Association between palliative care consultation and care transition outcomes among hospitalized advanced cancer patients

Abstract

Background

The period after hospital discharge is a high-risk phase for patients with advanced cancer, often involving acute-care use that reflects transitional care quality. The impact of inpatient palliative care (PC) consultation on short-term post-discharge outcomes, however, remains uncertain. We assessed whether inpatient PC consultation was associated with differences in 30-day post-discharge outcomes.

Methods

Using electronic medical records from a tertiary hospital linked with national claims data, we identified patients with lung, stomach, colorectal, liver, or pancreatobiliary cancer who died between 2018 and 2023. Those discharged alive after a hospitalization with inpatient PC consultation were matched 1:1 to patients without PC using propensity scores. Outcomes were 30-day emergency department (ED) visits, hospital readmissions, and intensive care unit (ICU) admissions, 30-day mortality, and total direct medical costs. Fine–Gray competing risk and generalized linear models were used for comparisons.

Results

Among 830 matched individuals, 30-day ED visits (45.8% vs 45.5%; adjusted odds ratio [aOR] = 0.95; 95% CI = 0.72 to 1.27) and readmission rates (69.6% vs 72.3%; aOR = 0.86; 95% CI = 0.63 to 1.18) were similar. Intensive care unit admission rates were substantially lower among patients receiving PC (1.9% vs 9.2%; aOR = 0.17; 95% CI = 0.07 to 0.37). The total 30-day medical costs were lower in the PC group (cost ratio = 0.65; 95% CI = 0.55 to 0.76). Thirty-day mortality was higher among patients who required PC (37.6% vs 16.1%).

Conclusion

Inpatient PC consultation was not associated with 30-day ED visits or hospital readmissions but was linked to substantially lower ICU admissions and reduced short-term medical costs.

Introduction

Care transitions constitute a particularly vulnerable period for patients with cancer, especially those with advanced disease, who often experience rapid changes in symptoms, functional decline, and prognostic uncertainty.^1,2 Unplanned healthcare encounters, such as hospital readmissions, emergency department (ED) visits, and intensive care unit (ICU) admissions, are common during this time and are widely used as indicators of transitional care quality in oncology.^3-5 These events can disrupt continuity of care, exacerbate symptom burden, and increase caregiver strain while also imposing substantial demands on acute care services and increasing end-of-life costs. As cancer care increasingly relies on shorter periods of hospitalization and outpatient management, ensuring stability during the early post-discharge period has become an essential component of high-quality care.^6,7

Palliative care (PC) consultations have become an integral part of hospital-based oncology practice, supporting symptom management, facilitating communication about prognosis and care goals, and assisting in discharge planning.⁸ Although inpatient PC is often positioned to bridge the transition from hospital to home,^2,9-11 most previous studies have focused on in-hospital processes and outcomes.^8,9,12-15 Consequently, evidence regarding its influence on early post-discharge care remains limited, and existing studies have reported mixed findings regarding outcomes such as ED visits and readmissions.^11-14,16 These gaps underscore the need to better understand how PC engagement during hospitalization affects the patterns and intensities of post-discharge care.

To address this gap, we evaluated the 30-day post-discharge clinical outcomes of patients with advanced cancer who did and did not receive inpatient PC consultations during their final hospitalization at a tertiary academic center. Our aim was to characterize the differences in acute care use and the patterns and intensities of early post-discharge care in a real-world clinical context.

Methods

Study design

This was a retrospective cohort study using a customized database linking electronic medical records (EMRs) from Seoul National University Hospital (SNUH) with national claims data from the Korean National Health Insurance Service (NHIS). Because NHIS captures healthcare utilization across all medical institutions in Korea, the linked EMR–claims database enabled follow-up of clinical events and healthcare utilization, regardless of where care was received.

Study setting

SNUH is a large tertiary academic referral center in South Korea with 1635 beds and approximately 428 000 inpatient admissions annually. Patients with cancer are primarily admitted to oncology-specific wards staffed by oncology specialists.

The hospital operates an established inpatient PC consultation service that provides approximately 1900 consultations on a yearly basis. In routine practice, PC consultation may be requested for symptom management, psychosocial support, complex communication, or discharge planning, with most referrals focused on goals-of-care discussions and care planning. Importantly, receiving inpatient PC consultation does not restrict subsequent healthcare use; patients retain full access to ED services, hospital readmission, and ICU care based on clinical needs. ED and ICU services are readily accessible within the Korean healthcare system and are commonly utilized among patients with advanced cancer. In addition, many hospitalizations for patients with advanced cancer occur as planned admissions from outpatient oncology clinics for symptom management or clinical evaluation rather than through the ED.

Study participants

The study population included patients with 1 of 5 major cancers—lung, liver, stomach, colorectal, or pancreatobiliary—who died between January 2018 and June 2023. These cancer types were selected because they represent the leading causes of cancer-related mortality in Korea and account for a large proportion of patients receiving hospital-based care near the end of life. Restricting the cohort to major solid tumors was intended to reduce heterogeneity in disease trajectories and patterns of healthcare utilization. Among 18 089 eligible decedents, 41 were excluded due to missing values for key variables (n = 23) or mismatched death dates between hospital and NHIS records (n = 18), resulting in a final analytic cohort of 18 048 patients. They were categorized into 2 groups according to their PC consultation status. For both groups, study subjects were selected from patients who were discharged alive after a hospitalization and subsequently died, allowing assessment of 30-day post-discharge outcomes. For the PC group, we regarded hospitalizations during which an inpatient PC consultation occurred and excluded those who did not receive it throughout their inpatient stay. PC exposure was defined based on inpatient consultation during the index hospitalization, and information on prior PC involvement was not available. For the non-PC group, we only considered patients who had been hospitalized at SNUH within 1-3 months before death without any PC referral. This time window was selected to capture hospital discharges occurring during a comparable late phase of illness and to ensure that both groups entered a similar post-discharge risk period. In addition, the non-PC group was restricted to patients with no record of PC consultation at SNUH prior to the index hospitalization. To ensure inclusion of patients discharged to a non-medical setting, those who had healthcare utilization records at medical institutions other than SNUH on the date of discharge were excluded, as such records indicate inter-facility transfer rather than return to the community. In addition, patients who stayed at SNUH <3 days were further excluded from both groups, constructing 1:1 propensity-score matched individuals for the analysis (Figure 1).

Figure 1.

Flowchart of the study.

Outcomes of interest

The primary outcomes were 30-day post-discharge ED visits, hospital readmissions, and ICU admissions, which were assessed as indicators of transitional care quality. Outcomes were defined as events occurring from the day after discharge (discharge date +1) through day 30; events occurring on the discharge date were excluded to avoid misclassification. For both groups, follow-up began at the time of hospital discharge, and outcomes were assessed over the same fixed 30-day period. Each outcome was analyzed as a binary variable indicating whether the event occurred within 30 days after discharge, and only the first occurrence of each outcome type was counted. Hospital readmission was defined as any inpatient admission recorded in the NHIS claims database within 30 days after discharge. In Korea, inpatient hospice care is delivered within hospital-based units and recorded as a hospitalization; therefore, hospice admissions were included.

Secondary outcomes included total direct medical costs incurred within 30 days after discharge and all-cause mortality during the same period. Total direct medical costs were calculated using NHIS claims data by summing all healthcare expenditures incurred within 30 days after discharge, including both insurer-covered payments and patient out-of-pocket payments. Given the advanced illness trajectory of this population, short-term mortality was used to contextualize patterns of post-discharge healthcare use rather than to evaluate care quality.

Statistical analysis

Descriptive statistics were used to summarize the baseline characteristics and post-discharge outcomes. Continuous variables are reported as medians with interquartile ranges, and categorical variables as frequencies and percentages. Group differences in baseline characteristics before and after matching were assessed using the Wilcoxon rank-sum test for continuous variables and the chi-squared test for categorical variables.

To minimize potential confounding between the PC and non-PC groups, propensity score matching was implemented utilizing a 1:1 nearest-neighbor algorithm without replacement, with a caliper of 0.2 standard deviation of the logit of the propensity score. The propensity score model included the following covariates: age, sex, household income level, residential area (metropolitan vs non-metropolitan), non-cancer Charlson Comorbidity Index, cancer type, year of death (2018-2019 as pre-COVID, 2020-2023 as COVID era), route of hospital admission (via ED vs other), and length of hospital stay. Covariate balance after matching was assessed using standardized mean differences, with absolute values below 0.1 considered indicative of adequate balance.

In the matched cohort, primary outcomes were compared between the PC and non-PC groups using logistic regression with robust standard errors to appropriately account for the paired nature of the matched data. Additionally, to mitigate potential bias arising from differential exposure time following hospital discharge, the interval from discharge to death (discharge-to-death) was further incorporated in the regression models as a categorical covariate (<15, 15-29, ≥30 days).

Fine–Gray sub-distribution hazard regression models were employed to analyze the primary outcomes, with death treated as a competing risk. This approach was used because the study population had a high chance of mortality during the observation period.¹⁷ The cumulative incidence function was estimated for each outcome. For the secondary outcome of all-cause mortality within 30 days following hospital discharge, Kaplan–Meier survival curves were presented, and between-group comparisons were conducted using the log-rank test. For the total direct medical costs within 30 days post-discharge, a generalized linear regression model with gamma distribution and the ratio of the 2 groups’ costs were compared. Among clinical outcomes for which the PC group exhibited lower utilization intensities, simulation-based sensitivity analyses were further employed to strictly evaluate the effect of PC consultation, accounting for generally poorer health status inherent to this group. All statistical analyses were performed using SAS (version 9.4), and a 2-sided P value of <.05 was considered statistically significant.

Ethics

The study was approved by the Institutional Review Board of SNUH (approval No. H-2305-088-1431), which waived the requirement for informed consent because de-identified data were used. All procedures complied with applicable data protection and privacy regulations.

Results

Study population and cohort construction

Among the 18 048 patients with advanced cancer who died between January 2018 and June 2023, 415 who received inpatient PC consultations and 415 without PC consultations were included in the final matched cohort (Figure 1). From the 4102 decedents in the PC group, 430 met the inclusion criteria for post-discharge follow-up and were eligible for matching. In the non-PC group (n = 13 946), 2002 patients met inclusion criteria. After applying 1:1 propensity score matching, 415 matched pairs were constructed for the analysis.

Baseline characteristics before and after matching

In the unmatched cohort, patients in the PC group were generally younger, more likely to reside in metropolitan areas, and had longer hospital stays. They were also more frequently admitted via non-emergency routes and more likely to have lung or stomach cancer, whereas liver and pancreatobiliary cancers were more common in the non-PC group. A greater proportion of patients with PC died during the COVID-19 era (2020-2023) (Table S1). After propensity score matching, the 2 groups were comparable across all baseline variables, including demographic, clinical, and admission-related factors. All standardized mean differences were below 0.1, indicating an adequate covariate balance (Table 1).

Table 1.

Baseline characteristics of matched advanced cancer patients.

Characteristics	PC group (n = 415), No. (%)	Matched Non-PC group (n = 415), No. (%)	P	SMD
Age (group)
<65	206 (49.6)	189 (45.5)	.24	0.08
≥65	209 (50.4)	226 (54.5)
Sex
Male	266 (64.1)	274 (66.0)	.56	0.04
Female	149 (35.9)	141 (34.0)
Household income
First quartile (lowest)	99 (23.9)	103 (24.8)	.96	0.04
Second quartile	68 (16.4)	63 (15.2)
Third quartile	73 (17.6)	74 (17.8)
Fourth quartile (highest)	175 (42.2)	175 (42.2)
Residence
Metropolitan	259 (62.4)	261 (62.9)	.89	0.01
Urban/suburban	156 (37.6)	154 (37.1)
Non-cancer CCI
0	48 (11.6)	40 (9.6)	.67	0.09
1	106 (25.5)	110 (26.5)
2	97 (23.4)	89 (21.5)
≥3	164 (39.5)	176 (42.4)
Cancer type
Lung	121 (29.2)	118 (28.4)	.98	0.05
Stomach	52 (12.5)	55 (13.3)
Colon	53 (12.8)	50 (12.1)
Liver	57 (13.7)	62 (14.9)
Gallbladder/pancreas	132 (31.8)	130 (31.3)
Year of death
2018-2019	98 (23.6)	88 (21.2)	.41	0.06
2020-2023	317 (76.4)	327 (78.8)
Route of admission
ED	95 (22.9)	90 (21.7)	.68	0.03
Non-ED	320 (77.1)	325 (78.3)
Length of hospital stay
<1 week	103 (24.8)	105 (25.3)	.87	0.01
≥1 week	312 (75.2)	310 (74.7)

Abbreviations: CCI = Charlson Comorbidity Index; ED = emergency department; PC = palliative care; SMD = standardized mean difference.

Thirty-day post-discharge clinical outcomes

In the matched cohort, 45.8% and 45.5% of the patients in the PC and non-PC groups, respectively, visited the ED within 30 days after discharge. The adjusted odds ratio (aOR) for ED visits was 0.95 (95% CI = 0.72 to 1.27) (Table 2). Hospital readmission occurred in 69.6% and 72.3% of the patients, respectively (aOR = 0.86; 95% CI = 0.63 to 1.18). In contrast, ICU admissions were substantially less frequent in the PC group (1.9% vs 9.2%), with a significantly lower aOR of 0.17 (95% CI = 0.07 to 0.37) (Table 2). Sensitivity analyses using alternative methods to adjust for time from discharge to death showed consistent patterns. While the associations for ED visits and readmissions remained non-significant, markedly lower ICU admission rates among the PC group persisted across all models (Table S2).

Table 2.

Thirty-day post-discharge clinical outcomes in the matched cohort.

Outcomes	PC group (n = 415), No. (%)	Matched non-PC group (n = 415), No. (%)	Crude OR (95% CI)	Adjusted ORa (95% CI)
ED visit	190 (45.8)	189 (45.5)	1.01 (0.77 to 1.33)	0.95 (0.72 to 1.27)
Hospital readmission	289 (69.6)	300 (72.3)	0.88 (0.65 to 1.19)	0.86 (0.63 to 1.18)
ICU admission	8 (1.9)	38 (9.2)	0.20 (0.09 to 0.42)	0.16 (0.07 to 0.37)

Abbreviations: CI = confidence interval; ED = emergency department; ICU = intensive care unit; PC = palliative care; OR = odds ratio.

^aAdjustments were made with time from discharge to death as a categorical variable (<15, 15-29, ≥30 days).

The cumulative incidence functions based on the Fine–Gray competing risk models further support these findings (Figure 2). The risk trajectories for ED visits (Figure 2A) and hospital readmissions (Figure 2B) 30 days after discharge were similar between the PC and non-PC groups. In contrast, the cumulative incidence of ICU admission was significantly lower in the PC group (Figure 2C). During the 30-day post discharge period, the PC group experienced more than 2-fold mortality compared to the non-PC group (37.6% vs 16.1%), which corresponded to a significantly higher risk shown in the Kaplan–Meier curve (HR = 2.78; 95% CI = 2.09 to 3.71) (Figure S1). In a simulation-based sensitivity analysis accounting for excess early mortality in the PC group, the adjusted estimate of ICU admission remained lower than that in the non-PC group (0.032 vs 0.092) (Figure S2, Table S3).

Figure 2.

Cumulative incidence of outcomes within 30 days after discharge. (A) Emergency department visits within 30 days after discharge. (B) Readmission within 30 days after discharge. (C) ICU admission within 30 days after discharge.

Thirty-day post-discharge medical costs

The total direct medical costs within 30 days after discharge were significantly lower in the PC group than in the non-PC group. The mean cost was 3784.99 USD (95% CI = 3367.00 to 4254.87) in the PC group and 5868.76 USD (95% CI = 5224.53 to 6592.44) in the non-PC group, corresponding to a cost ratio of 0.65 (95% CI = 0.55 to 0.76) (Figure 3). Even after making an adjustment for higher mortality in the PC group, the estimate still remained lower than in ICU admission (4150.93 vs 5868.76 USD) (Figure S2, Table S3), suggesting that the findings were robust to survival differences.

Figure 3.

Total direct medical costs of patients within 30 days after discharge. 1 USD = 1362.26 KRW (June 25, 2025). Dotted points represent outliers.

Discussion

This study compared 30-day post-discharge outcomes between patients with advanced cancer who received inpatient PC consultations and matched patients who did not. In this cohort, PC consultation during hospitalization was associated with a lower likelihood of ICU admission and reduced short-term medical costs, whereas the rates of ED visits and hospital readmissions were similar between the groups. These findings suggest that the influence of inpatient PC on transitional care may be more evident for high-intensity service use than for the overall frequency of acute care encounters.

The lower ICU admission rate among patients who received inpatient PC consultation suggests that PC may influence decisions regarding escalating care shortly after discharge. Rather than reducing the overall amount of acute care use, PC involvement appears to clarify expectations and support planning on how to respond to clinical changes when patients return home.¹⁰ The accompanying reduction in short-term medical costs aligns with this pattern as ICU-level care constitutes one of the most resource-intensive services near the end of life.^16,18 Taken together, these findings suggest that inpatient PC plays a specific role in reducing care intensity rather than frequency after discharge.^8,11,18

In contrast, ED visits and hospital readmissions occurred at similar rates between the groups. This finding is in line with prior studies reporting the mixed effects of inpatient PC on short-term acute care use,^{11-14,16,18-20} with some showing reductions in unplanned returns and others observing little change. One explanation is that these outcomes are strongly driven by the natural course of advanced cancer—acute symptom exacerbations, sudden functional decline, and infection-related events—which may require urgent evaluation regardless of prior PC involvement.¹¹ The relatively high rates of ED visits and readmissions in our cohort compared to earlier reports may also reflect the inclusion of patients with more advanced or unstable illness.^11,16,18,21 In this context, even well-coordinated discharge planning may have a limited capacity to prevent all ED visits or rehospitalizations, particularly within a short follow-up window.¹¹ Moreover, our data did not capture the destination or intent of readmission, so we could not differentiate potentially avoidable acute-care returns from transitions to inpatient hospices or other comfort-oriented settings, which have been frequently described in previous works.^20,22 When viewed collectively, these considerations suggest that ED visits and readmissions may be too heterogeneous and, therefore, should be understood cautiously when evaluating the influence of inpatient PC on transitional care quality.

The higher 30-day mortality observed in the PC group provides additional context for interpreting these patterns. Patients referred to PC during hospitalization are often at more advanced stages in their disease trajectory, which may contribute to ongoing symptom instability and the need for urgent evaluation after discharge, even when care is well coordinated. In this setting, ED visits or readmissions may reflect the natural progression of advanced conditions, rather than the shortcomings of transitional care. At the same time, the lower likelihood of ICU admission among PC recipients despite their limited survival suggests a different trajectory of care in the days and weeks after discharge, in which clinical deterioration is managed without escalation to intensive interventions.

Overall, these findings suggest that inpatient PC may help shape how care proceeds in the early period after discharge, particularly by reducing the likelihood of escalating intensive care services. Integrating PC consultations more deliberately into discharge planning may better support patients on their return home.^8,10 Future studies should explore the reasons underlying post-discharge ED visits and other unplanned encounters to clarify where additional support or coordination is most beneficial.

This study had several limitations. Although we used linked institutional and nationwide claims data, detailed clinical indicators at the time of hospitalization—such as symptom burden, functional status, cancer disease status, or the reason for admission—were not available. These factors may have influenced both the likelihood of receiving inpatient PC consultation and the risk of subsequent healthcare utilization, including ICU admission and medical expenditures. Although propensity score matching incorporated demographic, clinical, and healthcare utilization variables as proxies, residual confounding related to illness severity and clinical instability cannot be completely excluded. In addition, since the non-PC group was defined using hospitalizations occurring 1-3 months before death, the 2 different entry points assigned to each group may inadvertently introduce immortal time for the comparison cohort, which cannot be fully accounted using our matching strategy. Although this approach was intended to capture discharges during a comparable late phase of illness, differences in underlying prognosis between groups may remain, potentially biasing the estimates of healthcare utilization intensity. Also, we were unable to determine the specific reasons for each PC consultation, and previous studies have suggested that consultation intent may influence subsequent care trajectories.^19,23 Additionally, we did not evaluate the timing of inpatient PC consultations; some studies have suggested that earlier involvement may be associated with more favorable outcomes.²⁴ Information on individual documentation of do-not-resuscitate or life-sustaining treatment preferences during hospitalization was not available. Although such documentation in Korea does not automatically preclude ICU admission, the extent to which treatment-limitation decisions influenced ICU utilization could not be determined. Information on the destination or purpose of readmission was also unavailable, and the dataset did not include reasons for post-discharge ED visits, limiting insight into whether these encounters reflected unavoidable clinical deterioration or support gaps. In addition, the dataset did not include detailed information on care provided after discharge, such as home-based services, ongoing outpatient cancer treatment, or outpatient PC, limiting our ability to assess differences in post-discharge care coordination or support. Because claims data did not allow differentiation between acute-care readmissions and admissions for inpatient hospice or other comfort-oriented care, hospital readmission outcomes may have reflected a mixture of care types. As this study was conducted at a single academic center within the Korean healthcare system, patterns of care transition may differ in other settings. Finally, a sizeable proportion of eligible patients was excluded during propensity score matching, which may have affected generalizability despite improved covariate balance.

Despite these limitations, the use of linked clinical and nationwide claims data allowed for a detailed examination of 30-day post-discharge care patterns of rigorously matched patients with advanced cancer who were discharged alive after hospitalization. This study further provides evidence from an Asian healthcare setting, where data on transitional care after inpatient PC remain limited. In summary, inpatient PC consultation was associated with fewer ICU admissions and lower short-term medical costs after discharge, while overall acute-care use remained high in both groups (ED visits: 45.8% vs 45.5%; hospital readmissions: 69.6% vs 72.3%). Furthermore, 30-day post-discharge mortality was significantly higher among the PC group compared to the non-PC group (37.6% vs 16.1%). These findings help clarify how PC delivered during hospitalization may influence early post-discharge care, particularly by limiting escalation to intensive services during the transitional period.

Author contributions

Hak Jun Kim (Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Validation, Visualization, Writing—original draft, Writing—review & editing), Jin-ah Sim (Formal analysis, Project administration, Supervision, Validation, Writing—review & editing), Sun Young Lee (Funding acquisition, Project administration, Resources, Supervision, Writing—review & editing), and Shin Hye Yoo (Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Validation, Writing—original draft, Writing—review & editing)

Supplementary material

Supplementary material is available at JNCI Cancer Spectrum online.

Funding

This research was supported by grants from the Patient-Centered Clinical Research Coordinating Center (PACEN) funded by the Ministry of Health and Welfare, Republic of Korea (number: RS-2023-KH137917 and RS-2021-KH120239).

Conflicts of interest

All authors declare that there is no conflict of interest related to this work.

Data availability

The data used in this study are not publicly available, and the study participants were completely de-identified by the National Health Insurance Service in Korea. The study protocol and statistical analysis plan can be shared upon reasonable request from the corresponding author.