Social Drivers of Health and Pediatric Extracorporeal Membrane Oxygenation Outcomes

We conducted a retrospective single-center cohort study at Boston Children’s Hospital (BCH), a large quaternary, free-standing children’s hospital. All children (<18 years) supported on ECMO between January 2012 and September 2021 were included. For patients with multiple ECMO runs, only data from the first run were included. International patients and those without a documented address were excluded given inability to obtain SDoH information. The study protocol was approved by the Institutional Review Board (IRB-P00038604) with informed consent requirements waived.

Patient clinical and demographic data were obtained through linking the electronic medical record (EMR), our local BCH ECMO database, and institutional Extracorporeal Life Support Organization (ELSO) data. All patients requiring ECMO were identified through the BCH ECMO dataset. The institutional ELSO data export was linked to the BCH ECMO dataset using patient date of birth and ECMO cannulation date. This linked dataset was enriched with additional patient data including residential address and granular time-sensitive illness severity measures, such as pre-ECMO or post-ECMO diagnoses, laboratory values, vital signs, and ventilator settings via electronic search of the EMR using the date and time of ECMO cannulation (see Supplement A for additional details on automatic extraction from EMR). Patients were categorized into 4 distinct diagnostic cohorts: cardiac, neonatal respiratory, pediatric respiratory, and sepsis. Neonates were defined as less than 7 days old to account for specific respiratory diagnoses that occur postnatally. Based on prior research, a comprehensive list of International Classification of Diseases (ICD)-9 and 10 codes was compiled that fit the diagnostic categories (Supplemental Table 5).^34,35  Patients with ICD codes across multiple diagnostic cohorts were categorized using a previously established hierarchical tiered approach, with a patient having a cardiac code being assigned to the cardiac cohort, followed by patients with respiratory codes, then sepsis.^36–38 

The primary explanatory variable was the Child Opportunity Index (COI).^{26,28–32,39–41}  COI version 2.0 is a validated comprehensive composite index of census tract-level SDoH.²⁷  COI measures 29 variables across 3 domains, including education (eg, early childhood education, teacher experience), health and environment (eg, health insurance coverage, access to green spaces and healthy food, air pollution), and social and economic (eg, poverty rate and employment rate) (Fig 1). COI and its 3 domains are presented both as a score ranging from 0 to 100 (higher COI indicates healthier child development) and in quintiles of opportunity (very low, low, moderate, high, very high). COI was determined for each patient based on address assessed at the census tract level.⁴²  Other SDoH, including payer, language, use of an interpreter, distance from home to hospital, urban residency, and race and/or ethnicity, were obtained via the EMR. The patient’s race and/or ethnicity were assigned from BCH EMR data, noting this variable is self-identified, and is a social and power construct collected in predefined, socially-derived categories and input by hospital staff. Underrepresented racial and ethnic groups were defined as Black, Hispanic, Asian, or other groups. Interpreter use was used as a marker of nonproficiency in the English language. ArcGIS Pro v2.9 was used to geocode the BCH and patient residential address using the Esri StreetMap Premium 2020 address locator (USA.loc). Using the Esri Business Analyst 2021 road network and the network analysis closest facility tool was used to compute the driving distance (miles) and travel-time to the BCH location.⁴³  Urban-rural classification was determined using the National Center for Health Statistics scheme.⁴⁴  The primary outcome was in-hospital mortality, with secondary outcomes of ECMO run duration, ICU, and hospital length of stay (LOS).

Categorical variables were summarized using frequencies and percentages, and continuous variables with medians and either ranges or interquartile ranges as noted. Patient characteristics and outcomes were compared for patients with and without recorded COI using Fisher’s exact test for categorical variables and the Wilcoxon rank sum test for continuous variables. Patients for whom COI was not available were then excluded from all subsequent analyses with a sensitivity analysis performed to examine the impact of these missing variables. Characteristics were compared across Child Opportunity Index quintiles using tests of trend. ECMO run duration and length of stay outcomes were compared using the log-rank test; patients who died before discharge were censored at the time of death. Demographic information, diagnoses, SDoH, and ECMO variables were compared for patients who died in-hospital versus those discharged alive using Fisher’s exact test for categorical variables and the Wilcoxon rank sum test for continuous variables. A multivariable logistic regression model was created for risk adjustment when evaluating associations between social drivers of health and in-hospital mortality. Variables known to be associated with mortality or statistically significant at the 0.2 level were considered for inclusion in the model. Potential covariates were assessed for collinearity. A P value < 0.1 by the likelihood ratio test was required for retention in the final risk adjustment model; year of ECMO was forced into the model to account for any era effects. Odds ratios were estimated with 95% confidence intervals. The discrimination of the risk adjustment model was assessed using the c statistic and calibration using the Hosmer-Lemeshow test. Adjusted relationships between in-hospital mortality and individual social drivers were assessed by iteratively adding each SDoH variable to the risk adjustment model. As indiscriminate incorporation of pre-ECMO support and during ECMO variables will mask disparate outcomes by adjustment if certain populations have higher incidence of these variables because of biased decision-making or differential access, additional risk adjustment models were performed separating the prehospital patient factors and in-hospital clinical factors.⁷  Variation in ECMO use may contribute to inequitable health outcomes. To evaluate the extent to which differential institutional use of ECMO might contribute to findings, we examined associations between SDoH as explanatory variables and ECMO utilization. Acknowledging ECMO candidacy as a nuanced decision,⁴⁵  we used patients who died without ECMO as the comparison; a cohort with assumed high severity of illness.¹⁷  All patients with available COI data who died without ECMO in the 2 ICUs that offer ECMO support at our institution over the same time period were compared with the ECMO cohort.¹⁷  SAS version 9.4 (SAS Institute Inc., Cary, NC) and Stata version 16 (StataCorp, College Station, TX) were used for analysis.

During the study period, 564 children who had received ECMO support were identified in the institutional ELSO and local ECMO databases, 540 (96%) of whom had a calculable COI (Fig 2). No consistent trends were observed across COI quintiles (Table 1). Median age was 4 months, and 234 (43%) were female. Overall, 317 (59%) had at least 1 pre-existing comorbidity, 334 (62%) had cardiac diagnoses, 92 (17%) neonatal respiratory failure, 93 (17%) pediatric respiratory failure, and 21 (4%) sepsis. In-hospital mortality was 44%, with median ECMO run duration of 125 hours (interquartile range [IQR] 63–244), ICU LOS of 29 days (13–58), and hospital LOS of 40 days (17–80). Demographic and clinical characteristics of patients in decedent and survival cohorts are described (Table 2). Factors associated with in-hospital mortality included cardiac and sepsis diagnostic groups, pre-existing comorbidities, veno-arterial ECMO, central cannulation, higher precannulation lactate and arrest, and requirement for vasopressors or steroids during ECMO. Comparison of respiratory and cardiac cohorts shown in Supplemental Table 6.

A comparison of patients with and without COI available confirmed no significant differences in demographic, clinical or individual-level SDoH variables (Supplemental Table 7). Overall, median COI was 64 (IQR 32–81) and distribution among COI quintiles (Fig 3) was left-skewed with 52% of the cohort in the highest 2 quintiles. Likewise, scores in 3 COI subdomains were above population averages with a median score of 64 (40–80) in education, 71 (44–88) in health and environment and 60 (28,79) in social and economic (Table 3). No difference was observed between decedent and survivor cohorts examining COI overall (P = .70), subdomain scores, or between quintiles. Furthermore, there was no significant difference in ICU or hospital LOS, nor ECMO run duration by COI quintile (log-rank p .99, .62, .46 respectively) (Fig 5). Overall, 174 (34%) were from underrepresented racial and ethnic groups, 323 (60%) had public insurance, 57 (11%) required an interpreter, and 270 (54%) lived in an urban residence, a median of 67 miles (89 minutes) from home to hospital. Distance traveled to BCH shown in Fig 4. There was no statistically significant difference in any SDoH characteristic between ECMO survivors and decedents.

Prehospital clinical variables that met the significance-level threshold to include in the final model included age under 30 days, birth weight under 3 kg, cardiac and sepsis diagnoses, comorbidities, and year of ECMO. In-hospital clinical variables that met the significance-level threshold included vasopressors or steroids during ECMO and complications during ECMO support. Clinical variables of age, diagnostic cohort, comorbidity, vasopressor or steroid use, and complications on ECMO remained statistically significant in the multivariable analysis. The multivariable model had good discrimination (c = 0.78) and was well-calibrated (Hosmer-Lemeshow P = .20). COI, payer, need for interpreter, distance, and travel time from home to hospital, rural residence, and race and ethnicity were not statistically associated with in-hospital mortality in the univariate or multivariable analysis (Table 4). None of the SDoH modified the c-statistic nor significantly changed model calibration. Furthermore, there was no association with COI or individual SDoH and mortality in either the specific prehospital or in-hospital multivariate analyses (Supplemental Tables 8 and 9).

Investigation of ECMO utilization revealed a total of 587 children with COI data available who died in an ICU without ECMO support. There was no statistically significant difference in clinical variables, COI, insurance type, or race and ethnicity between those that died without ECMO and the ECMO-supported cohort (Supplemental Table 10).

In this single-center study evaluating relationships between SDoH and ECMO outcomes in children, we observed no association between pediatric ECMO outcomes and COI nor other SDOH, including race and ethnicity, payer, need for interpreter, urbanicity, and distance from the hospital. Although prior studies have examined the association of race and payer status with ECMO outcomes, we examine the association with SDoH using a multidimensional neighborhood-level index for the first time,⁷  in addition to the classically defined single measure social drivers. Our ECMO cohort had higher COI, reflecting relative social advantage of the catchment area compared with other regions.⁴⁰  Though no difference was observed in SDoH between ICU patients who died without ECMO and the cohort supported on ECMO. Overall, in-hospital mortality in our ECMO cohort was 44%. Clinical variables associated with mortality include low-birth weight neonates, diagnostic group, comorbidities, use of vasopressors or steroids, and neurologic complications on ECMO.

Our findings of no statistical association between measures of social opportunity and adverse ECMO outcomes is contextualized with a recent comprehensive scoping review identifying substantial inconsistency in presence and strength of associations between SDoH and ECMO outcomes across heterogeneous populations.⁷  Notably, overall, 59% of publications found no statistical association, although analyses were imprecise and confounder adjustments were inconsistently applied.⁷  This supports the complex interplay between social drivers and the intricate relationship with health outcomes.

Specifically, our findings contrast with extensive research demonstrating a negative impact of low SES and general pediatric health outcomes across childhood stages, diagnoses, and countries.^1,15,46–48  Explanations for the absence of an association between social opportunity and in-hospital mortality in our ECMO cohort may provide insights into the mechanism of impact of SDoH on general pediatric health outcomes. ECMO therapy is reserved for the most critically ill children, and SDoH may not have as significant an impact on outcomes once children are at the extreme end of illness severity. Supporting this theory are mixed results yielded in existing data on impacts of SDoH on outcomes in other critically ill pediatric populations.^46,49–51  Although the patient population investigated had higher COI than population averages aligned with the relative social advantage of region, an alternate explanation for the upward skew is survivor bias ∼ if patients with social disadvantage were more likely to die before admission. Although our analysis suggests differential institutional utilization does not explain the similar ECMO outcomes according to SDoH, disparities in ECMO utilization have been observed with challenges studying and quantifying differential utilization.^7,17  This avenue deserves further examination, as unjustified bias in healthcare decisions is well described and may drive or mask health inequities.^45,52–54  Another potential explanation for the lack of an association between SDoH and ECMO mortality is the highly protocolized nature of ECMO management. Complexity of clinical care on ECMO (including anticoagulation, afterload reduction, screening for complications, cannulation strategies, and weaning trials among others) necessitates standardization.⁵⁵  Uniformity of approach may mitigate any influence of SDoH; that is, a rigorous clinical protocol may serve as a protective factor ameliorating any potential implicit individual-level provider or institutional biases.^56,57 

Contrasting findings with adult ECMO data may suggest differential mechanisms for the influence of SES or social disadvantage between children and adults.⁷  Adult studies consistently demonstrated lower mortality for patients on ECMO with private insurance and those residing in higher-income neighborhoods.^18–21  Although, 3 pediatric studies using zip-code-derived median income in Healthcare and Utilization Project datasets, which variably incorporated other SDoH, indicated no influence of SES on pediatric ECMO outcomes.^22–24  Our comprehensive analysis demonstrated comparable findings in children using a multidimensional neighborhood-level index as well as investigation of payer, race, ethnicity, need for an interpreter, urban residence, and distance from home to hospital. Differential impacts of SDoH between pediatric and adult ECMO outcomes warrant consideration and is likely multifactorial, given the unique biologic, social, and environmental factors specific to children and their health. One potential explanation is improved health insurance access through the 1997 Senate Children’s Health Insurance Program, which provided additional insurance to more than 7 million children.^58,59  Another potential protective factor is the overall improved survival of neonates and children supported on ECMO compared with adults.^60–62  Children may also have more reversible forms of cardiopulmonary decompensation given fewer chronic comorbidities. Additionally, given the relatively smaller population of critically ill children to adults, and infrequent use of this resource-intensive therapy, pediatric ECMO is highly centralized in large, academic centers. In contrast, small community hospitals can offer critical care services, including ECMO, to adult patients. Consolidation of healthcare resources promotes quality care for rare disease processes at the expense of timely geographic access.^63–67  This may be responsible for adult patients with relative social disadvantage being susceptible to adverse ECMO outcomes through inequitable access to quality healthcare.⁵⁹  Differences in admitting hospital quality have been considered as a potential systems-level factor that may influence resuscitation care.^59,68,69 

Several limitations must be considered when interpreting these results, which need to be validated in larger prospective observational studies with multimodal evaluation of SDoH. This is a single-center study from a quaternary care center with a highly protocolized ECMO program with circuits managed by dedicated staff. This limits generalizability to other pediatric in-hospital settings; however, this permits granular data collection for risk adjustment. Because of sample size limitations and to avoid over-fitting the model, we a priori identified factors most likely to confound the relationship between explanatory variables and outcome. We studied the influence of SDoH on ECMO utilization by investigating a comparison cohort of patients who died without ECMO support, which intuitively has great illness severity. Although precedent for this approach exists,¹⁷  we were unable to differentiate patients in this cohort who may have benefited from ECMO and reasons why ECMO was not offered. Prehospital factors, referral patterns, and geographic access as well as nongeographic factors influencing hospital access navigation are critically important. In addition, there are numerous unmeasured nonbiological confounders that are potential drivers of, or maskers, of inequitable outcomes including structural drivers, such as systemic manifestations of racism. These factors will be important considerations for future study. Furthermore, SDoH measures may be imprecisely collected in the EMR. Finally, although multidimensional SDoH indices have considerable utility, they encompass different populations within a geographic region.

Nonetheless, findings from this study and generated hypotheses identify a need to better understand relationships between SDoH and outcomes across all pediatric conditions, including in critical illness. Questions remain as to what factors in pediatric ECMO care may be protective against negative influences of social disadvantage, including the role of protocolization and the impact of health resource centralization. These findings encourage an emerging research agenda toward socially-conscious research for ECMO^7,70  and align with numerous organizations acknowledging the complex influence of SDoH on child health through social, physical, behavioral, and biologic mechanisms, as well as recommendations for rigorous investigation into how social constructs impact pediatric health.^71–74  Centers should critically and transparently reflect on their own practices of ECMO utilization and outcomes with an equity diversity and inclusivity lens for accountability.⁷⁵ 

We observed no difference in pediatric ECMO outcomes according to SDoH, including the Child Opportunity Index, a neighborhood-level multidimensional index, as well as race and ethnicity, payer, need for interpreter, urbanicity, and distance from home to hospital. These findings contrast starkly with extensive evidence supporting negative impacts of social disadvantage and minoritization on general pediatric health outcomes and adult ECMO outcomes. Further investigation of the complex interplay between social drivers and intricate relationships with health outcomes is needed to better understand the possible drivers and protective mechanisms involved and may guide interventions targeted at mitigating disparities across different populations and conditions.

We thank Manasee Godsay, Stephanie Carlisle, and Steven Brediger for their contributions with data collection.

BCH

Boston Children’s Hospital

COI

Child Opportunity Index

ECMO

extracorporeal membrane oxygenation

ELSO

extracorporeal life support organization

EMR

electronic medical record

ICD

International Classification of Diseases

IQR

interquartile range

LOS

length of stay

SDoH

social drivers of health

SES

socioeconomic status

VA

veno-arterial