Logistic regression models showed that the odds of mortality decreased significantly for Latino children (odds ratio, 0.24; 95% CI, 0.06-0.88; P = .03) between the preintervention period (2007-2009) and postintervention period (2010-2012) but remained unchanged for the combined group of white and African American children (OR, 1.02; 95% CI, 0.73-1.43; P = .90).
eTable 1. Patient Characteristics: Caucasian Non-Latino
eTable 2. Patient Characteristics: African American
eTable 3. Patient Characteristics: Latino/Hispanic
Anand KJS, Sepanski RJ, Giles K, Shah SH, Juarez PD. Pediatric Intensive Care Unit Mortality Among Latino Children Before and After a Multilevel Health Care Delivery Intervention. JAMA Pediatr. 2015;169(4):383-390. doi:10.1001/jamapediatrics.2014.3789
Research on health equity has focused on documenting health care disparities or understanding factors leading to disparities, but limited efforts have focused on reducing health care disparities in children. Latino children have increased prevalence of acute and chronic conditions; they have limited access and other barriers to high-quality health care, including intensive care.
To determine whether pediatric intensive care unit mortality can be reduced by a multilevel health care delivery intervention.
Design, Setting, and Participants
Observational study of factors associated with pediatric intensive care unit mortality at a tertiary care metropolitan children’s hospital in Memphis, Tennessee. Participants were children younger than 18 years discharged from the pediatric intensive care unit during the 3-year preintervention period of 2007 to 2009 (n = 3891) and 3-year postintervention period of 2010 to 2012 (n = 4179).
Multilevel health care intervention to address the increased odds of mortality among Latino children.
Main Outcomes and Measures
The odds of mortality were analyzed over the 3-year preintervention period (2007-2009) using multivariable logistic regressions to control for age, sex, race/ethnicity, severity of illness, major diagnostic categories, diagnosed infections, and insurance status. Data from the postintervention period (2010-2012) were analyzed similarly to measure the effect of changes in health care delivery.
Unadjusted mortality rates for white, African American, and Latino children in 2007 to 2009 were 3.3%, 3.3%, and 8.6%, respectively. After controlling for covariates, no differences in the odds of mortality were observed between white children and African American children (odds ratio [OR], 1.0; 95% CI, 0.6-1.7; P = .97), but Latino children had 3.7-fold (95% CI, 1.8-7.5; P < .001) higher odds of mortality. A multilevel and multidisciplinary intervention was launched to address these differences. In the postintervention period, unadjusted mortality rates for white, African American, and Latino children were 3.6%, 3.2%, and 4.0%, respectively, with no differences observed after adjustment for covariates (OR, 0.7; 95% CI, 0.2-2.1; P = .49). The odds of mortality decreased between the preintervention period and postintervention period for Latino children (OR, 0.24; 95% CI, 0.06-0.88; P = .03) but remained unchanged for white and African American children (OR, 1.02; 95% CI, 0.73-1.43; P = .90).
Conclusions and Relevance
Latino children had higher odds of mortality, even after controlling for age, sex, severity of illness, insurance status, and other covariates. These differences disappeared after culturally and linguistically sensitive interventions at multiple levels. Local multilevel interventions can reduce the effect of health care inequities on clinical outcomes, without requiring major changes in health care policy.
Previous studies reported higher mortality rates among uninsured children admitted to the pediatric intensive care unit (PICU), but no differences were observed by age, sex, or race/ethnicity.1 Similar results were found for various diagnostic conditions, including asthma,2 pediatric trauma,3 and others, providing some of the impetus for policy initiatives4,5 and pilot models6 that led to health care reform.7
Latino children are the largest and fastest growing racial/ethnic minority group in the United States, constituting 22% of the total population younger than 18 years.8 Between 2000 and 2010, the number of Latino children grew by 4.8 million (or 39%).9 Their access to primary health care and health outcomes remains below national standards, partly because of a lack of health insurance10- 13 or cultural, socioeconomic, and other variables.8 Factors placing toddlers at high risk of nutritional deficiency were more prevalent among Latinos than other groups.14 Latino children also showed greater prevalence of chronic developmental conditions, poor oral health, asthma, obesity, diabetes mellitus, parasitic infestations, tuberculosis, and environmental toxin exposures.8 Each of these conditions may increase their risks of developing critical illness. Other studies15- 18 found that Latino children were more likely to receive delayed emergency care, develop appendicitis or perforated or complex appendicitis, have less analgesic and longer length of stay for abdominal pain, and manifest increased mortality and morbidity from systemic lupus erythematosis.
Not enough is known about race/ethnicity disparities in PICU care or their effect on outcomes of critically ill children.19 In this observational study, we first examined factors associated with mortality among PICU discharges from a metropolitan children’s hospital while controlling for previously reported confounding factors, including age, sex, major diagnostic categories, diagnosed infections, insurance status, and risk of mortality. Second, we evaluated the effect of a multilevel health care intervention aimed at reducing PICU mortality among Latino children.
Clinical and demographic data were analyzed for all patients discharged from Le Bonheur Children’s Hospital PICU from January 1, 2007, to December 31, 2012. The primary data source was the Virtual Pediatric Intensive Care Unit (VPICU; VPS, LLC),20 a multi-institutional database maintained by VPS, LLC and used with permission. Secondary data sources included other hospital, clinical, and administrative databases and the electronic medical record (EMR) of individual patients. Data were collected from all PICU patients younger than 18 years. Approval was obtained from the institutional review board at the University of Tennessee Health Science Center for publishing these data analyses.
Our principal outcome measure was mortality at PICU discharge. Clinical factors included race/ethnicity (white, African American, or Latino) and age group. Age groups were neonate (0-4 weeks), infant (1-12 months), early childhood (1-6 years), late childhood (7-12 years), or adolescent (13-17 years). Other clinical factors included were sex, severity of illness (SOI), diagnosed infections, major diagnostic categories (respiratory, cardiovascular, injury or poisoning, infectious, neurologic, and all others), and insurance status (private, public, self-pay or uninsured, or other [a sponsoring agency, liability cases, or patients with missing insurance information]). Insurance status was recorded at hospital admission and was supplemented by information from the hospital’s administrative database at hospital discharge. If patients arrived without insurance but were enrolled in public insurance during their hospital stay, they were classified as uninsured (Table 1). Race/ethnicity and language preferences were self-identified by parents or guardians when registering their child’s hospital admission. Diagnosed infections were defined as International Classification of Diseases, Ninth Revision coded discharge diagnoses of bacterial, viral, fungal, or parasitic infections, excluding seasonal upper respiratory tract viral infections, such as colds or influenza, except when diagnosed with novel influenza A(H1N1) or when accompanied by pneumonia. Major diagnostic categories were determined by the PICU attending physician and were entered into the VPICU; if indeterminate, the diagnostic category was added manually (by K.J.S.A.) after reviewing the EMR.
Mortality rates were adjusted using the Pediatric Risk of Mortality III (PRISM III) scores,21 computed from clinical variables measured within 12 hours of PICU admission. The PRISM III is an international standard of adjustments for risk of mortality related to SOI at the time of PICU admission. A web-based version of the VPICU was implemented in July 2009, which included a recalibration of the PRISM III scores. To maintain constant SOI adjustments throughout this study, we used the original PRISM III algorithm21 to recalculate the risk of mortality from the PRISM III scores between July 1, 2009, and December 31, 2012.
In the 2007 to 2009 preintervention period, higher odds of mortality were noted for Latino children. This prompted a detailed examination of factors related to Latino patients (health literacy, socioeconomic, and insurance status), health care delivery (access, quality, and timeliness), and the social determinants of health.22 These results were discussed in the hospital’s and health care system’s ethics committees, the hospital’s senior leadership council, and the quality council and with the entire physician staff, including residents and fellows, soliciting their input and participation in potential solutions. We launched a multilevel intervention to address these differences23 and studied the effect of this intervention over a 3-year postintervention period (2010-2012). The intervention included the following: (1) education of health care professionals regarding culturally competent care; (2) recruitment efforts to increase the number of bilingual staff24,25; (3) availability of 24-hour interpreter services in the emergency department and PICU; (4) translation of consent forms and educational materials for patients and families; (5) culturally sensitive end-of-life care discussions, with participation of palliative care services; (6) outreach efforts and contacts with the Latino community to remove barriers to health care access; and (7) help from the city government and local health department for preventive services. Components 1 through 3 were implemented in late 2009 and early 2010, components 4 and 5 throughout 2010 and 2011, and components 6 and 7 in 2011. Regular updates were prompted by the VPICU data reports, discussed in multidisciplinary forums (eg, the PICU council), or communicated electronically.
Univariable logistic regressions were used to calculate unadjusted mortality odds ratios (ORs) for each of the clinical variables (age group, sex, race/ethnicity, SOI, major diagnostic categories, diagnosed infections, and insurance status). Multiple logistic regression models examined the association of mortality with race/ethnicity, while adjusting for all covariates that were significantly associated with mortality in the univariable logistic regressions. Orthogonal contrasts were used to compare the odds of mortality between white children and African American children and to compare the odds of mortality for the combined group (white and African American children) with those of Latino children.
Additional logistic regression analyses were performed to test for differences in the odds of mortality between the preintervention (2007-2009) and postintervention (2010-2012) years of the study. These analyses were conducted after SOI adjustments based on the PRISM III scores.
To allow meaningful comparison, only white children (3319 [41.1%]), African American children (4392 [54.4%]), and Latino children (359 [4.4%]) were included in these analyses. Children of self-reported Asian, Native American, unknown, or mixed race/ethnicity were excluded. Patient characteristics and risks of mortality were comparable between the 2 study periods (2007-2009 vs 2010-2012) (Table 1 and eTables 1, 2, and 3 in the Supplement).
Unadjusted mortality rates were 3.3% for both white children and African American children, with no differences between the 2 groups (Table 2). Therefore, these groups were combined for comparison with Latino children, who had 2.6-fold higher odds of mortality (P < .001). Compared with the reference groups for these variables, higher mortality rates also occurred among neonates (P < .001), infants (P = .04), certain major diagnostic categories (P ≤ .002), those with diagnosed infections (P = .001), self-pay or uninsured individuals (P = .02), or participants with other insurance (P = .002). The contributions of these factors were further explored in multiple logistic regression models.
After controlling for age, sex, insurance status, diagnosed infections, and SOI based on the PRISM III scores, no differences were observed between the odds of mortality for white children and African American children (Table 3). These groups were combined for comparison with Latino children. After controlling for the same variables, Latino children had 3.7-fold higher odds of mortality than white and African-American children (95% CI, 1.8-7.5; P < .001). Diagnosed infections (P < .001) and other insurance status (P = .02) remained in the model as risk factors for increased PICU mortality in children. When entered individually or collectively, major diagnostic categories did not remain in the logistic regression models for mortality and did not explain the higher odds of mortality in Latino children.
Latino children had a higher prevalence of diagnosed infections than white and African American children (38.9% vs 32.1%, P = .05), but no significant interactions occurred in terms of mortality between Latino race/ethnicity and diagnosed infections. No interactions occurred between Latino race/ethnicity and various major diagnostic categories or age groups as factors associated with mortality. Therefore, the increased odds of mortality occurring in Latino children were independent of their distribution into certain age, sex, major diagnostic, or insurance status groups.
Unadjusted mortality rates were 3.6% for white children, 3.2% for African American children, and 4.0% for Latino children, with no significant differences between groups (Table 2). Higher mortality rates occurred among neonates (P < .001), certain major diagnostic categories (P < .001), and uninsured patients (P = .003) compared with their corresponding reference groups.
After controlling for age, sex, insurance status, diagnosed infections, and risk of mortality based on the PRISM III scores, no significant differences were observed in the odds of mortality between white children vs African American children (OR, 0.8; P = .48) or between white plus African American children vs Latino children (OR, 0.7; P = .49). No other variables, including age, diagnosed infections, or insurance status had any significant effect on the odds of mortality (Table 3).
Unadjusted odds of mortality decreased between the 2 study periods for Latino children (OR, 0.44) but not for white and African American children (Figure). After adjusting for SOI based on the PRISM III scores, the odds of mortality between the preintervention period and postintervention period decreased significantly for Latino children (OR, 0.24; 95% CI, 0.06-0.88; P = .03) but were unchanged for all children (OR, 0.92; 95% CI, 0.67-1.26; P = .60) and for the combined group of white and African American children (OR, 1.02; 95% CI, 0.73-1.43; P = .90).
Causes of health disparities among racial/ethnic minorities are complex, remain persistent, and exist across the life span.26,27 The findings of previous studies1- 4,6,8,10- 19,26,27 suggest that health disparities are associated with genetic, lifestyle, environmental, sociocultural, and health care delivery factors. We found 3.7-fold (95% CI, 1.8-7.5) higher odds of mortality among Latino children than among white or African American children, after controlling for age, sex, major diagnostic categories, diagnosed infections, and insurance status. Because of its excellent calibration and discriminatory functions,28- 30 the PRISM III currently is the standard, internationally used measure for calculating risk of mortality in PICU patients. Multilevel health care delivery interventions were designed to address the increased odds of mortality in Latino children.23 These disparities disappeared after these multidisciplinary and multilevel interventions were implemented. Our results indicate that health disparities are not intractable and that interventions focused on the health care delivery system can have a significant effect.31
The Institute of Medicine’s report Unequal Treatment: Confronting Racial and Ethnic Disparities in Health Care found that “minorities tend to receive a lower quality of health care than non-minorities, even when access-related factors, such as patients’ insurance status and income, are controlled.”32(p1) The 2010 National Healthcare Disparities Report (by the US Department of Health and Human Services) stated that “health care quality and access are suboptimal, especially for minority and low-income groups.”33(Summary) To realize its vision of a “nation free of disparities in health and healthcare,” the 5-year strategic plan for the Department of Health and Human Services identified the following 3 goals: (1) to achieve health equity; (2) to ensure access to quality, culturally competent care for vulnerable populations; and (3) to improve data collection and measurement.34,35
Our initial findings were consistent with other pediatric data available in 2009. For example, among children with acute appendicitis, Latino and African American children faced greater delays to surgery, longer lengths of stay, and higher hospital charges compared with white children, even after adjusting for SOI.16 Poorer outcomes also occurred in infant mortality rates (independent of birth weight differences36 or birth defects37) or for chronic childhood conditions, including attention-deficit/hyperactivity disorder, type 1 diabetes mellitus, and acute leukemia.38
At that time, few studies had investigated health disparities in outcomes of critically ill patients.32 For example, among pediatric trauma patients higher mortality rates occurred in African Americans and Latinos39 and in those without insurance,3,39 even after controlling for the type and severity of injury. Erikson et al40 found that race/ethnicity was not associated with hospital mortality among critically ill adults, but they examined an older data set (2001-2004) and excluded all patients with trauma, burns, or coronary surgery. Moreover, their results were inconsistent with previous studies.41- 43 Similarly, Epstein et al44 found no association of race/ethnicity or insurance status with PICU mortality. However, in their data only 31 of 75 institutions had collected race/ethnicity information, leading to a systematic bias. They were also blinded to the data sources and unable to correct errors or cross-check the data from each institution. Conversely, Nembhard et al45 reported increased mortality among African American and Latino infants with congenital heart defects compared with white infants.
Multicenter studies tend to conceal the true effect of race/ethnicity on health outcomes because of differences in case mix and components of health care delivery, clustering of patient outcomes by hospital, potential misclassification of race/ethnicity or other factors, and averaging of disparate outcomes across hospitals and health care systems.1,40,44 Consequently, single-center studies may be more likely to reveal health disparities,46 but they also prompt a thorough examination of local factors that can mobilize the support from key stakeholders (the community, hospital staff and leadership, and county government) to look for local solutions. We used more recent data than other studies,40,44,45 and we cross-checked our data across multiple hospital databases and the EMR, allowing us to correct any existing errors and minimize missing values.
If Latinos are less likely to receive care at high-technology hospitals,47 can higher mortality be associated with their being uninsured or uninsurable? About 20% of Latino children lacked health insurance in 2008 compared with 7% of white children and 12% of African American children.48 A lack of insurance is not the only factor why Latino children have limited health care access.11,49 Our logistic regression models adjusted for insurance status but still found that Latino race/ethnicity was associated with higher odds of mortality. Sensitivity analyses were repeated after exclusion of all patients classified as self-pay or uninsured and showed the same results. Like other institutions, we made concerted efforts to enroll patients into insurance programs sponsored by the state or federal government, resulting in fewer self-pay or uninsured patients (2.4% in our study compared with 6%-16% in other studies3,44).
Latino children with severe asthma had 3.3-fold higher odds of intubation50 and were less likely to receive asthma care plans at hospital discharge,51 placing them at greater risk of recurrence and complications. Other financial, linguistic, or logistic barriers may place these children at risk for receiving suboptimal health care.52,53 Important details and nuances in a child’s clinical presentation are often lost in translation. An inadequate history may be compounded by the absence of medical records or by prior exposure to household remedies and folk healers.10,52,54 Biased attitudes among health care professionals can also have profound clinical consequences, with inequities in prescriptions, history taking, care quality, and diagnostic evaluations.54 In 2007 to 2009, with an ongoing highly publicized national debate on the health care costs of illegal Latino immigrants,55- 57 a subconscious bias by health care professionals cannot be excluded.58
When faced with a poor prognosis, Latino parents may have limited understanding of end-of-life issues or their therapeutic options.59 Therefore, health care professionals have greater responsibility to ensure that a critically ill Latino child’s parents or guardians clearly understand the issues and options related to end-of-life care. Cultural factors also have an important part in the family’s approach to end-of-life care and their interactions with the health care team. We made efforts to identify, define, and address these cultural factors with each patient’s family.
Our multilevel intervention was designed to include the following patient-centered improvements: (1) initiatives that were responsive to sociocultural patient characteristics, (2) initiatives that were designed to modify hospital responsiveness, and (3) initiatives that enhanced the overall health care delivery system. This multilevel intervention did not address factors previously correlated with health disparities (genetic, lifestyle, and environmental issues) because they were outside the hospital’s direct sphere of control. The 3 levels of interventions chosen for this study were previously associated with creating a culturally competent system of care.60,61
Patient-centered interventions included translation of consent forms and educational materials for patients and families and culturally sensitive end-of-life care discussions, with participation of palliative care services. These interventions were designed to improve patient knowledge and awareness of the patient’s health condition. Interventions to alter hospital processes included the education of health care professionals regarding culturally competent care, recruitment efforts to increase the number of bilingual staff, and 24-hour interpreter services in the emergency department and PICU. These were structural investments made by the hospital and PICU leadership. Interventions designed to improve the responsiveness of the overall health care delivery system to the needs of Latino families included increased outreach efforts and contacts with the Latino community to remove barriers to health care access and obtain help from the city government and local health department for preventive services. These interventions attempted to create a more coordinated health care delivery system. While this study was not designed to capture the independent effects of these changes, future studies could be designed to test the efficacy of individual components of a multilevel health care intervention.62
Our study has several limitations. Although no distinctions were made in previous studies,1,16,37- 39,44,45,63,64 Latinos represent a heterogeneous group originating from different countries, and their children are exposed to various environmental, socioeconomic, or cultural risk factors and genetic susceptibilities.65 Future studies should examine specific risk factors associated with this cultural diversity. Our study used a retrospective database (the VPICU), with consequent limitations in study design. However, we minimized missing data and cross-validated our data with multiple clinical databases maintained by the hospital and, if needed, the EMR of individual patients, thereby limiting the effect of potentially imperfect race/ethnicity data.66 With significant reductions in mortality among Latino children, fewer deaths occurred in this group during the postintervention period. Therefore, these data represent only preliminary evidence that multilevel health care interventions can alter clinical outcomes. They need to be replicated in other populations and clinical settings.
In summary, Latino race/ethnicity was associated with 3.7-fold higher odds of mortality in PICU patients compared with white and African American race/ethnicity in the preintervention period 2007 to 2009, after controlling for age, sex, SOI, major diagnostic categories, diagnosed infections, and insurance status. These differences were eliminated after a multilevel intervention to provide patient-centered culturally sensitive environments, modify hospital responsiveness, and improve the processes of health care delivery.23,65 Although no causal inferences are implied, mortality differences between Latino children and other children disappeared in the postintervention period (2010-2012). Routine assessments should examine the effect of race/ethnicity on clinical outcomes of PICU patients, classify Latino patients based on their country or origin, and ensure the delivery of linguistically supportive and culturally sensitive pediatric intensive care.
Accepted for Publication: December 28, 2014.
Corresponding Author: Kanwaljeet J. S. Anand, MBBS, DPhil, Department of Pediatrics, University of Tennessee Health Science Center and Le Bonheur Children’s Hospital, 50 N Dunlap St, Room 351R, Memphis, TN 38103 (email@example.com).
Published Online: February 23, 2015. doi:10.1001/jamapediatrics.2014.3789.
Author Contributions: Dr Anand and Mr Sepanski had full access to all the data in the study and take full responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Anand, Sepanski.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: Anand, Sepanski.
Critical revision of the manuscript for important intellectual content: Giles, Shah, Juarez.
Administrative, technical, or material support: Giles.
Conflict of Interest Disclosures: None reported.
Additional Contributions: Karen K. Brinkley, RN, BSN, and D. Charlene Summerall, RN, BSN (Le Bonheur Children’s Hospital) maintained the VPICU database (VPS, LLC). Charu Sharma, MS, CRME, MHA (Le Bonheur Children’s Hospital) performed the initial data analyses. Stephanie Storgion, MD (University of Tennessee Health Science Center) and William May, MD (Le Bonheur Children’s Hospital) reviewed initial versions of the manuscript and provided comments. David Bertoch, MBA (Child Health Corporation of America) gave permission to use the Pediatric Health Information System data in our analyses. Christine Gall, MBA (VPS, LLC) gave permission to use the VPICU database for these analyses. No endorsement or editorial restriction on the interpretation of these data or the opinions of the authors was implied or stated by the organizations listed in this paragraph. No compensation was provided to the contributors.