Article Text
PDF
Abstract
Background Palliative care interventions have the potential to lower health service costs by reducing the intensity of treatments intended to have curative effect while concentrating on quality of life and, in due course, quality of death. A patient receiving treatment inspired by curative intent during the end stage of their life is potentially exposed to medical futility.
Aim To conduct a systematic review of the evidence for palliative interventions reducing health service costs without impacting on quality of care.
Method An electronic search of MEDLINE, EMBASE, AMED and CINAHL databases, augmented by hand-searching techniques, was performed. Only research where palliative care was the intervention or observation, and cost, together with either quality of life or patient satisfaction with care were outcome measures, was included in results.
Results Of 1964 sources identified, only 12 measured both cost and an appropriate quality outcome. Evidence supported existing research that palliative care interventions generally reduce health service costs. Evidence of concurrent improvement in quality-of-life outcomes was limited; little available evidence derives from randomised trial designs. Small sample sizes and disparate outcome measures hamper statistical assessments.
Conclusions Evidence that palliative interventions cut costs, without reducing quality of life, by minimising futile medical acts is limited. Further research, including both observational studies and controlled trials, should be conducted to collect empirical data in this field. Future research should examine palliative interventions earlier in chronic progressive illness, and incorporate standardised outcome measures to allow meta-analysis.
- Medical Futility
- Health Care Costs
- Costs and Cost Analysis
- Palliative Care
- Terminal care
Statistics from Altmetric.com
Introduction
A futile treatment is one that is very unlikely to produce an effective therapeutic outcome.1 This might arise where a treatment has low odds of success concurrent with high odds of burdensome side effects. It might also arise with multiple comorbidities where curing one ailment cannot materially improve a patient's overall health.
Multiple comorbidities are increasingly common towards the end of life;2 around 90% of deaths are ‘expected’ and occur after a period of life-limiting chronic illness.3 Typical trajectories, illustrated in figure 1, demonstrate how patients’ capacity to benefit from curative therapy generally reduces as their chronic illness progresses.4 By contrast, intensity of diagnostic testing and attempted therapy generally accelerate through the last years of life, and are at their greatest preceding death.5–12 These two trends are incongruous, with patients potentially exposed to burdensome side effects during a time when their capacity to benefit decreases.
Trajectories of decline at the end of life.4 Time is presented in arbitrary units on these diagrams, but the scale has been adjusted to demonstrate approximate relationships between the four trajectories.
Palliative interventions might be effective in reducing the incidence of futile treatment.13 In recent years, palliative care has offered solutions generally during the terminal stage of illness, after curative attempts have ceased. An emerging concept, illustrated in figure 2, integrates palliative care from first diagnosis of potentially life-limiting chronic disease, initially, alongside attempted curative therapy.14 In theory, palliative intervention improves the management of patients’ physical and psychological distress, and helps recognise that death is normally the culmination of a chronic process rather than a self-contained acute event;15 hence, intensity of attempted therapeutic treatment should reduce in proportion with patients’ reducing capacity to realise benefit.
Graphic representations of traditional and emerging concepts of palliative care.14 In the traditional concept, specialised care is provided after curative attempts have ceased. In the emerging concept, palliative care is integrated into end-of-life care from the onset of life-limiting disease, to ensure patients’ emotional and spiritual needs are considered alongside their physical needs.
Research comparing palliative with usual end-of-life care shows potential to reduce intensity of treatments toward end of life; a palliative approach generally reduces financial cost,16 ,17 a measure closely correlated to treatment intensity.6 Patient experience can also be improved by palliative care.18–20 However, these outcomes are generally studied separately21 and it is therefore possible that treatment intensity is reduced at the expense of complete patient care. Reducing the intensity of treatment and improving patient outcomes concurrently would imply palliative interventions successfully avoiding futile elements.
This review assesses the extent to which existing evidence supports the ability of palliative interventions to routinely reduce treatment intensity, thereby reducing cost and simultaneously improving patient outcomes at the end of life.
Method
Search strategy
A search strategy was designed to identify research examining the relationship between a palliative approach to end-of-life care, financial cost to the health system and patient quality of life.
MEDLINE (1966 to June 2011), EMBASE (1974 to June 2011), and AMED (1885 to June 2011) were interrogated using Ovid Interface; CINAHL (1937 to June 2011) was interrogated using EBSCO Host Interface. The Journal of Palliative Medicine and Archives of Internal Medicine were hand-searched (January 2010 to June 2011), as these titles returned the highest frequency of potentially relevant material during the first stage of literature search; documents from government, think-tank and charity sources, including The National Audit Office, King's Fund, Joseph Rowntree Foundation and WHO were also examined, and citation tracking was employed on any articles selected for inclusion to optimise capture of latest work.
Description of the full MEDLINE search is shown in table 1. No equivalent to ‘Medical Futility’ was included in EMBASE or AMED searches as no subject heading indexes this concept within either. The closest concept in EMBASE is ‘Treatment Outcome’; this heading, however, catalogues over 500 000 articles, and such volume was deemed impractical given time available for this project. These exceptions aside, the search strategy outlined for MEDLINE was replicated for all additional database searches, updating terminology for each search to ensure equivalent subject headings were captured.
Medline search strategy
Study selection
Studies were deemed suitable for inclusion if they explored differences between palliative and usual approaches to end-of-life care, measuring financial cost and either patient satisfaction or quality-of-life outcomes. Studies were excluded if they investigated specific treatments, rather than a general palliative approach, or if they measured only one aspect of quality of life, for instance, pain. Narrative review articles were also excluded. PICOS criteria are set out in table 2 for clarity.
PICOS summary
Search criteria were set by both authors. The literature search and review of all sources identified were conducted by IH. A flowchart summarising the papers retained and excluded at each stage in the selection process is provided in figure 3.
Stages of evaluation, inclusion and exclusion for studies considered during this systematic review. EOL, end of life; QOL, quality of life.
Assessment of sources’ quality
Sources retained for full reading were reviewed for quality against predetermined outcome criteria: ‘cost’ and either ‘patient quality of life’ or ‘satisfaction with care.’
Data were extracted per a form based on criteria set out in the Cochrane Handbook of Systematic Reviews. Date of publication, country of origin, type and setting of intervention, patients’ terminal illness, study design, recruitment process, sample size, statistical methodology and study duration were all noted. Values were extracted for financial cost per patient in intervention and control; amounts not reported in US$ were converted at the average exchange rate during their year of publication. Values were extracted for all reported quality-of-care outcomes together with a note of their scale or unit of measurement.
Studies not accounting for cost at the level of individual patient were excluded. Quality outcomes are more complicated to describe, as a number of factors are amenable to measurement. Patient satisfaction is generally a product of treatment quality and, therefore, a valid outcome measure.22 Sources have been excluded where quality of life is not measured either using a patient satisfaction scale, or a multidimensional quality-of-life scale encompassing more than one discreet measure. All studies reporting appropriate data for both outcomes were included in the review, and any methodological limitations noted in Results to allow assessment of their strength.
The final assessment of included sources appraised their heterogeneity and determined their potential for conducting a meta-analysis. Different timescales, disparate interventions and diverse research settings, in conjunction with different outcome measures utilised suggested that, for this study, narrative synthesis would be more appropriate than full meta-analysis. However, arithmetical average percentage cost saving, weighted by study size, was calculated by multiplying the saving demonstrated in each study by its sample size, then dividing the sum of these by the sum of all sample sizes.
Results
Characteristics of excluded studies
Literature searching identified 1964 potential sources including 321 duplicate records which were discarded. A further 1487 records were excluded following evaluation based on their titles and abstracts: 356 did not report appropriate quality of life or patient satisfaction measures, 183 did not report costs based on individual patient data and the remaining 948 were not original research comparing palliative and usual approaches with end-of-life care. An additional 25 potential sources were identified by hand searching.
The 181 remaining sources were read in full. A further 168 were excluded: 93 did not report sufficient quality of life or patient satisfaction measures, 28 did not report sufficient cost data and the remaining 48 did not research relevant interventions.
Characteristics of included studies
Characteristics of the remaining 12 studies are displayed in table 3. They comprise six randomised clinical trial studies, two prospective cohort studies and four retrospective case-controlled studies. Study sample size ranged from 28 to 1754, median value 247. Nine were conducted in the USA and one each from UK, The Netherlands and Canada. Two were published in the 1980s, three in the 1990s and the remaining seven in the 2000s.
Characteristics of studies included in review
Research settings spanned hospital-based services, including specialist palliative care and other medical disciplines, hospice services and primary care. Patients with cancer were most frequently represented within samples although patients with a diagnosis of organ failure were also included. The reported time between admission to study and death ranged from 76 to 242 days with weighted average value of 128 days.
Quality of included studies
A specific quality scale was not employed; such tools are not supported by empirical evidence.23 Full analysis is presented covering common areas where methodological challenges can occur: recruitment of participants, measurement of outcomes and reporting of results.
Selection bias in recruitment and randomisation
All included case-controlled studies24–27 recruited from patients already referred into the interventions they investigated. Consequently, there is a concern that their research samples may not accurately reflect the wider population of patients from which they are drawn. One26 confirmed all referred patients and potential case controls were contacted and listed separate refusal rates for each arm; however, a significantly higher refusal rate in the control arm of this study removed a significantly older and frailer group from the control. Impact from this was impossible to quantify, but it is clear that the control cannot be a complete reflection of the wider population, and this limitation may be equally present in the remaining case-controlled research. Of the two cohort studies, one28 achieved complete recruitment, as the homeless population under investigation had limited access to alternative treatment; the second29 relied on referral of potentially eligible patients from other physicians rather than independent screening. However, a complete breakdown of people refusing to participate, including common reasons, was provided.
Randomised controlled trial studies30–35 were variable in their reporting. One30 recruited by referral from other medical teams; all other groups screened hospital admissions in specified settings to identify patients presenting who might match their predefined inclusion profile. Three detailed recruitment and refusal to participate,30 ,32 ,33 whereas one31 did not specify losses at recruitment. The cluster-randomised trials34 ,35 recruited whole neighbourhoods, avoiding individual refusals at recruitment.
Computerised randomisation was utilised by more recently published studies,30 ,31 while older studies32 ,33 did not report their procedure. Cluster randomisation was based either on a patient's postcode falling within one of two districts35 or Primary Care Practice.34
Performance and detection bias from incomplete blinding
The two cluster-randomised trials34 ,35 blinded participants by delivering services additional to usual care, without patients being aware of the research in progress. One cohort trial29 researched palliative medicine practised by non-specialist physicians and successfully blinded patients. No remaining trials were blinded; indeed, patients often received information during informed consent processes sufficient to identify their membership of control or intervention arms. If any significant placebo effect arose where patients allocated to palliative intervention experienced greater relief from symptoms simply because they expected to, results would reflect this.
Only two included studies24 ,30 specified that researchers assessing quality-of-life outcomes were blinded to which arm their interviewees belonged. For all other included studies, there are concerns that researchers’ subjective judgement might conform to an expected result when using, for instance, the Karnofsky Scale,36 which has been argued as being open to interpretation.37
Reporting bias from incomplete outcome data
Nine studies reported on sample size identical to their initial sample. Losses in remaining studies were fully accounted for by insurance companies ceasing participation,35 death of participants before quality-of-life outcomes could be assessed,30 and unspecified, but nonetheless, documented withdrawals from study.30 ,33
Numerical values for quality-of-life outcomes were not disclosed within results of three included studies.24 ,33 ,34 Terms including ‘not significant’ or ‘improved’ were employed in place of quantitative information.
Overall assessment of quality of included studies
Studies were generally small in size. Six were non-randomised and two were cluster-randomised rather than randomised by individual patient. None of the 12 studies recruited participants by screening all admissions into their research setting, provided details of all refusals, then clearly utilised objective randomisation. Therefore, potential selection bias, where intervention and control samples were drawn from two distinct populations, undermines the strength of results. Inability to blind patients or researchers in many settings also leaves results open to performance bias.
However, the results of studies included here appear to report complete data even if numerical outcomes are sometimes reported qualitatively. Concerns surrounding selection and performance biases may simply reflect challenges inherent to research in palliative care.38 ,39
Impact of palliative care on financial cost
The impact of palliative intervention on healthcare costs ranged from a 77% cost reduction to a 9% cost increase. The overall reduction was 35% weighted for study size; a summary of cost savings reported is shown in table 4.
Summary of cost savings identified in included studies
Greatest cost reduction compared with usual care, 77%, was observed in a homeless shelter-based hospice where the patient typically had complicated comorbidities and drug dependency, but no access to primary care.28 Admission to this hospice allowed unusually efficient management of these patients’ care. Another significant saving was reported for a hospital-based palliative care team in New York.27 Patients in New York State generate high average spend during their last years of life;6 as greater spend is significantly driven by aggressive usual practices, palliative care would promote markedly different diagnostic and treatment regimens compared with usual care.40
Only one study associated palliative intervention with increased costs.35 The nature of usual care in The Netherlands may explain this; usual care is primary care led with general practitioner and community nursing services on-call at all times. This limits scope for cost savings compared with hospital-intensive systems. Indeed, the intervention described was an addition to existing services rather than the substitution generally described where standard practices are replaced by less intensive treatments.
The oldest study reports no meaningful cost saving.33 Intervention patients, allocated to hospice care, had a greater number of inpatient days than the usual care group who were admitted to hospital. Daily cost of staying in hospice and hospital are cited as similar, hence the minimal cost saving reported. The age of this study may be relevant: hospital costs have grown over time,41 ,42 at a greater rate than hospice costs,43 therefore, repeating the study today might show improved cost savings.
Impact of palliative care on patient outcomes
Impacts on patient care and satisfaction reported within studies included in this review are varied (table 5). Five studies report primary outcome measures for satisfaction with care, and seven on a range of care outcomes.
Summary of patient outcome measures reported within included studies
Seven studies24 ,26 ,30–34 report outcome measures of satisfaction with care. Three studies did not identify a statistically significant difference between patient satisfaction in palliative intervention or control arms;26 ,32 ,34 the remaining four24 ,30 ,31 ,33 identify better patient satisfaction in palliative intervention compared with control. Two24 ,30 utilise the Reid-Gundlach Satisfaction with Services Instrument,44 two32 ,33 use the Ware Scale,45 one uses an End of Life Family Interview46 and the remaining two26 ,34 use bespoke assessment tools. However, absence of numerical values in some cases, and variety of measurement tools employed, preclude further generalisation.
Two studies25 ,27 assess the effectiveness of symptom control, both reporting that average length of hospital stay was reduced by around 25% following palliative intervention. However, scope of these studies was confined to specific hospital admissions, therefore, while the evidence suggests beneficial effect from palliative intervention, it has limited scope compared with the whole course of chronic illness.
Remaining outcome measures reported by studies included physical and mental health and function. Two palliative intervention studies28 ,29 reported significant benefit in control of pain and physical distress, and two31 ,35 reported significant benefit to the emotional quality of life for patients’ families. There were no further statistically significant differences in any measured outcomes between palliative and usual care. No study reported evidence of usual care delivering significantly better outcomes than palliative care.
Discussion
This review suggests palliative care services can help reduce futile treatments at end of life. The reduction of 35% in healthcare costs indicates less intensive treatment regimens, and some evidence of concurrent improvement in patient satisfaction implies the foregone treatments have not foregone quality of life. However, little of the evidence is robust. Selection bias and attrition are principal concerns in research to date. Furthermore, around half the outcome measures assessed in underlying studies do not demonstrate a statistically significant difference between control and intervention, numerical values are not reported for all measures, and a variety of outcome measures are utilised such that meta-analysis cannot be performed.
Absence of statistical significance precludes substantial discussion about clinical relevance. Included studies do not generally state procedures ensuring sample size is sufficient to detect differences between intervention and control; greater recruitment to future trials may be required to secure statistical power. Another positive development would be the adoption of standard outcome measures across future research to allow results from multiple studies to be combined, thereby improving statistical power.47 A recent report identified 106 outcome measurement tools currently employed in palliative care research throughout Europe:48 94 of these were utilised in fewer than 10 studies each, making standardisation a high priority to improve comparability of research.49
More fundamental, however, is the nature of existing research into palliative care and its potential to reduce futile treatments. One-third of articles identified in this review were discarded because they discussed the ethics and theoretical burden of futile treatment. These contributions are valuable, yet answers to many concerns raised by such work will come most readily from gathering empirical data.50 Theoretically, there is no insurmountable impediment to recruiting into controlled trials aiming to generate such data;51 efficacy of an intervention is not proven until robust trials themselves provide evidence either way52—the concept referred to as equipoise.53 Some argue that equipoise should be subordinated to individual patients’ interest and informed consent.54 However, evidence from trials is the gold standard of evidence-based medicine; without such information, patients may not be properly equipped to determine their best interest.
Within research included in this review, the generally accepted point at which equipoise lies is referral to palliative care during terminal illness: the traditional model of palliative care from figure 2. This reflects wider research priorities.55 By restricting length of time over which differences between control and intervention are allowed to develop, any benefit inherent in the palliative approach may not be fully realised by patients in trials, hence not reported. Indeed, that the current evidence base suggests some benefit from palliative care over short timescales implies a need for research into earlier palliative intervention to determine impact over the whole course of chronic illness.
Notwithstanding the ultimate goal of controlled trials, prospective cohort studies also have potential to address some questions surrounding futile treatment in the last years of life. Longitudinal studies would be appropriate to investigate the relationship between patients’ interactions with palliative care services and outcomes over the course of chronic illness. It would also be possible to measure patients’ exposure to palliative medicine in primary care, or tertiary disciplines outside of palliative care, and determine associations with quality-of-life outcomes. This would inform the direction of subsequent controlled trials. A similar approach was employed by one of the studies included in this review,29 and there is potential to continue this avenue of work further.
Limitations
Limitations relate largely to the diverse outcome measures that hamper direct comparability between studies and insufficient sample sizes to ensure statistical power in all studies individually. As half the included evidence comes from observational studies, there is also the limitation that the reported effect could reflect an underlying selection bias instead of a valid outcome.
The use of a single reviewer allowed scope for error or bias in the selection and data extraction stages of this review. Volume of material identified by initial literature search was also relatively constrained, and it is possible that a broader search, incorporating more terms and greater hand searching, might have identified further relevant sources for inclusion.
Conclusions
Current research into palliative care demonstrates a general trend wherein a palliative approach can reduce futile treatment during the last years of life, although evidence is not strong. Generating strong empirical evidence is essential to confirm this potential strength of palliative care. Agreed standard outcome reporting measures incorporating both quality of life and patient satisfaction should be a key component of all future work in this area. Observational research is sufficiently robust to answer some questions, so prospective cohort studies should be considered along with randomised trials. Studies of early palliative care interventions are particularly indicated.
Acknowledgments
We would like to thank Elizabeth Evans, Anne Finucane and John Jungpa Park for their constructive inputs and advice on early drafts of this review.
References
Footnotes
-
Contributors SAM devised the research question. Both authors formulated the search strategy. IH performed the searches, identified eligible studies, appraised study quality, extracted data and drafted the first sections of text. Both authors contributed to the final draft. IH is the guarantor.
-
Competing interests None.
-
Provenance and peer review Not commissioned; externally peer reviewed.