Health Disparities of Critically Ill Children in Poverty: A Nationwide Population-Based Study

Background: There is a lack of nationwide studies on health disparity of critically ill patients under the National Health Insurance (NHI) System. We aim to evaluate health disparities in intensive care unit (ICU) admission, outcomes, and readmission after ICU discharge in an impoverished pediatric population. Methods: We conducted a retrospective cohort study using a national database of claims submitted to the Korean NHI and Medical Aid Program (MAP). MAP provides support for the population whose household income is lower than 40% of the median Korean household income, and we de�ned poverty as being a MAP bene�ciary. Patients between 28 days and 18 years old who were admitted to the ICU between August 1, 2010 and September 30, 2013, were included. Demographic characteristics, procedures, admission rates, and clinical outcomes were compared between the poverty and reference groups. Logistic regression model used to analyze hospital mortality and readmission with adjustment for patient characteristics, hospital type, and management procedures. Results: Out of 17,893 patients, 1,153 (6.4%) patients were in poverty. The age-standardized ICU admission rate was higher in the poverty group (126.9 vs. 80.2 per 100,000 person-years). There were more deaths among impoverished patients who were admitted to the ICU (11.8 vs. 4.3 per 100,000 person-years). Patients in the poverty group had a similar risk of adjusted in-hospital mortality to those not in the poverty group (odds ratio: 1.15, con�dence interval [CI]: 0.84–1.55) but a higher readmission rate (hazard ratio 1.25, CI 1.09–1.42). Conclusion: Pediatric patients in poverty were more likely to die in association with ICU admission. A high ICU admission rate rather than the high in-hospital mortality rate may cause the disparity in deaths. Further policies and studies are required to improve the health status of pediatric patients in poverty to decrease ICU admission. Trial registration: retrospectively


Introduction
Health disparity is a preventable risk factor for morbidity and mortality [1].Health disparities might come from disparities in pre-hospital health status or in-hospital healthcare [2].Poor health status of impoverished people is associated with an increased incidence of severe sepsis [3].In-hospital mortality rate could be affected by inadequate resource uses related to the poverty of patients in the intensive care unit (ICU) [4,5].Children are more likely to stay in ICUs at the end of life than adults do [6].For treatment of critically ill children, health disparity related to poverty is important to consider at ICU admission.
Previous research on health disparities in impoverished people has focused on a few speci c critical conditions [7][8][9][10].Few studies have reported on health disparities of critically ill patients in a whole ICU system [4,11].In addition, there was a lack of studies conducted under the National Health Insurance (NHI) system in which barriers of medical access are low.A nationwide study on health disparity of impoverished children is required to identify the incidence of ICU admission, healthcare outcomes during ICU admission, and health status after ICU discharge in this population without selection bias of hospitals, where the ratio of impoverished people might differ.Therefore, we sought to evaluate health disparities among all Korean pediatric patients in the ICU.Our study aims were to evaluate differences in (1) the incidence of ICU admission, (2) ICU management outcomes, and (3) readmission rates after ICU discharge between those who were in poverty and those who were not.

Study population and design
We conducted a retrospective cohort study using the national database from Health Insurance Review and Assessment (HIRA), which is a central o ce in the Korean Ministry of Health.Korea has a singlepayer national health system.The Korean NHI covers approximately 97% of Koreans, and the remaining 3% of Koreans who cannot afford national insurance are covered by the Medical Aid Program (MAP) [12].
MAP is a public assistance program to protect socially disadvantaged or low-income family whose household income is lower than 40% of the median Korean household income (example: for 1-person family < 0.4 $1302 = $521 per month and for 4-person family < 0.4 $3519 = $1407 per month).In this study, we de ned the poverty group as the MAP bene ciaries.Claims submitted for reimbursement to Korean NHI and MAP are reviewed by the HIRA service.Therefore, the HIRA database includes virtually all ICU admissions (including private hospitals) of all patients < 18 years of age in Korea between August 1, 2009, and September 30, 2014.
From this cohort, we identi ed ICU patients between 28 days and 18 years of age who had rst ICU admission.We did not enroll patients admitted in the rst study year (from August 2009 to July 2010) to limit participants to only new admissions within at least 1 year.We also excluded hospital discharges during the last study year (from October 2013 to September 2014) since we could not determine if the patients were re-admitted within 1 year or not.Then we excluded patients who changed insurance status during ICU admission (n = 34).The nal sample size was 17,893 patients who admitted between August 1, 2010, and September 30, 2013 (Additional le 1: Supplementary Fig. 1).The study was reviewed by the Institutional Review Board of Samsung Medical Center (# 2019-07-114) and informed consent was waived because we only accessed de-identi ed administrative data that had been previously collected.Detailed information regarding this study can be found elsewhere [13].

Study variables
We identi ed ICU admissions using the claim codes that hospitals submit for cost claims for ICU management of in-hospital stays to HIRA (codes AJ100-AJ590900).We considered ICU stays during the same hospitalization as a single ICU admission.Similarly, we considered the claim codes of hospital stay separated by same or less than 1 day as the same hospital admission.Diagnostic codes are based on the Korean Classi cation of Diseases 6th edition, which is the modi ed version of the International Classi cation of Diseases 10th revision adapted for use in the Korean health system [14].
Study outcomes were in-hospital mortality, hospital readmission, readmission to the ICU and emergency room (ER) visit within three months after discharge.In-hospital mortality was de ned as death prior to discharge.To obtain readmission information, we linked study participants' personal identi cation numbers to the 2010-2014 inpatient databases.
Hospitals were classi ed according to their capacity based on the number of hospital beds and the number of specialties as de ned by the Korean Health Law [16].A hospital is de ned as a healthcare institution with more than 30 inpatient beds.A general hospital is a hospital with more than 100 beds and more than seven specialty departments including Internal Medicine, Surgery, Pediatrics, Obstetrics and Gynecology, Anesthesiology, Pathology, and Laboratory Medicine.A tertiary hospital is a general hospital with more than 20 specialty departments that serves as a teaching hospital for medical students and nurses.Data on ICU length of stay, hospital length of stay, and total cost of hospitalization were also collected.

Statistical analysis
We obtained population estimates for each year of age, sex, and calendar year from the NHI Service based on our de nition of poverty.All analyses were conducted separately for the poverty and nonpoverty groups.We calculated standardized rates of ICU admission and mortality by the direct method [17] using the Korean standard population from 2010 to 2013 [18].Mean and standard deviation or median and interquartile range were used to describe the distribution of continuous variables.We used the chi-square test and Student's t-tests to compare categorical and continuous variables, respectively.
We used multivariable logistic regression analysis to compare the risk of in-hospital mortality between groups.Odds ratios with 95% con dence intervals (CI) were estimated using the model.We adjusted for age, gender, primary diagnosis, vasopressor, extracorporeal membrane oxygenation and mechanical ventilation.Since patients could be clustered by hospital, we used hospital as a strati cation factor in the logistic regression model.We estimated hazard ratios (HRs) with 95% CI for the cumulative incidence of readmission, readmission to the ICU, and ER visits associated with poverty after adjusting for other risk factors.We conducted subgroup analysis according to use of mechanical ventilation to evaluate the poverty effect in subgroups with different mortality rates.In the subgroup analysis, the same logistic regression model was used except adjustment of mechanical ventilation.
We considered a p-value < 0.05 to be signi cant.Statistical analyses were performed using SAS® Visual Analytics (SAS Institute Inc., USA) and STATA version 14 (StataCorp LP, College Station, TX, USA).

Results
The mean (standard deviation) age of study patients was 7.3 (6.1) years, and 57.9% were male.Among 17,893 patients, 1,153 patients (6.4%) were in poverty (Table 1).From August 2010 to September 2013, the age-standardized ICU admission rate was 82.4 per 100,000 person-years (80.2 per 100,000 personyears in non-poverty patients vs. 126.9per 100,000 person-years in poverty patients).Poverty patients had higher rates of ICU hospitalization at all ages, and rates were especially high among younger patients (Fig. 1).Compared to non-poverty patients, those who were in poverty were older and more likely to be male (57.7% vs. 59.9%)(Table 1).Patients in the poverty group were more likely to be admitted with injuries (22.9% vs. 17.8%),neoplasms (13.8% vs. 9.3%), neurologic diseases (10.7% vs. 9.1%), and infectious diseases (3.0% vs. 2.3%) than the non-poverty group.While 39.8% and 67.4% of ICU admissions occurred in tertiary hospitals among the poverty and non-poverty groups, respectively, and rates of mechanical ventilation and vasopressor use were higher in the non-poverty group than the poverty group (Table 1).The overall in-hospital mortality was 5.1% (6.0% in poverty patients and 5.1% in non-poverty patients; p = 0.223).The overall mortality of the subgroup with mechanical ventilation was 10.8%, while that of the subgroup without mechanical ventilation was 0.9%.There were 4.5 ICU deaths per 100,000 Koreans per year (4.3 deaths per 100,000 person-years in non-poverty patients vs. 11.8 deaths per 100,000 person-years in poverty patients; Fig. 1).Multivariable analysis showed that the two groups had a similar risk of in-hospital mortality (adjusted odds ratio, 1.15; 95% CI, 0.84-1.55;p = 0.38; Table 2).In the both subgroups with and without mechanical ventilation, the adjusted risk of in-hospital mortality was not statistically different between poverty and non-poverty group.Among patients who survived (n = 17,360), those in the poverty group were more likely to be re-admitted (adjusted HR 1.25; 95% CI 1.09-1.42)and visit the ER (adjusted HR 1.31; 95% CI 1.06-1.62)within three months after discharge than those in the non-poverty group were (Table 3).

Discussion
In this representative population-base cohort, we showed that the age-standardized population-based ICU admission rate of children in poverty (income of their household < 0.4 median household income of Korea) was 1.6 (126.9/80.2) times higher than that of children not in poverty.The age-standardized population-based mortality rate of children in poverty was 2.7 (11.8/4.3)times higher than that of children not in poverty.The adjusted in-hospital mortality of critically ill children admitted to the ICU was not different between the poverty and non-poverty groups.
The high age-standardized mortality in the impoverished population highlights a critical public health issue (Fig. 1).A child in an impoverished family has a high risk of death.Once children were admitted, the crude in-hospital mortality of children in poverty was not statistically different from that of the nonpoverty group (6.0% vs. 5.1%, p = 0.223).In addition, the difference in mortality rates between the two groups was not signi cant when we adjusted for demographics, hospital factors, and management procedures.The result of adjusted analysis was similar in the high mortality subgroup (mechanical ventilation) and low mortality subgroup (non-mechanical ventilation).It may suggest that there is no disparity in healthcare outcomes in ICU according to poverty.However, the admission rate was high in the poverty group across all age groups.Therefore, the age-standardized mortality of the impoverished population is likely due to high admission rate of the group rather than from high in-hospital mortality rate.We could not identify the cause of high admission rate in the poverty group, but poor pre-hospital health status could be an explanation.In previous studies, the severity scores of patients at ICU admission were higher in uninsured patients [5,11], which implies worse health status of them at ICU admission.Previous studies conducted at select hospitals were not able to identify disparities in admission rate and population-based mortality because of selection bias [8,11,19].Utilization of highvolume or top-ranked (low mortality) hospitals differed according to poverty status [20,21], and data from these renowned hospitals could bias admission rates of low-income patients.In our study, we included all hospitals where pediatric patients are admitted to the ICU, except for rare possible administrative losses of insurance claims.
Disparity in resource use was suspected as one of the causes of disparity in healthcare outcomes [5].The low incidence of mechanical ventilation (35.6% vs. 43.1)and vasopressor use (12.3% vs. 15.8%) in the poverty group might raise suspicion of a passive attitude toward treating patients in poverty.However, the rates of mechanical ventilation and vasopressor uses were different by age group in our previous study [13].The age distribution of the poverty group was different from that of the overall population, possibly due to the different age distribution of parents under MAP from the general population.When we strati ed age into 3 groups (infant, children, adolescent), there were no statistically signi cant differences in the incidence of mechanical ventilation and vasopressors according to poverty status (Additional le 2: Table S1).In addition, the incidence of other resource-heavy procedures such as transplantation, extracorporeal membrane oxygenation, and hemodialysis was similar between the poverty and nonpoverty groups in an NHI system (Table 1) (Additional le 2: Table S1).These ndings are contrary to the reported socioeconomic disparities in transplantation [22][23][24].The difference might originate from the different health insurance systems.
Low socioeconomic status (SES) could affect pre-hospital health status in various ways.Di culties in access to care and preventive health services may contribute to the high severity of illness and organ failure on hospital presentation [2].Other factors such as inadequate health behaviors, lack of parental education, unhealthy environmental factors, and low vaccination rates may contribute to the worse health status of the low SES population [2,25].In our study, ER visits and readmission rates after ICU discharge were higher in patients in poverty.Previous studies reported that the increased rate of ER visits in low SES patients is associated with differences in health status rather than health behaviors [26].Low SES is also associated with a high readmission rate [27,28].We suspect that the high ER visit and readmission rate among patients in poverty imply poor health status.Policies to improve the health status of patients in poverty might be requiried to decrease population-based mortality associated with ICU admission.

Table 1
Characteristics of pediatric patients in intensive care units according to poverty status.

Table 2
In-hospital mortality of pediatric patients in intensive care units according to poverty status.