Implications of Pediatric Palliative Consultation for Intensive Care Unit Stay

Introduction

For patients with serious illnesses, consultation with a palliative medicine specialist is recommended at diagnosis.^1–8 Yet, among children who died while hospitalized in the United States between 2001 and 2011, only 4% received specialty palliative care. ⁹ There are recognized barriers and facilitators to palliative care consultation for children in need, and the threshold to consult is affected by the knowledge and comfort of an individual provider as well as the available evidence to inform practice.^10–17 Benefits of pediatric palliative care (PPC) may include improved symptom control,^8,18–25 higher quality of life,^{20,21,26–31} as well as reduced costs and health care utilization.^{22,23,26–30,32–38}

Quantifying the value of PPC is important for supporting and expanding provision of these services, yet, prospective, randomized controlled trials are rare.^25,39 Analyses of PPC receipt are limited by nonrandomization,^{11,18–21,23,24,26,28–38,40–42} nonstandardization of the palliative intervention, and potential errors in pre–post study designs which lack a control group.^{24,27,34,35,42} In studies that distinguish early from late palliative care intervention, definitions of timing have typically been established retroactively from the date of death, potentially limiting the generalizability of the findings.^29,33,36,38

In-hospital outcomes of palliative care consultation may be especially subject to these biases in retrospective data analysis because patient acuity at admission is a strong confounder of the association between palliative care utilization and the course of the hospital stay. Particularly, admission to the intensive care unit (ICU) is an expensive endeavor that has been used in the PPC literature to describe high-intensity care at the end of life.^22,38,43 ICU length of stay (LOS), however, has rarely been reported in pediatric studies,^25,44 despite reduced ICU admission^45–48 and LOS^45,49–54 having been identified as drivers of cost savings associated with palliative care consultation in adults.

Despite evidence suggesting benefits of palliative consultation for reducing ICU stay among adults, ⁵⁵ the impact of specialty PPC on ICU LOS for children is unclear. Ananth et al. ³⁵ reported increased proportion of ICU admission following palliative consultation, while Hancock et al. ²⁵ demonstrated a nonsignificant reduction in ICU LOS among children randomized to receive PPC. To evaluate the potential impact of palliative care consultation, we propose analyzing ICU stay as a dynamic outcome over the course of hospitalization and evaluating its day-to-day association with receipt of PPC. We hypothesize that a time-varying measure of PPC (i.e., beginning to receive PPC during a hospital admission) would be associated with reduced likelihood of being in the ICU on a given day of hospitalization for a serious illness. Using a retrospective analysis of single-center data, we test this hypothesis among hospital admissions for diagnoses deemed to qualify patients for PPC.

Patients and Methods

This study was approved and a waiver of informed consent granted by the Institutional Review Board of Nationwide Children's Hospital (NCH; IRB17-01313), a 476-bed tertiary academic children's hospital in Columbus, Ohio. In 2016, NCH reported 18,183 inpatient discharges and 1,425,023 total patient visits for patients from all 50 United States and 52 foreign countries. ⁵⁶ We retrospectively reviewed hospital admissions of patients ages 0–21 years in 2015–2017 that lasted >1 night, so as to exclude overnight observations.

Hospitalizations were evaluated for inclusion in the study if an admission diagnosis was also a trigger, per expert opinion, for automatic referral to PPC. ⁵⁷ We focused on patients admitted with diagnoses for automatic referral, which could be clearly mapped to International Classification of Diseases, 9th or 10th revision (ICD9, ICD10) codes (Table 1). Eligible diagnoses were selected from malignant, pulmonary, genetic, neurologic/neuromuscular/neurodegenerative, metabolic/inclusion, infectious, orthopedic, renal, gastrointestinal, neonatal, and cardiac criteria, which were not mutually exclusive. We excluded diagnoses for which automatic referral to PPC is already practiced at our institution (i.e., single-ventricle physiology, heart transplant candidate/recipient, and bone marrow transplant candidate/recipient).

We excluded patients aged less than one year and patients with neonatal disease, as stay in the neonatal ICU was implied for this population. At NCH, the PPC team, in place since 2009, includes five attending physicians, two nurse practitioners, one chaplain, and one social worker. For patients with multiple admissions during the study period, the earliest admission was included in the analysis. During the index admission, we used a quality improvement database maintained by the PPC team to determine the earliest receipt of PPC and the quantities of visit types among PPC providers.

ICU stay and patient characteristics were initially compared between patients who did and did not receive palliative consultation during the index hospital admission, using unpaired t tests or rank-sum tests for continuous variables, and chi-square tests for categorical variables. For further analysis, data were transformed to a patient-day level ⁵⁸ (i.e., the data file contained one observation for each day a patient was in the hospital). ICU stay was assessed on each day during the index admission as a binary variable (0 = not in ICU, 1 = in ICU) reflecting the physical location of the patient in the closed-ICU model at our institution. Our measure of palliative care receipt focused on the earliest consultation, so we defined palliative care consultation status as zero on all days before the earliest palliative care consultation, and as one on all subsequent days. Covariates included patient age, gender, race/ethnicity, insurance coverage, proximity to the hospital (i.e., residing in the same county as NCH, a contiguous county, or another county or state), and, as a measure of complexity, whether the patient had previously undergone surgery (excluding minor procedures such as circumcision and tympanostomy). Due to significant differences found in the likelihood of palliative care consultation between patients with cancer and noncancer diagnoses at admission, multivariable analysis was performed separately for these groups.

Using the patient-day data file, we applied multivariable mixed effects logistic regression, where the outcome was the odds of being in the ICU on a given day. ⁵⁹ A patient-level random intercept was included to account for residual differences between patients in their risk of being admitted to the ICU during the hospital stay. This analysis included a control variable for days since hospital admission, to capture any trend toward increasing or decreasing likelihood of being in the ICU during prolonged hospital stay. The number of days since hospital admission was also interacted with the measure of receiving palliative care to determine whether the association of palliative care consultation with reduced likelihood of ICU stay was stronger at the beginning of a hospitalization. The significance test on the interaction coefficient was used to determine whether the association between palliative care consultation and likelihood of ICU stay meaningfully differed over the course of hospitalization. Analysis was performed in Stata/IC 14.2 (College Station, TX), and two-tailed p < 0.05 was considered statistically significant.

Results

Our analysis identified 777 eligible index admissions (total of 11,954 days in the hospital), of which 100 (13%) included palliative care consultation with a median of 6 (interquartile range [IQR]: 6, 13) PPC encounters: 4 medical (IQR: 2, 9), 1 chaplain (IQR: 0, 2), and 1 social worker (IQR: 0, 3) visits per admission. A total of 14 (2%) patients died during the index admission, of whom 11 (79%) received PPC and 13 (93%) were in the ICU on the day of death. Principal demographics of patients in the analysis included mean age 8 ± 6 years, male sex (55%), white race (71%), and commercial insurance (52%). Cardiac diagnoses were most frequent (29%) followed by gastrointestinal (22%) and malignant (20%) conditions (Table 1). Among all admissions, median hospital LOS was seven days (IQR: 4, 14) and 44% of admissions included a stay in the ICU, with a median total ICU LOS of four days (IQR: 2, 8).

For patients who received PPC, the earliest consultation occurred after a median of three days following admission (IQR: 1, 9) and 60% received ICU level of care during the index hospitalization. Among these 60 children who both received PPC and had an ICU stay, 13 received PPC before ICU admission, 4 received PPC on the same day as ICU transfer, and 43 received PPC after having been transferred to the ICU. Among all patients who received care in the ICU, those who also received PPC had longer median hospital LOS (38 days; IQR: 16, 91) and total ICU LOS (14 days; 6, 25) than those who did not receive PPC (8 days; IQR: 5, 15, and 3 days; IQR: 2, 6, respectively).

Table 2 summarizes patient characteristics, PPC consultation, and ICU stay according to cancer versus noncancer diagnosis. Patients with a cancer diagnosis tended to be older and were less likely to have previously undergone surgery. PPC receipt was much more common among patients with a cancer diagnosis (41% vs. 6% in the noncancer cohort). Among PPC patients, the first consult occurred after a median of two days in the cancer group compared to six days in the noncancer group. Considering outcomes during the hospital stay, hospital LOS was longer for cancer patients, but total ICU LOS did not differ by diagnosis type. In-hospital mortality was rare in both groups (2%).

Tables 3 and 4 illustrate the relationship between receipt of palliative care and the odds of being in the ICU, for the cancer and noncancer cohorts, respectively. The models accounted for patient-level differences that were not explicitly measured through the use of a random patient-level intercept. In the cancer cohort (Table 3), multivariable analysis found that receipt of palliative care within one day of admission was associated with 79% lower odds of being in the ICU (odds ratio [OR] = 0.21; 95% confidence interval [CI]: 0.13–0.34; p < 0.001). The association of palliative care consultation with lower odds of being in the ICU was partially attenuated (i.e., the OR approached 1) for each additional day since admission (interaction OR = 1.03; 95% CI: 1.02–1.04; p < 0.001), as shown in Figure 1. In the noncancer cohort (Table 4), receiving PPC within one day of admission was associated with 85% lower odds of being in the ICU (OR = 0.15; 95% CI: 0.08–0.26; p < 0.001). Although the interaction of PPC with days in the hospital was statistically significant (interaction OR = 1.008; 95% CI: 1.002–1.014; p = 0.015), it did not indicate substantive weakening of the association between PPC and likelihood of being in the ICU (Fig. 1). For example, the models shown in Tables 3 and 4 indicate that receiving PPC on day seven of a hospital stay predicted 74% lower odds of being in the ICU for cancer patients (95% CI: 60%–83%) and 85% lower odds of being in the ICU for noncancer patients (95% CI: 73%–91%).

OPEN IN VIEWER

FIG. 1.

Association of palliative care consultation with lower odds of being in the ICU. Solid = cancer cohort; Dashed = noncancer cohort. ICU, intensive care unit.

Two hundred eighty-one of the 777 patients (36%) had at least one additional eligible admission during the study period that met inclusion criteria. These patients had a total of 679 revisits—ranging from 1 to 11 subsequent admissions per patient—and the proportion involving ICU stays did not differ between initial visits (41%) or revisits (39%, p = 0.527). Among the 281 patients with revisits, the number of revisits did not differ between those who received PPC in their index admission (median: 2, IQR: 3, 4) and those who did not (median: 2, IQR: 3, 4; p = 0.447). The proportion of subsequent admissions that involved ICU stays also did not differ between those who received PPC in their index admission (38%) and those who did not (36%, p = 0.726).

Discussion

In this retrospective study, we identified a cohort of children admitted to a free-standing tertiary-care children's hospital who were eligible for PPC and found that receipt of PPC was associated with significantly reduced odds of being in the ICU on a given day. We identified this association by focusing on the time-dependent nature of ICU stay during a hospital admission—in other words, even though palliative care is typically provided to more ill patients at higher risk of prolonged stay in the ICU, PPC may prevent transfer to the ICU or facilitate transfer out of the ICU. Our finding of reduced odds of being in the ICU on days after the first PPC consultation has an important implication in demonstrating the value of PPC for in-hospital health care utilization outcomes.

Evidence on the value of PPC based on prospective randomization is available from only one study. ²⁵ Similar to the majority of the research on the impact of PPC, our study was conducted retrospectively. However, our study demonstrates the first use of a list of diagnoses suggested as triggers for PPC consultation to identify a cohort of hospitalized children who are eligible to receive PPC. Previous retrospective PPC studies have used same patient pre-/post-PPC design and generally did not identify a control group who did not receive PPC, as has been done in adult palliative care. ⁶⁰ Our diagnosis-based approach to constructing this cohort demonstrated internal validity, as it selected children with prolonged hospital stay, frequent need for ICU care, and greater proportion of patients receiving PPC, compared to reports of all pediatric in-hospital deaths.^9,38 With this approach of constructing a cohort of children recommended to receive PPC, our finding that only 13% of eligible children received consultation speaks to the persistent gap between the need for and receipt of specialty PPC, even in a hospital with an established PPC team.

Our study advances the paradigm for assessing the impact of specialty palliative consultation on utilization of the ICU. ⁶⁰ Without accounting for the timing of PPC consultation, patients in our cohort who received PPC had longer hospital (34 days vs. 6 days) and ICU (14 days vs. 3 days) stays than children who did not receive PPC. This difference in LOS was confounded, however, by days in the hospital and ICU before receipt of specialty palliative consultation. Our analysis on a patient-day level circumnavigated such confounding and indicated a significant association between having received PPC and reduced likelihood of being in the ICU on subsequent days; while accounting for patient-level differences that were not explicitly measured via a random patient-level intercept in the model. This evident benefit of PPC has not previously been described in pediatrics and has significant implications for reduced cost to the health care system. ⁶¹

As specialty palliative care can help control symptoms and elucidate goals to guide ongoing care, early consultation may have greater impact. With regard to our selected outcome of stay in the ICU, we did identify some attenuation of the PPC benefit (reduced odds of being in the ICU) for the cancer cohort when PPC was received later in the hospital stay, but not for children without an oncologic diagnosis. However, even in the cancer cohort, PPC was associated with reduced odds of being in the ICU well into a typical hospital stay among these patients (e.g., at three weeks since admission). We speculate that this may be related to PPC consultation controlling symptoms as well as facilitating transfer out of the ICU when consistent with patients and families' wishes and goals of care. ⁶²

Recognizing the calendar dates of our study period as arbitrary within the course of an individual's illness, we assessed eligible readmissions for differences from the index admission included in the primary analysis. While many children (36%) did have more than one eligible admission during the study period, there was not a statistically significant difference in the prevalence of ICU stay between the index admission and revisits. In addition, both the number of revisits and the proportion of revisits involving stay in the ICU were not different based on receipt of PPC during the index admission.

While our study demonstrates a new approach to quantifying the benefit of PPC for in-hospital outcomes such as ICU stay, it is limited by its retrospective, single-center nature. Since we aimed to define a cohort of children eligible to receive PPC, we used the diagnosis at the time of a hospitalization, but not at the time of the patient's first PPC encounter. ⁶³ Indeed, some patients may have first received PPC before the beginning of our study period, may have received PPC before a definitive diagnosis that met our inclusion criteria, or may have had their qualifying diagnosis omitted during admission coding/billing. These qualifying diagnoses reflect expert opinion but have not yet been used in other research, and, in some cases, were not clearly defined via ICD9/ICD10 codes. Furthermore, institution-specific use of various triggers for PPC may limit the generalizability of our findings. ¹⁵ Nonetheless, our use of coding and billing data to identify the cohort for this study represents a method of analyzing the benefits of PPC in comparison to a reference group that can be scaled to larger, multi-institutional data sets. Also, our data reflect only the receipt of specialty PPC and do not account for primary palliative care, earlier discussion about specialty PPC, and possible patient/family requests to delay or decline consultation. While our multivariable analyses revealed several other factors (such as gender) associated with the odds of being in the ICU, it is important to note that these findings were specific to the cohort of children deemed eligible to receive PPC, and do not necessarily represent predictors of ICU utilization among all hospitalized children. Finally, our analysis was limited by the nonuniformity of PPC, as the specific services provided by PPC vary by patient.

Conclusion

Among a cohort of hospitalized children deemed eligible to receive specialty palliative care, we found the likelihood of being in the ICU on a given day declined after PPC consultation. We propose that this finding reflects the role of PPC in helping to avoid ICU admission, to facilitate transfer out of the ICU when consistent with goals of care, and/or when symptoms are palliated such that intensive care is not required. Furthermore, this result mitigates confounding by severity of illness, which tends to bias group comparisons of ICU LOS between children who do and do not receive palliative care. Given the challenges of conducting randomized prospective studies in children eligible for PPC, future work may adopt similar methodology to quantify the value of PPC both in reference to the time before receiving PPC and among children who were eligible for PPC services, but did not receive them.