Journal Logo

Feature Articles

Racial/Ethnic Minority Children With Cancer Experience Higher Mortality on Admission to the ICU in the United States*

Leimanis Laurens, Mara PhD1,2; Snyder, Kristen MD3; Davis, Alan T. PhD4,5; Fitzgerald, Robert K. MD1,2; Hackbarth, Richard MD1,2; Rajasekaran, Surender MD, MPH1,2

Author Information
Pediatric Critical Care Medicine: October 2020 - Volume 21 - Issue 10 - p 859-868
doi: 10.1097/PCC.0000000000002375
  • Free
  • Infographics

Abstract

PICUs care for some of the most ill and vulnerable children in the United States. Similar to the general inpatient population, racial and socioeconomic disparities could exist in the PICU population resulting in differential survival outcomes. These disparities have been studied in non-ICU populations for decades, but only recently they have been considered in ICU patients (1,2).

The Centers for Disease Control and Prevention estimates that 15,000 children receive a diagnosis of cancer every year (3). A diagnosis of childhood cancer not only disrupts the well-being of the child, but places an enormous fiscal and psychosocial strain on the family. At admission to the ICU, pediatric patients with cancer experience higher mortality rates of 6.8% compared to 2.4% for all other diagnoses (4). The epidemiology, experience, and outcomes of these children and their families need to be better understood so as to inform their care and harness health promotion efforts.

Social factors such as race, ethnicity, and socioeconomic status (SES) are found to associate with disparities in cancer survival (5–7). National databases have long been used to study the epidemiology of cancer (8,9), including its distribution in children of differing race/ethnicity (5). The Virtual Pediatric Systems (VPS) database has been used previously to conduct studies on outcomes in the PICU (10). The VPS database allows centers to compare their clinical outcomes with others to improve quality, benchmark with peers, and establish best practices (10–12). Therefore, the database lends itself well to examine outcomes in the various categories of patients needing PICU care.

Developing an understanding of how race/ethnicity interacts with a complex disease such as cancer, although important, is fraught with difficulty. Cancer itself is a heterogeneous disease, encompassing differing diagnoses each with their own risk stratification, treatment algorithms, and prognoses. Thus, any effect attributed to race/ethnicity may stem from biologic determinants, social determinants, and potentially their interaction (13). Certain racial/ethnic groups have been found to have a disproportionate risk for mortality due to certain cancers, for example, Hispanic patients diagnosed with hematologic cancers (14–16). In spite of the limitations of using race, ethnicity, or culture as predictive or explanatory variables in healthcare research, there remains a need to examine their relationship to the outcomes of critically ill patients (13). Even if not causal, race/ethnicity will function as an indicator of socioeconomic or genetic vulnerability allowing for stratified risk reduction in patients with diseases such as cancer.

Studies performed on general ICU pediatric patients indicate that race/ethnicity and SES may impact their ability to survive critical illness (2). Given the well-documented racial/ethnic disparities in healthcare among pediatric patients with cancer (5–7), we hypothesized that racial/ethnic minority patients will have higher mortality compared with Caucasian patients when admitted to the PICU. To test this hypothesis, we used data from the VPS database to assess if survival differed among race/ethnic groups after controlling for severity of illness and other high-risk variables such as mechanical ventilation (MV), type of cancer, and stem cell transplant (SCT).

MATERIALS AND METHODS

Sites, Study Design, and Sample

This retrospective multicenter cohort analysis was conducted using the VPS database, the largest international registry of PICU patients. VPS contains data from more than 1,000,000 PICU admissions originating from 135 participating centers. Data for patients with oncologic diagnoses at admission from January 2009 to November 2018 were extracted. For consistency, we analyzed data from level 1 U.S. PICUs with 24/7 in-house attending coverage, an accredited PICU fellowship training program with unit sizes of more than 15 beds, and a minimum of 600 admissions per year. A waiver of consent and Health Insurance Portability and Accountability Act authorization were granted by our local Institutional Review Board at Spectrum Health. All data received from VPS were de-identified.

Inclusion Criteria

The study included admissions from birth to greater than 18 years old if they required a PICU admission, and their cancer status was described as an active or significant ongoing problem. PICUs in some centers will care for patients more than 18 years old if they are still receiving their care from Pediatric Oncologists. We excluded perioperative admissions that had less than 24 hours of PICU admission. Additional codes generated were related to cancer type, mass or tumor location, and history of chemotherapy or radiation. De-identified data variables included age, sex, race/ethnicity, geographic region as identified by the Census Bureau of the United States (South, West, Midwest, Northeast), primary and secondary diagnoses, reason for admission, discharge diagnosis, Pediatric Risk for Mortality (PRISM) III score, cancer type (solid tumor, brain tumor, and hematologic malignancy), SCT status, interventions and procedures, mortality, disposition (general care/step down vs home), limitation or withdrawal of care, and insurance type (Medicaid vs private insurance). The sample size and analyses include unique and multiple readmissions of the same patients. The VPS database does not mention the timing of the cancer diagnosis, only the actual diagnosis, and whether it is an active or significant ongoing problem.

Data Collection

The primary outcome variable was mortality. Independent variables included cancer type (solid tumor vs brain tumor vs hematologic), SCT status, age, race/ethnicity (Caucasian, African American, Hispanic, American Indian or Alaska Native, Asian/Indian/Pacific Islander, Native Hawaiian, or Other Pacific Islander and Mixed/Other). The VPS designations for race were extracted and used as a variable for race along with geographic region (South, West, Midwest, Northeast), respiratory support at admission (e.g., invasive MV), and payer type.

In classifying race/ethnicity, we chose the three main variables of interest as Caucasian, African American, and Hispanic. The other races were in disproportionately smaller numbers and, as such, were combined into a separate group called “AmerInd/Asian/Indian.” This category included the following races: American Indian, Alaska Native, Asian, Asian/Indian/Pacific Islander, Native Hawaiian, or Other Pacific Islander. Mixed/Other was a VPS-designated option and was used as a separate race variable. We classified patient encounters as Caucasian with the assumption that these patients were non-Hispanic Caucasian. We excluded from the study patient encounters if no entry was made regarding race (577 subjects; 4.7%).

The analysis was performed by using mixed-model analysis to control for each center’s unique effect, as well as controlling the effect of subjects with repeat admissions. Unique patient identifiers permitted detection of readmissions to the same PICU, but not to another PICU. Each admission is a discrete event that is initiated when patients are admitted to the PICU and ends when a discharge order is placed and thus independent of any distinct time interval. We controlled for severity of illness using PRISM III scores and SCT status. The VPS database makes no distinction on the type of SCT, whether auto or allogenic transplant or the timing of the transplant. The application of a procedure during the PICU admission was not used as an indicator of severity because we felt that it was more important to focus on the pathology rather than on the procedure. For example, shock/sepsis was considered as being more important than a central venous line (CVL) placement, particularly as children with oncologic disease will often present to the PICU already having a Broviac or a port and thus not always need an additional CVL for resuscitation, placed by the intensivist.

Analysis

Variables were selected for clinical utility from what is available within the VPS database and univariate analysis done to test for fidelity. Summary statistics were calculated. Quantitative data are expressed as the median and the interquartile range (25–75th percentile), whereas nominal data are expressed as a percentage. Quantitative data in the analyses were first transformed due to their non-normal distributions. The inverse hyperbolic sine transformation was used for MV days and PRISM scores, whereas the PICU length of stay (LOS) was log transformed. Comparisons between the groups’ admission demographics were performed using mixed-effects regression. One set of equations analyzed a transformed quantitative variable (PRISM III score, MV days, or PICU LOS) as the dependent variable and race as the independent variable, whereas the other equations used binomial variables as the dependent variables. All equations included the subject as a random effect because some subjects had multiple readmissions. Additional mixed-effects logistic regression analyses were performed. One set examined the dependent variable mortality, and the independent variables race/ethnicity, readmissions, transformed PRISM III score, transformed PICU LOS, transformed MV days, SCT status, sex (reference: female), hematologic cancer, solid tumor (not brain), and brain tumor in specific subgroup of patients (e.g., intubated subjects, subjects with acute lymphoblastic leukemia [ALL]). The independent variables were chosen based on the desire to have the most clinically appropriate predictors of mortality included in the model. The second set of models evaluated mortality as the dependent variable, with the same independent variables described previously. This was run for all subjects, as well as for subsets from each of the four geographic regions (South, West, Midwest, Northeast). An additional model included payer status as a surrogate for SES (Medicaid, yes vs no) as an independent variable. The random effects for the models included the PICU site and subject. Mixed-effects regression is a validated strategy for evaluating clustered data when individuals in one cluster (e.g., the PICU site, subjects with multiple readmissions) tend to be more similar to one another than to other individuals (17). Data were available for all study variables, with the exception of payer status, which was missing for 57% (13,212/23,128) of the admissions. The single analysis involving the use of the variable payer status was performed with the 9,916 subjects with complete data. No multiple imputation technique was used for this analysis. Zip codes could have provided an additional indicator of SES but were not available for our analysis. Statistical significance was determined at a p value of less than 0.05. Analyses were performed using Stata v. 15.1 (StataCorp, College Station, TX).

RESULTS

Demographic and Clinical Data

Oncology patients made up 23,128 admissions out of a total of 775,426 PICU admissions (3.0%). There were 12,232 unique patients, with 10,896 admissions having at least one readmission (range, 1–118). Demographics for the total sample are presented in Table 1. Overall, there were slightly more male admissions (54.9%). The admissions were mostly Caucasian (n = 11,897; 51.4%), with Hispanic (n = 5,391; 23.3%) and African American (n = 2,995; 13.0%) as the second and third largest race/ethnic groups, respectively. The African American, Hispanic, and Amerind/Asian/Indian groups had higher rates of readmissions compared with Caucasians. In terms of geographic representation, the majority of admissions reported were from the South (n = 8,322; 36.0%), followed by the West (n = 7,900; 34.2%), Midwest (n = 5,297; 22.9%), and the Northeast (n = 1,609; 7.0%). Caucasians made up the majority of admissions overall in every region except for the West, where the Hispanic ethnic group (38.9%) made up most of the admissions. There were regional differences; in the South, Hispanics had the second-highest admission rate (22.3%), whereas African Americans were slightly less (20%). PRISM III scores were significantly different between races with Hispanic patients having significantly higher scores than Caucasians (Table 1). All race/ethnicity groups had comparable rates of limitation of care and withdrawal of support.

TABLE 1. - Patient Admission Demographics of Total PICU Admissions for 2011–2018 (n = 23,128)
Variable Caucasian African American Hispanic Amerind/Asian/Indiana Others pb
Proportion of all admissions, n (%) 11,897 (51.4) 2,995 (13.0) 5,391 (23.3) 1,500 (6.5) 1,345 (5.8)
Age, n (%)
 Birth to < 2 yr 1,457/11,897 (12.3) 351/2,995 (11.7) 529/5,391 (9.8) 181/1,500 (12.1) 169/1,345 (12.6) < 0.001
 2 to < 6 yr 2,975/11,897 (25.0) 743/2,995 (24.8) 1,258/5,391 (23.3) 496/1,500 (33.1) 355/1,345 (26.4) 0.055
 6–18 yr 6,420/11,897 (54.0) 1,656/2,995 (55.3) 3,081/5,391 (57.2) 722/1,500 (48.1) 735/1,345 (54.7) < 0.001
 ≥ 18 yr 1,045/11,897 (8.8) 245/2,995 (8.2) 523/5,391 (9.7) 101/1,500 (6.7) 86/1,345 (6.4) 0.057
% female, n (%) 5,251/11,897 (44.1) 1,358/2,995 (45.3) 2,434/5,391 (45.2) 775/1,500 (51.7) 605/1,345 (45.0) 0.248
Pediatric Risk for Mortality III,c median (IQR) 3 (0–9) 3 (0–9) 5 (0–10) 3 (0–8) 3 (0–9) < 0.001
Readmission, n (%) 5,409/11,897 (45.5) 1,490/2,995 (49.8) 2,575/5,391 (47.8) 833/1,500 (55.5) 589/1,345 (43.8) < 0.001
PICU length of stay,c median (IQR) 2.6 (1.1–5.3) 2.7 (1.2–5.9) 2.8 (1.2–6.0) 2.0 (0.9–5.0) 2.6 (1.1–5.8) < 0.001
Mechanical ventilation days,c median (IQR) 0 (0–0) 0 (0–0.03) 0 (0–0) 0 (0–0) 0 (0–0) < 0.001
Regions, n (%)
 West (n = 7,900) 2,883/7,900 (36.5) 420/7,900 (5.3) 3,076/7,900 (38.9) 995/7,900 (12.6) 526/7,900 (6.7) < 0.001
 South (n = 8,322) 4,102/8,322 (49.2) 1,667/8,322 (20.0) 1,854/8,322 (22.3) 310/8,322 (3.7) 389/8,322 (4.7) < 0.001
 Midwest (n = 5,297) 3,913/5,297 (73.9) 702/5,297 (13.3) 263/5,297 (5.0) 135/5,297 (2.5) 284/5,297 (5.4) < 0.001
 Northeast (n = 1,609) 999/1,609 (62.1) 206/1,609 (12.8) 198/1,609 (12.3) 60/1,609 (3.7) 146/1,609 (9.1) 0.017
Patient admission disposition, n (%)
 General care/step down 9,093/11,897 (76.4) 2,302/2,995 (76.9) 4,204/5,391 (78.0) 980/1,500 (65.3) 1,025/1,345 (76.2) 0.197
 Home 1,560/11,897 (13.1) 330/2,995 (11.0) 536/5,391 (9.9) 402/1,500 (26.8%) 159/1,345 (11.8%) < 0.001
 Death 727/11,897 (6.1) 245/2,995 (8.2) 428/5,391 (7.9) 69/1,500 (4.6) 88/1,345 (6.5) < 0.001
 Other 517/11,897 (4.4) 118/2,995 (3.9) 223/5,391 (4.1) 49/1,500 (3.3) 73/1,345 (5.4) 0.107
Limitations of care, n (%) 605/10,636 (5.7) 180/2,778 (6.5) 335/5,055 (6.6) 72/1,426 (5.1) 78/1,136 (6.4) 0.130
Withdrawal of care, n (%) 439/10,371 (4.2) 104/2,666 (3.9) 184/4,959 (3.7) 36/1,332 (2.7) 39/1,109 (3.5) 0.085
aAmerican Indian or Alaska Native, Asian, Asian/Indian/Pacific Islander, Native Hawaiian, or Other Pacific Islander.
bA mixed-effects logistic regression model was run for each variable listed in the first column: this variable was the dependent variable, race/ethnicity was the indepepndent variable (reference group: Caucasian), and the random effect was the subject; the only exceptions were the dependent variables > 18 yr and Northeast, for which the models did not converge; in these two cases, logistic regression was performed, with clustering on the subjects to control for nonindependence of the subjects (due to multiple readmissions for some subjects).
cData are expressed as the median, with the interquartile range (25–75th percentile) in parentheses.

Oncologic Diagnosis in the PICU

PICU oncologic admissions were subdivided into three major groups; solid, brain, and hematologic malignancy. The majority of PICU admissions had a diagnosis of a solid tumor (55.9%; 12,920/23,128). In studying hematologic malignancies 46.6% (10,780/23,128), ALL was the most common (6,170/10,780; 57.2%), followed by acute myeloid leukemia (AML) (3,187/10,780; 29.6%). Brain neoplasms constituted 11.7% of admissions (2,700/23,128).

Secondary malignancies (n = 1,449; 6.3%) and oncologic admissions following SCT (n = 1,159; 5.0%) made up a lesser proportion of the admissions. The majority of Caucasians and African Americans admissions in the PICU, as well as those in the Mixed/Other group, had solid tumors, whereas the majority of Hispanic and the Amerind/Asian/Indian admissions had hematologic malignancies. Over 35% of Hispanic PICU subjects had ALL when compared with only 24.9% of Caucasians, 17.5% of African Americans, and 27.8% of the Amerind/Asian/Indians. There were no statistically significant differences in SCT rates among the various races.

Critical Illness and PICU Procedures

Pulmonary disease made up 31.5% (n = 7,291) of admissions to the PICU. The Amerind/Asian/Indian patients (496/1,500; 33.1%) had the highest percentage of pulmonary disease, compared with the other groups. The next largest diagnosis was shock/sepsis (n = 4,535; 19.6%). A higher number of Hispanic patients than any other race/ethnicity group required admission for shock/sepsis (n = 1,333/5,391; 24.7%). Acute kidney injury was documented in 7.5% (n = 1,746) of admissions, with all races demonstrating a similar distribution.

We analyzed PICU procedures done during their admission. Although each individual patient may have had multiple instances of a specific procedure, the procedure was only counted once for each patient during their PICU stay. With the exception of MV, there was no statistical difference in the rates of procedures because they were found to be were equally applied among all the races and ethnic groups. The level of acuity in patients with oncologic diagnosis is high with an increased need for life support devices as one in four of them needed one procedure or the other, which included endotracheal intubation (n = 5,515; 23.8%), arterial catheter (n = 4,849; 21.0%), and hemodialysis/plasmapheresis catheters (n = 1,414; 6.1%). A smaller fraction required additional supports such as high-frequency oscillatory ventilation (n = 426; 1.8%) or extracorporeal membrane oxygenation (n = 93; 0.4%).

Mortality in the PICU

The study found that 1,610 (7.0%) of the 23,128 oncology admissions resulted in mortality. Table 2 shows mortality rates between racial/ethnic groups in specific subsamples. MV was more likely to be associated with mortality in Hispanics compared with Caucasians. Likewise, Hispanics had significantly higher odds ratios (ORs) for mortality (OR, 1.39; 95% CI, 1.13–1.70) when they had pulmonary disease compared with Caucasians. African Americans with shock/sepsis had higher mortality compared with Caucasians (OR, 1.56; 95% CI, 1.11–2.20).

As shown in Table 2 and Figure 1, African Americans and Hispanics with hematologic malignancies suffered higher mortality compared with Caucasians (OR, 1.49; 95% CI, 1.14–1.95 and OR, 1.29; 95% CI, 1.04–1.61, respectively). Significantly more African American and Hispanic patients had mortality from solid tumors compared with Caucasian patients (OR, 1.35; 95% CI, 1.05–1.72 and OR, 1.32; 95% CI, 1.04–1.67, respectively). Table 2 also shows the mortality related to specific tumors by each racial/ethnic classification which were listed in the VPS database.

TABLE 2. - Mortality in the PICU by Racial/Ethnic Groups Related to Clinical Diagnoses, Interventions, and Tumor Types
Mortality Rate if They Have the Following: Caucasian, n (%) African American, n (%) Hispanic, n (%) Amerind/Asian/Indian,an (%) Others, n (%) pb
Infections 366/3,403 (10.8) 102/828 (12.3) 224/1,848 (12.1) 45/367 (12.3) 39/378 (10.3) < 0.001
Mechanical ventilation 615/2,786 (22.1) 201/812 (24.8)* 324/1,202 (27.0)* 59/394 (15.0) 73/321 (22.7) < 0.001
Pulmonary disease 572/3,663 (15.6) 169/974 (17.4) 344/1,767 (19.5)** 56/496 (11.3) 63/391 (16.1) < 0.001
Acute kidney injuryc 227/859 (26.4) 66/238 (27.7) 134/474 (28.3) 26/92 (28.3) 20/83 (24.1) < 0.001
Sepsis with or without shock, septic shock/etc 237/2,201 (10.8) 80/490 (16.3)* 170/1,333 (12.8) 33/255 (12.9) 24/256 (9.4) < 0.001
High-frequency oscillatory ventilation 141/226 (62.4) 34/55 (61.8) 65/105 (61.9) 11/17 (64.7) 10/23 (43.5) 0.772
Extracorporeal membrane oxygenation 33/51 (64.7) 8/18 (44.4) 11/20 (55.0) 1/1 (100) 2/3 (66.7) 0.361d
Dialysis catheter 170/705 (24.1) 46/169 (27.2) 97/394 (24.6) 19/82 (23.2) 19/64 (29.7) < 0.001
Type of tumor
 Hematologicc 421/5,276 (8.0) 127/1,136 (11.2)*** 279/2,935 (9.5)* 46/846 (5.4) 48/587 (8.2) < 0.001
 Solidc,e 365/6,908 (5.3) 140/1,925 (7.3)* 181/2,604 (7.0)* 33/699 (4.7) 46/784 (5.9) < 0.001
 Brain tumorsc 58/1,450 (4.0) 23/341 (6.7) 43/570 (7.5)* 11/177 (6.2) 9/162 (5.6) < 0.001
Acute lymphoid leukemia/etcc 215/2,967 (7.3) 49/525 (9.3) 150/1,901 (7.9) 24/417 (5.8) 29/360 (8.1) < 0.001
Acute myeloid leukemia/etcc 170/1,551 (11.0) 61/440 (13.9) 110/779 (14.1)*** 22/239 (9.2) 16/178 (9.0) < 0.001
Hodgkin’s/etcc 44/679 (6.5) 17/183 (9.3) 25/271 (9.2)* 1/78 (1.3) 5/47 (10.6) < 0.001
Nephroblastoma/etcc 45/643 (7.0) 11/206 (5.3) 7/159 (4.4) 3/57 (5.3) 3/69 (4.4) 0.018
Secondary malignancies 80/754 (10.6) 31/190 (16.3) 31/311 (10.0) 8/98 (8.2) 9/96 (9.4) < 0.001
Patients with SCTf 92/670 (13.7) 14/102 (13.7) 36/237 (15.2) 11/72 (15.3) 13/78 (16.7) < 0.001
Patients with SCT and hematologic tumorc,f 77/377 (20.4) 10/50 (20.0) 31/172 (18.0) 10/57 (17.5) 11/49 (22.5) < 0.001
SCT = stem cell transplant.
aAmerican Indian or Alaska Native, Asian, Asian/Indian/Pacific Islander, Native Hawaiian or Other Pacific Islander
bp value is for mixed-model analysis, in which mortality was the dependent variable, race/ethnicity (reference variable: Caucasian), inverse hyperbolic sine-transformed Pediatric Risk for Mortality III score, inverse hyperbolic sine-transformed mechanical ventilation days, natural log-transformed PICU length of stay, SCT status, sex (reference variable: female), and cancer type (heme, solid not brain, brain tumor; reference variable: heme) were the independent variables, and PICU facility and subjects were the random variables; reported higher mortality when compared with Caucasian: *p < 0.05; **p < 0.01; ***p < 0.005.
cCancer type not included in the mixed-model analysis.
dAmerind/Asian/Indian exhibited collinearity and was dropped from the model.
eExcluding brain tumors.
fSCT not included in the mixed-model analysis.

Figure 1.
Figure 1.:
PICU cancer deaths described by cancer type and race/ethnicity. Analyses were performed according to cancer type and performed using mixed-model analysis. Mortality was the dependent variable; race/ethnicity (reference variable: Caucasian), inverse hyperbolic sine-transformed Pediatric Risk for Mortality III score, inverse hyperbolic sine-transformed mechanical ventilation days, natural log-transformed PICU length of stay, stem cell transplant status, sex (reference variable: female), hematologic cancer, solid tumor (not brain), and brain tumor were the independent variables; and PICU facility and subjects were the random variables; reported higher mortality when compared with Caucasian: *p < 0.05; **p < 0.005; Amerind/Asian/Indian: American Indian or Alaska Native, Asian, Asian/Indian/Pacific Islander, Native Hawaiian, or Other Pacific Islander.

A multivariate analysis including all of the subjects in a mixed-effects logistic regression model was used to look at predictors of mortality (Table 3). We controlled for PRISM III severity of illness, SCT status, solid tumor (not brain), brain tumor, hematologic tumor, days of MV, PICU LOS, readmissions, and sex. Hispanics and African Americans had significantly higher odds of mortality (ORs of 1.24 and 1.37, respectively), compared with Caucasians (Table 3). The PRISM score was a significant predictor of mortality. A 10% increase in the PRISM score was associated with a 7.1% increase in the odds of death. The PICU LOS was significantly and inversely related to the risk of mortality. We interpreted this as indicating that the patients who did not survive had shorter ICU lengths of stay than those who survived. In addition, subjects with longer days of MV were significantly less likely to survive.

TABLE 3. - Mixed-Effects Logistic Regression Analysis of Predictors of Mortality in the PICUa
Variable OR (95% CI) P
Race/ethnicity
 Amerind/Asian/Indianb 0.86 (0.64–1.16) 0.321
 African American 1.37 (1.14–1.66) 0.001
 Hispanic 1.24 (1.05–1.47) 0.011
 Others 1.09 (0.83–1.44) 0.519
Pediatric Risk for Mortality III 1.97 (1.86–2.10) < 0.001
Stem cell transplant 1.78 (1.41–2.25) < 0.001
Hematologic cancer 1.97 (1.43–2.71) < 0.001
Solid tumor (not brain) 1.89 (1.37– 2.59) 0.001
Brain tumor 1.97 (1.35–2.88) < 0.001
Mech ventilation days 2.33 (2.16–2.52) < 0.001
Sex 1.01 (0.89–1.14) 0.893
Readmission 1.63 (1.41–1.87) < 0.001
PICU length of stay (d) 0.77 (0.74–0.85) < 0.001
OR = odds ratio.
aMixed-effects logistic regression model, where mortality was the dependent variable; race/ethnicity (reference: Caucasian), inverse hyperbolic sine-transformed Pediatric Risk for Mortality III score, inverse hyperbolic sine-transformed mechanical ventilation days, natural log-transformed PICU length of stay, stem cell transplant status, sex (reference: female), hematologic cancer, solid tumor (not brain), and brain tumor were the independent variables; and PICU facility and subjects were the random variables.
bAmerican Indian or Alaska Native, Asian, Asian/Indian/Pacific Islander, Native Hawaiian, or Other Pacific Islander.

We excluded payer status as an independent variable because it was unavailable for 13,212 subjects (57%). However, we were interested in analyzing the effect of payer status in the model looking at predictors of mortality, so we ran an additional multivariate analysis which included payer status (in addition to the other variables described for the analysis in Table 3), using the data available for the 9,916 subjects. In this separate analysis, payer status by itself was not predictive of mortality (OR, 0.95; 95% CI, 0.79–1.15).

Regional Analysis of Mortality

Admissions were examined to determine survival rates for oncology patients in the PICU at a regional level. The region where the PICU is located did relate to mortality and outcome. Using the same independent variables described previously, a subanalysis was performed individually for each region (Table 4). In the South, African American admissions were significantly more likely to result in mortality compared with Caucasians (OR, 1.46; 95% CI, 1.16–1.85). In the West, Hispanic admissions were significantly more likely to die than Caucasians (OR, 1.43; 95% CI, 1.08–1.90). There were no significant differences between any of the racial ethnic groups and Caucasians in either the Midwest or lower reporting Northeast.

TABLE 4. - Logistic Regression Analysis of Predictors of Mortality in the PICU by Regiona
South West Midwest Northeast
Variable OR (95% CI) P OR (95% CI) p OR (95% CI) p OR (95% CI) p
Race/ethnicity
 AmerInd/Asian/Indianb 0.81 (0.48–1.37) 0.431 1.08 (0.70–1.68) 0.719 0.58 (0.23–1.48) 0.257 0.62 (0.07– 5.68) 0.674
 African American 1.46 (1.16–1.85) 0.002 0.99 (0.53–1.82) 0.965 1.12 (0.76–1.66) 0.559 1.36 (0.63–2.92) 0.432
 Hispanic 1.18 (0.91–1.52) 0.211 1.43 (1.08–1.90) 0.014 0.94 (0.49–1.82) 0.859 1.96 (0.92–4.18) 0.083
 Others 1.29 (0.84–2.00) 0.245 1.19 (0.73–1.94) 0.475 0.87 (0.46–1.65) 0.672 0.92 (0.34–2.49) 0.867
Pediatric Risk for Mortality III 1.64 (1.51–1.79) < 0.001 2.23 (1.97–2.53) <0.001 2.53 (2.15–2.99) < 0.001 2.53 (1.83–3.48) <0.001
Stem cell transplant 1.92 (1.35–2.75) <0.001 1.74 (1.17–2.57) 0.006 1.64 (0.95–2.86) 0.078 2.26 (0.55–9.28) 0.259
Hematologic cancer 1.66 (0.98–2.81) 0.059 1.76 (1.01–3.06) 0.046 3.69 (1.84–7.40) < 0.001 1.28 (0.25–6.70) 0.765
Solid tumor (not brain) 1.38 (0.82–2.31) 0.224 2.00 (1.15–3.43) 0.036 4.11 (1.85–7.47) < 0.001 1.86 (0.35–9.77) 0.463
Brain tumor 1.52 (0.82–2.31) 0.170 1.98 (1.04–3.82) 0.014 3.72 (1.68–10.05) 0.002 0.51 (0.03–8.77) 0.644
Mech ventilation days 2.41 (2.15–2.70) < 0.001 2.39 (2.04–2.80) <0.001 2.55 (2.10 –3.11) < 0.001 1.54 (1.16–2.05) 0.003
Sex 0.94 (0.78–1.12) 0.475 0.96 (0.76–1.22) 0.753 1.33 (1.01–1.75) 0.044 0.86 (0.49–1.48) 0.580
Readmission 1.51 (1.24–1.85) < 0.001 1.38 (1.04–1.84) 0.025 2.11 (1.51–2.94) < 0.001 3.04 (1.37–6.76) 0.006
PICU length of stay 0.73 (0.65–0.82) < 0.001 0.83 (0.74–0.94) 0.003 0.72 (0.61–0.86) < 0.001 1.26 (0.96–1.67) 0.101
OR = odds ratio.
aLogistic regression models by region, where mortality was the dependent variablel race/ethnicity (reference: Caucasian), inverse hyperbolic sine-transformed Pediatric Risk for Mortality III score, inverse hyperbolic sine-transformed mechanical ventilation days, natural log-transformed PICU length of stay, SCT status, sex (reference: female), hematologic cancer, solid (not brain) tumor, and brain tumor were the independent variables; and PICU facility and subjects were the random variables.
bAmerican Indian or Alaska Native, Asian, Asian/Indian/Pacific Islander, Native Hawaiian, or Other Pacific Islander.

DISCUSSION

The course and eventual outcome of diseases, such as cancer, are determined by more than just the clinical characteristics of the malady itself. Individual genetic, social, and environmental influences should not be underestimated (5,18,19). All these factors link to a person’s race and ethnicity. The primary focus of this study was to assess whether a relationship exists between a patient’s race/ethnicity and survival in critically ill children with cancer in the ICU. The effect of race on outcome has been studied before in the general PICU population (10,20,21); however, this is the first report to focus specifically on oncology patients admitted to the PICU. It is also the first to note regional differences in the outcome of these patients.

This was a retrospective study using data acquired from the VPS database over a nearly 10-year period from 2009 to 2018. As in previous studies of cancer outcomes in the PICU, PRISM III scores along with relevant cancer characteristics were used to control for severity of illness (22–24). The primary risk factors for mortality were African American race, Hispanic ethnicity, SCT status, number of readmissions, and the need for MV. The major finding of our study is that African American and Hispanic patients with cancer are more likely to die than Caucasians during ICU admission even after accounting for severity. Although not statistically significant, there were higher rates of limitation of care documented among Hispanic and African American patients. Those rates could mean that compared with Caucasians, African Americans and Hispanics are choosing to limit care more often thus passing away in the PICU increasing their mortality rates. It is more likely that they were being confronted more often with advanced stage of critical illness or cancer leading to increased rates of limitations in care. Despite the Amerind/Asian/Indian group having the highest admissions rate for pulmonary disease, Hispanic patients had significantly higher PRISM scores at admission and were more likely to succumb to pulmonary disease. African Americans were more likely to die of shock/sepsis than other groups. Overall, Hispanic and African American children were 1.2 and 1.4 times more likely to die in the PICU than Caucasian children.

The higher rate of readmissions and greater mortality in Hispanic and African American children suggest a more complicated and protracted ICU course that could be from a combination of factors both genetic and socioeconomic. The relative contribution of genetic factors versus socioeconomic influences is difficult to determine and beyond the scope of this study. However, it is highly unlikely that genetic factors alone fully explain the differences noted in mortality. Our findings suggest that African American and Hispanic children with cancer are presenting to the PICU at a more advanced stage of illness. Clinical presentation at a higher severity of critical illness or more advanced stage of cancer makes treatment more challenging and the outcomes uncertain. Issues of access to care with subsequent delay in diagnosis could explain worse outcomes in these minority groups. Historically, the delivery of healthcare is not equal among all patients with nearly 21% of all children living in poverty (25,26). African American or Hispanic children are two to three times more likely to live in poverty than Caucasian children (27), and poverty affects access to healthcare (28). Families with limited resources tend to be higher utilizers of emergency department care as has been shown in Hispanic pediatric patients with cancer (29). This implies that these families may present for care later in the course of an illness. A 2013 study from the state of Georgia showed that when more restrictive immigration legislation is enacted, Hispanic children are disproportionally affected with their parents being less likely to seek medical attention (30). Barriers to accessing healthcare most certainly affect outcome. Delay in seeking medical attention results in increased mortality, secondary to infections, when Hispanic children with cancer are admitted to the PICU (31).

In addition to overall outcome disparities, we found that there are regional differences in mortality based on race/ethnicity in these children. In the South, African Americans were 1.5 times as likely to die as Caucasians. In the West, Hispanics had 1.4 times the mortality risk of Caucasians. The reasons for this are not completely clear, but it has been suggested that your zip code is more important than your genetic code (32). It is known that issues of social and environmental injustice disproportionately affect African Americans, Hispanics, and other minorities (33). In impoverished urban and rural regions, children are more likely to be exposed to environmental toxins and socioeconomic stressors (34,35). Furthermore, child poverty is clustered geographically, and race/ethnic composition associates most strongly with poverty in rural farming areas. The South in particular has a very high regional poverty rate of over 16% (26). The West and South have a higher proportion of Hispanic farm laborers compared with the Midwest (35). They and their families are exposed to the well-documented carcinogenic and epigenetic effects of pesticides, air pollutants, and other environmental toxins (19,36,37). These children may be exposed to herbicides and pesticides in a variety of ways, such as in utero exposure, direct contact in farm fields, and the closer proximity of their housing to sprayed fields leading to secondary exposure to chemicals on parental clothing or wash basins. Exposure burdens such as this could affect not only the frequency of cancer, but its prognosis as well.

As the VPS database is not designed to account for SES, the one variable that we were able to identify as a possible surrogate for SES was insurance payer type. Unfortunately, this was not a required data field and was missing in more than half of the patients. On analysis, payer type by itself was not predictive of mortality for any of the minorities with cancer in the PICU. Even if insurance type was a required data field, it is unlikely that it would be a reliable surrogate for SES. Studies from countries with universal public health insurance show that disparities in healthcare persist based on other socioeconomic indicators (38,39).

There are some other important limitations to the study. The VPS database does not contain prehospital data that might provide stronger linkage to inequity (e.g., delay in diagnosis, time from illness onset to presentation to care). In addition, the VPS database includes data only from participating centers and there are regional differences in participation rates. The Northeast is 2.5 times denser than the higher reporting South with an average of 345.5 people per square mile, yet it accounted for only 6.9% of all admissions in our data sample (40). This could lead to potential biases that may affect some of our observations especially ones regarding regional mortality rates. The included ICUs are on the VPS website https://www.myvps.org/vps-community. Even though limitations in care rates were similar between all races, no long-term post-hospital care data are available. To be clear, these data address mortality only in the PICU and there is the potential that certain race/ethnic groups differ in the rates of discharge from the general oncology floor directly to hospice care or pass away at home. The VPS database does not collect data outside the ICU experience of each patient, and this could indeed confound our findings. Cancer itself is not a single entity, but a group of diseases with individual courses, treatments, and outcomes that differ from one another. Finally, assignment of race and ethnicity in the database might not accurately reflect the family’s self-identified race or ethnicity as others have noted (41).

CONCLUSIONS

This study demonstrates that both African American and Hispanic children with cancer have higher adjusted PICU mortality than Caucasian children. The root causes for these differences in survival are difficult to establish and beyond the scope of this article with its retrospective design being a limitation. We will need well-designed prospective studies to confirm our findings and get to the exact causes for the differences we observed in our study. Increased participation with national databases such as VPS in all geographic regions of the United States and the inclusion of more socioeconomic data elements to ICU and oncology databases may move us closer to that goal. In the future, this could lead to the development of both short- and long-term treatment strategies that emphasize earlier detection of critical illness in minority patients with cancer.

ACKNOWLEDGMENTS

The VPS data was provided by Virtual Pediatric Systems, LLC. Their assistance was invaluable to the completion of our manuscript. No endorsement or editorial restriction of the interpretation of these data or opinions of the authors by VPS has been implied or stated. We acknowledge the assistance of Brittany Essenmacher in the preparation of this manuscript.

REFERENCES

1. Riley C, Maxwell A, Parsons A, et al. Disease prevention & health promotion: What’s critical care got to do with it?. Transl Pediatr. 2018. 7:262–266
2. Epstein D, Reibel M, Unger JB, et al. The effect of neighborhood and individual characteristics on pediatric critical illness. J Community Health. 2014. 39:753–759
3. Siegel DA, Li J, Henley SJ, et al. Geographic variation in pediatric cancer incidence - United States, 2003-2014. MMWR Morb Mortal Wkly Rep. 2018. 67:707–713
4. Zinter MS, DuBois SG, Spicer A, et al. Pediatric cancer type predicts infection rate, need for critical care intervention, and mortality in the pediatric intensive care unit. Intensive Care Med. 2014. 40:1536–1544
5. Kehm RD, Spector LG, Poynter JN, et al. Does socioeconomic status account for racial and ethnic disparities in childhood cancer survival?. Cancer. 2018. 124:4090–4097
6. Kahn JM, Keegan TH, Tao L, et al. Racial disparities in the survival of American children, adolescents, and young adults with acute lymphoblastic leukemia, acute myelogenous leukemia, and Hodgkin lymphoma. Cancer. 2016. 122:2723–2730
7. Winestone LE, Getz KD, Miller TP, et al. The role of acuity of illness at presentation in early mortality in black children with acute myeloid leukemia. Am J Hematol. 2017. 92:141–148
8. Grovas A, Fremgen A, Rauck A, et al. The National Cancer Data Base report on patterns of childhood cancers in the United States. Cancer. 1997. 80:2321–2332
9. Rice HE, Englum BR, Gulack BC, et al. Use of patient registries and administrative datasets for the study of pediatric cancer. Pediatr Blood Cancer. 2015. 62:1495–1500
10. Epstein D, Wong CF, Khemani RG, et al. Race/ethnicity is not associated with mortality in the PICU. Pediatrics. 2011. 127:e588–e597
11. Marcin JP, Pollack MM. Review of the acuity scoring systems for the pediatric intensive care unit and their use in quality improvement. J Intensive Care Med. 2007. 22:131–140
12. Pollack MM, Patel KM, Ruttimann UE. PRISM III: An updated Pediatric Risk of Mortality score. Crit Care Med. 1996. 24:743–752
13. Mays VM, Ponce NA, Washington DL, et al. Classification of race and ethnicity: Implications for public health. Annu Rev Public Health. 2003. 24:83–110
14. Kahn JM, Cole PD, Blonquist TM, et al. An investigation of toxicities and survival in Hispanic children and adolescents with ALL: Results from the Dana-Farber Cancer Institute ALL Consortium Protocol 05-001. Pediatr Blood Cancer. 2018. 65. doi:10.1002/pbc.26871
15. Li Y, Newton JG, Getz KD, et al. Comparable on-therapy mortality and supportive care requirements in Black and White patients following initial induction for pediatric acute myeloid leukemia. Pediatr Blood Cancer. 2019. 66:e27583
16. Chow EJ, Puumala SE, Mueller BA, et al. Childhood cancer in relation to parental race and ethnicity: A 5-state pooled analysis. Cancer. 2010. 116:3045–3053
17. Kirkwood BR, Sterne JAC. Goodgame F, Moore K. Summary of approaches to the analysis of clustered data. In Essential Medical Statistics. 2003Second EditionMalden, MA: Blackwell Publishing; 368–370
18. Gupta S, Wilejto M, Pole JD, et al. Low socioeconomic status is associated with worse survival in children with cancer: A systematic review. PLoS One. 2014. 9:e89482
19. Pogribny IP, Rusyn I. Environmental toxicants, epigenetics, and cancer. Adv Exp Med Biol. 2013. 754:215–232
20. Anand KJ, Sepanski RJ, Giles K, et al. Pediatric intensive care unit mortality among Latino children before and after a multilevel health care delivery intervention. JAMA Pediatr. 2015. 169:383–390
21. Turner D, Simpson P, Li SH, et al. Racial disparities in pediatric intensive care unit admissions. South Med J. 2011. 104:640–646
22. Thomas NJ, Spear D, Wasserman E, et al.; CALIPSO Study Investigators and the Pediatric Acute Lung Injury and Sepsis Investigators Network. CALIPSO: A randomized controlled trial of calfactant for acute lung injury in pediatric stem cell and oncology patients. Biol Blood Marrow Transplant. 2018. 24:2479–2486
23. Haase R, Lieser U, Kramm C, et al. Management of oncology patients admitted to the paediatric intensive care unit of a general children’s hospital–a single center analysis. Klin Padiatr. 2011. 223:142–146
24. Heying R, Schneider DT, Körholz D, et al. Efficacy and outcome of intensive care in pediatric oncologic patients. Crit Care Med. 2001. 29:2276–2280
25. Smedley BD, Stith AY, Nelson AR; Institute of Medicine (US). Committee on Understanding and Eliminating Racial and Ethnic Disparities in Health Care. In: Unequal Treatment: Confronting Racial and Ethnic Disparities in Health Care. 2003, Washington, DC: National Academies Press; pp 52–54
26. McCarty AT. Child poverty in the United States: A tale of devastation and the promise of hope. Sociol Compass. 2016. 10:623–639
27. The Annie E. Casey Foundation. 2019 KIDS COUNT Data Book: 2019 State Trends in Child-Well-being. 2019. Baltimore, MD: Annie E. Casey Foundation;
28. Agency for Healthcare Research & Quality. Access and Disparities in Access to Health Care. Available at: https://www.ahrq.gov/research/findings/nhqrdr/nhqdr15/access.html. Accessed June 30, 2019
29. Mueller EL, Hall M, Carroll AE, et al. Frequent emergency department utilizers among children with cancer. Pediatr Blood Cancer. 2016. 63:859–864
30. Beniflah JD, Little WK, Simon HK, et al. Effects of immigration enforcement legislation on Hispanic pediatric patient visits to the pediatric emergency department. Clin Pediatr (Phila). 2013. 52:1122–1126
31. Smits-Seemann RR, Pettit J, Li H, et al. Infection-related mortality in Hispanic and non-Hispanic Children with cancer. Pediatr Blood Cancer. 2017. 64. doi:10.1002/pbc.26502
32. Ritchie D. Our zip code may be more important than our genetic code: Social determinants of health, law and policy. R I Med J (2013). 2013. 96:14
33. Rauh VA, Landrigan PJ, Claudio L. Housing and health: Intersection of poverty and environmental exposures. Ann N Y Acad Sci. 2008. 1136:276–288
34. Jagai JS, Messer LC, Rappazzo KM, et al. County-level cumulative environmental quality associated with cancer incidence. Cancer. 2017. 123:2901–2908
35. Hernandez T, Gabbard S. Findings From the National Agricultural Works Survey (NAWS) 2015-2016. Available at: https://www.doleta.gov/naws/research/docs/NAWS_Research_Report_13.pdf. Accessed September 20, 2019
36. Collotta M, Bertazzi PA, Bollati V. Epigenetics and pesticides. Toxicology. 2013. 307:35–41
37. Alvarez CH. Norton-Smith KG: Latino Destinations and Environmental Inequality: Estimated Cancer Risk from Air Toxics in Latino Traditional and New Destinations. Socius. 2018; 4:1–11
38. Tolkkinen A, Madanat-Harjuoja L, Taskinen M, et al. Impact of parental socioeconomic factors on childhood cancer mortality:A population-based registry study. Acta Oncol. 2018. 57:1547–1555
39. Adam M, Rueegg CS, Schmidlin K, et al.; Swiss Paediatric Oncology Group; Swiss National Cohort Study. Socioeconomic disparities in childhood cancer survival in Switzerland. Int J Cancer. 2016. 138:2856–2866
40. U.S. Census Bureau (2018). American Community Survey 1-year estimates. Available at: http://censusreporter.org/profiles/02000US1-northeast-region/. Accessed November 21, 2019
41. Magaña López M, Bevans M, Wehrlen L, et al. Discrepancies in race and ethnicity documentation: A potential barrier in identifying racial and ethnic disparities. J Racial Ethn Health Disparities. 2016. 4:812–818
Keywords:

cancer; health inequality; pediatric intensive care unit; Pediatric Risk for Mortality III; pediatrics; race

Copyright © 2020 by the Society of Critical Care Medicine and the World Federation of Pediatric Intensive and Critical Care Societies