Early palliative care for adults with advanced cancer

Markus W Haun; Stephanie Estel; Gerta Rücker; Hans‐Christoph Friederich; Matthias Villalobos; Michael Thomas; Mechthild Hartmann

doi:10.1002/14651858.CD011129.pub2

Early palliative care for adults with advanced cancer

Authors' declarations of interest

Version published: 12 June 2017 Version history

https://doi.org/10.1002/14651858.CD011129.pub2

Collapse all Expand all

Abstract

available in

Background

Incurable cancer, which often constitutes an enormous challenge for patients, their families, and medical professionals, profoundly affects the patient's physical and psychosocial well‐being. In standard cancer care, palliative measures generally are initiated when it is evident that disease‐modifying treatments have been unsuccessful, no treatments can be offered, or death is anticipated. In contrast, early palliative care is initiated much earlier in the disease trajectory and closer to the diagnosis of incurable cancer.

Objectives

To compare effects of early palliative care interventions versus treatment as usual/standard cancer care on health‐related quality of life, depression, symptom intensity, and survival among adults with a diagnosis of advanced cancer.

Search methods

We searched the Cochrane Central Register of Controlled Trials (CENTRAL), MEDLINE, Embase, the Cumulative Index to Nursing and Allied Health Literature (CINAHL), PsycINFO, OpenGrey (a database for grey literature), and three clinical trial registers to October 2016. We checked reference lists, searched citations, and contacted study authors to identify additional studies.

Selection criteria

Randomised controlled trials (RCTs) and cluster‐randomised controlled trials (cRCTs) on professional palliative care services that provided or co‐ordinated comprehensive care for adults at early advanced stages of cancer.

Data collection and analysis

We used standard methodological procedures as expected by Cochrane. We assessed risk of bias, extracted data, and collected information on adverse events. For quantitative synthesis, we combined respective results on our primary outcomes of health‐related quality of life, survival (death hazard ratio), depression, and symptom intensity across studies in meta‐analyses using an inverse variance random‐effects model. We expressed pooled effects as standardised mean differences (SMDs, or Hedges' adjusted g). We assessed certainty of evidence at the outcome level using GRADE (Grading of Recommendations Assessment, Development, and Evaluation) and created a 'Summary of findings' table.

Main results

We included seven randomised and cluster‐randomised controlled trials that together recruited 1614 participants. Four studies evaluated interventions delivered by specialised palliative care teams, and the remaining studies assessed models of co‐ordinated care. Overall, risk of bias at the study level was mostly low, apart from possible selection bias in three studies and attrition bias in one study, along with insufficient information on blinding of participants and outcome assessment in six studies.

Compared with usual/standard cancer care alone, early palliative care significantly improved health‐related quality of life at a small effect size (SMD 0.27, 95% confidence interval (CI) 0.15 to 0.38; participants analysed at post treatment = 1028; evidence of low certainty). As re‐expressed in natural units (absolute change in Functional Assessment of Cancer Therapy‐General (FACT‐G) score), health‐related quality of life scores increased on average by 4.59 (95% CI 2.55 to 6.46) points more among participants given early palliative care than among control participants. Data on survival, available from four studies enrolling a total of 800 participants, did not indicate differences in efficacy (death hazard ratio 0.85, 95% CI 0.56 to 1.28; evidence of very low certainty). Levels of depressive symptoms among those receiving early palliative care did not differ significantly from levels among those receiving usual/standard cancer care (five studies; SMD ‐0.11, 95% CI ‐0.26 to 0.03; participants analysed at post treatment = 762; evidence of very low certainty). Results from seven studies that analysed 1054 participants post treatment suggest a small effect for significantly lower symptom intensity in early palliative care compared with the control condition (SMD ‐0.23, 95% CI ‐0.35 to ‐0.10; evidence of low certainty). The type of model used to provide early palliative care did not affect study results. One RCT reported potential adverse events of early palliative care, such as a higher percentage of participants with severe scores for pain and poor appetite; the remaining six studies did not report adverse events in study publications. For these six studies, principal investigators stated upon request that they had not observed any adverse events.

Authors' conclusions

This systematic review of a small number of trials indicates that early palliative care interventions may have more beneficial effects on quality of life and symptom intensity among patients with advanced cancer than among those given usual/standard cancer care alone. Although we found only small effect sizes, these may be clinically relevant at an advanced disease stage with limited prognosis, at which time further decline in quality of life is very common. At this point, effects on mortality and depression are uncertain. We have to interpret current results with caution owing to very low to low certainty of current evidence and between‐study differences regarding participant populations, interventions, and methods. Additional research now under way will present a clearer picture of the effect and specific indication of early palliative care. Upcoming results from several ongoing studies (N = 20) and studies awaiting assessment (N = 10) may increase the certainty of study results and may lead to improved decision making. In perspective, early palliative care is a newly emerging field, and well‐conducted studies are needed to explicitly describe the components of early palliative care and control treatments, after blinding of participants and outcome assessors, and to report on possible adverse events.

PICOs

Population

Intervention

Comparison

Outcome

The PICO model is widely used and taught in evidence-based health care as a strategy for formulating questions and search strategies and for characterizing clinical studies or meta-analyses. PICO stands for four different potential components of a clinical question: Patient, Population or Problem; Intervention; Comparison; Outcome.

See more on using PICO in the Cochrane Handbook.

Plain language summary

available in

Early palliative care for adults with advanced cancer

Review question

What is the evidence for the effects of early palliative care on quality of life, survival, depression, and symptom intensity in people with advanced cancer?

Background

Frequently, cancer is diagnosed at a late stage, and the disease might have progressed through anticancer treatment. Patients can choose to start or continue anticancer treatment at the potential cost of side effects. Standard care means that all patients are offered palliative care towards the end of life. However, patients may be able to receive palliative care a lot earlier. This approach, which is known as early palliative care, begins at the time of, or shortly after, the diagnosis of advanced cancer. Often, early palliative care is combined with anticancer treatment such as chemotherapy or radiotherapy. Early palliative care, whether provided by the attending oncologist or by specialist teams, involves empathetic communication with patients about their prognosis, advance care planning, and symptom assessment and control.

Study characteristics

In October 2016, we searched for clinical trials on early palliative care in adults with advanced cancer. We included seven studies and found 20 ongoing studies. Most of the studies included participants older than 65 years of age on average, diagnosed with different tumour types and receiving treatment in tertiary care centres in North America. Most of these studies compared early palliative care with standard oncological (cancer) care. All studies were funded by government agencies.

Key results

When evaluated together in a meta‐analysis, studies showed that in patients with advanced cancer, early palliative care may slightly increase quality of life. It may also decrease symptom intensity to a small degree. Effects on survival and depression are uncertain. A single study reported side effects (adverse events), for example, more pain and reduced appetite. For the remaining six studies, information about side effects was not published, but trial authors told us they had not observed any.

Certainty of the evidence

We rated the certainty of the evidence using four levels: very low, low, moderate, and high. Evidence of very low certainty means that we have little confidence in the results. Evidence of high certainty means that we are very confident in the results. We found that certainty of the evidence was low for health‐related quality of life and symptom intensity, and was very low for depression and survival. We downgraded certainty of the evidence for various reasons, for example, problems in the way studies were carried out, differences between studies, and the small number of studies. We remain uncertain about the effects of early palliative care; therefore we have to interpret the results with caution. When published, ongoing studies may provide more evidence, and this may affect the certainty of the results.

Authors' conclusions

Implications for practice

All stakeholders shall be advised that besides the seven included studies, we identified 20 ongoing studies and 10 studies awaiting assessment. Therefore, the evidence base for early palliative care in cancer is growing, and conclusions remain preliminary.

For people with advanced cancers

Available evidence of very low to low certainty suggests that patients with advanced cancers could benefit from early palliative care with respect to small improvements in quality of life and symptom intensity. At this point, effects on survival and/or on depressive symptoms remain uncertain. Nevertheless, to improve quality of life and reduce symptoms, patients could approach their attending physician and request referral to palliative care at an early stage of disease.

For clinicians

From a practitioner's perspective, some previous reviews have reported definitive success of early palliative care interventions for improving quality of life, controlling bodily symptoms and depressive symptoms, and prolonging life. However, according to our results, these claims were likely to be at least premature for the entire group of patients with advanced cancer. Besides studies favouring such outcomes, we also detected a study with possibly negative effects on symptoms and survival. More research is needed before solid conclusions regarding routine care can be drawn. Included studies were heterogeneous in many aspects. Although we found some possibly clinically relevant evidence for the effectiveness of early palliative care in terms of quality of life and symptom intensity, the certainty of this evidence was low to very low. Results of our review do not support that early palliative care leads to prolonged survival in general. Therefore, at this point, clinicians could consider early palliative interventions on a case‐by‐case basis to address quality of life alongside symptom intensity and counsel patients adequately (Peppercorn 2011), but refrain from claiming that these interventions will have an additional impact on survival, or that they offer the only way to target quality of life. The patient should be informed adequately and his or her wishes should be respected during treatment planning.

For policy makers

Access to additional specialised palliative care teams is currently limited and availability of services is often absent even in developed countries (Kelley 2015). Hence, policy makers face the challenge of systematically introducing early palliative care into environments with potentially limited available resources. At this point, we have found no evidence that specialised palliative care teams (as part of integrated care) are superior to those providing a generic palliative care approach (co‐ordinated care). In addition, cost utility of early palliative care remains unclear at this point. However, findings of our review do support strong implementation of elements of early palliative care in clinical routines. These elements may consist of advanced communication for identification of patient priorities, care co‐ordination towards symptom control, and comprehensive psychosocial care potentially involving caregivers (Janssens 2016).

Implications for research

General

With only seven studies included, we clearly need additional sufficiently powered and well‐designed studies. Especially with respect to effect estimates of outcomes other than health‐related quality of life and symptom intensity, we are in need of larger (i.e. multi‐centred) studies to establish robust evidence. Besides uniform and significant effects, we found that studies differed in average effect size or even in the direction of effects. To explain this heterogeneity with respect to entity, interventions, dose, and study methods further, we need to continue to work on an even clearer, evidence‐based definition for early palliative care (Lee 2015). A clearer definition would constitute the foundation for establishing and comparing interventions across studies, and first efforts would stem from qualitative studies on core interventional elements (Hui 2015a; Jacobsen 2011; Janssens 2015; Yoong 2013). In general, we consider it essential to better describe training as well as therapist adherence. It is equally important to provide more information on the usual care provided locally. To ensure clear interpretation of findings, we should provide a thorough and extensive description of both experimental and control conditions.

Against the background of evidence presented here, it has to be considered that early palliative care in cancer is still a relatively new treatment approach that has so far almost exclusively been evaluated in the context of tertiary care contexts; and is not a clearly defined and homogeneous type of intervention; but that research is important because early palliative care may have the potential to improve current clinical practice in advanced cancer diseases.

Design

Interventions should be described under the different models proposed for early palliative care, and frequency and duration of treatment should be stated. For strengthening the internal validity of effect estimates, future studies need to be rigorous in both design (ideally controlling for palliative care skills/training of oncologists/palliative care physicians and high‐ vs low‐volume centres) and delivery, and should be based on sufficient power. Specifically, investigators in future studies should use all available measures to control for selection bias (i.e. to ensure adequate allocation concealment), performance bias (i.e. to blind study participants), detection bias (i.e. to blind outcome assessors), and selective reporting (i.e. to report studies as indicated in the preregistration). It is most important that investigators provide detailed descriptions of the several components of both intervention and control conditions. Notwithstanding, for ethical and disease‐inherent reasons, conducting RCTs and restricting attrition are major challenges in palliative care (Wee 2008). With respect to setting, interventions should be expanded beyond high‐volume tertiary referral hospitals in Western countries. It has been shown that clinical expertise and centre volume impact treatment effect (Choudhry 2005). Specifically, treatment in comprehensive cancer centres is often linked with superior survival (Wolfson 2015). Although research on the transferability of early palliative care interventions to more naturalistic contexts has already commenced, we would encourage investigators to focus first on rigorous RCTs that follow conventional designs to determine internal validity, substantiate findings, and increase the certainty of evidence. Concerning homogeneity of samples, it might be worthwhile to investigate 'tumour homogeneous' samples to better account for specific disease trajectories and patient characteristics (e.g. male gender and young age in patients with lung cancer, as recently demonstrated by Nipp 2016) that are likely to specifically impact the effectiveness of an early palliative care approach. Only in a second step, that is, when certainty of effect estimates is higher, may pragmatic studies looking at implementability of early palliative care be initiated (Treweek 2009). If early palliative care proves effective in the future, we regard continuation of studies along this pragmatic‐explanatory continuum as crucial (Loudon 2015; Thorpe 2009).

Measurement (endpoints)

Concerning measurements, health‐related quality of life and symptom intensity have emerged as appropriate outcomes that are possibly sensitive to change and can be recommended for routine collection. In addition, affective symptoms should be assessed, as they constitute a particular salient distress factor in patients with advanced cancer (Haun 2014; Mehnert 2014). Compared with these endpoints, survival is controversial, as it is not the primary aim of palliative interventions. However, in terms of further advancements, information on how the intervention may work and on essential components should be derived. Moreover, future studies need to harmonise measurements with respect to applied scales and predefined time points. The most common follow‐up for primary outcomes currently occurs at 12 weeks. Blinding of outcome assessment is essential, as is its explicit reporting in publications.

Summary of findings

Open in table viewer

Summary of findings for the main comparison. Early palliative care for adults with advanced cancer

Clinical question: Should early palliative care be preferred over treatment as usual for improving health‐related quality of life, depression, and symptom intensity in patients with advanced cancer?
Patient or population: patients with advanced cancer Settings: mainly outpatient care in Australia, Canada, Italy, and the USA Intervention: early palliative care Comparison: treatment as usual Follow‐up: at 12 weeks or mean difference in repeated measurement results for longitudinal designs
Outcomes	*Anticipated absolute effects (95% CI)**		Relative effect (95% CI)	Number of participants (studies)	Certainty of the evidence (GRADE)	Comments
Outcomes	Risk with treatment as usual	Risk with early palliative care	Relative effect (95% CI)	Number of participants (studies)	Certainty of the evidence (GRADE)	Comments
Health‐related quality of life (HRQOL), SD units: measured on FACIT‐Pal, TOI of FACT‐Hep, TOI of FACT‐L, FACT‐G, McGill Quality of Life, FACIT‐Sp. Higher scores indicate better HRQOL. Follow‐up: range 12 weeks to 52 weeks	HRQOL score improved on average 0.27 (95% CI 0.15 to 0.38) SDs more in early palliative care participants than in control participants		‐	1028 (7 RCTs)	⊕⊕⊝⊝ LOW^1,2,3	By conventional criteria, an SMD of 0.2 represents a small effect, 0.5 a moderate effect, and 0.8 a large effect (Cohen 1988)
Health‐related quality of life (HRQOL), natural units: measured on FACT‐G (from 0 to 108)	Baseline control group mean score at 70.5 points^a	HRQOL score improved on average 4.59 (95% CI 2.55 to 6.46) points more in early palliative care participants than in control participants	‐	1028 (7 RCTs)	⊕⊕⊝⊝ LOW^1,2,3	Calculated by transforming all scales to the FACT‐G in which the minimal clinically important difference is approximately 5 and the SD in the cancer validation sample was 17.0 (Brucker 2005)
Survival: estimated with the unadjusted death hazard ratio	Study population^b		HR 0.85, 95% CI 0.56 to 1.28	800 (4 RCTs)	⊕⊝⊝⊝ VERY LOW^1,4,5,6
	61 per 100	56 per 100 (41‐71)	HR 0.85, 95% CI 0.56 to 1.28	800 (4 RCTs)	⊕⊝⊝⊝ VERY LOW^1,4,5,6
Depression, SD units: measured on CES‐D, HADS‐D, PHQ‐9. Higher scores indicate higher depressive symptom load. Follow‐up: range 12 weeks to 52 weeks	Depression score improved on average ‐0.11 (95% CI ‐0.26 to 0.03) SDs more in early palliative care participants than in control participants		‐	762 (5 RCTs)	⊕⊝⊝⊝ VERY LOW^1,2,4	By conventional criteria, an SMD of 0.2 represents a small effect, 0.5 a moderate effect, and 0.8 a large effect (Cohen 1988)
Depression, natural units: measured on CES‐D (from 0 to 60). Higher scores indicate higher depressive symptom load	Baseline control group mean score at 13.8 points^c	Depressive symptoms score improved on average ‐0.98 (95% CI ‐2.31 to 0.27) points more in early palliative care participants than in control participants	‐	762 (5 RCTs)	⊕⊝⊝⊝ VERY LOW^1,2,4	Calculated by transforming all scales to CES‐D and applying an SD of 8.9 from baseline control group score in Bakitas 2009
Symptom intensity, SD units: measured on ESAS, QUAL‐E Symptom Impact Subscale, SDS, RSC, LCS of FACT‐L, HCS of FACT‐Hep. Higher scores indicate higher symptom intensity. Follow‐up: range 12 weeks to 52 weeks	Symptom intensity score improved on average ‐0.23 (95% CI ‐0.35 to ‐0.1) SDs more in early palliative care participants than in control participants		‐	1054 (7 RCTs)	⊕⊕⊝⊝ LOW^1,2,3	By conventional criteria, an SMD of 0.2 represents a small effect, 0.5 a moderate effect, and 0.8 a large effect (Cohen 1988)
Symptom intensity, natural units: measured on ESAS (from 0 to 900). Follow‐up: range 12 weeks to 52 weeks	Baseline control group mean score at 286.3 points^c	Symptom intensity symptoms score improved on average ‐35.4 (95% CI ‐53.9 to ‐15.4) points more in early palliative care participants than in control participants	‐	1054 (7 RCTs)	⊕⊕⊝⊝ LOW^1,2,3	Calculated by transforming all scales to the ESAS and applying an SD of 154.0 from baseline control group score in Bakitas 2009
Adverse events	See comment	See comment	Not estimable	1614 (7 RCTs)	See comment	Most often, study authors did not address assessment or findings on adverse events in their study publications. However, on request, authors of 6 studies described the tolerability of early palliative care as very good. A single study mentioned adverse events only in the early palliative care group, i.e. higher percentage of participants with severe scores for pain and poor appetite along with higher level of unmet needs (Tattersall 2014)
*Risk in the intervention group (and its 95% confidence interval) is based on assumed risk in the comparison group and relative effect of the intervention (and its 95% CI) ^aApproximate average of baseline control group FACT‐G scores across 4 included studies (Bakitas 2009; Bakitas 2015; Maltoni 2016; Temel 2010) ^b12‐Month follow‐up control group risk in the largest study reporting on survival (Bakitas 2009) ^cBaseline control group CES‐D score in the largest study reporting on depression (Bakitas 2009) CI: confidence interval; GRADE: Grading of Recommendations Assessment; HR: unadjusted death hazard ratio; SD: standard deviation; SMD: standardised mean difference
GRADE Working Group grades of evidence High certainty: We are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect. Very low certainty: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect.
¹We downgraded 2 points owing to very serious limitations in study quality (high risk of bias across studies) ²We decided against downgrading for indirectness, although 2 studies were conducted exclusively in patients with metastatic pancreatic and advanced lung cancer, respectively (Maltoni 2016; Temel 2010). We decided against downgrading for inconsistency, as we did not detect significant heterogeneity ³We decided against downgrading for imprecision, as the optimal information size (OIS) criterion was met, and the 95% confidence interval around the difference in effect between intervention and control excludes zero ⁴We downgraded 1 point for imprecision, as the optimal information size (OIS) criterion was met, but the 95% confidence interval around the difference in effect between intervention and control includes zero. The 95% confidence interval fails to exclude harm ⁵We decided against downgrading for important inconsistency (large I²) because we had downgraded by 3 points already ⁶We decided against downgrading for indirectness, as only a single study was conducted exclusively in patients with advanced lung cancer (Temel 2010)

Background

Research has led to remarkable improvements in cancer treatment, but at the time of diagnosis, some patients still have a reduced life expectancy. Incurable cancer can pose an enormous challenge for patients, their families, and medical professionals, and can affect patients' quality of life in many ways (Addington‐Hall 1995). Interventions tailored to improve the physical and psychological well‐being of people with cancer are of utmost importance. Palliative care comprises an "approach that improves the quality of life of patients and their families facing the problem associated with life‐threatening illness, through the prevention and relief of suffering by means of early identification and impeccable assessment and treatment of pain and other problems, physical, psychosocial and spiritual" (WHO 2013). Interdisciplinary care and caregiver support assist healthcare professionals in delivering the essential elements of palliative care by managing the patient's quality of life and controlling symptoms (Hui 2013a). However, although early access is inherent in the definition of palliative care, usual practice is still limited to the terminal phase of illness.

Description of the condition

With an incident rate of 14.9 million cases and 8.2 million deaths in 2013, malignant neoplastic diseases remain one of the leading causes of death worldwide (Global Burden of Disease Cancer Collaboration 2015). Globally, the most common entities and causes of cancer‐related mortality, measured as disability‐adjusted life‐years (DALYs), are breast cancer in women and lung cancer in men. Cancer incidence has been estimated to increase yearly by 1%, with the growing population worldwide and the demographic shift towards an ageing population in developed countries serving as the paramount factors for future cancer burden (Boyle 2008).

Despite significant progress in our understanding of the risk factors for cancer, development of methods for early identification of some cancers or precancerous diseases, and sound advances in the treatment of many cancers previously deemed fatal (e.g. breast, prostate, melanoma, Hodgkin's disease), cancer continues to cause the premature death of many individuals (particularly cancers of the pancreas, lung, brain, and stomach) (Prigerson 2015; Quaresma 2015). At the time of diagnosis, chances of curative treatment are often minimal owing to advanced disease. The American Cancer Society defines advanced cancer as "cancers that cannot be cured", and metastatic cancer as tumours that "have usually spread from where they started to other parts of the body" (American Cancer Society 2013). However, not all advanced cancers are metastatic. For example, brain tumours may be considered advanced because they are not curable, and life‐threatening, even in the absence of metastasis. In addition, the survival rate of patients remains very poor, especially for metastatic lung cancer and for pancreatic and biliary tract malignancies.

Because death is anticipated in many of these cases, it is essential that appropriate treatment plans are developed to improve survival, while aiming for a subjectively worthwhile quality of life. Both symptom control and disease‐modifying therapy are needed in these situations. By causing a major decline in physical efficiency and persistent chronic pain, advanced cancer regularly puts the physical and psychological integrity of patients at high risk. In many cases, appropriate execution of necessary medical treatments and of the daily routine at home demands continuous familial and often additional external support. Symptoms such as pain, fatigue/drowsiness, low appetite and/or anorexia‐cachexia syndrome, dysphagia, nausea, diarrhoea, constipation, shortness of breath, and mental confusion are often independent prognostic factors for predicting life expectancy in people with recently diagnosed incurable cancer (Trajkovic‐Vidakovic 2012). In addition, patients and their caregivers may be concerned about burdensome existential ruminations leading to psychological distress on both sides, with long‐term risk of severe impairment in physical and psychological health among patients and caregivers, as well as declining resources of social support (Mehnert 2014; Singer 1999; Sklenarova 2015). Such developments within the family often promote conflict about responsibilities regarding decision making concerning therapeutic and everyday challenges. Economic consequences frequently comprise, for example, reduced family income or considerable out‐of‐pocket medical spending, leading to financial hardship for patients and their families (Elkin 2010; Zhang 2009). Owing to these strains, professional support gains extraordinary importance in alleviating physical discomfort and in contributing to improved quality of life among patients.

Description of the intervention

Palliative care is provided to reduce suffering and improve quality of life among patients and their caregivers. In recent years, the term 'early palliative care' was introduced to differentiate palliative care treatments applied early in the course of a life‐threatening disease from palliative care delivered mainly with high symptom burden or in the terminal phase of illness, as was the established clinical practice. In cases of advanced cancer, early palliative care is provided alongside active disease treatment such as chemotherapy or radiotherapy.

A typical treatment protocol for investigators in early palliative care trials encompasses communication with the patient about illness and prognosis, symptom assessment and management, support for coping, and regular follow‐ups. According to the latest consensus definition of palliative care, such treatment is called 'early' when it is administered within eight weeks of diagnosis of advanced cancer (Ferrell 2017). Other qualitatively identified elements include relationship and rapport building, development of coping skills, understanding of the illness, and discussion of available cancer treatments, including end‐of‐life planning (Yoong 2013). A prerequisite for palliative care in such an early situation is readiness of health care professionals to engage in coherent and empathetic communication with the patient (de Haes 2005; Dowsett 2000; Meyers 2003; Morrison 2004; Sinclair 2006). Early palliative care commonly is focussed on outlining realistic and attainable goals of treatment (van Mechelen 2013) and facilitating patient choices by providing adequate information and assessment of patient values and preferences with regard to advance care planning (Levy 2016). The inherent belief is that symptoms can be prevented or can be managed more easily when treated early, thereby improving the patient's quality of life. Most treatments involve education, evidence‐based methods used for symptom control, and psychosocial support. In essence, early palliative care is based on a proactive attitude and usually is provided to patients without high symptom burden or unmet psychosocial needs.

Researchers have identified the following models of palliative care (Hui 2015a).

Solo practice model: This model ascribes responsibility for cancer diagnosis and treatment as well as palliative care exclusively to the primary oncologist.
Co‐ordinated care model: As is often observed in common clinical practice, the primary oncologist in collaboration with the primary nursing team offers and co‐ordinates supportive/palliative care. As part of this so‐called congress model, primary providers refer patients to various specialists, who address domains of palliative care (other physicians, clinical nurse specialists, social workers, chaplains, psychotherapists, and clinical psychologists or psychiatrists).
Integrated care model: In this model, oncologists routinely refer patients to specialist palliative care teams early in the disease trajectory, rather than excluding involvement of other specialists.

Regardless of the model selected, early palliative care can be delivered across a breadth of settings, including community centres, hospitals, and inpatient hospice units. Community hospice services may also support patients at an earlier stage of disease in the day care/outpatient setting.

Comparator arms in early palliative care trials generally consist of usual oncology care. This may include referral to or application of palliative measures at any time along the disease trajectory as initiated by an oncologist, patient, or family member. However, referral to or application of palliative measures are not usually offered actively to all patients.

How the intervention might work

With a focus on intensified doctor‐patient communication, early palliative care may lead to higher levels of social support and may increase the likelihood of acceptance of the diagnosis and illness severity. These effects, along with the augmented satisfaction of the patient‐physician relationship, may improve the patient's openness to symptom control and psychosocial interventions, thereby reducing distress. Reduced distress itself is associated with improved quality of life and is consistently associated with survival (Gotay 2008; Irwin 2013; Pinquart 2010). Furthermore, patients and family members undergoing early palliative care are better informed about treatment directives and end‐of‐life decisions, which promotes higher self‐efficacy and a greater sense of control of decisions with respect to a person's individual values (McClain 2003). On the one hand, better symptom control and psychosocial function could promote better adherence with reasonable treatment plans. On the other hand, palliative care is linked to less aggressive cancer treatment, such as reduced use of questionable chemotherapy and less treatment time in intensive care units (Earle 2008). This tendency to de‐escalate treatment intensity in final, irreversible health conditions, together with extension of outpatient and community palliative care services, is important for patients' well‐being as well as to socioeconomics (Lowery 2013; Smith 2003).

Why it is important to do this review

Evidence for the effects of late palliative care is ambiguous because the time required to establish beneficial effects may be too short (El‐Jawahri 2011; Gomes 2013; Higginson 2010; Zimmermann 2008). Palliative interventions applied early, around the time of diagnosis of incurable advanced cancer, may be more favourable for improving symptom and disease management (Levy 2016), leading some investigators to believe that a paradigm shift has occurred (e.g. Kamal 2016; Kelley 2010; Schenker 2015). To date, although several reviews on early palliative care interventions for patients with advanced cancer have been published (Bauman 2014; Davis 2015; El‐Jawahri 2011; Gomes 2013; Greer 2013; Higginson 2010; Hui 2015b; Parikh 2013; Salins 2016; Smith 2012; Tassinari 2016; von Roenn 2011; Zambrano 2016; Zhi 2015; Zimmermann 2008), to our knowledge, no meta‐analysis has been carried out. An overview of interventions applied within this framework has not been provided, and uncertainty remains about the general impact of such interventions on patient‐ and caregiver‐related outcomes.

Objectives

Methods

Criteria for considering studies for this review

Types of studies

We included randomised controlled trials (RCTs) and cluster‐randomised controlled trials (cRCTs).

Types of participants

Patients were eligible if they had been given the diagnosis of a malignant tumour entity at an advanced stage (as assessed by the oncologist and based on disease stage and tumour type) and without curative treatment options (i.e. owing to metastatic disease or inoperability, or both). In accordance with information provided by the American Cancer Society (American Cancer Society 2013), we defined advanced cancers as "cancers that cannot be cured," and that, in the case of metastatic cancer, "have usually spread from where they started to other parts of the body." For all malignant entities, limited prognosis can be a common disease consequence and, therefore, constituted the main eligibility criterion for inclusion of participants in this review. Participant survival had to be estimated at two years or less. We did not include disabled long‐term survivorship patients, although such patients may also be in need of early palliative care. Assessment of prognosis had to be based on disease stage as an objective clinical indicator, in conjunction with the clinician's estimation provided by the primary oncologist. We considered for inclusion only studies of adults, 18 years of age and older, and we excluded studies of adults given the diagnosis during childhood and of people already in the terminal phase of illness (predicted survival of less than three months with eligibility for hospice care) at study enrolment.

Types of interventions

As defined in a previous review (Zimmermann 2008), we included all types of professional palliative care services that provided or co‐ordinated comprehensive care for patients at early advanced stages of cancer. Investigators had to state explicitly early palliative care intent, or this had to be reflected in the sample composition, that is, most participants had to be enrolled shortly after diagnosis of advanced disease. In addition, care had to be multi‐dimensional (i.e. the intervention had to target at least the "physical" and "psychological" domains of quality of life). We excluded studies evaluating the impact of only one domain of quality of care (e.g. medication for pain, psychological interventions for depression). We did not include stand‐alone palliative therapies provided to modify the disease to prolong life (e.g. palliative chemotherapy) or relieve symptoms (e.g. palliative radiotherapy). We applied no restrictions on type of delivery (inpatient, outpatient) or place of consultation (clinic, patient's home). The active comparator was treatment as usual/standard cancer care (i.e. no systematic palliative treatment or delayed or late palliative care).

Types of outcome measures

We included two 'Summary of findings' tables as presented in the PaPaS Author Guide (AUREF 2012) and as recommended in the Cochrane Handbook for Systematic Reviews of Interventions, Chapter 4.6.6 (Higgins 2011a). The 'Summary of findings' table includes outcomes of quality of life, survival, depression, symptom intensity, and adverse events (see summary of findings Table for the main comparison).

Primary outcomes

We included the following primary outcomes and corresponding measures.

Health‐related quality of life (e.g. measured by Functional Assessment of Cancer Therapy (FACT), City of Hope Quality of Life Questionnaire, European Organization for Research and Treatment of Cancer Quality of Life Questionnaire (EORTC QLQ‐C30), McGill Quality of Life Questionnaire, 36‐Item Short Form Health Survey (SF‐36), or the Supportive Care Needs Survey (SCNS)).
Survival.
Depression (e.g. measured on Beck Depression Inventory (BDI), Hamilton Rating Scale for Depression (HAM‐D), Hospital Anxiety and Depression Scale (HADS), Patient Health Questionnaire (PHQ‐9), or Centre for Epidemiological Studies ‐ Depression Scale (CES‐D)).
Symptom intensity (e.g. measured on Edmonton Symptom Assessment Scale (ESAS), or on Brief Pain Inventory (BPI)).

Outcomes had to be measured through self‐report questionnaires, participant records or interviews that determined measures of adequate psychometric properties. Scales had to reflect continuous or time‐to‐event data for survival. The most relevant time points used to measure outcome were "medium term" (one to four months after initiation of early palliative care) for self‐rated outcomes, and "long term" for survival.

Secondary outcomes

We assessed three categories of secondary outcomes.

Caregiver burden as a caregiver‐related outcome (e.g. measured on Caregiver Strain Index (CSI), Supportive Care Needs Survey for Partners & Caregivers (SCNS‐P&C), BDI, HAM‐D, PHQ‐9, or CES‐D).
Healthcare utilisation (e.g. measured as length of hospital stay in days, number of outpatient attendances, direct or indirect medical resource use) as an economic outcome.
Harms/adverse events (measured as the binary outcome "Any adverse event: yes/no").

During compilation of the review, we made the post‐protocol decision to report results for outcomes that had not been prespecified in the protocol (namely, place of death, problems with medical interactions, satisfaction with care, and illness and understanding of prognosis). We did this so we could provide a more comprehensive overview of outcomes available for early palliative care.

Search methods for identification of studies

We prepared a highly sensitive literature search strategy by which to identify eligible studies. Joanne Abbott (JA), the Information Specialist for the Cochrane Pain, Palliative and Supportive Care Group, and Maria‐Inti Metzendorf, of the Library of the Medical Faculty of Mannheim, Heidelberg University, supported drafting of search strategies. Two review authors (MWH and SE) documented the search process and the records, assessed potentially relevant studies, and made the final selection for inclusion and data extraction. We resolved disagreements by discussion and, in the case of eight studies, by consultation with an arbiter (MH).

Electronic searches

We searched the following electronic databases without language restrictions.

CENTRAL (Cochrane Central Register of Controlled Trials) via the Cochrane Library (2016, Issue 9).
MEDLINE (Medical Literature Analysis and Retrieval System Online) via OvidSP 1946 to 11 October 2016.
Embase (Excerpta Medica dataBASE) via OvidSP 1974 to 11 October 2016.
PsycINFO via OvidSP 1887 to 11 October 2016.
Cumulative Index to Nursing and Allied Health Literature (CINAHL) via EBSCO 1937 to 11 October 2016.
OpenGrey (European Association for Grey Literature Exploitation ‐ EAGLE) (www.opengrey.eu) via EXALEAD 1985 to 11 October 2016.

We performed free‐text search of titles, abstracts, and keywords, as well as medical subject headings (MeSH), during searches. We ran the search from the earliest publication date possible in each database. We tailored searches to individual databases. Please see Appendix 1 for the full MEDLINE search strategy in OVID, and see Appendix 2, Appendix 3, Appendix 4, Appendix 5, and Appendix 6 for all other search strategies.

In addition, we searched citations of key review authors (Marie A Bakitas, Jennifer S Temel, Martin H N Tattersall, Camilla Zimmermann) via Web of Science and the "related article" feature of PubMed.

The Information Specialist for the Cochrane Pain, Palliative and Supportive Care Group conducted searches in CENTRAL, MEDLINE, Embase, and PsycINFO via the OvidSP interface, which was not available at Heidelberg University Hospital.

Searching other resources

Trial registers

We searched the metaRegister of Controlled Trials (mRCT) (www.controlled‐trials.com/mrct), clinicalTrials.gov (www.clinicaltrials.gov), and the World Health Organization (WHO) International Clinical Trials Registry Platform (ICTRP) (apps.who.int/trialsearch/) on 11 October 2016, to identify additional completed or ongoing studies (search term: "palliative").

Reference lists

We checked the reference lists of 15 relevant reviews (Bauman 2014; Davis 2015; El‐Jawahri 2011; Gomes 2013; Greer 2013; Higginson 2010; Hui 2015b; Parikh 2013; Salins 2016; Smith 2012; Tassinari 2016; von Roenn 2011; Zambrano 2016; Zhi 2015; Zimmermann 2008) and all potentially relevant records.

Correspondence

We contacted six authors of main studies and 21 investigators who were known to be carrying out research in this area for additional studies and to provide unpublished data.

Data collection and analysis

Selection of studies

During database searches, we downloaded all retrieved records, including abstracts, and compiled them by using the reference manager Endnote X6. We removed duplicate records electronically through Endnote X6 and manually after checking study authors, titles, and abstracts. As the next step, two review authors (MWH and SE) independently assessed search results and excluded records that obviously did not fulfil inclusion criteria. Raters linked together multiple reports of the same study. For remaining studies marked as potentially relevant by either review author, we obtained full‐text documents and checked respective studies for eligibility. To ensure reproducibility of judgements regarding studies to be included, two unblinded raters (MWH and SE) again independently assessed full‐text documents and agreed on which studies should be included in the review. To formally measure agreement between raters with regard to study inclusion, we calculated the simple kappa statistic to determine whether eligibility criteria should be reconsidered. When raters disagreed regarding study eligibility, we reached consensus or referred the question to an arbiter (MH). At this stage, we compiled a list of excluded records along with the primary reason for each exclusion (see Characteristics of excluded studies).

Data extraction and management

To extract data from each study, we set up and pilot‐tested a data collection form and prepared coding instructions in accordance with the checklist proposed by Cochrane (Higgins 2011b, Table 7.3.a).

Two unblinded review authors (MWH, as topic area specialist, and SE, as methodologist) independently extracted data from published study reports and recorded them on a standard data collection form. We collected information on study design, setting, participants (including sample sizes), intervention details, outcomes (including time points), results, and missing data, as well as on risk of bias. We collated onto a single collection form data from multiple reports of the same study.

For meta‐analysis of continuous outcome variables using standardised mean differences (SMDs), we extracted mean values and standard deviations of outcome measurements, as well as the number of participants included in each intervention group. For cluster‐randomised trials, we applied the methods described in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011d). We registered effect estimates with confidence intervals and P values.

For time‐to‐death outcomes, we obtained estimates of log hazard ratios and their standard errors (Tierney 2007). Again, we resolved disagreements by discussion and, when necessary, by consultation with an arbiter (MH). For completion of study details and missing numerical results, we contacted study authors when necessary. One review author (SE) entered data suitable for pooling into Cochrane software Review Manager version 5.3 (RevMan 2014), and a second review author (MWH) verified entries. We report the characteristics of included studies in sufficient detail in the Characteristics of included studies table. We contacted study authors to request unpublished data for meta‐analysis when required.

Assessment of risk of bias in included studies

Two review authors independently assessed risk of bias for each study using the criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions and resolved disagreements by discussion (Higgins 2011c). When applying the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) system, we gave major priority to rating the certainty of evidence by assessing study outcomes with regard to risk of bias (Balshem 2011; Guyatt 2011a). We applied the Oxford Quality Score as the basis for eligibility (Jadad 1996), limiting inclusion to randomised controlled and cluster‐randomised trials. Blinding of personnel was neither mandatory for inclusion nor necessary for risk of bias assessment because blinding usually is not feasible in palliative care studies (Piggott 2004); however, we assessed blinding of outcome assessment while assessing risk of bias (Movsisyan 2016a). Furthermore, high attrition rates did not automatically lead to exclusion, as these are to be expected in palliative care studies. Recruitment is a major challenge in the palliative care context, justifying more pragmatic methods such as cRCTs (median attrition rate at 40%, according to Zimmermann 2008). Rather, we regarded differences between intervention and control groups with reference to incomplete outcome data (Guyatt 2011a) as crucial criteria for ascribing risk of attrition bias. We decided to omit sample size as a criterion for risk of bias, as recommended in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011c). Rather than assigning a priori judgement that small studies were at high risk of bias, we explored this assumption in our review by investigating small‐study bias and by assessing sample size when grading the evidence for each outcome.

Furthermore, two unblinded independent review authors (MWH and SE) conducted a domain‐based evaluation by completing a 'Risk of bias' assessment for each included study, using a data collection form that included seven specific domains: random sequence generation, allocation concealment, blinding of participants, blinding of outcome assessment, incomplete outcome data, selective reporting, and other bias. For cRCTs, we assessed risk of bias with regard to recruitment bias, baseline imbalance, loss of clusters, incorrect analysis, and comparability with individually randomised studies. We assessed the following for each study.

Random sequence generation (checking for possible selection bias): We assessed the method used to generate the allocation sequence as having low risk of bias (any truly random process, random number table, computer random number generator) or unclear risk of bias (method used to generate sequence not clearly stated). We used a non‐random process to exclude studies that were at high risk of bias (odd or even date of birth; hospital or clinic record number).
Allocation concealment (checking for possible selection bias): The method used to conceal allocation to interventions before assignment determines whether the intervention allocation could have been foreseen in advance of, or during recruitment, or changed after assignment. We assessed methods as having low risk of bias (telephone or central randomisation, consecutively numbered sealed opaque envelopes) or unclear risk of bias (method of allocation concealment not clearly stated). We excluded studies that did not conceal allocation and were therefore at high risk of bias (open list).
Blinding of participants (checking for possible performance bias): We assessed the methods used to blind study participants from knowledge of which intervention a participant received. We assessed methods as having low risk of bias (study states that it was blinded and describes the method used to achieve blinding); unclear risk of bias (study states that it was blinded but does not provide an adequate description of how this was achieved). We considered studies in which participants were not blinded to have high risk of bias.
Blinding of outcome assessment (checking for possible detection bias): We described all measures used, if any, to blind outcome assessors from knowledge of which intervention a participant received. If applicable, we provided information related to whether the intended blinding was effective.
Incomplete outcome data (checking for possible attrition bias): We assessed differences between intervention and control groups with reference to incomplete outcome data. Bias resulted only if the number lost was imbalanced between groups. However, large loss to follow‐up in relation to the number of events always led to suspicion of a serious threat of bias (Guyatt 2011a).
Selective outcome reporting (checking for possible reporting bias): We assessed outcomes reported and compared them with outcomes listed in initial study registrations or in published protocols. We suspected reporting bias if study reports failed to include results for prespecified key outcomes.
Other bias: This included stopping early for benefit, use of non‐validated outcome measures, carryover effects in cross‐over studies, and recruitment bias in cRCTs.

Measures of treatment effect

To account for use of different scales across studies, we used SMDs as effect measures for continuous data for the primary outcomes health‐related quality of life, depression, and symptom intensity. We analysed time‐to‐event data (survival duration) as death hazard ratio under the proportional hazards assumption that hazard ratio was constant across the follow‐up period. With regards to secondary outcomes, we included results for caregiver burden as well as for healthcare and resource use as provided in the narrative review of a single study.

Unit of analysis issues

Unit of analysis issues may arise because in early palliative care studies, results may be presented for several periods of follow‐up, and because in cRCTs, groups of participants instead of individual participants are randomised. We addressed the issue of several periods of follow‐up by restricting measurement to a single point for each outcome, as described in Primary outcomes. For cRCTs, we adjusted sample sizes before calculating pooled effect estimates, if corresponding analyses did not properly account for the cluster design (e.g. by applying multi‐level modelling, performing variance components analyses, or using generalized estimating equations) (Higgins 2011d). In cases of more than two parallel intervention arms, we planned to consider only two arms (preferably early palliative care arm vs standard care).

However, we did not find studies that included more than two arms.

Dealing with missing data

Whenever possible, we asked the original investigators to provide missing data. In palliative care settings, missing data may not be missing at random but may often indicate poor outcomes (i.e. death) (Hui 2013b; Hussain 2016). Thus, a simple replacement method did not seem adequate. We ultimately conducted available case analyses and included data only on cases for which results are known (Higgins 2011d).

Assessment of heterogeneity

We investigated variation in effects observed across studies by including a Chi² test within forest plots, with regards to the total number of identified studies. For further quantification of inconsistency across studies, we calculated the I² statistic, which directly reflects the percentage of variability in effect estimates that is due to heterogeneity rather than to chance (Deeks 2011). Higher percentages suggest greater observed heterogeneity. We expected heterogeneity due to different scales, patient populations, clinical settings, and types of interventions. As studies assessed the same outcomes but measured them in a variety of ways (e.g. different psychometric scales for health‐related quality of life), we applied the SMD as the summary statistic in meta‐analysis. For health‐related quality of life, higher scores reflected benefit, but for depression and symptom intensity, higher scores indicated harm, and lower scores suggested benefit. To explore heterogeneity, we conducted a categorical subgroup analysis for 'models of early palliative care (solo practice, co‐ordinated care, integrated care)'.

Assessment of reporting biases

We performed comprehensive database and manual searches, including searches of grey literature, to reduce the risk of reporting bias. As appropriate test power was not ensured owing to an insufficient number of included studies (< 10), we refrained from creating funnel plots and conducting Egger's test for funnel plot asymmetry (Egger 1997; Sterne 2011). In applying random‐effects estimates of the intervention effect, we decided not to exclude small studies, as this might have led to an inappropriate reduction of studies in a field that is just emerging. We expect that in future updates of the review, when more studies on early palliative care have been published, we will be able to explore reporting biases by comparing fixed‐effect and random‐effects estimates or L'Abbé plots as a visual method of assessing differences in results of individual studies. Nevertheless, in case of small‐study effects, we explored probable explanations, compared intervention effects, and cautiously considered sample size when grading and discussing the evidence for each outcome (Roberts 2015).

Data synthesis

To clarify the evidence for early palliative care interventions, we performed meta‐analyses based on an inverse variance random‐effects model with a sufficiently homogeneous group of studies, as planned (DerSimonian 1986). For quantitative synthesis, we combined study results on health‐related quality of life, survival (unadjusted death hazard ratio), depression, and symptom intensity across studies using Review Manager 5.3 (RevMan 2014). Statistical analysis of study findings in meta‐analyses included a combination of pair‐wise comparisons regarding differences in anticipated continuous primary patient‐related outcome data between early palliative care and the control condition. We expressed pooled effects as SMDs, or Hedges' adjusted g.

For cRCTs, we also conducted analysis at the level of the individual according to the generic inverse variance method. Provided that analysis in the primary study accounted for clustering of data (e.g. by multi‐level modelling), we extracted direct estimates of the required effect measure. We controlled for unit of analysis error, which often leads to inadequate weighting of the cRCT in the meta‐analysis due to overly precise confidence intervals. Therefore, we adjusted sample sizes to "effective sample sizes" for studies that did not properly account for the cluster design, as explicated in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011d). To check for herd effects and contamination, we interpreted cRCT results in conjunction with effects identified by the individual RCTs included in the meta‐analysis.

Certainty of the evidence

Two review authors (MWH, MH) independently rated certainty of evidence for the primary outcomes. We used the GRADE (Grades of Recommendation Assessment, Development and Evaluation) system to rank certainty of the evidence using GRADEprofiler Guideline Development Tool software (GRADEPro GDT 2015) and the guidelines provided in Chapter 12.2 of the Cochrane Handbook for Systematic Reviews of Interventions (Schünemann 2011). In determining the level of certainty regarding whether estimates of effects were correct, we first addressed risk of bias for individual studies. Individual studies achieved low risk of bias when most or all criteria attained a low level of risk and any violations were not crucial. Studies that exhibited one violation of crucial importance (i.e. high risk of bias) with regard to a point estimate provided evidence of limited certainty and therefore were downgraded in the next step (Guyatt 2011a). Second, we applied the GRADE system of rating the certainty of evidence for each outcome across studies (Balshem 2011).

The GRADE approach applies five considerations (study limitations, consistency of effect, imprecision, indirectness, and publication bias) to assess certainty of the body of evidence for each outcome. The GRADE system uses the following criteria for grade of evidence assignment.

High certainty: we are very confident that the true effect lies close to that of the estimate of the effect;
Moderate certainty: we are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of effect, but there is a possibility that it is substantially different;
Low certainty: our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect;
Very low certainty: we have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect.

We decreased grade if we noted:

serious (‐1) or very serious (‐2) limitations in study quality;
important inconsistency (‐1);
some (‐1) or major (‐2) uncertainty about directness;
imprecise or sparse data (‐1); or
high probability of reporting bias (‐1).

As suggested by the GRADE Working Group (Guyatt 2011a), we were generally conservative in downgrading and considered risk of bias in the context of other limitations. We made close‐call situations explicit. For transparency, we explained in footnotes in the 'Summary of findings' table (summary of findings Table for the main comparison) rationales for downgrading according to the GRADE system.

'Summary of findings' table

We included a 'Summary of findings' table to present the main findings in a transparent and simple tabular format. In particular, we included key information regarding certainty of the evidence, magnitude of effect of interventions examined, and the sum of available data on key outcomes.

Subgroup analysis and investigation of heterogeneity

For prespecified explanatory variable 'models of early palliative care (solo practice, co‐ordinated care, integrated care)', we conducted a categorical subgroup analysis to identify organisation as a potential effect modifier on the basis of seven included studies. Because we included an insufficient number of studies (n = 2; Maltoni 2016;Temel 2010), which drew on a population with a homogeneous malignancy, we decided that we would not conduct a similar analysis for the second hypothesised explanatory variable 'samples with a single type of tumour versus samples with various tumour types'. Owing to an insufficient number of included studies (n < 10), we did not perform a meta‐regression to explore a dose effect of the intervention on outcome variables. This decision reflects accordance with current interpretations of guidelines for systematic reviews and meta‐analyses, "discouraging statistical investigations such as subgroup analyses and meta‐regression, rather than simply adopting a cautious approach to their interpretation, unless a large number of studies is available" (Thompson 2002). Nevertheless, we evaluated heterogeneity by computing the I² statistic as described above and interpreted results with regard to the direction of effect across studies. We regarded an I² statistic exceeding 75% as considerable (Higgins 2003). Eventually, in the discussion, we extensively commented on risk of bias findings and degree of heterogeneity within each outcome comparison.

Sensitivity analysis

Owing to an insufficient number of included studies (n < 10), we did not perform sensitivity analysis by re‐running the meta‐analysis while excluding one study at a time to identify outlying studies. Consequently, we did also not incorporate results of the risk of bias assessment in sensitivity analyses limited to high‐quality studies. However, we conducted a sensitivity analysis on study design as a covariate (RCT vs cRCT) to investigate the robustness of the pooled effect estimate. We highlight the observational character of results of this sensitivity analysis and avoided presenting definitive conclusions for early palliative care, as it represents a still emerging interventional approach with few studies published to date.

Results

Description of studies

Results of the search

Searches of six databases (see Electronic searches) yielded 21,475 records. Searches of other resources (trial registers, systematic reviews, conference proceedings, journal eTOC alerts, contact with content experts) revealed 1719 additional records that appeared to meet the inclusion criteria. We therefore obtained a total of 23,190 records.

Once we had removed duplicates, we had a total of 16,999 records. We excluded 16,886 records on the basis of reviews of reviews of titles and abstracts. We obtained the full text for the remaining 113 records. We excluded 21 studies (29 records) (see Characteristics of excluded studies) and added 10 studies (20 records) to Characteristics of studies awaiting classification. We identified 20 ongoing studies (26 records) (see Characteristics of ongoing studies).

Use of a simple kappa statistic for interrater variation with regard to study inclusion amounted to κ = 0.60, which indicated good agreement (Higgins 2011b). In summary, we included seven studies reported in 38 references (31 full papers and seven trial registry entries, one per study), ranging from one to 11 full papers per study (Bakitas 2009, six full papers; Bakitas 2015, five full papers; Maltoni 2016, two full papers; McCorkle 2015, one full paper; Tattersall 2014, one full paper; Temel 2010, 12 full papers; Zimmermann 2014, four full papers). For a further description of our screening process, see the study flow PRISMA (Preferred Reporting Items for Systematic Reviews and Meta‐Analyses) diagram (Liberati 2009), depicted in Figure 1.

Figure 1

Screening process diagram (as recommended by the PRISMA statement).

Included studies

Designs

Five of the seven studies were prospective RCTs with participants as units of randomisation and a single intervention and a single comparator arm (Bakitas 2009; Bakitas 2015; Maltoni 2016; Tattersall 2014; Temel 2010). Bakitas 2015 applied a so‐called fast‐track RCT design, randomising participants to receive concurrent palliative care with standard oncology care shortly after diagnosis of advanced or progressive disease (early group), or three months later (delayed group). Two studies were cluster‐randomised trials treating oncology clinics as units of randomisation and participants as units of inference (McCorkle 2015; Zimmermann 2014).

Sample sizes

Unadjusted sample sizes varied between 120 and 461 participants. Recruitment length ranged from 22 to 51 months. In total, we analysed data from studies involving 1614 participants. Bakitas 2009 and Bakitas 2015 provided data for 320 caregivers in total. Maltoni 2016 measured family satisfaction with end‐of‐life care but has not yet published caregiver results. Zimmermann 2014 provided data for 151 caregivers in total (McDonald 2016). Six of the seven studies were guided by power calculations (details in Characteristics of included studies): Bakitas 2009 powered on quality of life, symptom intensity, and depression, and Maltoni 2016 and Temel 2010 on the Trial Outcome Index (TOI), that is, pancreatic/lung cancer‐specific symptom intensity and physical and functional well‐being. Bakitas 2015 powered on quality of life and depression, whereas Zimmermann 2014 powered on quality of life only. Tattersall 2014 powered on a 0.5 standard deviation (SD) but did not indicate a primary outcome. McCorkle 2015 did not provide a power calculation. Four of the six studies drawing on power calculations were underpowered at recruitment stage, most commonly owing to slower enrolment than was projected. Two studies (Temel 2010; Zimmermann 2014) reached adequate power by amending the protocol.

Study populations

Five of the seven studies investigated populations with heterogeneous tumour entities (Bakitas 2009; Bakitas 2015;McCorkle 2015; Tattersall 2014;Zimmermann 2014). In contrast, Temel 2010 focussed on exclusive enrolment of participants with metastatic non‐small cell lung cancer, whereas Maltoni 2016 focussed on exclusive enrolment of participants with metastatic pancreatic cancer. Four studies investigated caregivers along with participants (Bakitas 2009; Bakitas 2015; Maltoni 2016; Zimmermann 2014). Mean age ranged from 60 to 67 years. Across all studies, investigators included slightly higher numbers of male compared with female participants, except in two studies, in which women constituted the majority of participants (McCorkle 2015, Zimmermann 2014). In five studies, approximately two‐thirds of all participants were married or were living with a partner. This proportion was slightly lower in two studies (McCorkle 2015; Temel 2010). In three studies, the vast majority of participants (> 85%) stated that they had received nine or more years of education. A similar proportion (> 75%) was unemployed. Three studies did not provide data on education levels of nine years or below nor on employment status (Maltoni 2016; Tattersall 2014; Temel 2010).

Setting

Five studies took place in the United States (US) (three in predominantly rural areas in New Hampshire, Connecticut, and Vermont; one in the metropolitan area of Boston; one at Yale‐New Haven). One study was conducted in Toronto, Canada (metropolitan area); one in Italy (multiple sites); and one in Sydney, Australia (metropolitan area) (see Characteristics of included studies table for details). Although two US studies recruited from National Cancer Institute‐designated (comprehensive) cancer centres solely, and one recruited from a tertiary referral hospital, the remaining two US studies additionally recruited from a Veterans Affairs Medical Center. Both the Canadian study and the Australian study recruited from tertiary referral hospitals. For the Maltoni 2016 trial, investigators recruited most participants from palliative clinics of tertiary centres, and a minority were enrolled in smaller community cancer centres (unpublished data received upon study author request).

Early palliative care interventions

Solo practice model

We did not identify any studies providing early palliative care based on a solo practice model.

Co‐ordinated care model

Three studies followed the co‐ordinated care model in establishing an advanced practice nurse as a co‐ordinator and in linking care from different specialist disciplines (Bakitas 2009; Bakitas 2015; McCorkle 2015). In the ENABLE II study, Bakitas 2009 provided outpatient palliative care. Specifically, two advanced practice nurses with palliative care specialty training provided case management and education via a manualised, telephone‐based approach for participants in the intervention group. The intervention comprised four initial structured educational and problem‐solving telephone sessions provided on a weekly basis (education manual: Charting your Course: An Intervention for People and Families Living With Cancer) and at least monthly telephone follow‐up sessions thereafter until the participant died or the study ended. Investigators applied problem‐solving management on the basis of systematic distress assessment using the Distress Thermometer and a cut‐off > 3. When concerns were identified, participants were encouraged to contact oncology or palliative care clinical teams.

In the ENABLE III study (Bakitas 2015) for outpatient palliative care, all participants received usual oncology care directed by a medical oncologist. The intervention comprised anticancer and symptom control treatments and consultation with oncology and supportive care specialists, including a clinical palliative care team, which was provided whenever requested, regardless of group assignment. The intervention followed a telehealth concurrent palliative care model, commencing within 30 to 60 days of an advanced cancer diagnosis, cancer recurrence, or progression. The model was based on an initial in‐person, standardised outpatient palliative care consultation with a board‐certified palliative care clinician and six structured weekly telephone coaching sessions provided by an advanced practice nurse, again using a manualised curriculum Charting Your Course: An Intervention for Patients With Advanced Cancer. Sessions one to three focussed on problem solving, symptom management, self‐care, identification and co‐ordination of local resources, communication, decision making, and advance care planning. Sessions four to six comprised Outlook, a life‐review approach that encourages participants to frame advanced illness challenges as personal growth opportunities; after completion of the six Charting Your Course sessions, monthly follow‐up calls reinforced prior content and identified new challenges or care co‐ordination issues.

In the McCorkle 2015 study for outpatient palliative care, all participants received usual oncology care directed by a medical oncologist. This study was based on a 10‐week standardised intervention for the experimental group delivered by different members of each team, which included monitoring participants' status, providing symptom management, executing complex care procedures, teaching participants and family caregivers, clarifying the illness experience, co‐ordinating care, responding to the family, enhancing quality of life, and collaborating with other providers. Advanced practice nurses at the clinics initially contacted participants within 24 hours and maintained weekly phone and scheduled in‐person contacts (five clinic visits and five telephone calls). Members of each disease‐specific multi‐disciplinary team worked together as a palliative care unit, with each member taking on different functions to ensure that all components of the intervention were addressed. Furthermore, the clinic advanced practice nurse oversaw co‐ordination and implementation.

Integrated care model

Four studies followed the integrated care model. In the Maltoni 2016 study, participants assigned to the interventional arm underwent systematic symptom assessment during an appointment scheduled with a palliative care specialist, who applied a predefined checklist during the consultation. Topics on the checklist were adapted from the Temel 2010 trial. Participants met with a member of the palliative care team within two weeks of enrolment and were seen thereafter every two to four weeks until death. Appointments and interventions were oriented by general palliative care guidelines introduced by the US National Consensus Project. The palliative care specialist who regularly saw interventional arm participants prescribed drugs and requested other interventions tailored to participants' physical, psychological, and spiritual needs. However, recommendations made by the palliative care expert on decision‐making processes had to be shared with the attending oncologist.

The study by Tattersall 2014 provided outpatient palliative care via meetings between the participant and a palliative care nurse consultant member of the hospital palliative care team. The nurse outlined available palliative care services, including advice about symptom control, and offered to arrange review by a palliative care physician. Contact details for the palliative care service were provided. The palliative care nurse offered to telephone the participant monthly to check on that individual's well‐being; if the participant preferred, the nurse provided contact details for the participant's use. Standard oncological care was given according to the oncologist’s recommendations.

In the Temel 2010 study, participants in the intervention group met with a member of the palliative care team, which consisted of board‐certified palliative care physicians and advanced practice nurses, within three weeks after enrolment and at least monthly thereafter in the outpatient setting until death. Additional visits with the palliative care service were scheduled at the discretion of participant, oncologist, or palliative care provider. General guidelines for palliative care visits in the ambulatory setting were adapted from the National Consensus Project for Quality Palliative Care and were included in the study protocol. Investigators paid specific attention to assessing physical and psychosocial symptoms, establishing goals of care, assisting with decision making regarding treatment, and co‐ordinating care on the basis of individual needs of the participant. All participants continued to receive routine oncological care throughout the study period.

In a cRCT that examined care provided in outpatient clinics, as well as inpatient and home care, Zimmermann 2014 followed a multi‐disciplinary approach to address physical, psychological, social, and spiritual needs. At outpatient clinics, participants consulted with palliative care physicians and nurses during routine visits once monthly and more often if necessary. Routine structured symptom assessment conducted during every visit was combined with routine psychosocial assessment and discussion of goals of care, of participant and family support needs, and of participant and family coping and psychological distress. Advance care planning was discussed according to participant and family readiness. Palliative care nurses provided routine telephone follow‐up after each visit. A 24‐hour on‐call service was explained during the first visit and was provided throughout the study. The hospital service included direct access to the palliative care unit for symptom management and follow‐up by the palliative care team when the participant was admitted to non‐palliative care unit services. Within home care, community care access centre services were explained and offered during the first visit, and need was reassessed at each visit. The availability of a home palliative care physician was explained during the first visit, and this service was offered when Eastern Cooperative Oncology Group (ECOG) performance status exceeded a score of 3, or when the participant requested the service.

Comparators

Active comparators in the included studies constituted free access to all oncology and supportive services, including referral to other palliative care services (Bakitas 2009), and usual oncology care directed by a medical oncologist. This consisted of anticancer and symptom control treatments and consultation with oncology and supportive care specialists, including a clinical palliative care team (Bakitas 2015). In the Maltoni 2016 study, participants assigned to the standard arm were scheduled to meet with the palliative care team only when participants themselves, their families, or the attending oncologist requested an appointment.

In the Australian study and in one US study, control participants were referred to the palliative care service when recommended by the oncologist (McCorkle 2015; Tattersall 2014). Similarly, in the study by Temel 2010, participants in the control condition met with the palliative care service only on their own, their family's, or the oncologist's request. In the fifth study, the control group followed an approach that mainly addressed physical symptoms and was provided in outpatient clinics, as hospital service, or as in‐home care (Zimmermann 2014).

Outcomes

Six of the seven studies listed quality of life as a primary outcome, although Bakitas 2015 did not differentiate between primary and secondary outcomes and also targeted quality of life as a study outcome. Only Temel 2010 and Zimmermann 2014 named quality of life as single primary outcome. The remaining four studies listed more than one primary outcome, including symptom intensity, resource use, depression, unmet needs, emotional distress, health distress, self‐rated health, functional status, and/or survival (Bakitas 2009; Maltoni 2016; McCorkle 2015; Tattersall 2014). Investigators included the following as secondary outcomes: depression, anxiety, self‐efficacy, uncertainty, survival, participant interaction with nurses and doctors, quality of care, family satisfaction with (end‐of‐life) care, caregiver burden, aggressiveness of (end‐of‐life) care/number of lines of chemotherapy, experience of end‐of‐life care, use of healthcare resources, and place of death.

Funding sources

The US studies (Bakitas 2009; Bakitas 2015; Temel 2010) were funded by the National Cancer Institute (NCI), the American Society of Clinical Oncology (ASCO), philanthropic gifts, and the National Institute for Nursing Research (NINR). The Australian study (Tattersall 2014) was funded by the National Health and Medical Research Council (NHMRC). The US cRCT (McCorkle 2015) was funded by the NINR, and the Canadian cRCT (Zimmermann 2014) by the Canadian Cancer Society and the Ontario Ministry of Health and Long Term Care. The Italian study Maltoni 2016 was funded by the Italian Ministry of Health.

Excluded studies

We excluded from the review 21 studies that we had initially rated as potentially relevant. We excluded most of these studies because interventions did not demonstrate genuine early palliative care intent (n = 10). We excluded other studies owing to the absence of a multi‐dimensional approach (n = 7), other than predefined primary outcomes (n = 1), quasi‐experimental design (n = 1), withdrawal from the study (n = 1), and implementation instead of clinical study design (n = 1). For an overview of excluded studies, please refer to the Characteristics of excluded studies table.

Studies awaiting classification

We identified 10 studies that had been completed at the time of the search. However, these studies are awaiting classification, as they have not yet been published. For an overview of studies awaiting classification, please refer to the Characteristics of studies awaiting classification table.

Ongoing studies

We identified 20 ongoing studies. For an overview of these studies, please refer to the Characteristics of ongoing studies table.

Risk of bias in included studies

We assessed risk of bias using the Cochrane 'Risk of bias' tool (see Figure 2 and Figure 3) (Higgins 2011c). In formulating summary assessments of risk of bias for each important outcome (across domains) within and across trials, we applied the approach introduced by Higgins 2011e (Figure 2; Figure 3). Across trials, we identified high risk of bias for all outcomes (for health‐related quality of life and symptom intensity due to selection (insufficient allocation concealment), performance, detection, attrition, and reporting biases; for survival due to selection (insufficient allocation concealment), performance, and attrition biases; and for depression due to selection (insufficient allocation concealment), performance, detection, and reporting biases).

Figure 2

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Figure 3

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Allocation

Random sequence generation

All seven studies were randomised and adequately described the method used to generate the random sequence; therefore, we judged these studies to be at low risk of bias for this domain. Most studies applied computer‐generated random numbers. We did not identify any studies at high or unclear risk of bias for this domain.

Allocation concealment

Authors of four studies adequately described allocation concealment of the sequence in the main publication (Bakitas 2009; Maltoni 2016; Tattersall 2014; Zimmermann 2014). For the Temel 2010 study, we received information from the principal investigator upon request. For Bakitas 2015 and McCorkle 2015, risk of bias remained unclear owing to insufficient information. We considered two studies to be at low risk of bias for this domain (Maltoni 2016; Tattersall 2014), although we noted high risk of bias for the three remaining studies: Bakitas 2009 and Temel 2010 did not conduct allocation concealment. As investigators randomised clusters before obtaining consent of individuals, we classified the Zimmermann 2014 study to be at high risk for this domain. Of note, study authors discussed this limitation in the main publication of the study.

Blinding

As explicated in the Methods, we did not include blinding of personnel in our risk of bias assessment owing to infeasibility and inappropriateness in the context of palliative care. We considered this infeasibility of personnel blinding to be a methodological factor that applied similarly to all included studies. However, we included blinding of participants and blinding of outcome assessment.

Blinding of participants

In terms of blinding of participants, we judged six studies (Bakitas 2009; Bakitas 2015; Maltoni 2016; McCorkle 2015; Tattersall 2014; Temel 2010) to be at high risk of bias for this domain. In the Zimmermann 2014 trial, investigators ensured blinding of participants within the framework of a cluster‐randomised trial, so we judged this trial be at low risk of bias.

Blinding of outcome assessment

With respect to blinding of outcome assessment, five of the seven primary reports on included studies provided no details of this. Thus, we judged these studies (Bakitas 2009; Maltoni 2016; McCorkle 2015; Tattersall 2014; Temel 2010) to be at unclear risk of bias for this domain. Zimmermann 2014 did not blind investigators. We considered this study to be at high risk of detection bias. Bakitas 2015 explicitly mentioned that outcome assessors were blinded. We considered this study to be at low risk of detection bias.

Incomplete outcome data

Six of the seven included studies reported attrition rates in intervention and control groups that were approximately identical. As a characteristic of patient populations with advanced cancer, these rates were rather high. The most important reasons given across studies were decline in performance status, death, exhaustion, or cognitive impairment. We judged these six studies to be at low risk of bias for this domain. We had difficulty rating risk of attrition bias for the Zimmermann 2014 study, in which higher attrition in the intervention group reached borderline significance over the control group. Thus, in a close‐call situation, we decided to assume that risk for attrition bias was low. One study (Tattersall 2014) reported seriously high attrition across intervention and control groups, indicating high risk of attrition bias.

Selective reporting

We observed few inconsistencies between outcomes listed in study registrations and those reported in publications for three studies. However, all key outcomes were reported and some included studies were published only recently, so we judged these three to be at low risk of bias for this domain. Bakitas 2009, Maltoni 2016, and Temel 2010 reported outcomes in accordance with the a priori study registration. Zimmermann 2014 did not provide information on results for registered secondary outcomes of Caregiver Quality of Life Index‐Cancer (CQOL‐C) and the Short Form (SF)‐36 Survey. However, researchers reported all key outcomes, and we favoured low risk of bias.

The Bakitas 2015 publication did not differentiate between primary and secondary outcomes. However, investigators reported all key outcomes. We made a close‐call decision favouring low risk of bias. Tattersall 2014 was not registered until after recruitment had been opened, and so we judged this study to be at unclear risk of bias for this domain.

For McCorkle 2015, we detected high risk of reporting bias: Uncertainty (MUIS‐C) described as a single primary outcome in clinicaltrials.gov registration was reported by study authors as a secondary outcome. Published results of this study included additional secondary outcomes that had not been preregistered.

Other potential sources of bias

Overall, we judged six studies to be at low risk of other bias, and one study to be at unclear risk for this domain.

In sum, we did not identify other potential sources of major bias in five studies, and so we judged them to be at low risk of bias for this domain. All five studies measured participant characteristics and outcomes at baseline, and four studies found no substantial differences between intervention and control groups.

For McCorkle 2015, results showed statistically significant differences between arms at baseline with respect to age, gender, and comorbidity. Given this baseline imbalance, recruitment bias may be present. We made a close‐call decision favouring low risk of other bias, as high risk of selection bias was already detected.

In Zimmermann 2014, results revealed imbalance between intervention and control groups at baseline, exhibiting a tendency for higher outcome measure scores (for FACIT‐Sp at P = 0.03; for ESAS at P < 0.001; for FAMCARE‐P16 at P < 0.001) in the intervention group. For quality of life and family satisfaction with care, this implied that improvements were more difficult to attain for the intervention group and therefore negatively biased effect size. For symptom intensity, worsening might have been more likely for the control group, entailing a positively biased effect size. Also a larger number of participants with genitourinary cancers were included in the control group at baseline. However, we judged this study to be at low risk of bias.

Tattersall 2014 found baseline differences between groups in time since initial cancer diagnosis (shorter time for intervention group) and in oncologists’ estimate of likely survival (better prognosis for intervention group). However, researchers controlled for these variables in their analyses, and in the light of a close‐call situation, we decided not to rate down for imbalance bias but stated unclear risk of bias.

For Bakitas 2009, Bakitas 2015, Maltoni 2016, and Temel 2010, we found no evidence indicating other potential sources of bias.

Effects of interventions

See: Summary of findings for the main comparison Early palliative care for adults with advanced cancer

We report here synthesis results for the following prespecified primary outcomes: health‐related quality of life, survival, depression, and symptom intensity. For all outcomes across all seven included studies, data could be incorporated into syntheses as long as the given study measured the outcome. Apart from pooled effect estimates, we also report results for individual studies. At this point, we underline that we have to interpret pooled effect estimates with caution owing to low certainty of the current evidence.

For prespecified secondary outcomes (caregiver burden, healthcare utilisation, and harms/adverse events), we could not find a sufficient number of studies, warranting synthesis in meta‐analysis. Instead, we report data in a narrative format. We found data on all prespecified outcomes of interest, although pooling of effects was possible only for primary outcomes. For an overview of results, please see summary of findings Table for the main comparison, as well as the risk of bias assessment presented in Figure 2.

Concerning time points, we identified five RCTs as relying on predefined time points for the outcomes of health‐related quality of life, depression, and symptom intensity at 12 weeks (Bakitas 2015; Maltoni 2016; McCorkle 2015; Temel 2010; Zimmermann 2014). Therefore, we calculated SMDs for these studies on the mean difference at 12 weeks. Two studies applied mixed‐effects models for repeated measures on longitudinal data (Bakitas 2009; Tattersall 2014). For these studies, we used resulting mean differences over time in calculating SMDs.

Primary outcome: health‐related quality of life

Pooled data from seven studies (five RCTs, two cRCTs), with 1028 analysed participants (sample size for available case analysis at T1) available for the relevant comparison, showed that those receiving early palliative care had significantly higher quality of life than those receiving usual care (SMD 0.27, 95% CI 0.15 to 0.38) (Analysis 1.1; Figure 4). The effect size is small by conventional criteria. We combined different scales measuring this outcome of interest across studies by applying SMDs. Positive SMDs reflect benefit (better quality of life); negative SMDs indicate harm (lower quality of life). We found that researchers used seven different scales for measuring health‐related quality of life as an outcome in the included studies (Functional Assessment of Chronic Illness Therapy for Palliative Care, FACIT‐Pal, in Bakitas 2009 and Bakitas 2015; Trial Outcome Index, TOI, of the Functional Assessment of Cancer Therapy‐Hepatobiliary, FACT‐Hep, in Maltoni 2016; Functional Assessment of Cancer Therapy‐General, FACT‐G, in McCorkle 2015; McGill Quality of Life, McGill QOL, in Tattersall 2014; TOI of the Functional Assessment of Cancer Therapy‐Lung, FACT‐L, in Temel 2010; and Functional Assessment of Chronic Illness Therapy for Spiritual Well‐Being, FACIT‐Sp, in Zimmermann 2014). Zimmermann 2014 additionally used the Quality of Life at the End of Life, QUAL‐E, on which, in contrast to findings for the FACIT‐Sp, the difference between groups in change scores at 12 weeks was borderline significant (P = 0.05).

Figure 4

Forest plot of comparison: 1 Health‐related quality of life, outcome: 1.1 Health‐related quality of life.

Within the GRADE approach, we downgraded the certainty of evidence for health‐related quality of life to low owing to high risk of bias at study level across studies (‐2 points due to very serious limitations in study quality: high risk of bias for selection (insufficient allocation concealment), performance, detection, attrition, and reporting biases) (summary of findings Table for the main comparison).