Palliative care and health service use for people with life-limiting conditions

Various factors affect the need and demand for palliative care, including diagnosis and the nature of illnesses, illness trajectories, and the complexity of individual needs. From a population-based monitoring and research perspective there is currently no agreed diagnostic definition on who is eligible and should receive palliative care. However, various studies have used epidemiological approaches, based on mortality and hospital admission data, to estimate the size of the population in need of palliative care (defined here as those who would benefit from palliative care). 

Over the last 2 decades various population-level approaches have been developed for estimating the need of palliative care. The approaches vary in their complexity and sometimes rely on data not readily available. The mortality condition-specific approach in estimating the size of the palliative care population developed initially by Rosenwax (2005) and expanded/refined by Murtagh (2014) is the most widely cited as it relies on routinely available administrative data (mortality and hospital data), in particular the minimal estimate as it only relies on mortality data. Methods that use symptom prevalence data (Higginson, Gomez-Batiste and Sleeman (2019)) rely on disease/symptom/pain level data that are not routinely available at the national population level and may not be practical to collect (see Murtagh et al. (2014) for further details on these various methods). 

Another approach to identifying those in need of palliative care is to use the 20 conditions most often associated with symptoms requiring palliative care according to the Lancet Commission on Palliative Care and Pain Relief (Sleeman). This uses multipliers to account for varying degrees of physical and psychological suffering typically experienced by those with different life-limiting conditions. This approach is useful for predicting current and future need of palliative care; however, it is not suitable for an epidemiological approach to examine service use for a defined population who would benefit from palliative care. 

The Rosenwax and Murtagh methods were considered the most appropriate given the availability of data in the NIHSI data asset and the aims of the project. These methods were compared to identify the preferred approach for estimating population-based need for palliative care. To identify whether the narrow (Rosenwax) or broader (Murtagh) method would be best suited for identifying a population in need of palliative care, the likelihood of receiving specialist palliative care for each population was compared using a range of statistical measures. We estimated the number of deaths based on the 8 to 10 selected underlying causes of death using the Rosenwax and Murtagh methods (referred to as the minimal estimate by both researchers), as well as the maximal estimate (referred to as predictable deaths population). 

Other methods categorise the population in need of palliative care into 3 “end of life disease trajectories” (Seow et al (2017)), which provides a conceptual overview of disease progression and specific clinical approaches at the end of life, at different time points, provided through different health-care services or professionals. However, this method may not include all life-limiting conditions that should be considered in estimating the palliative care need population. Further, many people with life-limiting illness have multiple comorbidities and may not experience one of the 3 end of life trajectories, but instead a combination of these trajectories.

B.1 Narrow and broad definitions to define palliative care need population

There is value in assessing the population in need of palliative care using a narrow and broader range of conditions (recorded as the underlying cause of death), as it is likely that the true number of people in need of palliative care would lie within this range.

As shown in Table B1, the number of deaths was substantially higher using the Murtagh than Rosenwax methods – 109,059 compared with 61,218 deaths, or 80% and 45% of all predictable deaths, respectively. 

In the Rosenwax method, neoplasms (or cancer) accounted for 71% of deaths, however accounted for a much smaller proportion (39%) in the Murtagh method due to the inclusion of a broader range of chronic conditions. Inclusion of additional chronic heart/ cerebrovascular diseases, dementia and additional respiratory conditions accounted for, respectively, 67%, 20% and 9% of the difference in the number of deaths between the Murtagh and Rosenwax methods. Inclusion of more chronic renal and liver conditions (in the Murtagh method) resulted in marginal increases in the number of deaths, as the number of deaths from these conditions tends to be relatively small. 

*Total predictable ('maximal') cohort: aged 40-115 years at death, male/female only, excluding accidental, external and childbirth causes of death.

B.2 Most appropriate method for identifying the palliative care need population

To identify whether the narrow (Rosenwax) or broader (Murtagh) method would be best suited to this project for identifying a population in need of palliative care we assessed which method resulted in a higher likelihood of receiving SPC, using a range of statistical measures.  

As shown in Table B2, the Murtagh method captures 85% of deceased individuals that received SPC in the last year of life, compared with 66% for the Rosenwax method (sensitivity measure). The Rosenwax method uses a smaller number of non-malignant conditions, and does not capture advanced chronic conditions, including heart or, cerebrovascular diseases, and dementia. 

These conditions (particularly dementia) are being increasingly recognised as associated with palliative care needs. This is confirmed in the sensitivity analysis that highlights that 34% of people who received SPC in the last year of life did not die from the narrow (Rosenwax) list of conditions. Many of these presumably had chronic heart/cerebrovascular diseases, dementia, or respiratory conditions. This suggests that the Rosenwax method applied to the single underlying cause of death may underestimate the population in need of SPC. However, it provides a cohort with a higher likelihood of receiving SPC than a cohort with broader inclusion criteria.

Not all people with the broad (Murtagh) conditions would, however, require input from SPC, as shown by the lower likelihood of receiving SPC in the Murtagh compared with the Rosenwax methods – specificity of 22% compared with 68%, respectively. Similarly, the Murtagh method also had lower positive predictive value – 41% received a SPC service in their last year of life, compared with 56% for the Rosenwax method. 

Although these statistical measures do vary by underlying causes of death, the overall patterns in the Murtagh and Rosenmax methods were similar – malignancy has a strong positive predictive value and relatively high odds ratio of receiving SPC, and most non-malignancies (except perhaps for Murtagh-defined liver disease) do not (Table B2). This highlights that SPC was developed and is still largely delivered to those with advanced cancer. 

These measures highlight that the broad (Murtagh) definition is more likely to include people who did not receive (or need) SPC, while the narrow (Rosenwax) method may exclude people who did receive SPC. Further, the broad (Murtagh) definition captures 80% of all predictable deaths compared with 45% in the Rosenwax method. Given this, the Murtagh method is likely to provide a more realistic estimate of a population who may benefit from a palliative approach to care and is used as a starting point for defining the palliative care need population.

Note that the studies by Rosenwax and Murtagh have derived the palliative care need population from conditions that most often generated a need for palliative care. Therefore, it cannot be assumed that all people who died from these conditions required specialist palliative care or did not receive appropriate palliative care from other health professionals (outside of specialist palliative care). 

* Statistically significant at the p<0.05 level.

B.3 Limitations with mortality data for identifying palliative care need

The above-mentioned studies, particularly the Murtagh minimal estimate method, have been widely used and provide valuable insights on how to identify a population who may benefit from palliative care. However, using cause of death approaches may not always be an accurate reflection of diagnosis, which is a major consideration in referral for palliative care. Further, mortality studies may not capture the illness trajectory over a longer period, including that palliative care services required in the last few weeks of life may differ to those needed earlier in the illness trajectory that may extend over many years. Further, these mortality approaches rely on accurate reporting of causes of death on the death certificate, which is particularly challenging for older people with multiple complex comorbidities. 

Inclusion of symptom/pain/condition prevalence over the illness trajectory would provide a more accurate estimate of a population that would potentially benefit from palliative care. This reflects the increasing recognition that palliative care needs (physical symptoms, psychological distress, family and social support, information, and practical needs) are determined by more complex and interacting factors than diagnosis alone. However, there is currently limited national data across all settings that capture this information. In the absence of such data, this study has used the mortality-condition approach that has been used in other epidemiological studies for estimating the population in need of palliative care.