on the same day if it was more convenient and acceptable for the patient. If the patient declined to participate in the study then the reason for this (if known) was documented on the screening log.

Patients without capacity to consent

For patients without capacity, a personal consultee was sought for advice. For patients with no personal consultee, the advice of a nominated consultee was sought. The nominated consultee was usually another doctor working in the hospital/hospice (who was not involved in the research), a social worker, a chaplain or the patient’s general practitioner (GP).

In a similar manner to the approach adopted for patients with capacity to consent, consultees of patients without capacity to consent were advised that it was usual practice to wait for 24 hours before giving assent. If the consultee gave telephone advice for the patient to be included in the study but they were unable to visit the unit to provide written evidence of assent, then verbal agreement was initially deemed sufficient to allow the research team to enrol the patient in the study and start data collection. However, in these circumstances an assent form was posted to the consultee to be signed and returned to the research team within 2 weeks of the patient being enrolled in the study. If no signed assent form was received then the patient was withdrawn from the study and all data were destroyed.

Patients with fluctuating capacity to consent

If a patient who temporarily lacked and then recovered capacity to consent was included in the study, then, when they recovered capacity, they would be informed about their involvement in the study and would have the opportunity to withdraw or confirm participation.

For the comparison between PiPS-B model predictions and clinical prediction of survival

The primary model of interest for this research was PiPS-B. To show at least a 5% improvement in correct predictions using PiPS-B compared with clinicians’ predictions, assuming 80% power and 5% significance level and using a McNemar’s test, 1267 patients with complete PiPS-B data would be required. The formula and the software used for these calculations were based on the work produced by Machin et al. 36 It was estimated that to obtain 1267 complete PiPS-B data sets it would be necessary to recruit 1334 patients with capacity to consent (assuming 5% missing data). Furthermore, to recruit 1334 patients with capacity to consent, it was estimated that it would be necessary to recruit approximately 1778 patients in total (assuming 25% of patients would lack capacity to consent).

For the validation of the PiPS models

To validate predictions from a risk model it has been recommended that the validation data should have at least 100 events. 37 The validation data for PiPS2 involved several centres. There is no guidance on sample size calculation for multicentre prognostic validation data. We expected clustering of patients within centres to be minimal, based on other studies in community care. 38 However, to be conservative, we inflated the number of events required in the validation data to 150. Assuming an event rate of 17.8%, based on the original study, we estimated that we would require 843 patients to validate the PiPS-B model. In fact, to have sufficient participants to compare PiPS-B with CPS, we planned to recruit approximately 1778 patients, most of whom would be able to provide data for the validation of both PiPS-A and PiPS-B. Thus, the proposed sample size was more than adequate to validate both models. Using similar arguments, the proposed sample size was also more than sufficient to validate the other prognostic models (i.e. PPI, PPS, PaP and FPN).

Descriptive analysis

Predictors and outcomes were summarised using descriptive analysis. Categorical predictors have been reported as raw numbers and percentages. Continuous variables have been summarised using mean or median and standard deviation (SD) or interquartile range (IQR) as appropriate. The percentage of values missing for each predictor has also been presented. The survival times of patients have been summarised using median and IQRs, and Kaplan–Meier graphs.

Primary analyses

Secondary analyses

As part of the secondary analyses, the PiPS models’ predictions for the 2-week and 2-month cut-off points were combined to produce a categorical prediction of survival (days, weeks or months+) and were compared with clinicians’ estimates and the corresponding observed values with respect to their accuracy. Linear-weighted k has also been used to compare the performance of the clinicians with that of the models.

The PPI, PPS, PaP and FPN prognostic models were also assessed as part of the secondary analyses. The calibration of these prognostic models were assessed using the CS,40 based on a logistic model for binary outcomes and the Cox model for survival outcomes. 42 Model discrimination was assessed using the c-statistic for binary outcomes and Harrell’s concordance index for survival outcomes. 41 The performance measures estimated for the various models were compared descriptively. The predictions made by these prognostic models were also compared with the corresponding observed outcomes and clinician predictions (where available or where possible).

As multiple comparisons are performed in this report, a Bonferroni correction43 has been used to adjust the significance level for the secondary analyses. The amended significance level of 0.05/7 (0.0071) has been used for the secondary analyses.

Bias owing to missing data was investigated and multiple imputation based on chained equations44 was used to impute missing predictor values.

Performance of the 14-day and 56-day models

Before considering the global performance of the PiPS-A and PiPS-B models (which involved assessing how well the combined 14-day and 56-day models performed using the decision rule), each model was assessed independently.

Calibration of 14-day and 56-day models

The calibration of the prognostic models was assessed using the CiL and CS based on a logistic regression model fitted to the validation data using the predicted log-odds as the only predictor. The CiL and CS are shown in Table 12.

TABLE 12 - Calibration in the large and CS for PiPS-A and PiPS-B models

Variable	CiL (95% CI)	CS (95% CI)
PiPS-A
14 days (‘2 weeks’)	–0.037 (–0.168 to 0.095)	0.981 (0.872 to 1.09)
56 days (‘2 months’)	0.109 (0.002 to 0.215)	0.946 (0.842 to 1.05)
PiPS-B
14 days (‘2 weeks’)	–0.202 (–0.364 to –0.039)	0.840 (0.730 to 0.950)
56 days (‘2 months’)	0.152 (0.030 to 0.273)	0.914 (0.808 to 1.02)

A risk model is considered to be well calibrated if the CiL is close to zero and the CS is close to 1. The PiPS-A14 and PiPS-A56 were both well calibrated. The PiPS-B56 model was well calibrated. PiPS-B14 showed some degree of overfitting (CS 0.840), with predictions slightly higher for the 57–74% risk group.

The calibration of both the PiPS-B14 and PiPS-B56 models has also been assessed by comparing the observed and the predicted proportions of events (Figures 4 and 5, respectively).

FIGURE 4.

Observed and predicted proportion of events using PiPS-B14 model. n = 1497; proportion of events = 1238 out of 1498 (82.7%). One participant was removed from the analysis because their PiPS-B prognostic index was an outlier. The data have been divided into 10 equal-sized groups (deciles) to produce the graph.

FIGURE 5.

Observed and predicted proportion of events using PiPS-B56 model. n = 1498; proportion of events = 727 out of 1498 (48.5%).

Performance of the combined PiPS-A and PiPS-B models using the decision rule

When the outputs from the 14- and 56-day models were combined using the decision rule previously described (see also Appendix 7), the result was a prediction about whether the patient would survive for days (0–13 days), weeks (14–55 days) or months+ (> 55 days).

Kaplan–Meier survival curves for PiPS-A and PiPS-B models

The performance of the PiPS-A and PiPS-B models (using the decision rule for combining the 14-day and 56-day models) was further assessed by plotting Kaplan–Meier survival curves for each of the three risk groups: days, weeks and months+. The results are shown in Figures 6 and 7. The Kaplan–Meier curves were also redrawn (Figures 8 and 9) to illustrate the ability of the PiPS-A and PiPS-B models to discriminate between the three risk categories at the clinically relevant time points of 14 and 56 days.

FIGURE 6.

Kaplan–Meier curve by PiPS-A predictions.

FIGURE 7.

Kaplan–Meier curve by PiPS-B predictions.

FIGURE 8.

Kaplan–Meier curve by PiPS-A predictions, censored at 56 days. Reference lines show day 14 and day 56.

FIGURE 9.

Kaplan–Meier curve by PiPS-B predictions, censored at 56 days. Reference lines show day 14 and day 56.

Discriminatory ability of PiPS-B using imputed data

The discriminatory ability of the PiPS-B models was assessed using the c-statistic. Separate c-statistics were calculated for the weeks and the months+ models. For the PiPS-B 14-day model the c-statistic was 0.836 (95% CI 0.810 to 0.861). For the PiPS-B 56-day model the c-statistic was 0.808 (95% CI 0.787 to 0.829). The results are very similar to those obtained using the complete data sets without imputation. The discrimination of both models was excellent. 55

Accuracy of PiPS-B risk categories (using the decision rule and with imputed data) against agreed multiprofessional estimates of survival

McNemar’s test was used to compare the proportion of overall patient deaths predicted correctly by PiPS-B risk categories (in the imputed data set) with the corresponding proportion predicted correctly by AMPES. There was no statistically significant (p = 0.875) difference between the proportion of patient deaths predicted correctly by PiPS-B risk categories (using the decision rule) and the corresponding proportion predicted correctly by AMPES. The 2 × 2 tables are not presented because the results are based on the calculation of the p-value across 10 imputed data sets.

Palliative Prognostic score

The PaP score categorises patients into one of three risk groups to predict 30-day survival: A (> 70% probability of surviving 30 days), B (30–70% probability of surviving 30 days) and C (< 30% probability of surviving 30 days). Higher scores predict shorter survival. The PaP score is generated by applying a weighted score to each of six variables (i.e. CPS, KPS, anorexia, dyspnoea, total white blood cell count and lymphocyte percentage). There were survival data available for 1592 patients with PaP scores recorded. There were 794 (49.1%) study participants in risk group A, 655 (41.1%) in risk group B and 143 (9.0%) in risk group C.

Kaplan–Meier survival curves

The discrimination of the PaP was investigated by plotting survival curves for each risk group, which are shown in Figure 10.

FIGURE 10.

Kaplan–Meier survival curves for PaP risk groups. This figure has been reproduced with permission from Stone et al. 59 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original figure.

This figure has been reproduced with permission from Stone et al. 59 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original figure.

Feliu Prognostic Nomogram

The FPN uses five variables (ECOG performance status, serum albumin concentration, LDH concentration, lymphocyte counts and TTD) to predict the probability of survival at 15, 30 and 60 days.

Palliative Prognostic Index

The PPI is calculated using five clinical variables (PPS, oral intake and the presence or absence of dyspnoea, oedema and delirium). The model stratifies into three groups: survival of < 3 weeks (score of > 6), survival of < 6 weeks (score of > 4) and survival of > 6 weeks (score of ≤ 4). In our patient population, 501 (49.2%) participants had a PPI score of > 6, 428 (23.4%) had a score of > 4 and 900 (49.2%) had a score of ≤ 4.

Kaplan–Meier survival curves for Palliative Prognostic Index

The discrimination of the PPI was investigated by plotting Kaplan–Meier survival curves for each risk group, which are shown in Figure 11.

FIGURE 11.

Kaplan–Meier survival curves for PPI predictions by risk group. This figure has been reproduced with permission from Stone et al. 59 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original figure.

This figure has been reproduced with permission from Stone et al. 59 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original figure.

Palliative Performance Scale  

The PPS is a measure of functional status and is one of the variables included in the PPI score. The PPS is scored by health-care professionals and consists of 10 categories. A score of 10% represents the poorest functional status (totally bed-bound, unable to do any activity, mouth care only and drowsy or in a coma). A score of 100% represents the best functional status (full ambulation, normal activity and work with no evidence of disease, able to take care of themselves, normal intake of food and full conscious level).

Discriminatory ability of the Palliative Performance Scale

The discriminatory ability of the PPS as a prognostic tool was assessed by plotting the Kaplan–Meier survival curve for each of the PPS levels (Figure 12).

FIGURE 12.

Kaplan–Meier curve by PPS level. This figure has been reproduced with permission from Stone et al. 59 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original figure.

This figure has been reproduced with permission from Stone et al. 59 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original figure.

The median survival for each PPS level is shown in Table 32. With the exception of PPS level 100% (for which there were too few participants to judge), the median survival of each group increased in the expected direction as the performance status improved.

TABLE 32 - Summary statistics of the survival of participant separated by the PPS levels

Too few participants to judge survival times.

PPS score (%)	n	Survival time (days)
		25%	50%	75%
10	60	1	2	4
20	78	3	6	16
30	138	7	20	42
40	273	10	24	69
50	493	19	40	97
60	376	28	65	172
70	265	44	99	284
80	109	85	186	477
90	33	135	252	568
100	2a	161	161	–

Desire for detailed patient prognostic information

The majority of patients and carers expressed a desire for detailed information from doctors about life expectancy. Few reported that they had received this information, however, and reported that what information was given was not expressed clearly. A number of patients who said that they had asked about their life expectancy commented that doctors tended to be vague, overoptimistic and unwilling to deliver news that was considered bad or uncertain:

Whenever I have asked doctors about my prognosis, nearly everybody’s been vague, to be honest. I know that I’m going, but I want to know, even now, have I got a couple of weeks, a couple of months, the end of the year? I certainly know it isn’t anything further than that, but nobody dare tell me. I think doctors tend to be optimistic, I’d rather they would be more realistic.

P8: male patient aged 63 years (hospice)

Doctors are reluctant to give estimates of survival, their approach is very uncommunicative. When the prognosis is negative, you have to really probe and ask lots of questions, doctors would rather get out of giving you negative information altogether. I think it’s because they don’t want to give people bad news.

P5: male patient aged 61 years (hospice)

The specialist said to me it won’t be that long. Now, what does long mean, it’s a piece of string, isn’t it? I think they’re [doctors] very reluctant to say, ‘oh you’ve got 6 months’, and then you drop dead the following day. I would say the specialist was being extremely vague. He said, ‘it will kill you’, but he didn’t say when.

P3: male patient aged 76 years (hospice)

It is noteworthy that the first quotation seems to confirm our choice of presenting information in terms of days, weeks or months. A number of carers commented that, even if it was not shared with the patient, they would have preferred more detailed prognostic information. They described that they found clinicians reluctant to give a definite answer:

Nobody told him whether he’d got a week, a month or a year to live. They [the oncologist] just said the cancer was bad and we were just in shock because nobody could tell us what time my husband had left, the oncologists are reluctant to say, but I think as individuals you need some idea.

C18: female carer aged 56 years (community)

I was told that Mum had less than a year to live and that was 12 months ago . . . Mum wouldn’t want to know how long she has, plus she gets a bit confused, but the family would like more detailed information about her prognosis.

C2: female carer aged 67 years (hospice)

The main reason that both patients and carers gave for wanting more detailed prognostic information was to make plans and sort out family affairs. This related to sorting out individual and family finances and making funeral arrangements:

I said right from the start, if it’s terminal, tell me. I’d rather know so plans can be made and things can be put into place. I mean, I’m going to leave a wife behind so I wanted to get myself and my finances sorted out.

P29: male patient aged 69 years (community)

My husband has just turned 40 and is now on his seventh month without any treatment. The hardest part is the not knowing how long he has got. I think if I had a rough idea I think that would make things a little bit easier for me, with regards to organising his affairs and planning his funeral. But I am still in limbo, it could be next week or it could be 6 months.

C7: female carer aged 35 years (hospice)

A minority of patients explained that they did not want to discuss their life expectancy with clinicians. The main reason was that they preferred not to feel anxious or upset during the time they had left; in the following example, the patient was very elderly and did not have much time left irrespective of her advanced disease:

I would rather not know when I am going to go . . . I’ve lived a long life and I think I’d probably be more anxious, if I knew exactly when.

P21: female patient aged 92 years (hospice)

However, where patients did not want detailed prognostication, there was evidence in the carers’ data that patients were happy for relatives to receive it:

My wife doesn’t want to know what time she has left, she doesn’t, but we decided that I should be given detailed prognostic information.

C19: male carer aged 63 years (community)

One community care patient explained that his reluctance to discuss his prognosis was because he considered that he was still fighting his cancer, although the quotation perhaps also reveals his insight into his condition and the tension this represents for many patients between knowing and not knowing:

My doctor asked if I wanted to know my prognosis, but I said I didn’t. I know I have a life-limiting condition and I am still fighting the cancer, I am not ready to give up and think about my prognosis just yet. It is not the right time for me to be discussing my life expectancy.

P28: male patient aged 62 years (community)

The quotations on this theme clearly indicate that, for most patients, prognostic information is considered helpful, but there is likely to be a minority of patients who, for a number of reasons, do not wish to know.

Acceptability of PiPS predictor models

All participants considered that the PiPS models were acceptable for use in clinical practice and a useful aid in helping clinicians to more accurately predict life expectancy:

I think the tool is useful as it gives an indication of time left for people, however accurate it might be, that doctors can then pass on. At least it’s not this big open space that I have at the moment, the tool would help narrow the time frame, at least to give patients like me some idea.

P7: female patient aged 71 years (hospice)

I think developing this tool is massively needed. I’ve found the hardest part of going through all of this is the not knowing. I’m always asking, ‘can you tell me how long?’, and everybody says the same thing: ‘sadly, we can’t’. We were given a rough estimation, like, obviously 12 months being the longest survival rate from the oncologist, and we are actually past that now.

C7: female carer aged 35 years (hospice)

Patients and carers also considered that the PiPS model could help doctors initiate sensitive conversations with patients and carers about prognostication, potentially helping them to overcome their reported reluctance to do so (as cited in the context of the previous theme):

At the end of the day, doctors leave patients to ask about prognostic information because they [doctors] don’t want to volunteer the uncertainty. Any tool that helps must be useful for all concerned, especially in helping doctors explain this sort of information to patients.

P22: male patient aged 73 years (hospice)

I think the tool would be useful to help doctors start that sort of conversation about time left for people. I have felt that it is treated as a big secret, as if they [doctors] feel embarrassed to tell you or don’t know what to say. But I think if you’ve got something saying, look, we’ve had a look at this tool and it says maybe it’s months not years, you know, a bit of power to your elbow is always useful, isn’t it?

P15: female aged 61 years (hospice)

These are insightful comments from the patients and a theme that resonates with doctors’ concerns. We return to this theme when considering the views of the health-care professionals.

Presentation of sensitive information

Patients and carers agreed that the most appropriate way of receiving prognostic information was during face-to-face discussions and that this should be verbal rather than written:

I’ve had a number of face-to-face discussions with my doctors about my cancer and prognosis. I know talking about dying is a sensitive topic, it is part of life, I’ve accepted it.

P27: female patient aged 72 years (community)

I don’t think it is a good idea to show patients any written information about their prognosis, I think it could make people feel very anxious. I also think showing patients’ information about their probability of survival is hard for people to grasp. It could be very stressful for some, especially if they have a more negative outlook on things.

C1: female carer aged 42 years (hospice)

When having these conversations, patients and carers preferred doctors to convey prognostic information truthfully, honestly and sensitively. Participants confirmed that life expectancy presented as days, weeks or months was most meaningful to them:

I’d just like to know truthfully how quickly this is going to affect me . . . how soon am I going to get to the stage where I can’t swallow at all; what’s going to happen, when I’m going to die? I want to be well informed about the different stages.

P26: female aged 85 years (community)

Rather than being given statistical information about my husband’s time left, I’d rather doctors say to me, it’s days or weeks, rather than there’s a 50% chance he will still be here in 2 months. It’s just clearer to understand that way for me.

C7: female carer aged 35 years (hospice)

I’d personally go for the actual weeks, days, years, sort of number rather than percentages or a significant event, it’s more informative.

P27: female patient aged 72 years (community)

These data lend support to our presentation of PiPS results as days, weeks or months, although, overall, there appears to be misunderstanding among patients and carers about the ability of PiPS to predict longer survival. The scope and limits of PiPS would need to be clearly explained by clinicians.

Challenges and difficulties with predicting length of survival

All clinicians commented on the complex nature of estimating length of survival for patients with advanced cancer and many found it difficult. A number of clinicians described how difficult it was to predict survival in ‘well’ patients who had not started to deteriorate. The following quotes are from two trainee doctors:

In the past I’ve found it extremely difficult to give prognostic information, especially when working in a hospital, as it seems more unpredictable and you are not really working in a specific palliative care environment, you do sometimes feel a bit lost with these kinds of conversations.

H15: trainee doctor (F2)

I think it’s difficult to predict exactly when someone is going to die and I think you do have to sort of explain about uncertainties to people because I know from talking to patients, in the past the oncologist may have said ‘oh, you’ve got 6 months’, and then 6 weeks later patients find themselves in a hospice and they’re dying.

H6: trainee doctor (F1)

Interestingly, this view was supported by more senior clinicians:

I find predicting length of survival extremely difficult. I think once there’s a change in somebody’s health condition it’s certainly a lot easier, because then you’ve got a reason to suspect that they may well be deteriorating. If somebody continues in a stable disease phase without the presence of imaging, I think it’s difficult to prognosticate.

Consultant (palliative care), ID number H24

If somebody has other aspects of their general condition that shows they’re in decline, then it’s easy to give an estimation of time left because you’ve got some benchmarks to say, this person is deteriorating. Other than that, I think it’s just . . . we don’t have any more discernible information, really, that would allow us to give an accurate prediction.

Consultant (palliative care), ID number H27

If there’s nothing in the patient’s disease trajectory which would suggest a change in their condition, then I think it’s very difficult to prognosticate, so I think you’d struggle to give an accurate prognosis if somebody’s condition doesn’t change. If somebody’s condition deteriorates, then it’s the rapidity of that decline that gives you the best guide. And then it is just a guide.

Consultant oncologist, ID number H20

A number of senior clinicians said that they would generally explain to patients that accurately predicting length of survival was, in many cases, an impossible task owing to the amount of uncertainties involved:

I’ll often say it’s not like on TV, we’re not good enough to estimate accurately how long an individual is going to live, and then we talk about months or weeks, or . . . I’ll often tag onto the end of . . . so let’s say I think it’s likely to be weeks, because they’re going down quite quickly; I say how many weeks that is I’ve honestly no idea, things could flatten off and it might be quite a lot of weeks and that might turn into a few months, or things could get worse and we might be talking about a very small number of weeks.

Consultant oncologist, ID number H28

I say that, even with all of our experience, we’re often no better at predicting the future than relatives and patients are themselves. But patients and relatives don’t often get that. I’ve had one patient’s relative think I was withholding information from him, and that of course I must know, I’m the consultant, I must know how long the person has, and got really very angry with me. But I genuinely wasn’t certain with that patient.

Consultant (palliative care), ID number H13

These quotations are from interviews with senior clinicians with considerable experience of observing deterioration in patients with advanced cancer. It is interesting, therefore, that in the validation study, clinicians’ estimates were found to be quite good, with around 60% accuracy. There appears to be a mismatch between perceptions and reality. Clinicians’ lack of confidence in their ability to predict survival using clinically observed parameters might explain the next theme, which is the language they used with patients and families when discussing prognosis.

Language used when discussing prognosis

During the interviews, clinicians were asked about the type of language they used when discussing prognostication with patients and carers. The majority of clinicians, irrespective of experience or seniority, explained that they tended to avoid giving detailed prognostic information with specific time frames. Clinicians said that this was usually because they actually did not know, as these quotations illustrate:

Patients do ask but I keep it quite vague, especially just trying to sort of judge how much a patient wants to sort of talk to you about it. Yes, I wouldn’t just, answer just says weeks, or whatever, I think you just have to be quite kind and gentle. Also some patients just don’t want to know about prognosis. We just talk more in terms of sort of the rates of deterioration, that’s how I talk about it a bit more, you know, things are progressing over days, weeks, months, that kind of time frame.

H25: trainee doctor (F2)

Rather than giving an approximate time frame, I might say, as you’ve seen, things have been going downhill and if there are things that needs to be done, such as, you know, getting your financial, social affairs in order, I wouldn’t delay them any longer. I refrain from giving any specific time, like you’ve got 2 weeks or you’ve got – I don’t know – exactly 2 months, because I don’t think you’re going to know exactly how long it’s going to be.

Consultant (palliative care), ID number H31

Generally I try and keep the answer vague, I mean, you’re obviously able to give an indication as to when someone’s obviously breathing their last or when someone is very unwell, but when someone is in that stage where they’re kind of on a knife edge, I always say, well, it could be the next few weeks, but at the same time we’ve all been surprised and seen people last for months and months. So it’s about discussing the uncertainness involved in predicting survival.

Consultant (palliative care), ID number H30

In the third quotation, when predicting survival the clinician balances bad news with hope that the patient could live much longer. Referring back to the patient and carer data, this may be perceived as ‘vague’ rather than honest and kind. In the following quotation, the clinician describes how he explains to patients the process by which predictions are made and how observed changes support estimates or ‘guesses’. The process is shared with the patient and carer to some extent:

I’m careful to, I think, really try to avoid any certainty. And I usually, before I start talking about what my guess might be, I often talk about the fact that we don’t know, and that we guess, and how we guess. And I, sort of, try to give them a little bit of an understanding of how we’re prognosticating; that we’re using what we know about their illness, yes, and their scans, but actually, we’re using how they are, how things are changing, how they’ve been changing, and we use that to see how we’re going to predict.

Consultant (palliative care), ID number H23

Another clinician discussed how he avoids giving a number to patients because it will inevitably be wrong. He also considered that a vague prediction is probably more helpful:

I tend not to put numbers on it, I tend to think about things in terms of, you know, days, weeks, months, years. But I don’t usually put a number on it because that immediately means that I’m going to be wrong, if I put a number on it. And I tend to use that, sort of, language. Also, I think that actually, there’s no way of being sure, and often that, sort of, vague idea of a guess is often, for a lot of people, more helpful than adding numbers to the guess, is what I tend to say.

Consultant (palliative care), ID number H7

Junior clinicians also described how they avoided giving numbers to patients and preferred to convey vaguer estimations of prognosis:

I’ve always avoided numbers and will talk generally; well, it may be months. I talk about things will then change and then we might find that you’re changing and we’re looking at weeks, but nobody can say when that change is occurring, whether it’s going to occur tomorrow, next week, or in several months’ time. So, trying to get across that it’s a very inexact science.

H5: trainee doctor (F2)

I acknowledge that the prognostic information that I’m giving to patients is relatively vague and I apologise for that, but generally patients understand, and I don’t use numbers because I think patients and relatives pick up on the number and they focus on that number. Whereas actually, you’ve invariably got a scale of a very short time to a very long time and you’ve no idea where the patient is on that scale.

Specialty trainee doctor, ID number H16

A number of senior clinicians further considered that giving specific time frames could cause psychological harm to patients and carers, especially once patients had survived beyond their maximum given prediction. The following quotations illustrate this:

So, one of the questions that almost everyone wants to know is how long they’ve got, and I think it probably is the hardest question to answer because none of us know. I tend to be quite guarded and careful about how I answer that question, and usually probably far too vague for their liking, mainly because I’ve seen the negative effects of people having been given a very clear time scale, often from a hospital clinic they’re given x number of months, and more often than not they might exceed that and then they feel like they’re living on borrowed time, and actually, psychologically, for them and their family that’s often more damaging.

H9: GP

I’ll really try and avoid anything that might give them [patients] a number. And that’s, I suppose, because I’ve had many patients who feel like they’ve been given a number and then it causes them problems, because they’re counting on that. Others aren’t bothered so much by it, because then they get to the point, and once, when they outlive it, it can be a positive thing, but for some of them, they’re just waiting to die.

Consultant (palliative care), ID number H29

I don’t ever use numbers, because it fixes people’s focus onto a particular time scale, and rather than making what they can and enjoying each day for its own right and, you know, having an eye to the things that they need to be sorting out. I think that’s the sense of a much more vague time line from that point of view is far more helpful for most people.

Consultant (palliative care), ID number H31

Insights from this theme suggest that the ‘vague’ language that some patients and carers dislike is used by clinicians because they consider it to be a more honest response when so many factors are at play. There is also some reticence about giving a number in case a patient becomes fixated on it, which is a theme we pick up later when we discuss barriers to using PiPS.

Reasons for overestimating prognosis with patients

During the interviews, clinicians were asked about the accuracy of estimating length of survival for patients with advanced cancer and about existing studies suggesting that clinicians tend to give inaccurate and overoptimistic prognostic information. All agreed that the main reason they felt clinicians may give an overestimate of length of survival was that colleagues prefer to convey kinder, optimistic information that was perceived as less harmful. The following two quotes are from GPs:

I suppose you don’t want to cause harm if patients are in that good place, but also you’re then thinking, well, am I going to cause more harm by not addressing this now if patients are in a state of denial, and when it does hit them . . . those kinds of things. I suppose breaking any bad news, you do it at a rate the patient can cope with and there’s . . . I suppose there’s no rush, it’s not going to affect any change in the management, more how you’re handling the circumstances, in a way.

H12: GP

I don’t often get any negative criticism off patients, but when I have maybe a couple of times due to this factor where I’ve tried to sensitively discuss end of life and maybe the patient’s not been ready for it. Once I was on a home visit and the next time I saw the patient he was like, are you going to tell me I’m going to die again? And obviously that’s how he perceived it and I wasn’t aware . . . maybe I could have done it better or more sensitively. But he wasn’t ready for that discussion and it was me that was leading it instead of the patient. You can cause a lot of harm if you’re not careful, just by your manner and how you communicate with patients.

H11: GP

The following quotation further illustrates that this kinder, optimistic approach is also adopted by more junior clinicians:

It’s about all sorts of complex reasons why doctors may over estimate prognosis, really. It’s trying not to upset patients, not be a failing doctor, trying to understand what death and prognosis means to patients and what else is going on in people’s lives.

Specialty trainee doctor, ID number H17

Other clinicians also considered that it was better not to challenge patients’ or carers’ perceptions of the patient’s disease trajectory and described not wanting to reduce people’s hope if they were using denial as an effective coping mechanism:

It’s not that clinicians are overestimating prognosis it may be that sometimes, we can see that a patient is operating at a high level of denial, we wouldn’t necessarily challenge if we thought that was a useful coping mechanism for them, any challenge may cause stress and harm to the patient, but it might be that we then have to subtly bring along the patient and family’s understanding over time to try and prepare them for what’s going to happen.

Consultant (palliative care), ID number H14

In the next quotation, the clinician describes his experience of rapid deterioration in patients when such coping mechanisms are removed:

If they’re [patients] coping quite well with their perception of actually time left and we as clinicians change that quite drastically, that’s when you can do harm, and it can seem that people deteriorate very quickly then. So there’s that difficult line to tread between giving patients more optimistic information which may give patients time to still function, so they can plan things that are important, compared with taking away hope and doing harm.

Consultant (palliative care), ID number H24

I think when they [oncologists] have given prognostic indicators, I think they generally err on the side of saying they’ve got slightly longer than maybe they have to patients. They are trying not to destroy people’s hope. I think everybody wants to give people good news, don’t they, and in this scenario there isn’t much good news.

H19: GP

This theme therefore presents a tension. Clinicians withhold detailed estimations of survival because experience suggests they can get it wrong and that it can erode hope, and some consider denial to be an effective coping strategy. Yet the majority of patients want to know how long they have left. There appears to be a mismatch between what patients want and clinicians’ perception of this. This leads to the next theme, which presents their views on whether or not PiPS would help.

Clinicians’ acceptability of PiPS models

When asked about the acceptability of the PiPS models, clinicians gave a number of positive responses. The majority considered the model to be a useful algorithm that could offer a more scientific approach to estimating patients’ prognoses than a clinical judgement or hunch:

I think it’s useful to have it [PiPS] there and offer it to patients and say there is a scientific tool that can give you a better idea about what time you have left, which gives you closer numbers than I can, which might help your planning.

H4: GP

I think where PiPS would sit for me as a tool is if a patient asked me what I thought their prognosis would be, because currently I’ve got only my experience to use; whereas I wonder how it would work in my clinic. So, if, for example, a patient asked me, what do you think my prognosis is, we would have our usual probable discussion exploring what he thought, what they thought; and then I think I would probably say, there is an online tool that actually is more accurate than me, if you want us to use that tool we can do it now if you like.

Consultant oncologist, ID number H28

It would appear from the validation study that PiPS-B risk categories are as accurate as AMPESs, if not more so. Anticipating this as a possible outcome, we asked clinicians whether or not PiPS would still be beneficial. Clinicians stated that PiPS could help improve their confidence in making accurate predictions of life expectancy:

Even if PiPS is no more accurate than a clinician’s estimate, it would be useful for use in clinical practice, especially with less experienced staff, as when you’re a junior doctor you don’t necessarily have the experience to be confident in identifying a dying patient, so often you’re very hopeful. I think sometimes when you’re a junior doctor, you could err on the side of false reassurance. So I think because of the lack of experience, I think it’s more difficult for junior staff to prognosticate accurately.

Consultant (palliative care), ID number H7

I think we’re a little bit uncomfortable when we are asked the question about what time is left, because we know that it’s an estimate. We don’t want to be completely wrong, I suppose, and yet we understand that something sudden can happen at any point, can’t it? So, I think the PiPS tool perhaps gives you more confidence in making a prediction.

Consultant oncologist, ID number H20

I think for a lot of doctors having something more solid, where we can say actually it’s looked at blood tests, it’s looked at different clinical markers and it gives you something more firm to back you up and give you some numbers. We know that we can never be 100% sure, but what this tool does is gives you a bit more confidence. Maybe more doctors would have the conversation if they felt more confident about what they were saying, and actually that would be much better if doctors started talking about it more, that would be a really good thing.

Consultant (palliative care), ID number H14

I think for you to be confident with a consistent approximation, so a consistent level of knowledge that you can apply to how long, PiPS would be very useful, because it would give you some assurance that your approach every time is the same to the information that you’re being given. So you can use the tool to measure how accurate your own predictions are.

Specialist palliative care nurse, ID number H21

Importantly, and echoing findings from the patient and carer interviews, clinicians agreed that PiPS would be a useful communication prompt that could be used when conveying prognostic information. It could make starting, and holding, difficult conversations easier:

When communicating prognostic information it would be nice to wrap it around something, and I’m presuming with something like PiPS you can almost wrap it round, you know, we’ve filled in a tool, what we do is we look at you, we look at lots of things, such as your symptoms, we take them into consideration, and what the tool shows us that you’re very poorly and then I would maybe wait for the ‘how long?’ question.

Specialist district nurse, ID number H22

I mean PiPS is another way of starting this sort of difficult conversation which I do think a lot of clinicians have the gut reaction to shy away from. It is important to talk to people and to give them an idea, because I think the worst possible case scenario is these people who lose their relatives and it’s a terrible shock because nobody has had the balls to actually front it with them, so I think it could have a place.

H12: GP

A number of participants further discussed that PiPS would be helpful for encouraging clinicians to give a more realistic prediction by using it in a way that takes the responsibility for making that prediction away from clinicians themselves:

Sometimes oncologists might overestimate because they’re trying to be nice and maybe we [GPs] do that as well. So it might actually help a doctor who has that tendency to be more optimistic to use PiPS. I suppose then it’s not the doctor that’s giving the bad news, it’s like the tool, it’s the computer results that are saying this, which a doctor may find useful.

H10: GP

These points are particularly important when considering the use of PiPS in clinical practice, because, as the interviewees indicate, even if PiPS were to be no more accurate that clinician estimates, it might provide clinicians with more confidence and a framework around which to initiate a discussion of prognosis. For the clinician, using a prognostic tool like PiPS would be more akin to giving a patient the results of a test. This would remove some of the direct responsibility for the ‘bad news’ from the clinician and may make it easier for them to broach the subject. Nonetheless, the clinician will still need excellent communication skills to convey this information in a sensitive way and, therefore, if PiPS is to be used in this way it is clear that clinicians will need to develop good communication skills for this approach to be successful.

Facilitators for use in clinical practice

All clinicians commented that the online PiPS prognosticator tool was user friendly and quick and easy to use:

The tool is quick and easy to use, I could put one of our junior doctors in front of PiPS, and they would complete it, without any training really.

Consultant (palliative care), ID number H14

I find the online tool quite straightforward to use, once everything is entered it is just a case of submit and the results appear.

Consultant (palliative care), ID number H27

I think PiPS is very straightforward. It’s available online, so you can put the data in. I have no difficulty at all with it.

Consultant oncologist, ID number H20

A number of clinicians considered that PiPS could be beneficially used as an educational training tool, especially for less experienced staff:

It would be interesting for junior staff like me to use PiPS that are new to palliative care. Say, OK, for your first month in your job, try and plug in the details of the patients that we’ve actually got the relevant details for, and then just get a feel yourself on how that’s matching up to your own reality and clinical judgement of what time patients have left.

H6: trainee doctor (F1)

Well, PiPS would be useful in clinical practice, if you found the tool gives the same prediction as an experienced clinician on more or less every occasion, then your junior staff should be trained to that standard; and that would be a way of saying, right, let’s look at a patient, OK, tell me what your gut feeling is, assess your patient. Then you could say, well, why do you feel that, and I feel this . . . OK, let’s have a look on the PiPS tool, let’s see if you are right or not . . . that would be something that would help with learning and development, wouldn’t it?

Specialist district nurse, ID number H22

Participants further commented on how the PiPS tool could help inform decision-making in relation to treatment options and/or discharge planning:

If it’s [PiPS] no more accurate than clinical judgement, it’s probably got a good basis for training, I guess. You may have a particular patient and you may think to yourself, I wonder what the numbers are re[garding] length of time. And then you could use the tool and see how close you get, to help build up your experience.

H10: GP

Even if it’s no more accurate than clinicians’ estimate, I would still use it, especially in the hospital setting. The reason being I think one of the things . . . or at least from a palliative care point of view, is that it could aid any MDT [multidisciplinary team] discussions as sometimes it is difficult to convince other clinicians about a patient’s prognosis. If we can use PiPS, then at least we could use this as an evidence base to say why we think the particular patient has got weeks or months.

Consultant (palliative care), ID number H30

Interestingly, a trainee doctor (F2) described that PiPS could support doctors attempting to convince more senior colleagues about a patient’s prognosis. The following quotation illustrates this point:

I think PiPS is a useful tool, especially for us junior palliative care staff working in a hospital setting, trying to convince more senior clinicians about a patient’s prognosis. It also aids the decision about stopping curative treatment or not and thinking what is best for the patient and the time they have left.

H5: trainee doctor (F2)

Clinicians further highlighted that PiPS would be helpful when commissioning care packages for patients:

PiPS is definitely helpful, with commissioning care packages, so commissioners like statistics, and certainty, because they don’t want to think that they’re going to be funding continuing health care for ages and ages, so being able to put some statistics in from a prediction, I think is helpful in the right people getting continuing health care, when we’re looking at arranging discharges and care packages.

Consultant (palliative care), ID number H27

A few participants considered that PiPS would be a helpful tool for use outside specialist palliative care, such as in the community. For example, the GP in the next quotation commented that PiPS would be useful in helping colleagues estimate length of survival for patients:

The value, I suppose, comes down to if we do have to diagnose somebody in primary care and they say, well, how long have they got, and the usual answer at GP level at the moment is, I don’t really know, why don’t you go and see the specialist and they will give you more information. At least then if we had PIPS in place we could run the patient’s details through it to generate an estimate.

H9: GP

Overall, therefore, there are a number of reasons why PiPS might be adopted in routine clinical practice.

Barriers to use in clinical practice

A minority of participants considered that it was not appropriate to use the PiPS tool, especially if blood tests were needed to improve the accuracy of the prediction. This related to both the practicalities of organising blood tests for patients with advanced cancer and the ethics implications of drawing blood from very unwell patients:

I tend not to run a lot of blood tests on patients that are palliative and cancer based. I am not sure how ethical it is to be taking bloods with patients that have advanced cancer, especially if I am seeing them at home. It is about making them comfortable, not sticking them with needles.

H4: GP

I think what does sit with me a little is to say to patients, well, we might be able to work out how long you are going to live but you’ve got to wait a couple of weeks to have a blood test done, because it’s a couple of weeks’ wait to get them back for their bloods.

Consultant oncologist, ID number H20

I don’t think PiPS is a waste of time. I think it’s potentially a waste of resources but wouldn’t say it’s a waste of time, but I’d say it’s a waste of blood tests and cost.

H5: trainee doctor (F2)

A few GPs considered that introducing PiPS into clinical practice could be time-consuming, both completing the tool and finding time to support patients emotionally:

We’re saturated at the moment in primary care, so to have to do something else like complete and run a PiPS estimate could be time-consuming. Also then you’ve got to find time to give patients the result. You’re grappling with time to deal with the results and to give patients time to talk and that kind of thing, like making time in supporting the patient emotionally.

H11: GP

Another, potentially more important, barrier relates to previous discussion about clinicians preferring to rely on their clinical judgement and avoiding prognostic discussions with patients and carers. The latter was due to difficulties with balancing a kinder, more optimistic stance with being realistic with patients considered not ready for receiving devastating news. If PiPS is perceived to potentially do more harm than good, it is unlikely to be used:

All the statistics doesn’t fit well with everyone and I just would always trust my own clinical judgement because that’s what I’ve always done. And if using a tool doesn’t benefit the patient in any way then it won’t be used. If you don’t actually feel that it’s going to actually help, it’s just another box-ticking exercise.

H9: GP

I think people do feel that if they’re told they’ve got 3 to 6 months they’ll come to my clinic and say, ‘I’m now 6 months and a week, what am I going to do now, what does that mean now?’. I’ve certainly looked after a few people who have been stuck because they’re either dying or just here waiting because somebody said they should have died. So, they’re left with, well what does that mean, where am I now, what do I do with life while I wait? It can feel much more that they’re waiting for something rather than just getting the most out of life, which is clearly what we’re aiming to help them do.

Consultant (palliative care), ID number H11

Clinicians’ attitudes to existing prognostic models

Clinicians were shown existing prognostic models (PPI, PPS, PaP and FPN) and asked to comment on their impressions of their usefulness. Clinicians considered these prognostics tools to be cumbersome and difficult to use compared with PiPS. They liked being able to complete PiPS online:

Compared with PiPS, the PPS tool scores are too wide-ranging and the PPI I suppose it’s not bad, it’s not as easy to do as running the PiPS on the computer one. I suppose the [FPN] is similar in a little way because it’s looking at some of the same parameters in the bloods and then it’s giving you the probability scores a bit like a percentage, 0.9–90%. I suppose it’s how user friendly it is in that format there, it doesn’t look that easy to use. Yeah, so it’s a lot easier to be able to press a button on the computer and get the results.

H12: GP

I think all these tools [PPI, PPS, PaP and FPN] are a bit cumbersome unless you’re of a certain mind and you like your stats and your figures . . . you’re going to have a low uptake of clinicians using them, whereas if it’s a click of a button we have the answer it’s a lot easier for people to use.

H11: GP

For some of these existing tools we’ve not got the influence of those recognised markers in the bloods, so I would have less confidence in it. They also just look more difficult to use than PiPS.

Consultant (palliative care), ID number H14

Others described the outcome from the tools other than PiPS as vague:

The PPS tool is rather vague and a bit wishy-washy. I probably wouldn’t find it useful. For the PPI tool the information is more detailed there, it makes a bit more sense in the sense it’s less vague in re[garding] the time frames.

H4: GP

The PPS results are a little wide-ranging to explain to the patient, therefore it is not very useful really. It wouldn’t be easy explaining to patients that they have a 30–70% chance of surviving 30 days. Patients prefer less statistical information.

Consultant (palliative care), ID number H24

Clinicians clearly found PiPS easier to use and were happier with the way the results are presented by PiPS than by other prognostic indicators that are available.

Patients’ and carers’ information needs and acceptability of PiPS

The majority of patients and carers clearly expressed a desire for detailed prognostic information but found clinicians vague, overoptimistic and unwilling to deliver accurate information about length of patient survival. The main reasons for wanting detailed information were to put finances in order and to make funeral plans. All patients and carers considered that the PiPS models/tools were acceptable for use in clinical practice and were a potentially useful aid in helping clinicians more accurately predict life expectancy and initiate sensitive conversations about prognostication with patients and carers. Participants confirmed that life expectancy presented as days, weeks or months was most meaningful to them.

Clinicians’ perspectives

Clinicians found estimating length of survival complex and often challenging and the process of conveying prognostic information to patients and carers a difficult and uncomfortable task. Clinicians, regardless of seniority, explained that they avoided giving specific time frames in discussions, either because they did not know or because they did not want the discussion to have a negative impact on the patient or carer. All clinicians described the challenges and difficulties they faced in predicting length of survival and admitted to being vague and avoiding the discussion of time frames with patients and carers. Clinicians considered that PiPS might have a place as an educational training tool, especially for training less experienced staff. Clinicians further commented on how PiPS could help inform decision-making in relation to treatment options, discharge planning and admission to hospices, or when commissioning care packages for patients. Clinicians said that even if PiPS was no more accurate than a clinician’s estimate they would still regard it as a potentially beneficial tool that could help improve their confidence in making predictions about survival. Clinicians considered that PiPS could be a useful communication aid when conveying prognostic information to patients and carers. Clinicians further discussed that PiPS might be a helpful way of encouraging clinicians to give more realistic predictions and could possibly be used in a way that took the responsibility of making predictions away from clinicians themselves.

We identified a few barriers to using PiPS in clinical practice. The first was the practical and ethics difficulties of organising blood tests for patients with advanced cancer. Two GPs considered that introducing PiPS into clinical practice could be time-consuming in relation to both completing the tool and finding time to communicate the results to patients and families. This barrier was not mentioned by their hospice colleagues. Other barriers related to clinicians preferring to rely on their own clinical judgement or wanting to avoid prognostic discussions with patients and carers.