One of our major successes in workstream 1 was the refinement and validation of the IPOS. This represents a major step forward internationally for palliative care outcome measurement, for several reasons. First, we have demonstrated that this is a valid and reliable outcome measure, both in patient self-report and staff proxy-report versions. There are very few measures to assess well-being and health status in advanced illness which include both patient-report and proxy-report versions, with detailed evidence on how these correlate.73 Proxy-report is often required in palliative care as patients become too ill or fatigued to self-report. Second, although there are pre-existing psychometrically robust symptom measures (such as the Edmonton Symptom Assessment Scale97 and the Memorial Symptom Assessment Scale98) and quality of life measures (such as EORTC QLQ-C15-PAL77,99) for palliative care, there are no measures for advanced illness which capture the full range of symptoms/concerns which affect people with advanced illness. The IPOS reflects not just their physical or psychological symptoms, but also their information needs, family distress and support needs, and practical concerns. Third, our IPOS validation study included a high proportion of people with poor functional status, strengthening our conclusions for the advanced illness population. These are often the most difficult participants to recruit into studies, but our close working with participating sites helped to maximise recruitment and ensure that we included participants across the full range of functional status.
The IPOS is now freely available on our measures webpage: https://pos-pal.org/. It is already widely used both nationally and internationally; we have a network of UK-based palliative care services who link into an Extension of Community Healthcare Outcomes (ECHO) webinar network hosted by Hospice UK to support implementation and use, and there are already 12 translated versions of the IPOS (following our manual for translation and cultural adaption) in Czech, Estonian, French, German, Greek, Hindi, Italian, Korean, Polish, Portuguese, Swedish and Turkish, for use both in the UK and internationally. At least seven further translations are in process.
In addition, we were able to complete further work on the palliative Phase of Illness measure during Workstream 1, which was essential for the later casemix work. Preliminary work in the UK100 and Australia59 had suggested that palliative Phase of Illness is important in determining casemix within palliative care. However, little psychometric or clinical testing had been undertaken; we were able to show that Phase of Illness has clinical utility as a measure of overall palliative need, capturing additional information beyond other measures such as the AKPS and PCPSS.74 A limitation of this work was that we did not have any scope to analyse how Phase of Illness and the IPOS relate to each other; it was too early in validation of the IPOS to use the measure in this preliminary work. Nevertheless, our work confirmed the importance of including palliative Phase of Illness in workstreams 3 and 4 and that it reflected a distinct (and different) dimension of palliative care needs.
A further achievement was being able to derive and test a patient experience measure. VoC is brief and easy to use on a large scale with patients receiving palliative care across different settings.75 Patient, PPI and clinical feedback on this measure has been good; it has been clinically adopted across a range of palliative care services (e.g. St Christopher’s Hospice) and it too is freely available at https://pos-pal.org/.
Through discussion with our PPI group, it became much clearer to us how uncertainty influences patient experience during advanced illness by affecting patients’ information needs, preferences and future priorities for care. Assessment of these three factors is a useful starting point to guide clinical assessment and shared decision-making, and we considered how to include them in workstreams 3 and 4. However, it proved difficult to incorporate these findings into workstreams 3 and 4; this work was already complicated, and finding a workable way to add this dimension to our data capture (such that it would help inform the casemix classification) proved unfeasible.
In workstream 2, we conducted and analysed a large number and diverse range of interviews exploring stakeholders’ perspectives on measuring complexity. A major success from this workstream was the production of a novel framework for understanding complexity in palliative care. The findings from this piece of work were very well received by clinicians and researchers working in palliative care. The Field-Weighted Citation Impact in Scopus (Elsevier, Amsterdam, the Netherlands) is already 4.16 (as of 1 August 2020), indicating 316% more citations than average for the field.
We had several challenges with recruitment in workstream 2, particularly with recruiting a representative sample of the various stakeholder groups, but we overcame these to achieve a balanced sample by approaching additional sites and participants. Our study was able to represent views from stakeholder groups that are usually under-represented in complexity research, including patients and families, allied health professionals and national leads. To develop a well-integrated and meaningful theoretical framework to understand complexity, we presented the findings to internal and external experts at different stages of our data analysis. This meant going through several iterations to develop and refine a comprehensive and meaningful framework. Although this required more time and effort, the resulting output was a more integrated and useful framework to describe complexity in palliative care. In terms of informing workstreams 3 and 4, this work was invaluable in helping to ensure we measured psychological symptoms, social concerns, information needs (reported as highly relevant for pre-existing complexity), and practical concerns were identified.
Our work on models of specialist palliative care was, to the best of our knowledge, one of the first attempts at deriving empirical criteria to distinguish different specialist palliative care models. Using mixed methods in a sequential approach, we developed a set of criteria from these primary data to characterise and distinguish different specialist palliative care services in the UK. These criteria comprised setting, type of care, size of service, diagnoses accepted, disciplines, mode of care, types of interventions, out-of-hours characteristics, external education provision, use of outcome/experience measures, and bereavement provision, plus the purpose of the team, who funds/manages the team, ability to self-refer and discharge processes. These criteria capture the key differentiating components between different models of specialist palliative care across settings (i.e. hospice inpatient care, hospital-based care and community-based care) and, to our knowledge, enable these different models of care to be described and compared accurately for clinical and commissioning purposes for the first time. This study also provides the foundational work for improved research on which components of a model of care are most effective. A strength of this work is that it sought expert consensus from ‘real world’ professionals to identify the key criteria to characterise and differentiate these highly varied models of specialist palliative care.81 Our criteria, taken together, provide a defined and workable way to characterise and distinguish different models of specialist palliative care. These criteria are not exhaustive and they are not intended to be; however, they do help to discriminate between different models. We recognise that they do not reflect all models of palliative care across England but they provide a starting point in terms of evidence on which to build. In recent years, and particularly since this programme began, there has been a growing number of different models of palliative care developing,101–103 especially in community settings;104 distinguishing them in both practice and research is important.
In workstreams 3 and 4 we have for the first time, to our knowledge, demonstrated robustly which casemix variables are associated with higher or lower costs for specialist palliative care across different settings of care, and tested these classes. A casemix classification was developed for each of the three settings: hospital advisory, inpatient hospice and community specialist palliative care. Casemix classes are presented in the casemix classification and are based on the clinical criteria of pain, other physical symptoms, psychological symptoms, family distress, palliative Phase of Illness, functional status and living situation. This is a major achievement. Some work was previously undertaken on this by NHS England and Public Health England using similar casemix criteria,58 but there was limited evidence of how the classes were derived, and classification and regression tree analyses were not applied.
In our work on the casemix classification, it became clear that each participating site had very different models of care. A challenge was attempting to use the multiple criteria meaningfully in the work to develop the casemix classification. There were too many models of care and too many variables to describe them for any attempt to stratify or adjust for the different models. To some extent, we allowed for differing models of care in our sample size calculation (and we achieved our sample size), but the wide range of different models of care we encountered across participating sites was unexpected. When fitting ‘site’ as a variable into the CARTs to explain costs for inpatient hospice and community episodes of specialist palliative care, we found that site (as a proxy variable for model of care) explained a notable amount of variance in the outcome and more than the patient-level complexity factors. In all settings, there seemed to be a clustering of sites offering more traditional models of care versus sites with innovative/new care models. Owing to the limit on the number of sites per model of care in our design, however, we did not explore this independent variable in the trees and deliberately restricted the CART to patient–level complexity levels only. In some ways, this was helpful; the casemix classification is derived from patient-level variables alone; it is not the model of care which dictates the casemix class. However, clearly the model of specialist palliative care delivered does – at some level – influence costs. We need far more evidence about which models of care (when properly characterised) are most cost-effective.
Lastly, two factors had a major impact on the completion of the final report and the preparation of the remaining two main papers for publication; the Chief Investigator left the programme to take up a new position – with a part-time transition period during 2018 and 2019 – and the COVID-19 pandemic interrupted and severely delayed the completion of this report.