The Association of High-Quality Hospital Use on Health Care Outcomes for Pediatric Congenital Heart Defects in a Universal Health Care System

ABSTRACT

Introduction

Congenital heart disease (CHD) has an incidence of 0.8% to 1.2% worldwide, making it the most common birth defect. Researchers have compared high-volume to low-volume hospitals and found significant hospital-level variation in major complications, health resource utilization, and mortality after CHD surgery. In addition, researchers found critical CHD patients at low-volume/non-teaching facilities to be associated with higher odds of inpatient mortality when compared to CHD patients at high-volume/teaching hospitals (odds ratio 1.76). We examined the effects of high-quality hospital (HQH) use on health care outcomes and health care costs in pediatric CHD care using an instrumental variable (IV) approach.

Materials and Methods

Using nationwide representative claim data from the United States Military Health System from 2016 to 2020, TRICARE beneficiaries with a diagnosis of CHD were tabulated based on relevant ICD-10 (International Classification of Diseases, 10^th revision) codes. We examined the relationships between annual readmissions, annual emergency room (ER) use, and mortality and HQH use. We applied both the naive linear probability model (LPM), controlling for the observed patient and hospital characteristics, and the two-stage least squares (2SLS) model, accounting for the unobserved confounding factors. The differential distance between the patient and the closest HQH at the index date and the patient and nearest non-HQH was used as the IV. This protocol was approved by the Institutional Review Board at the University of Maryland, College Park (Approval Number: 1576246-2).

Results

The naive LPM indicated that HQH use was associated with a higher probability of annual readmissions (marginal effect, 18%; 95% CI, 0.12 to 0.23). The naive LPM indicated that HQH use was associated with a higher probability of mortality (marginal effect, 2.2%; 95% CI, 0.01 to 0.03). Using the differential distance of closest HQH and non-HQH, we identified a significant association between HQH use and annual ER use (marginal effect, −14%; 95% CI, −0.24 to −0.03).

Conclusions

After controlling for patient-level and facility-level covariates and adjusting for endogeneity, (1) HQH use did not increase the probability of more than one admission post 1-year CHD diagnosis, (2) HQH use lowered the probability of annual ER use post 1-year CHD diagnosis, and (3) HQH use did not increase the probability of mortality post 1-year CHD diagnosis. Patients who may have benefited from utilizing HQH for CHD care did not, alluding to potential barriers to access, such as health insurance restrictions or lack of patient awareness. Although we used hospital quality rating for congenital cardiac surgery as reported by the Society of Thoracic Surgeons, the contributing data span a 4-year period and may not reflect real-time changes in center performance. Since this study focused on inpatient care within the first-year post-initial CHD diagnosis, it may not reflect the full range of health system utilization. It is necessary for clinicians and patient advocacy groups to collaborate with policymakers to promote the development of an overarching HQH designation authority for CHD care. Such establishment will facilitate access to HQH for military beneficiary populations suffering from CHD.

INTRODUCTION

Congenital heart disease (CHD) has an incidence of 0.8% to 1.2% worldwide, making it the most common birth defect.¹ Approximately 40,000 children undergo surgery for CHD in the United States each year.² In addition, CHD is the leading cause of death related to congenital anomalies, representing approximately 40% of deaths in children with birth defects in the world.³ The relationship between hospitals’ pediatric cardiac surgical volume and mortality has been studied often.^4–7 Researchers have compared high-volume to low-volume hospitals and found significant hospital-level variation in major complications, length of stay, and mortality after CHD surgery.² In addition, researchers found management of critical CHD at low-volume/non-teaching facilities to be associated with higher odds of inpatient mortality when compared to that of CHD patients at high-volume/teaching hospitals (odds ratio 1.76).⁸

Analyzing center-specific performance for pediatric CHD care is difficult due to the diversity and complexity of the diagnoses and procedures, varying risk factors, and the infrequency of the procedures.⁹ Moreover, utilizing treatment variation from observational databases for outcome estimation can cause unsuitable deductions if treatment is designated in a manner that is related to expected outcomes.¹⁰ A Center of Excellence (CoE) designation has been recommended as a manner to create a network of high-quality care and to influence hospital and provider selection for patients.¹¹ Similar to a CoE, a high-quality hospital (HQH) is an institution that provides effective and safe care with a culture of excellence, resulting in optimal health.¹² Nevertheless, previous studies of CoE programs have been limited to surgical procedures and have had mixed results.¹¹ These inconsistencies may result from how hospital quality of care is measured, variables are controlled, or outcomes are measured.

This article builds on the previous research by addressing the endogeneity of CHD severity by estimating models using instrumental variable (IV). An IV is used to generate unbiased, consistent estimates when unmeasured factors exist and are associated with the treatment and outcome variables. We used the differential distance between the nearest HQH and the nearest non-HQH for CHD care as an IV to compare clinical outcomes and resource utilization for HQH use versus non-HQH use related to pediatric CHD care for TRICARE beneficiaries.

This study aimed to compare clinical outcomes (annual readmissions, annual emergency room (ER) utilization, and mortality) among TRICARE beneficiaries who underwent treatment at HQH to those who did not use HQH for pediatric CHD care (ages 0 to 17). We hypothesized that the probability of annual readmissions, annual ER use, and mortality would be lower at HQH compared to non-HQH.

METHODS

Study Design

We conducted a retrospective cross-sectional study with US data from 2016 to 2020.

Study Population

Since there is a particular interest in what facility an initial corrective surgery takes place, this study included all pediatric inpatient CHD care received within the first year post-CHD diagnosis in US hospitals from 2016 to 2020. TRICARE beneficiaries with a diagnosis of CHD were tabulated based on relevant ICD-10 (International Classification of Diseases, 10^th revision) codes.¹³ The study population was categorized into four types of CHD: single ventricle CHD, moderate-complex CHD, simple CHD, or other CHD based on the hierarchy of the complexity of CHD¹³ (Supplemental Table S1). Records with other supplemental insurance as well as those lacking at least one year of treatment or follow-up care at a military facility outside of the United States were excluded. Furthermore, children with patent ductus arteriosus and/or isolated congenital anomalies of the peripheral or cerebral vascular system as well as those with cardiac transplantation were excluded to enhance cohort homogeneity.¹⁴ Finally, patients were excluded if there were missing data on variables of interest (8.4% for race and <0.01% for differential distance).

Data Source

We used the US Military Health System (MHS) Data Repository (MDR) to identify MHS beneficiaries aged 0 to 17 years who received CHD care from 2016 to 2020. The MDR provides information at the patient encounter level, including diagnoses, treatments, patient demographic characteristics, and the facility in which care was delivered. The MDR includes data for beneficiaries who use US MHS facilities that are operated by the DoD, also known as direct care for military treatment. On the other hand, purchased care is defined as beneficiaries who receive care in the civilian private sector that is funded by TRICARE. This protocol was approved by the Institutional Review Board at the University of Maryland, College Park (Approval Number: 1576246-2).

Study Outcomes

For the analysis of outcomes, each outcome of interest was modeled separately to include annual readmissions, annual ER use, and mortality. The dependent variable annual readmission was designated as the primary dependent variable and was defined as more than one admission post 1-year of diagnosis of a CHD classified as yes (1) or no (0).¹⁵ A consecutive hospitalization within 24 hours of the index hospitalization was counted as a single hospitalization and elective surgeries were excluded.¹³ The secondary dependent variable, ER use, was determined by reported ER utilization for any emergent or urgent admission post 1-year of diagnosis for a CHD and classified as yes (1) or no (0). The tertiary dependent variable, mortality, was determined by whether a patient died at home, died in a medical facility, or did not recover during admission 1-year post CHD diagnosis and was classified as yes (1) or no (0).

Independent Variable: HQH Use

The independent variable, HQH use, was a dichotomous indicator for HQH utilization categorized as (1) yes or (0) no. This study categorized HQH based on the Society of Thoracic Surgeons (STS) Congenital Heart Surgery Database (CHSD). This database is the largest congenital and pediatric cardiac surgical data registry in the world and includes the analysis of outcomes and the improvement of quality for congenital heart surgeries.¹⁶ The STS CHSD covers hospitals’ results over a 4-year period and presents respective observed-to-expected (O/E) mortality ratios for operative mortalities, which incorporates a risk adjustment to account for differences in case mix, serving as a platform for benchmarking performance and improving quality for CHD care.¹⁷ Hospitals with an overall O/E mortality ratio of equal to or less than 1 were designated as HQH, since these hospitals had fewer deaths than expected based on the case mix treated at the hospital.¹⁸ Hospitals with an O/E mortality ratio greater than 1 were categorized as non-HQH. The STS CHSD data from the 2015 to 2018 period were used to classify hospitals, totaling 42 HQH within the United States. In addition, 984 hospitals that did not report to the STS for CHD care were categorized as non-HQH.

Covariates

Covariates were defined at the time of index admission post-initial CHD diagnosis. To adjust for predisposing characteristics, the following covariates were used: age (1 to 30 days, 31 days to less than 1 year, 1 year to 5 years, 6 years to 17 years), gender (male and female),⁴ sponsor (also known as the military member), marital status (currently married, not married to include divorced and widowed),¹⁹ and self-declared race (White, Black, or Other race to include Asian, Pacific Islander, or Native American) by the sponsor. Sponsor race was used as an imputation of dependent (i.e., child) race which is a tested method of imputation for this population.¹⁴ To adjust for enabling resources, the following covariates were used: type of TRICARE insurance (Prime, Other insurance to include Extra, Standard, Select, Young Adult, or Unenrolled), sponsor service branch (Army, Air Force, Marine Corps, Navy, Other service),¹⁴ sponsor pay grade/rank to include Junior Enlisted (E1-E4), Senior Enlisted (E5-E9), Officer, and Other rank (cadets, midshipmen, officer candidates, reserve officer training corps members, and warrant officers),¹⁴ and provider state. To adjust for need characteristics, the following covariates were used: complexity of CHD (simple, moderate-complex, single ventricle, other),¹³ common comorbid conditions observed during ambulatory visits and/or hospitalizations (yes or no),^13,20 presence or absence of a genetic syndrome (Down’s, Noonan’s, DiGeorge, Holt-Oram, Turner, and Williams-Beuren syndromes) (yes or no),^13,20,21 and prematurity (gestational age less than 37 weeks) (yes or no).^4,21 Patients with multiple CHD diagnoses were assigned to the complexity of CHD category associated with the most complex diagnosis.

Statistical Analyses

The analyses were performed in a sequential two-step process: naive linear probability model (LPM) and the IV approach.

Naive LPMs

The probabilities of annual readmissions, annual ER use, and mortality were analyzed using LPMs. The abovementioned naive LPMs did not consider whether a patient’s HQH use was endogenous, because some patients use an HQH due to the higher severity of the disease (which is likely to be unobserved by the econometrician). When endogeneity exists, unobserved characteristics can lead to bias in the results.²²

IV Approach

We employed an IV approach to address the issue of endogeneity because of the unobserved selection bias between the HQH and non-HQH groups. This study used two-stage least squares (2SLS) IV estimation, previously used in health care, to predict the medical center used.^10,23 The IV was defined as the differential distance between (1) the patient and the closest HQH at the index date and (2) the patient and nearest non-HQH that could have been used at the index date. Differential distance between the closest medical center of interest and the closest counterfactual medical center has been used as an IV in previous research.^10,24,25

For differential distance to be a valid instrument, it must (1) be related to treatment (HQH status of CHD medical center) also known as the “relevance assumption” and (2) be related to outcomes only through treatment (HQH status of CHD medical center) and therefore unrelated to unobserved determinants of the outcome variables also known as the “exclusion restriction assumption.”

In this study, the distance from the patient’s residence to HQH and distance from the patient’s residence to non-HQH were based on straight-line distances between the patient zip code centroid and the hospital centroid zip code.²⁶ First, patients were grouped based on the median value of the differential distance for the sample (3.3 miles), resulting in a binary variable (differential distance within 3.3 miles [yes or no]). Next, differential distance was specified with three binary variables to represent four different patient groups based on every 25th percentile of differential distance across the sample (differential distance groups of less than or equal to −61.9 miles, between −61.9 to 3.3 miles, between 3.3 and 62 miles, greater than 62 miles).

The first stage of 2SLS estimates the endogenous variable (e.g., the hospital, HQH versus non-HQH used) as a function of the IV and patient characteristics. The IV was defined as the differential distance between (1) the patient and the closest HQH at the index date and (2) the patient and nearest non-HQH hospital that could have been used at the index date.

In the model, |${X_i}$| represents the binary endogenous patient HQH use for individual i. DD represents the IV (differential distance). |${Z_i}$| represents a vector of variables that measure patient characteristics (e.g., age, sponsor race, gender, sponsor marital status, type of insurance, sponsor pay grade [rank], sponsor service branch, complexity of CHD, genetic syndrome, common comorbid condition, prematurity, and state fixed effects). We also included the year-fixed effects (⁠|${e_t}$|⁠).

In the first stage of the 2SLS, the ordinary least squares (OLS) method was used to estimate HQH use choice. If more severe CHD patients tend to choose HQH, the OLS coefficient on HQH use will be a biased and inconsistent estimate of the true causal relationship. For this reason, an IV model (Equation 2) was estimated, which assumes that HQH use has the reduced form (Equation 1) where differential distance (DD) is the distance between the closest HQH and the closest non-HQH for CHD care.

In the three outcome models, |${Y_i}$| represents the probabilities of annual readmissions, annual ER use, and mortality for individual i. In the second stage, the predicted variables from the first stage are entered as regressors to estimate outcome variables. |${Z_{i\,}}$|and |${e_t}$| include the same independent variables listed in Equation 1. The IVREGRESS procedure in Stata 17.0 was estimated using clustering at the hospital level to account for correlation in outcomes within the same hospital. All analyses were performed using Stata statistical software (Version 17.0 StataCorp, College Station, Texas).

RESULTS

Descriptive Analyses

Among 11,775 children with CHD identified over the study period (2016 to 2020), 10,851 patients met the inclusion criteria. For the annual readmission model, a consecutive hospitalization within 24 hours of the index hospitalization was counted as a single hospitalization and elective surgeries were excluded (N = 10,120). Supplemental Table S1 presents the covariate distribution by HQH status and covariate distribution by median differential distance (3.3 miles) of the sample. Of the population, 82.9% did not use an HQH at any point for CHD care and 17.1% used an HQH at some point for CHD care. In contrast to grouping patients by center choice, grouping patients by median differential distance produced a more balanced distribution of covariates.

2SLS models varied in the number of patient groups used to specify differential distance. There were two patient groups specified: (1) within a median differential distance (yes or no) or (2) differential distance groupings based on 25, 50, 75, and 100 percentiles. Table I contains the partial F-statistic for the first-stage model of the CHD center of choice post 1-year CHD diagnosis (HQH or non-HQH) detailed in Equation 1. The Staiger and Stock F-statistics for the differential distance instrument exceeded the threshold standard of 10.²⁷ The Hausman test statistics for the overidentified models were not statistically significant. For these reasons, the IV was determined to be strong and valid across all differential group specifications and outcome variables (see Table I).

Naive LPMs and 2SLS Models

Table II presents the estimates of the effects of HQH use on health care outcomes. We present two sets of results: (1) the naive LPM that treated patient HQH used as exogenous and (2) the 2SLS that treated patient HQH used as endogenous. To compare the models that were adjusted for endogeneity to models that were not adjusted for potential endogeneity, we contrasted the marginal effects of the 2SLS estimates to the marginal effects of the LPM (without IV). Table II lists the IV and non-IV regression results for the entire sample. The non-IV results contain marginal effects results for LPM in which the regressor of interest was HQH use. The non-IV marginal effect for annual readmissions suggested that HQH use increased the probability of more than one hospital admission 1-year post CHD diagnosis (marginal effect, 18%; 95% CI, 0.12 to 0.23). Although statistically insignificant, the non-IV marginal effect for annual ER use suggested that HQH use increased the probability of annual ER use 1-year post CHD diagnosis (marginal effect, 2%; 95% CI,−0.01 to 0.04). Lastly, the non-IV marginal effect for mortality suggested that HQH use increased the probability of mortality 1-year post CHD diagnosis (marginal effect, 2%; 95% CI, 0.01 to 0.03).

The 2SLS specifications were estimated with the full set of measured confounders. Patients with more than one admission post 1-year CHD diagnosis represented 13.7% of the sample. The 2SLS estimates revealed that HQH use was not significantly associated with annual readmissions post 1-year CHD diagnosis (marginal effect,−4%; 95% CI, −0.15 to 0.07). Although the relationship was not statistically significant at the 0.05 level and the magnitude of the coefficient was small, the coefficient was negative, which suggests that HQH use did not increase the probability of more than one annual readmission post 1-year CHD diagnosis. ER use post 1-year CHD diagnosis represented 7.86% of the sample population. The 2SLS estimates indicated that HQH use was significantly associated with a decreased probability of annual ER use post 1-year CHD diagnosis (marginal effect,−14%; 95% CI, −0.24 to −0.03). Mortalities represented 3.18% of the sample population. The 2SLS estimates revealed that HQH use was not significantly associated with mortalities post 1-year CHD diagnosis (marginal effect,−2%; 95% CI, −0.07 to 0.03).

A consolidated table of the effect of HQH status on outcomes (annual readmissions [yes/no], annual ER use [yes/no], and mortality [yes/no]) based on the 2-group and 4-group differential distance IV specification is found in Supplemental Table S2. Results across the 2-group and 4-group differential distance IV specifications were very similar. A consolidated table of the effect of HQH status on outcomes based on CHD complexity is found in Supplemental Table S3.

A comparison of the results of the IV and non-IV models indicated that relying on the association of HQH use to clinical outcomes (as in the LPM) results in biased estimations of the impact of HQH use on clinical outcomes.

Sensitivity analysis

The outcome variables were estimated using the bivariate probit model, within Stata 17.0, with the “seeming unrelated” syntax with SEs clustered at the hospital level, finding similar results to the 2SLS approach (see Table III). The effects of HQH use are presented. In the bivariate probit model, there was a distinct difference in the P-values associated with annual readmissions in the case of the bivariate probit estimates. There was no statistically significant effect of HQH use on annual ER use, unlike the 2SLS estimates, after controlling for patient-level covariates and adjusting for possible selection bias. The results from the annual readmissions outcome equation suggested that there was no statistically significant effect of HQH use on annual readmissions after controlling for patient-level covariates and adjusting for possible selection bias. The results from the mortality outcome equation suggested that there was no statistically significant effect of HQH use on mortality after controlling for patient-level covariates and adjusting for possible selection bias. Furthermore, the value of rho was examined to determine if the outcomes and selection equations were correlated. For all outcomes of interest (annual readmissions, annual ER use, and mortality), the 95% CI for the value of rho and the likelihood-ratio tests indicated that rho was statistically different from zero, so there was significant selection bias in HQH status and the outcome variables.

DISCUSSION

The current study provides evidence that after controlling for patient-level and facility-level covariates and adjusting for endogeneity, (1) HQH use did not increase the probability of more than one admission post 1-year CHD diagnosis, (2) HQH use lowered the probability of annual ER use post 1-year CHD diagnosis, and (3) HQH use did not increase the probability of mortality post 1-year CHD diagnosis.

Of the population, 82.9% did not use an HQH at any point for pediatric inpatient CHD care and 17.1% used an HQH at some point for pediatric inpatient CHD care. Similar to previously published work, compared to low-volume surgery centers, high-volume surgery centers represented 82% of CHD hospitalizations.^28,29 Low-volume centers have been associated with higher readmissions when compared to high-volume centers.²⁹ In our study, HQH use did not increase the probability of more than one admission post 1-year CHD diagnosis. Although CHD patients are a notable high-utilization group due to their longer lengths of stay in the hospital,²⁹ our study found that HQH use was associated with the lowered probability of annual ER use post 1-year CHD diagnosis. This may be due to enhanced outpatient care coordination, more efficient post-discharge monitoring strategies, or a lower rate of complication. An inverse surgery volume to mortality relationship has been explained in congenital cardiac surgery.^4,30,31 Although our study found that the association between HQH and mortality was insignificant, the coefficient was negative, which suggests that HQH helped to reduce mortalities. Similar to other studies that compared high-volume surgery centers to low-volume surgery centers and used volume as a quality proxy,²⁹ we found that 54.6% of the sample treated at HQH had either a moderate or severe CHD diagnosis.²⁸ Since HQH are more likely to treat more severe CHD cases,²⁸ the study’s models may not be appropriately controlling for the severity of the case mix which could create biased results. Furthermore, some patients who may have benefited from utilizing an HQH for CHD care did not, alluding to potential barriers to access, such as health insurance restrictions or lack of patient awareness.

This study’s results should be considered with several limitations. Since 84.7% of hospitals participate in public reporting via STS but not all US centers participate, the inability to fully classify hospitals as an HQH versus a non-HQH may be a limitation of this study. Although we used hospital-quality rating for congenital cardiac surgery as reported by the STS, the contributing data span a 4-year period may not reflect real-time changes in center performance. Since this study focused on inpatient care within the first-year post-initial CHD diagnosis, this study’s findings may not reflect the full range of health system utilization. This study focused on purchase care and did not capture medical care received at a military treatment facility. Admissions, ER use, and/or mortalities experienced at a military hospital were thus excluded from the analysis. All the limitations of 2SLS apply.

CONCLUSION

After controlling for patient-level and medical center-level characteristics to adjust for potential selection bias, HQH use for CHD care resulted in less health resource utilization and did not jeopardize the survival of CHD patients. Moreover, the use of the IV to randomize treatment assignments is warranted. Additionally, our findings highlight that education about the importance of care in an HQH is imperative at the point of initial CHD diagnosis, particularly for those with more complex CHD diagnoses. It is necessary for clinicians and patient advocacy groups to collaborate with policymakers to promote the development of an overarching HQH designation authority for CHD care. Proper HQH accreditation and resource allocation will not only help to identify standardized quality metrics and best practices but can also be used to further regionalization efforts. Such establishment will facilitate access to HQH for military beneficiary populations suffering from CHD.

ACKNOWLEDGMENTS

The authors would like to acknowledge the contributions of their colleagues. The authors have obtained written permission from all persons named in the acknowledgement.

CLINICAL TRAIL REGISTRATION

Not applicable.

INSTITUTIONAL REVIEW BOARD

This study was approved by University of Maryland, College Park Institutional Review Board (IRB) (Approval Number: 1576246-2). Exempt Studies.

INSTITUTIONAL ANIMAL CARE AND USE COMMITTEE

Not applicable.

Not applicable.

INDIVIDUAL AUTHOR CONTRIBUTION STATEMENT

AJE analyzed the data and drafted the original manuscript. TK, LF, JC, and DY assisted with the design of this research and reviewed the original manuscript. NYC and PB reviewed and edited the manuscript.

INSTITUTIONAL CLEARANCE

Institutional clearance approved.

SUPPLEMENTARY MATERIAL

SUPPLEMENTARY MATERIAL is available at Military Medicine online.

FUNDING

No funding to report.

CONFLICT OF INTEREST STATEMENT

None declared.

DATA AVAILABILITY

The data that support the findings of this study are protected by the Military Health System and will require approval for the Defense Health Agency to be released.

REFERENCES

Author notes