Data for this study were obtained through the North Carolina Congenital Heart Defects (NCCHD) Surveillance Network. This database is composed of 50,429 patients with CHD, as identified by International Classification of Diseases, Ninth Revision (ICD-9), code recorded in the electronic medical record for a healthcare encounter between 2008 and 2013 at one of five comprehensive or affiliated satellite CHD centers (Duke, University of North Carolina, East Carolina University, Wake Forest, and Atrium Health). For the analysis reported here, infants undergoing congenital heart surgery between 2008–2012 identified by procedure code at less than one year of age were evaluated. Follow-up data were collected through 2013. Electronic health records were linked to state birth and death certificate records and to the Society of Thoracic Surgeons Congenital Heart Database. Patients undergoing patent ductus arteriosus ligation and heart transplantation were excluded. Patient RI was determined using 2010 census block group level data with address at the time of index hospitalization. Type of CHD was divided into severe, shunt, valve, or other based on ICD-9 code (supplementary Table S5). The groupings of severe, shunt, and valve were predetermined by the CDC and Prevention Congenital Heart Defect Surveillance Network using a previously described method that categorizes CHD diagnostic codes into mutually exclusive hierarchical groups integrating both hemodynamic severity and basic anatomical characteristics.^26,27 This study was approved by the Duke Health Institutional Review Board and funded by the CDC grant NU50DD004933-01.

Exposures of interest included patient race, RI, and EI. Race was collected from electronic health records and recorded as White, Black, or Other. Other race included the following races: Asian, American Indian or Native Alaskan, and Native Hawaiian or other Pacific Islander. Given small number of patients in each group included in the other category in the study population, the decision was made to combine these into one category. Ethnicity was collected from the electronic health record as Hispanic or non-Hispanic and not included with race data. RI and EI were determined by matching a patient to their census block group and calculating RI and EI using previously describe methodology.¹⁷ The reference group for RI was individuals who are a race other than NHB (non-NHB). The reference group for EI was college education.

Outcomes assessed during index encounter include LOS (days) and surgical complications (major and other). Outcomes assessed during 1-year post discharge include days alive and out of hospital, readmission, death, number of outpatient encounters, having at least one emergency department (ED) visit, and having at least one unplanned inpatient admission at participating institutions. LOS of index encounter was calculated using initial admission date to discharge date of first encounter in which patients had a qualifying ICD-9 or CPT code. See supplementary table S4 for full list of ICD-9and CPT codes. Surgical complications were identified by ICD-9 codes and chosen based on cardiac specific complications from the Society of Thoracic Surgeons Congenital Heart Surgery Database.²⁸ See supplementary table S6 for full list of complications and ICD-9 codes. Days alive and out of hospital were calculated by subtracting days admitted from the number of days in one year. Death included any type of mortality during the study period. Number of outpatient encounters was calculated as the total number of outpatient cardiology visits the patient had during the study period. Having at least one ED visit and unplanned inpatient admission were dichotomized into no ED visits/ admissions or one or more visits/ admissions.

Demographics of the cohort, including age, sex, race (White, Black or African American, Other, unknown), ethnicity (Hispanic or Non-Hispanic), and insurance type (Medicaid, private, self-pay, or other), and outcomes were summarized by RI and EI tertiles and by race. Tertile 1 represented those with the least isolation and tertile 3 represented those infants with the highest degree of isolation. For continuous variables, mean and standard deviation are presented. For categorical variables, counts and percentages are presented.

The association between RI and EI with outcomes of interest was assessed. For analyses of complications, readmission, binary resource utilization and death, a logistic regression model was used. For analyses of LOS, days alive and out of hospital and number of outpatient encounters, a negative binomial model was used. The relationship between RI, EI, and outcomes was evaluated for non-linearity using natural cubic splines. Non-linear relationships were modeled using two linear piece-wise splines in the final models. Adjusted analyses were performed utilizing the following variables: age (days), sex, race, ethnicity, disease severity, and preterm birth (gestational age <37 weeks). Death was not modeled in the adjusted setting due to small event counts. Rate ratios and odds ratios with 95% confidence intervals (CI) were presented for every 0.10 unit increase in the evaluated index.

Additional models were fit to evaluate the potential for an interaction between race and RI. For the purpose of describing the cumulative effect of race and RI on outcomes, RI was dichotomized using the median value, and patients with Black or White race were grouped into four classes (Black patients with RI< median, Black patients with RI ≥ median, White patients with RI < median, and White patients with RI ≥median). All models included both RI and race. Rate ratios and odds ratios were presented with Black or White race and RI < median as the reference categories. Adjustment variables were the same as those previously specified.

Of the 50,429 unique patients in the database, 6,816 had a procedure code of interest. An index procedure occurred in infants less than one-year-old in 1,558 infants and 1,307 had the procedure performed between 2008–2012. Of these, 1,217 infants had available racial and education isolation data and were included in the final cohort (Supplementary Figure S1).

The median age of the infants in the cohort was 17 days (Table 1). The cohort had 43.3% female infants. The majority of infants had severe CHD (63.4%) defined by ICD-9 code. There was a diverse racial make-up with 22.2% Black, 57.4% White, 13.6% Hispanic 3.2% other, and 17.2% unknown. Many infants had Medicaid (34.3%) followed by private insurance (24.1%). Few patients were self-pay (0.5%) or other insurance (2.0%); however, there was a high rate of missing data for insurance status. The mean RI index was 0.22 (SD 0.17). The mean EI index was 0.74 (SD 0.15).

The median LOS across the cohort was 16 days during the index hospitalization (Table 1). The rate of major complications was 18.3%, however there was a higher rate of other complications (41.1%). There was a 18.2% rate of unplanned readmission one-year post discharge. The median number of days alive and out of hospital during the year post-discharge was 361 days. One-year mortality was 6.8%. In terms of resource utilization in the year post-discharge, mean number of outpatient encounters was 5.6. Most infants had at least one unplanned inpatient encounter (54.5%). There were modest rates of at least one ED encounter (19.8%).

The mean ages and sex distribution were similar across races. There was a higher percentage of Medicaid insurance in Black infants (40.7%) compared to the White (27.0%) and Other race (25.6%) infants. CHD severity was similar across races. RI was higher in Black infants (0.36) compared to White (0.17) and Other race (0.21) infants. EI was higher in Black infants compared to infants of White and Other races (Supplementary Table S1).

Black infants had a longer LOS at index hospitalization (42.1 days) than White infants (27.0 days,) (Supplementary Table 1). Infants of Other races had a mean LOS 35.4 days. The rates of other complications were higher in Black infants (47.9%) and other race infants (51.3%) than in White infants (38.1%). Black infants had the fewest out of hospital days in the first year with a mean of 310.3 days compared to White (338.2 days) or Other race infants (350.5 days,). Black infants had an increased rate of 1-year death (12.1%) compared to White (5.1%) and Other race (2.6%) infants. Black infants had a lower mean number of outpatient encounters (mean 4.5) compared to White (mean 5.9) or Other (mean 5.3) race infants.

In unadjusted analysis, Black patients compared to White patients had an increased median LOS (RR 1.49, 95% CI 1.25–1.78) and increased odds of other complications (OR 1.49, 95% CI 1.07–2.08) and death (OR 2.18, 95% CI 1.21–3.93) (Table 2). After adjusting for racial isolation, age, sex, ethnicity, disease severity, and preterm birth, LOS remained significant with Black infants having increased LOS (RR 1.63, 95% CI 1.38 – 1.91) while other complication (OR 0.99, 95% CI 0.66–1.47) and at least one ED encounter (OR 1.51, 95% CI 0.98–2.34) were no longer significant. Death was not modeled in adjusted setting due to small counts.

Infants had similar age, sex, amount with Hispanic ethnicity, and disease severity across RI tertiles (Supplementary Table S2). There was a lower percentage of private insurance in the most isolated tertile (tertile 3) compared to the least isolated tertile (tertile 1). The mean LOS and 1-year morality rate were highest in the most isolated tertile (tertile 3) (36.3 days and 9.0%, respectively). The mean number of outpatient visits was highest in the least isolated tertile (tertile 1) (6.5 in tertile 1 vs 4.7 in tertile 3).

In unadjusted analysis there was an increased LOS (up to RI 0.3), mortality, odds of having at least one ED visit and at least one inpatient encounter (up to RI 0.3) with increasing RI (Figure 2). There was a decrease in the number of outpatient encounters and odds of having at least one inpatient encounter with increments in RI greater than 0.3. When adjusting for age, sex, race, ethnicity, disease severity and LOS, an increased odds of at least one unplanned inpatient encounter remained significant (OR= 1.39, 95% CI 1.20–1.63 per 0.1 increment in RI). Additionally, there remained decreased number of outpatient encounters at increments in RI up to 0.3 (RR=0.81, 95% CI 0.71–0.93) and decreased odds of at least one unplanned inpatient encounter for increments above an RI of 0.3 (OR 0.82, 95% CI 0.69 – 0.96). There was no difference in LOS or number of outpatient encounters for increments in RI above 0.3 or risk of major or other complications.

There was a similar age, sex, amount with Hispanic ethnicity, and preterm birth rate across EI tertiles (Supplementary Table S3). There was a higher percentage of infants with private insurance in the least educationally isolated tertile (tertile 1) and higher percentage of infants with Medicaid insurance in the most isolated tertile (tertile 3). Hospital readmissions and mortality rate were highest in the most isolated tertile (20.3% and 8.6%, respectively). This group also had the lowest number of days alive and out of hospital (321.9 days).

In unadjusted analysis there was an increased risk of longer LOS (RR1.05, 95% CI 1.01–1.09), other complications (OR 1.10, 95% CI 1.02–1.19), hospital readmission (OR 1.14, 95% CI 1.03–1.27) and at least one ED visit with increased educational isolation at EI above 0.8 (OR 1.85, 95% CI 1.28–2.10) (Figure 3). After adjustment for age, sex, race, ethnicity, disease severity, preterm birth and length of index hospitalization, increased risk of at least one ED visit remained significant with increased educational isolation at EI above 0.8 (OR 1.82, 95% CI 1.23–2.70). There was no difference in odds of major surgical complications, mortality, number of outpatient encounters, and having at least one unplanned inpatient encounter.

When assessing the interaction between patient race and racial isolation we found little evidence that race modifies the association of RI with outcomes (all p-values but one was not statistically significant, data not included). When we compared patient outcomes based on race and RI, Black infants with greater than or equal to median RI had the highest mean LOS (RR= 1.69, 95% CI 1.42–2.02) relative to White infants with less than median RI, and the highest odds of mortality (OR 3.40, 95% CI 1.84–6.32), and at least one inpatient encounter (OR 1.62, 95% CI 1.16–2.26) (Table 3). While overall race was not associated with other complications (p=0.06), Black infants with RI greater than or equal to the median had increased odds of complications (OR 1.50, 95% CI 1.08–2.10). These infants also had decreased number of outpatient encounters (OR 0.66, 95% CI 0.49–0.89). After adjustment, associations with both LOS (RR 1.67, 95% CI 1.43–1.95) and number of outpatient encounters (OR 0.65, 95% CI 0.47–0.88) remained statistically significant. Additionally, White infants with greater than or equal to the median RI had a decreased number of outpatient encounters in both unadjusted and adjusted analysis (OR 0.71, 95% CI 0.53–0.94) compared to White infants with less than median RI.