Hospital variation in post-operative cardiac extracorporeal membrane oxygenation use and relationship to post-operative mortality

Abstract

Objective:

It is unclear how extracorporeal membrane oxygenation use varies across paediatric cardiac surgical programmes and how it relates to post-operative mortality. We aimed to determine hospital-level variation in post-operative extracorporeal membrane oxygenation use and its association with case-mix adjusted mortality.

Methods:

Retrospective analysis of 37 hospitals contributing to the Pediatric Cardiac Critical Care Consortium clinical registry from 1 August 2014 to 31 December 2019. Hospitalisations including cardiothoracic surgery and post-operative admission to paediatric cardiac ICUs were included. Two-level multivariable logistic regression with hospital random effect was used to determine case-mix adjusted post-operative extracorporeal membrane oxygenation use rates and in-hospital mortality. Hospitals were grouped into extracorporeal membrane oxygenation use tertiles, and mortality was compared across tertiles.

Results:

There were 43,640 eligible surgical hospitalisations; 1397 (3.2%) included at least one post-operative extracorporeal membrane oxygenation run. Case-mix adjusted extracorporeal membrane oxygenation rates varied more than sevenfold (0.9–6.9%) across hospitals, and adjusted mortality varied 10-fold (0–5.5%). Extracorporeal membrane oxygenation rates were 2.0%, 3.5%, and 5.2%, respectively, for low, middle, and high extracorporeal membrane oxygenation use tertiles (P < 0.0001), and mortality rates were 1.9%, 3.0%, and 3.1% (p < 0.0001), respectively. High extracorporeal membrane oxygenation use hospitals were more likely to initiate extracorporeal membrane oxygenation support intraoperatively (1.6% vs. 0.6% low and 1.1% middle, p < 0.0001). Extracorporeal membrane oxygenation indications were similar across hospital tertiles. When extracorporeal cardiopulmonary resuscitation was excluded, variation in extracorporeal membrane oxygenation use rates persisted (1.5%, 2.6%, 3.8%, p < 0.001).

Conclusions:

There is hospital variation in adjusted post-operative extracorporeal membrane oxygenation use after paediatric cardiac surgery and a significant association with adjusted post-operative mortality. These findings suggest that post-operative extracorporeal membrane oxygenation use could be a complementary quality metric to mortality to assess performance of cardiac surgical programmes.

Introduction

Extracorporeal membrane oxygenation is a life-saving therapy for a small percentage of patients after cardiac surgery. Appropriate extracorporeal membrane oxygenation deployment is key to optimising outcomes for children with cardiovascular disease. Type of heart disease, preoperative patient status, quality of preoperative imaging, preoperative decision-making, intraoperative care, technical quality of the operation, and the quality of post-operative intensive care all impact extracorporeal membrane oxygenation use. Hospitals likely adopt different thresholds for deploying extracorporeal membrane oxygenation after paediatric cardiac surgery, but differences in frequency of extracorporeal membrane oxygenation use across centres could alternatively indicate variation in quality of perioperative care by a paediatric cardiac surgical programme. Given the risks of extracorporeal membrane oxygenation, such as stroke and other organ injuries, it is reasonable to presume that limiting extracorporeal membrane oxygenation after paediatric cardiac surgery is desirable and low extracorporeal membrane oxygenation use rates may generally represent higher quality care.

However, many questions remain about variation in post-operative extracorporeal membrane oxygenation use and how this relates to overall quality and outcomes. Understanding the relationship between case-mix, extracorporeal membrane oxygenation use, and mortality is crucial when considering post-operative extracorporeal membrane oxygenation rate as a potential marker of programmatic quality. Extracorporeal membrane use is not necessarily an indication of worse performance in any individual case. Hospitals would ideally observe low extracorporeal membrane oxygenation use and low mortality. If this relationship were universal, it would be reasonable to conclude that low extracorporeal membrane oxygenation use is a marker of high programmatic quality. However, it is also possible that low mortality hospitals use extracorporeal membrane oxygenation more frequently to achieve lower mortality. Determining these associations between extracorporeal membrane oxygenation use and mortality, and how extracorporeal membrane oxygenation is deployed at low- and high-use hospitals, could provide hospitals with benchmark data and opportunities to improve outcomes. Previous investigation into this relationship has been limited. ^1,2

In this context, we aimed to explore hospital-level variation in extracorporeal membrane oxygenation use and its association with post-operative mortality after paediatric cardiac surgery. We used the Pediatric Cardiac Critical Care Consortium clinical registry to calculate case-mix adjusted post-operative extracorporeal membrane oxygenation use and mortality rates and compared these metrics within and across hospitals.

Materials and methods

Data source

Pediatric Cardiac Critical Care Consortium collects data on all patients with cardiac disease admitted to the cardiac ICU of participating hospitals as previously described. ³ At the time of this analysis, 37 hospitals were submitting cases to the registry. Each participating centre has a trained data manager who completes a certification exam and who collects and enters data in accordance with the standardised Pediatric Cardiac Critical Care Consortium Data Definitions Manual. ⁴ The Pediatric Cardiac Critical Care Consortium registry shares common terminology and definitions with applicable data points from the International Pediatric and Congenital Cardiac Code, Society of Thoracic Surgeons Congenital Heart Surgery Database, and American College of Cardiology Improving Pediatric and Adult Congenital Treatment Registry, as previously described. ³ Participating centres are audited on a regular schedule, and audit results suggest complete, accurate, and timely submission of data. ⁵ The University of Michigan Institutional Review Board provides oversight for the Pediatric Cardiac Critical Care Consortium Data Coordinating Center; this study was approved with waiver of informed consent (HUM00114217). Additionally, the Children’s Hospital of Philadelphia Institutional Review Board reviewed this study and deemed it not human subjects research as only de-identified data was provided to the investigators (Institutional Review Board #16-012833).

Patient population

Hospitalisations for all patients, including adults, undergoing cardiac surgery and recovering in a Pediatric Cardiac Critical Care Consortium cardiac ICU from 1 August 2014 to 31 December 2019 were included. Hospitalisations were included if a patient underwent a Society of Thoracic Surgeons index cardiac surgery with or without cardiopulmonary bypass at any time during the admission, exclusive of patients <2.5 kg undergoing isolated patent ductus arteriosus repair. ⁶ We also included those cardiac operations that could not be classified into Society of Thoracic Surgeons–European Association for Cardiothoracic Surgery mortality categories in our analyses because, from our previous work, these operations had a high extracorporeal membrane oxygenation use rate and we felt it important to characterise their outcomes. ⁷ For our calculation of post-operative in-hospital mortality, we modified our previously validated mortality case-mix adjustment model by including Society of Thoracic Surgeons–European Association for Cardiothoracic Surgery non-classifiable operations in order to be consistent with our study cohort and case-mix adjustment model for extracorporeal membrane oxygenation use. ⁸

Data collection

Variables included in this analysis are strictly defined in the Pediatric Cardiac Critical Care Consortium data definitions manual. Demographics, non-cardiac comorbidities, and other patient factors necessary for risk adjustment were collected relative to the index operation for each hospitalisation. Surgical complexity was characterised using Society of Thoracic Surgeons–European Association for Cardiothoracic Surgery mortality categories as above. The time of extracorporeal membrane oxygenation initiation relative to the patient’s index operation was calculated. Extracorporeal membrane oxygenation runs initiated intraoperatively during the index procedure or at any time in the post-operative period were included.

Statistical analysis

A hospitalisation was the primary episode of analysis. Patient characteristics and outcomes are described using standard measures of central tendency based on the distribution of the data. The two outcomes of interest were (1) case-mix adjusted extracorporeal membrane oxygenation use and (2) case-mix adjusted post-operative in-hospital mortality, both at the hospital level.

In a prior analysis, we identified patient, operative, and illness severity factors associated with post-operative extracorporeal membrane oxygenation use. ⁷ The supplemental material lists these variables. Each of these independent variables was assessed prior to or immediately after the index operation. We fit a two-level multivariable logistic regression model with hospital random effect to obtain the expected probability of post-operative extracorporeal membrane oxygenation use based on the included covariates. We averaged the expected probability of post-operative extracorporeal membrane oxygenation use within each hospital to obtain hospital-level expected extracorporeal membrane oxygenation use and calculated an adjusted rate based on observed to expected.

Similarly, we created a case-mix adjusted model for in-hospital mortality based on previously published literature and calibrated within the Pediatric Cardiac Critical Care Consortium database using a similar method. ^8,9 The supplemental material lists the included variables. This model was used to calculate expected mortality at the hospital level.

Each model was validated using bootstrap resampling (1000 samples) to obtain bias-corrected confidence intervals for the odds ratios corresponding to each model covariate. From the bootstrapping, we also calculated an optimism-corrected c-statistic for the final model. Model calibration was assessed using the Hosmer–Lemeshow X ² test (p >0.05 demonstrating adequate fit). The final model for adjusted post-operative extracorporeal membrane oxygenation use showed excellent model discrimination (optimism-corrected C-statistic of 0.87) and was well-calibrated (Hosmer–Lemeshow X ² = 10.5, p = 0.23, Supplemental Figure 1(a). Similarly, the final model for post-operative in-hospital mortality showed excellent model discrimination (optimism-corrected C-statistic of 0.86) and was well-calibrated (Hosmer–Lemeshow X ² = 14.81, p = 0.0630, Supplemental Figure 1(b).

From our original bootstrapping, we obtained bias-corrected 95% confidence intervals around the adjusted rates for each metric at each hospital. Hospitals whose confidence intervals did not cross the overall observed rates were considered to demonstrate statistically significant lower-than-expected or higher-than-expected extracorporeal membrane oxygenation use or mortality given that hospital’s mix.

We then compared aggregated post-operative extracorporeal membrane oxygenation use and mortality rates across hospitals to determine the association between extracorporeal membrane oxygenation use and mortality. Hospitals were grouped into tertiles—low, middle, and high—based on adjusted post-operative extracorporeal membrane oxygenation use as previously described when comparing two aggregated hospital-level post-operative outcomes. ^10–12 Cut-offs for the tertiles were based on creating low and high extracorporeal membrane oxygenation—use groups that were clinically different from each other with ranges that had ≥2% difference in extracorporeal membrane oxygenation use rates. The low tertile was an extracorporeal membrane oxygenation use rate ≤ 2.5%, middle 2.6–4.4%, and high ≥4.5%. Similarly, we compared observed and adjusted post-operative mortality across extracorporeal membrane oxygenation use tertiles within hospitalisation subgroups that included extracorporeal membrane oxygenation and those that did not.

Finally, we performed two sensitivity analyses. First, we calculated extracorporeal membrane oxygenation use rates with and without extracorporeal cardiopulmonary resuscitation cases. We explored this because extracorporeal cardiopulmonary resuscitation cases do not reflect elective extracorporeal membrane oxygenation deployment across hospitals, a potential explanation of variation in extracorporeal membrane oxygenation use. Second, we analysed the data with and without a surgical volume variable to determine whether this associated with extracorporeal membrane oxygenation use.

Analyses were performed using SAS Version 9.4 (SAS Institute, Cary, NC) or STATA Version 14 (Stata Corp, College Station, TX), with statistical significance at a p value of <0.05.

Results

Our cohort included 43,640 cardiac surgical hospitalisations, of which 1397 (3.2%) included at least one post-operative extracorporeal membrane oxygenation run. Median surgical case volume for Society of Thoracic Surgeons index operations was 221 (interquartile range 157–323) across 37 participating hospitals. Median annual postoperative extracorporeal membrane oxygenation caseload was seven (interquartile range 5–13) cases.

Table 1 shows characteristics of the study population. Post-operative extracorporeal membrane oxygenation runs were more likely to occur after Society of Thoracic Surgeons–European Association for Cardiothoracic Surgery 4, Society of Thoracic Surgeons–European Association for Cardiothoracic Surgery 5, and non-classifiable operations. Cardiopulmonary bypass (CPB) times for index operations were nearly double for extracorporeal membrane oxygenation hospitalisations compared to hospitalisations that did not include extracorporeal membrane oxygenation (185 vs. 102 minutes).

Table 1. Patient characteristics

STS = Society of Thoracic Surgeons; STAT, Society of Thoracic Surgeons–European Association of Cardiothoracic Surgeons Mortality Categories; IQR = interquartile range; CICU = cardiac intensive care unit.

* Composite of cardiopulmonary resuscitation, shock at time of surgery, hepatic dysfunction, stroke within 48 hours of surgery, and/or renal failure requiring dialysis.

Variation in extracorporeal membrane oxygenation use and mortality across hospitals

Figure 1(a) shows case-mix adjusted post-operative extracorporeal membrane oxygenation use rates by hospital. We observed more than sevenfold variation in adjusted extracorporeal membrane oxygenation use across hospitals with rates ranging from 0.9% to 6.9%. There were six hospitals (16.2%; hospitals 1, 3, 4, 8, 9, 13) that had lower-than-expected extracorporeal membrane oxygenation use. There were nine hospitals (24.3%; hospitals 28, 29, 31–37) that had statistically significant higher-than-expected extracorporeal membrane oxygenation use.

Figure 1. Case-mix adjusted ECMO use (a) and surgical mortality (b) by hospital in post-operative surgical patients. 95% CI. Hospital ID numbers are the same in both figures. In each figure, green dots represent hospitals with lower-than-expected ECMO rates, and red dots represent hospitals with higher-than-expected ECMO rates. Blue horizontal line shows the overall mean. ECMO=extracorporeal membrane oxygenation; CI=confidence interval.

Figure 1(b) shows case-mix adjusted mortality rates by hospital. Rates ranged from 0% to 5.5%. There were five hospitals (13.5%; hospitals 2, 5, 12–14) that had lower-than-expected case-mix adjusted mortality in the setting of lower-than-expected extracorporeal membrane oxygenation use, and three hospitals (8.1%; hospitals 29, 35, 36) had higher-than-expected case-mix adjusted mortality in the setting of higher-than-expected extracorporeal membrane oxygenation use. There was one hospital (37) that had higher-than-expected extracorporeal membrane oxygenation use with lower-than-expected case-mix adjusted mortality.

Extracorporeal membrane oxygenation use and mortality across hospital tertiles

Aggregate adjusted extracorporeal membrane oxygenation use rates across hospital tertiles were 2.0%, 3.5%, and 5.2% (p < 0.0001 for all comparisons). Figure 2(a) shows overall case-mix adjusted mortality rates across hospital tertiles grouped by case-mix adjusted extracorporeal membrane oxygenation use rates. Expected mortality rates across the low, middle, and high extracorporeal membrane oxygenation use tertiles were 2.2–2.4%, suggesting that case mix was similar. The low extracorporeal membrane oxygenation use hospitals had an overall adjusted mortality rate of 1.9% compared to the middle extracorporeal membrane oxygenation use (3.0%) and high extracorporeal membrane oxygenation use hospitals (3.1%), p < 0.0001. Figure 2(b) shows case-mix adjusted mortality rates across extracorporeal membrane oxygenation use tertiles with extracorporeal cardiopulmonary resuscitation cases excluded. The same relationship of overall case-mix adjusted extracorporeal membrane oxygenation use and mortality persists. Supplemental Figures 2(a) and 2(b) show observed and case-mix adjusted mortality rates for extracorporeal membrane oxygenation and non-extracorporeal membrane oxygenation hospitalisation subgroups. Observed and case-mix adjusted post-operative mortality for the extracorporeal membrane oxygenation cohort are significantly higher than for hospitalisations without extracorporeal membrane oxygenation, consistent with extracorporeal membrane oxygenation mortalities largely accounting for differences in overall mortality.

Figure 2. Case-mix adjusted ECMO use rates and case-mix adjusted surgical mortality across hospital tertiles of ECMO use (a) and sensitivity analysis of case-mix adjusted ECMO use rates and case-mix adjusted surgical mortality across hospital tertiles excluding ECPR cases (b). ECMO = extracorporeal membrane oxygenation; ECPR=extracorporeal cardiopulmonary resuscitation.

Supplemental Table 1 shows the timing of initial extracorporeal membrane oxygenation cannulation and extracorporeal membrane oxygenation indications by extracorporeal membrane oxygenation use tertiles. Intraoperative extracorporeal membrane oxygenation initiation was more likely at high extracorporeal membrane oxygenation use hospitals (1.6% high, 1.1% middle, 0.6% low, p < 0.001). Extracorporeal cardiopulmonary resuscitation rates were similar across all tertiles. Supplemental Table 2 shows patient and operative characteristics by extracorporeal membrane oxygenation use tertiles. Society of Thoracic Surgeons–European Association for Cardiothoracic Surgery category frequencies were clinically similar across all tertiles. Post-operative mechanical ventilation rate was higher at low extracorporeal membrane oxygenation use hospitals (71.8% vs. 62.2–65.3%, p < 0.0001). CPB times were longer in extracorporeal membrane oxygenation patients compared to CPB times in the overall cohort regardless of tertile. High extracorporeal membrane oxygenation use hospitals had greater numbers of both extracorporeal cardiopulmonary resuscitation and non-extracorporeal cardiopulmonary resuscitation extracorporeal membrane oxygenation runs.

Relationship with surgical volume

Figure 3 shows a funnel plot of case-mix adjusted extracorporeal membrane oxygenation use rate along the continuum of hospital surgical volume. Table 2 shows the variables associated with mortality from logistic regression with hospital case-mix adjusted extracorporeal membrane oxygenation use rate, hospital surgical volume, patient, and operative factors included. Case-mix adjusted extracorporeal membrane oxygenation use rate was associated with mortality (Odds Ratio (OR) 1.15 95% confidence interval 1.05–1.26). Annual average surgical volume was not associated with mortality.

Figure 3. Funnel plot for case-mix adjusted ECMO use rate based on hospital surgical volume. ECMO = extracorporeal membrane oxygenation.

Table 2. Logistic regression model of mortality

Discussion

We found wide variation in case-mix adjusted post-operative extracorporeal membrane oxygenation use after paediatric cardiac surgery in a multicentre cohort. We also demonstrated an association between post-operative extracorporeal membrane oxygenation use and mortality with low-extracorporeal membrane oxygenation use hospitals demonstrating significantly lower case-mix adjusted mortality than high-extracorporeal membrane oxygenation use hospitals. The totality of these findings support using adjusted post-operative extracorporeal membrane oxygenation use rate as an indicator of perioperative quality of care at paediatric cardiac surgical programmes.

Our findings confirm those of two previous studies demonstrating extracorporeal membrane oxygenation use variation in post-operative cardiac patients with this analysis filling important gaps. Using the Society of Thoracic Surgeons database and adjusting for patient characteristics and case mix, Mascio et al. found 15-fold variation in post-operative extracorporeal membrane oxygenation use at 80 hospitals. ² Adjusted extracorporeal membrane oxygenation rates were also compared to the hospital’s surgical volume with no significant relationship found. The association between extracorporeal membrane oxygenation use and mortality was not evaluated in Mascio’s study.

Another analysis, using administrative data from the Pediatric Health Information System database, also demonstrated varying rates of case-mix adjusted extracorporeal membrane oxygenation use. ¹ Across 43 hospitals, the authors illustrated substantial variation in extracorporeal membrane oxygenation use rates and mortality. They found a linear relationship between extracorporeal membrane oxygenation use and extracorporeal membrane oxygenation surgical mortality. They concluded that overall case-mix adjusted surgical mortality decreased from 5.0% to 3.5% in the presence of extracorporeal membrane oxygenation; however, to make this conclusion, they assumed that all patients who survived on extracorporeal membrane oxygenation would have died in the absence of extracorporeal membrane oxygenation, which was an acknowledged major limitation of their study. Further, the limitations of administrative data with regard to accurately classifying paediatric cardiac surgical procedures and analysis of associated outcomes have been previously described by Pasquali et al. ^13,14

Our data come from the Pediatric Cardiac Critical Care Consortium clinical registry which includes granular clinical data allowing for the use of risk-adjustment methodology previously validated in the Society of Thoracic Surgeons Congenital Heart Surgery Database and in other Pediatric Cardiac Critical Care Consortium analyses. ^8,15 We were able to adjust for several demographic and clinical factors relevant to both extracorporeal membrane oxygenation use and mortality. Our results extend the findings of Bratton et al. and suggest that the highest performing programmes have both low adjusted mortality and low adjusted extracorporeal membrane oxygenation use.

There are several reasons why post-operative extracorporeal membrane oxygenation use would vary across paediatric cardiac surgical hospitals. Consensus criteria do not exist for post-operative extracorporeal membrane oxygenation support. The degree to which a perioperative team tolerates physiologic derangement, compromised oxygen delivery, vasoactive support, and organ impairment likely differs within and across institutions. The availability of ICU resources may also impact the decision to place a patient on extracorporeal membrane oxygenation. However, it is also possible that extracorporeal membrane oxygenation rates reflect the overall quality of care, including pre-, intra-, and post-operative phases. As such, extracorporeal membrane oxygenation use rates may be an indicator of the overall quality of a paediatric cardiac surgical programme including cardiology, anaesthesia, perfusion, surgery, and intensive care teams. Supporting this, analyses excluding extracorporeal cardiopulmonary resuscitation runs showed variation in extracorporeal membrane oxygenation use and increased mortality in high extracorporeal membrane oxygenation use centres confirming that mortality differences across tertiles were driven by both extracorporeal cardiopulmonary resuscitation and non-extracorporeal cardiopulmonary resuscitation indications. Further, high extracorporeal membrane oxygenation use hospitals performed more intraoperative initiation of extracorporeal membrane oxygenation, suggesting that the mechanism of differences in extracorporeal membrane oxygenation use may be in part due to pre- and intraoperative care. It is reasonable to hypothesise that annual hospital surgical volume would affect extracorporeal membrane oxygenation use and outcomes; however, our analyses did not show that surgical volume impacted differences in either extracorporeal membrane oxygenation use or mortality, consistent with previously published data. ²

Hospital #37 had greater-than-expected extracorporeal membrane oxygenation use yet low case-mix adjusted mortality, contrary to the prevailing relationship in our results. We sought to understand their unique practice in more detail. The team at this hospital described their philosophy to deploy extracorporeal membrane oxygenation early with a substantial number of intraoperative cannulations. There are no formal criteria for extracorporeal membrane oxygenation deployment, but their practice is aimed at avoiding extracorporeal cardiopulmonary resuscitation. The extracorporeal cardiopulmonary resuscitation rate at their centre is <10%, much lower than the cohort extracorporeal cardiopulmonary resuscitation rate of 31.6% [personal communication]. Staffing includes an in-house attending cardiac intensivist 24/7. The surgical team is not in-house at night. They also favour early cardiac catheterisation to identify and address residual lesions in extracorporeal membrane oxygenation patients. Catheterisation is usually performed within 1–2 days of cannulation with the catheterisation team often able to perform necessary interventions. Most who undergo catheterisation on extracorporeal membrane oxygenation survive to decannulation and to discharge.

Our methods provide benchmark data for hospitals to compare extracorporeal membrane oxygenation use rates relative to what is expected for their patient populations. Pediatric Cardiac Critical Care Consortium data are available on a real-time reporting platform to participating sites and are presented in an internally unblinded fashion, thus allowing hospitals to engage in collaborative learning with high performing peers. Given the relationship to post-operative mortality, this can be a complementary quality metric for hospitals looking to assess programme performance across all perioperative phases. Incorporating adjusted extracorporeal membrane oxygenation use rates into our quality armamentarium is consistent with calls in the scientific and clinical communities to develop non-mortality metrics of perioperative quality of care.

As with any analysis using clinical registry data, we were limited to study variables that exist in the Pediatric Cardiac Critical Care Consortium database. We utilised the variables available to us but cannot rule out residual confounding even though we used models based on well-validated methods. The lack of certain data points, such as patient status immediately prior to extracorporeal membrane oxygenation, is particularly relevant to studying adjusted mortality of extracorporeal membrane oxygenation patients; to properly risk adjust, this outcome would require detailed data measuring severity of illness and medical management at the time of extracorporeal membrane oxygenation initiation. Linkage between the Extracorporeal Life Support Organization database or other data sets and the Pediatric Cardiac Critical Care Consortium registry could support such an analysis. There is a need to understand the mechanistic drivers of increased extracorporeal membrane oxygenation use, such as residual lesions and need for reintervention, which were not done in this study. Pediatric Cardiac Critical Care Consortium is currently studying unplanned re-operation and how that impacts outcomes. Additionally, mortality is only one important outcome for these complex patients. Other outcomes to be considered in follow-up studies include short- and long-term morbidity, most importantly neurologic injury, neurodevelopment, and functional status. Finally, we don’t know the mechanism(s) of increased mortality shown here as this data set does not allow us to show causation. This deserves investigation in a follow-up study.

Pediatric Cardiac Critical Care Consortium hospitals may not be representative of all hospitals that perform cardiac surgery as our hospitals’ case volumes are greater on average than those in the Bratton study ¹ . Also, all of the hospitals in this study have cardiac ICUs and perform the full spectrum of cardiac surgical complexity which might not generalise to other types of hospitals. Finally, we are not able to provide a clear mechanism for why some hospitals are able to achieve better outcomes. To improve the lives of children with critical cardiovascular disease, we must fill this knowledge gap with further study.

Conclusions

We have shown that post-operative extracorporeal membrane oxygenation use varies greatly across Pediatric Cardiac Critical Care Consortium hospitals and that low extracorporeal membrane oxygenation use hospitals have significantly lower case-mix adjusted post-operative mortality. Thus, adjusted extracorporeal membrane oxygenation use rate is a complementary quality metric to post-operative mortality for assessing perioperative care at children’s hospitals. The highest quality care ideally results in both low mortality and appropriately low extracorporeal membrane oxygenation use. Several hospitals in Pediatric Cardiac Critical Care Consortium demonstrate that this is achievable. In future work, we will aim to identify organisational factors and care processes of hospitals who achieve these results, disseminate that knowledge, and engage in quality improvement efforts to develop measures of appropriate extracorporeal membrane oxygenation use, thus improving outcomes across hospitals in order to improve the lives of children with cardiovascular disease.