Impact of palliative care on end-of-life care and place of death in children, adolescents, and young adults with life-limiting conditions: A systematic review

Abstract Objective To determine the impact of palliative care (PC) on end-of-life (EoL) care and the place of death (PoD) in children, adolescents, and young adults with life-limiting conditions. Method Eight online databases (PubMed, Medline, EMBASE, Cochrane Library, CINAHL, Airiti, GARUDA Garba Rujukan Digital, and OpenGrey) from 2010 to February 5, 2020 were searched for studies investigating EoL care and the PoD for pediatric patients receiving and not receiving PC. Results Of the 6,468 citations identified, 14 cohort studies and one case series were included. An evidence base of mainly adequate- and strong-quality studies shows that inpatient hospital PC, either with or without the provision of home and community PC, was found to be associated with a decrease in intensive care use and high-intensity EoL care. Conflicting evidence was found for the association between PC and hospital admissions, length of stay in hospital, resuscitation at the time of death, and the proportion of hospital and home deaths. Significance of results Current evidence suggests that specialist, multidisciplinary involvement, and continuity of PC are required to reduce the intensity of EoL care. Careful attention should be paid to the need for a longer length of stay in a medical setting late in life, and earlier EoL care discussion should take place with patients/caregivers, especially in regard to attempting resuscitation in toddlers, adolescents, and the young adult population. A lack of robust evidence has identified a gap in rigorous multisite prospective studies utilizing data collection.


Introduction
Life-limiting conditions are those for which there is no possibility of cure, where death is inevitable at some point in time during childhood or young adulthood (McNamara-Goodger and Feudtner, 2012;. Pediatric palliative care (PC) is a multidisciplinary approach that improves the quality of life of this population and their families through preparing them for an anticipated death through providing physical, psychosocial, and spiritual supports, as well as managing complex conditions during the end of life (EoL; Chambers, 2018). The number of children, adolescents, and young adults with life-limiting conditions is increasing as a result of increased survival and earlier recognition of life-limiting diagnoses. The United Kingdom data suggest that 66.4 per 10,000 individuals in pediatric populations aged from 0 to 19 years were living with life-limiting conditions in 2017, and the prevalence is likely to rise to 84.2 per 10,000 by 2030 (Fraser et al., 2020). However, it was estimated that only 18.6% of them actually received PC before death (Widger et al., 2016).
The majority of this population goes through a prolonged period of inpatient admission, primarily in intensive care unit (ICU) settings during EoL, and die in the hospital (Gao et al., 2016;DeCourcey et al., 2018). Only 58% of children, adolescents, and young adults with cancer actually died at home as their place of preference (Stilwell et al., 2020). A growing body of evidence demonstrates that receiving early PC improves the quality of life and symptom control, facilitates earlier advanced care planning or planned withdrawal of ventilator support outside the ICU setting, and enables a choice in the place of death (PoD) outside of hospital (Abel et al., 2013;Laddie et al., 2014;Mitchell et al., 2017). Nevertheless, several patient and family-related factors influence the EoL decision-making and the PoD (Allen, 2014;Foster et al., 2016). Thus, there is limited scientific evidence by which to illustrate the impacts of PC on children, adolescents, and young adults with life-limiting conditions in terms of their healthcare circumstances and outcomes during the EoL.
Recent systematic reviews suggest that access to PC is associated with reduced time in hospital and less invasive treatments (Marcus et al., 2020;Tayloy et al., 2020;Zuniga-Villanueva et al., 2020). However, these reviews specifically evaluated the outcomes of specialist PC services, which were developed mainly in the USA, Canada, the UK, and across Europe (Knapp et al., 2011). Thus, the evidence base is not yet able to offer a universal PC model, especially in the case of non-Western countries. Furthermore, these reviews aggregated the results for patients at earlier stages of their illness. The evidence of the impact on EoL care and the PoD, including studies on PC without further subdivisions of the general and specialist levels of services, remains unclear. The aim of this study is to determine the impact of PC on EoL care and PoD for children, adolescents, and young adults with life-limiting conditions.

Methods
A systematic review was conducted to evaluate the academic literature related to PC in order to determine if the use of PC compared with not using it impacts on EoL care and the PoD. This systematic review followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses for Protocol (PRISMA-P) guidelines (Moher et al., 2015).

Inclusion and exclusion criteria
We included studies if their target population was children up to 25 years of age with a life-limiting condition; the exposure of interest was self-described as "palliative care" and/or comprised at least two components of PC, including general and/or specialist PC, which means PC delivered by a healthcare team who are/are not PC specialists, as defined by the National Consensus Project for Quality Palliative Care (2018). The researchers provided a comparison between PC exposure and at least one of the outcomes of admission in terms of acute care beds, length of hospital stay, receiving cardiopulmonary resuscitation (CPR) before death, PoD, and the intensity of EoL care. Publications in English, Mandarin, or Bahasa were considered based on the language competencies of the reviewers.
Studies were excluded where the type of intervention was exclusively on only one component of comprehensive PC (e.g., providing opioid and sedation medications or advanced care planning) or included a mixed sample of adults aged over 25 years. Assessments of EoL care were not included if data in the last year of life could not be fully determined. Outcomes of the presence of limitations related to resuscitation orders were not included if the congruence of the presence of limitations of resuscitation orders and the actual CPR attempted at the time of death could not be fully determined. Cross-sectional descriptive studies and qualitative studies were excluded.

Search strategy
The following online databases were searched: Medline, EMBASE, PubMed, Cochrane Library, CINAHL, Airiti, GARUDA Garba Rujukan Digital, and OpenGrey from 2010 to February 5, 2020, where a series of pediatric PC in various settings had been implemented over time. The search terms used included the recognized Medical Subject Heading (MeSH) and Embase subject headings (Emtree) with various combinations of the terms children, adolescents, young adults, palliative, end-of-life, place of death, and specific diagnoses from the International Classification of Disease (ICD10) coding framework (Fraser et al., 2012) (Supplementary  Tables 1-5 in the Supplementary File for search strategy).

Screening and selection
Retrieved references were imported using the EndNote. S.-C.L. and D.Y. screened and selected all articles by title and abstract independently using a prespecified screening and selection tool. Disagreements were discussed with M.-C.H. until a consensus was reached.

Quality assessment
The quality of the individual studies was critically graded by applying the Joanna Briggs Institute (JBI) appraisal tools based on appropriate checklists. The checklists included assessment of the methodological quality of the study and any possibility of bias in its design, procedures, or analyses, scored as "yes" (2), "unclear" (1), "no" (0), or "not applicable." The summary scores for each study were derived by calculating the total scores of relevant items. S.-C.L. and D.Y. independently assessed all papers. Any disagreement was resolved through a discussion with M.-C.H.. Scores were expressed as a percentage, where >80% was considered strong; >70% was considered good; >50% was considered adequate, and ≤50% was considered limited. We did not exclude studies based on a quality assessment.

Data extraction
Data extraction was conducted by one reviewer (S.-C.L.), using an extraction form piloted on three eligible studies. Key study characteristics, including the country of origin, setting, sample data, participant characteristics, intervention vs. control, outcomes, and measurement tools, were extracted. For this study, home and community PC was defined as any type of the availability of palliative home care services, 24-h on-call service, or a 24-h outreach pediatric PC team.
If the intervention effects were completely reported, pooled effect estimate (e.g., odds ratio and 95% confidence intervals) would be calculated using the random effects model in RevMan 5.4.1 where possible. Fisher's method for combining p-values or vote counting based on the direction of effects was used if there was minimal reporting of the data beyond p-values and the direction of effect, or the type of effect measure varied across the studies (McKenzie and Brennan, 2019).

Study selection
A total of 6,461 studies were identified through electronic database searches, and seven studies were identified through reference lists. After duplicates were removed, title/abstract screening, and full-text screening, 15 studies met the eligibility criteria. A detailed illustration of the study selection process is presented in Figure 1.

Risk of bias
The critical appraisal scores ranged between 45% and 95%. Seven of the 15 studies were judged to be of strong quality (Thienprayoon et al., 2013;Smith et al., 2015;Ullrich et al., 2016;Widger et al., 2018 (Tables 2 and 3).

Study characteristics
Characteristics of the included studies are presented in Table 1. Of the 15 studies, nine were conducted in the USA, and one each was conducted in Canada, France, Germany, New Zealand, Singapore, and the UK. The studies were published from 2013 to 2019 as cohort studies (n = 14) and case series (n = 1).

Outcome characteristics
Data on demographics, hospital admission, LoS, CPR at the time of death, PoD, and intensity of EoL care were extracted from national databases, regional databases, medical records, and surveys. Significance was defined as p < 0.05.

Number of admissions in acute care beds
The combination of p-values in three studies suggests that there was strong effect of PC in reducing the number of hospital admissions in at least one study ( p < 0.001) (Chong et al., 2018;Revon-Rivière et al., 2019;Zernikow et al., 2019). Children and adolescents aged < 18 who received home-based PC had fewer  Shih-Chun Lin et al. hospital admissions during their last year of life, but no differences were found in the number of emergency visits (Chong et al., 2018). Studies also varied in terms of whether or not they found differences in the number of hospital admissions in the last month of life for children, adolescents, and young adults with cancer (Revon-Rivière et al., 2019; Zernikow et al., 2019) (Tables 2 and 3).
A consistent finding across studies was that children, adolescents, and young adults who received inpatiant hospital PC, either with or without the provision of home and community PC, were less likely to be admitted to the ICU (Keele et al., 2013;Widger et al., 2018;Revon-Rivière et al., 2019). Among children, adolescents, and young adults who died for any reason, those who received PC had a 71% reduction in risk of ICU admission (Keele et al., 2013); for those who died with cancer, receiving PC was associated with a 6.25-fold decrease in the odds of ICU admission (OR, 0.16; 95% CI 0.13-0.20) (Widger et al., 2018;Revon-Rivière et al., 2019). If those who died due to treatment-related causes (e.g., acute toxicities) were excluded, there was a 3.33-fold decrease in the odds of ICU admission during the last month of life. However, general PC had no impact (Widger et al., 2018). These results are presented in Table 4.

Length of hospital stay
The combination of p-values in five studies suggests that there was an effect in reducing the LoS in the EoL in at least one study ( p = 0.011, five studies) (Keele et al., 2013;Smith et al., 2015;Osenga et al., 2016;Chong et al., 2018;Spraker-Perlman et al., 2019). Even an analysis restricted to studies where during intervention, only inpatient hospital PC was delivered, suggested an increased likelihood of shorter LoS in the PC group in at least one study ( p = 0.0167, three studies) (Keele et al., 2013;Smith et al., 2015;Spraker-Perlman et al., 2019) (Table 5).
However, of the five studies examining LoS, in two of them, no differences in terms of cumulative LoS were observed in the last admission, last month, or last year of life [40% (95% CI 5-85%), p = 1.625] (Keele et al., 2013;Chong et al., 2018). Only one study excluded those who were hospitalized <5 days before death (Keele et al., 2013). Four of five studies did not exclude patients who died quickly after hospital admission, which may have created bias since those who received PC were more likely to have complex, critical illnesses with longer disease trajectories; on the contrary, those with short LoS or unexpected death usually lacked access to PC (Keele et al., 2013;Osenga et al., 2016). Therefore, the association between PC and LoS tended to be biased.

CPR at the time of death
Of the three studies examining CPR at the time of death, all but one showed that patients who underwent hematopoietic stem cell transplants were less likely to undergo CPR at the EoL ( p = 0.03) (Ullrich et al., 2016). However, studies on infants, or adolescents, and young adults, with a median age of 1.3 months and 17.3 years, respectively, found no difference in terms of CPR attempted at the time of death (Osenga et al., 2016;Snaman et al., 2017). Even though two of the three studies did find the median resuscitation discussion to be 5 days earlier ( p < 0.001) or increased do-not-resuscitate (DNR) orders in place in patients followed by a PC team (AOR = 7.92, 95% CI 2.02-31.12) (Osenga et al., 2016;Ullrich et al., 2016). This result is presented in Table 6.

Place of death
Even though PC was accessible in inpatient, home, and community settings, the examined studies varied in terms of whether the patients in the PC group were less likely to die in a medical setting (i.e., hospital and ICU), or more likely to die in the community (i.e., home, hospice, and home hospital services).  (Table 7). In the case of children in the PICU, those who died after receiving PC at discharge who had either transferred hospitals or had moved into the community had an 8.06 times greater chance of not dying in a medical setting compared with those who were not discharged for PC . However, if restricted to cancer patients, the possibility of dying in the hospital was comparable regardless of whether there was PC in inpatient, home, and community settings (Thienprayoon et al., 2013;Vern-Gross et al., 2015;Ullrich et al., 2016;Widger et al., 2018). Although home hospital services were accessible, and EoL discussions occurred earlier (median time from first discussion to death was 204 days earlier), there was no significant decrease in Y, yes; U, unclear; N, no; N/A, not applicable. Y=2; U=1; N = 0 points Q1= Were the two groups similar and recruited from the same population? Q2 = Were the exposures measured similarly to assign people to both exposed and unexposed groups? Q3 = Was the exposure measured in a valid and reliable way? Q4 = Were confounding factors identified? Q5 = Were strategies to deal with confounding factors stated? Q6 = Were the groups/participants free of the outcome at the start of the study (or at the moment of exposure)? Q7 = Were the outcomes measured in a valid and reliable way? Q8 = Was the follow-up time reported and sufficient to be long enough for outcomes to occur? Q9 = Was follow up complete, and if not, were the reasons to loss to follow up described and explored? Q10 = Were strategies to address incomplete follow-up utilized? Q11 = Was appropriate statistical analysis used? inpatient deaths after the implementation of PC (Vern-Gross et al., 2015). However, if the intervention included the delivery of hospice care, PC was associated with being more likely to die at home (Thienprayoon et al., 2013;. In at least one study, there was strong evidence suggesting that PC in reduces the proportion of ICU deaths in all hospital deaths ( p = 0.0017, two studies) (Ullrich et al., 2016;Snaman et al., 2017). Both with and without the delivery of inpatient hospice, home, and community PC, all four studies showed that inpatient hospital PC decreased the proportion of patients dying in an ICU (Keele et al., 2013;Ullrich et al., 2016;Snaman et al., 2017;. Furthermore, Ullrich et al. (2016) reported that there were no differences in receiving less or more than 1 month of PC before death in a cure-oriented stem cell transplantation setting.

Intensity of EoL care
Two studies reported results based on the intensity of EoL care summary scores (Widger et al., 2018;Revon-Rivière et al., 2019) (Table 8). Both defined high-intensity EoL care (HI-EoL) as the occurrence of at least one of the following four indicators: intrahospital intravenous chemotherapy <14 days prior to death, ≥1 ICU admission, >1 emergency room visit, or >1 hospital admission in the last month of life. General PC was associated with a 1.67-fold decrease in the odds of HI-EoL (Widger et al., 2018), whereas specialist inpatient hospital PC, either with or without the provision of home and community PC, was associated with a 3.57-fold decrease in the odds of HI-EoL care during the last month of life for children aged 0-25 who died as a result of cancer (OR, 0.28; 95% CI 0.20-0.40) (Widger et al., 2018;Revon-Rivière et al., 2019). However, if those who died due to treatment-related causes were included, general PC alone was not associated with a lower intensity of the EoL care. These patients were more likely to have had hematologic malignancies and were significantly less likely to have accessed specialist PC services (Widger et al., 2018).

Discussion
This review examined the impact of PC on EoL care and PoD in children, adolescents, and young adults with life-limiting conditions based on different models of care. The results demonstrated that those who receive PC are less likely to be admitted to ICUs,  are less likely to receive HI-EoL care, and are less likely to die in ICUs. However, there was less evidence of effects of PC on LoS in the EoL, receiving CPR at the time of death, and the proportion of hospital and home deaths. Several possible explanations exist, including the fact that the symptom burden may be better managed in a medical setting than in the community during EoL (Clark et al., 2016;Zernikow et al., 2019) or that access to PC may have an impact on decisions to receive less intervention-focused care aimed at extending life (e.g., ICU care) but not on medical care during the EoL. This also suggests that intensivists and those providing acute care need to incorporate effective PC communication skills and improve team cohesiveness with the primary clinical team and specialist PC team (Richards et al., 2018).
Regarding the circumstances of death, mixed results were observed for the impact of CPR attempts at the time of death in the PC group. There was a lack of homogeneity across these studies in terms of the ages and diagnoses included in the samples, model of care, and the time between PC involvement and death. Although all studies indicated finding PC to have benefits in terms of promoting DNR orders in place (Osenga et al., 2016;Ullrich et al., 2016;Snaman et al., 2017), its impact on reductions in the number of pediatric patients undergoing CPR at the time of death at the request of the patient and/or caregiver showed little consensus. Comparable to previous studies (Mack et al., 2015;Johnston et al., 2017), studies in this review revealed that studies on toddlers, as well as adolescents and young adults with cancer, indicated that PC has little impact on reducing the number of cases with CPR attempts at the time of death (Osenga et al., 2016;Snaman et al., 2017). Several possible explanations exist, including the fact that adolescents and young adults have better competency related to expressing their desire to accept invasive treatments out of a desire to sustain their lives (Hinds et al., 2005). In goals of care and level of hope in patients receiving stem cell transplants may differ from those with complex chronic conditions, which has implications for EoL care decision-making.
Although previous studies have shown that an increase in home-based PC resources may reflect an increase in home deaths (Webber et al., 2019;Håkanson et al., 2017), the impact of inpatient, home, and community PC on increasing the number of home deaths is inconsistent in this review. However, studies where the intervention included the delivery of hospice care reported PC to be associated with being more likely to die at home as compared with no PC (Thienprayoon et al., 2013;. Complex symptoms may be better dealt with in hospitals or hospices than would be the case at home. The differences in the impact of PC on PoD may also related to the distance from a tertiary center (Kassam et al., 2017), different health system infrastructures, policies, socio-economic conditions (Håkanson et al., 2017), and cultural differences.

Limitations of the included studies
Although this study adds to the literature through a systematic review intended to provide evidence for the association of PC with five important outcomes, the results should be interpreted cautiously given the lack of experimental evidence and the heavily reliance on unadjusted estimates based on observational studies. In 11 studies, there were concerns about the clarity of the inclusion and exclusion criteria or about similarities in the characteristics of the two groups under consideration (Chang et al., 2013;Keele et al., 2013;Friedrichsdorf et al., 2015;Smith et al., 2015;Vern-Gross et al., 2015;Osenga et al., 2016;Snaman et al., 2017;Chong et al., 2018;Widger et al., 2018;Revon-Rivière et al., 2019;Zernikow et al., 2019). Of the 14 cohort studies, outcomes were adjusted for demographic and clinical variables in the analyses in only five of them (Smith et al., 2015;Osenga et al., 2016;Widger et al., 2018;Revon-Rivière et al., 2019). Possible confounding by family preference was not controlled for in any of the included studies. The duration of time from receiving PC to death was stated in only one study (Ullrich et al., 2016); therefore, there were concerns as to whether the follow-up time of the PC involvement in the intervention group was sufficiently long enough for outcomes to occur. This review was unable to reveal differences in the impact of PC among children, adolescents, and young adults referred to PC based on whether it occurred early and late in their illness trajectory.
There were concerns about the representativeness of samples. Subjects in seven of the 15 studies (46.7%) comprised only patients who had been hospitalized and died in hospital in order to collect accurate EoL care data (Keele et al., 2013;Osenga et al., 2016;Ullrich et al., 2016;Snaman et al., 2017;Chong et al., 2018;Revon-Rivière et al., 2019;Spraker-Perlman et al., 2019). Subjects in seven of the 15 studies (46.7%) used a single-site design (Friedrichsdorf et al., 2015;Smith et al., 2015;Vern-Gross et al., 2015;Osenga et al., 2016;Ullrich et al., 2016;Snaman et al., 2017;Spraker-Perlman et al., 2019). The primary diagnoses of the subjects varied, with the majority having neonatology/chromosome disorders, respiratory diseases, and hematology/oncology diseases. The distribution of diagnosis among the groups of subjects may have been different from the general pediatric population in different locations. There was a concentration of studies from the USA. Although studies involving Chinese and Bahasa populations were included, and Chinese and Indonesian databases were searched, no studies conducted using these languages were found. EoL care reflects child and family preferences, which is highly relationship-based (Hinds et al., 2005) and culturally related. Therefore, more population-based cohort studies assessing the impact of PC are needed in non-Western countries.

Strengths and limitations of the review
This was the first systematic review to demonstrate the impact of PC on EoL care and PoD in children, adolescents, and young adults with life-limiting conditions. Since this review focused on studies published from 2010 to 2020, this increased the homogeneity of the progress of curative treatments and the recognition of the need for PC in the pediatric population. Only studies comparing those who actually received and did not receive PC were included, as opposed to comparing those with and without access to such services. This ensured that all subjects in the intervention groupf did receive PC. No included studies compared groups using a before-after design. This decreased the potential impact on the EoL care outcomes due to the progression of the illness. This review was not restricted to a specific PC model in order to allow a comparison of impacts across different components of intervention while also limiting the outcome of hospital admission and LoS to the last year of life. This widened the scope of the included studies and enriched the data synthesis, but the outcome may not be generalizable to a population at the early diagnosis stage. Outcomes in this review may have been influenced by differences in the characteristics of the intervention and control groups. Patients who died due to any cause were considered as being in life-limiting conditions and were included in this review. For those who had better prognoses and received a significant amount of intensive care, but subsequently died with iatrogenic complications, caregivers and healthcare providers may have been less aware of the availability of specialist PC resources, or patients and caregivers may have been less likely to be referred for such care (Ullrich et al., 2016;Widger et al., 2018). These groups may have increased ICU admissions, HI-EoL, been denied PC team involvement, or died in the hospital, which led to a concern related to possible referral bias.

Implications for clinical practice and research
Our results provide insights into PC resource allocation throughout the last year of life. To the best of our knowledge, there were Table 7. Description of the included papers comparing place of death for people with PC involvement vs. those without it (n = 10) only two articles to date comparing HI-EoL care between oncology pediatric patients with and without PC. One randomized controlled trial examining the impact of PC on infants with prenatal single-ventricle diagnoses was excluded from this review because the outcomes did not meet the last year of life criteria (Hancock et al., 2018). More research is needed related to developing suitable indicators for the purpose of investigating the intensity of EoL care among non-cancer pediatric patients. Due to concerns about the heterogeneity of the populations, interventions, outcome measures, and study designs, it was not appropriate to conduct a further meta-analysis.

Conclusion
In this work, PC was associated with a reduction in ICU admissions, ICU deaths, and HI-EoL care in the last month of life. The findings reveal an enormous ongoing gap in the understanding of whether pediatric PC is generally useful in reducing hospital admissions and LoS during EoL, reducing futile CPR, and avoiding hospital death. Based on this review, we make the following recommendations: first, specialist and multidisciplinary PC as well as the continuity of PC are required to reduce HI-EoL care. Second, careful attention should be paid to the need for a longer LoS in a medical setting late in life, and earlier EoL care discussion should take place with patients/caregivers especially in regard to attempting CPR in toddlers, adolescents, and the young adult population. Third, a lack of robust evidence led to identifying a gap in rigorous, comprehensive, and multisite prospective studies of data collection focusing on the impact of PC on EoL care and PoD for pediatric patients between 0 and 25 years of age. In addition, there may be reason to believe the patterns of utilization vary in different areas around the world, which necessitates the inclusion of cultural factors in future data collection.
Supplementary material. The supplementary material for this article can be found at https://doi.org/10.1017/S1478951520001455.