Secondary Logo

Journal Logo

Adult Circulatory Support

Volume–Outcome Relationships in Extracorporeal Membrane Oxygenation: Retrospective Analysis of Administrative Data From Pennsylvania, 2007–2015

Huesch, Marco D.

Author Information
doi: 10.1097/MAT.0000000000000675

Abstract

In patients with deterioration of respiratory or cardiac function, use of short-term extracorporeal membrane oxygenation (ECMO) is commonly accepted as a standard of care.1 This technology has been rapidly adopted internationally2 and in the United States, with increasing numbers of institutions performing these procedures on increasing numbers of patients.3

In this context of rapid market entry and increasing volumes, an open question is whether this rapid growth in ECMO has led to better outcomes as “practice makes perfect” in surgical fields at the individual or operator level.4–6 In other fields such as cardiac surgery,7,8 learning may be so fast that there are no apparent returns to greater experience. Moreover, in related critical care settings, wide variability in the use of intensive care has been found.9 Such variability could mask any effects of learning in ECMO.

In the first large empirical investigation in ECMO, Barbaro et al.10 used a large international registry-based study and found that ECMO in institutions with greater current volume was associated with significantly lower mortality. A study by McCarthy et al.11 using only US data found paradoxically better mortality among smaller annual volume facilities.

However, current volume is also a proxy of scale, an independent predictor of outcomes.4,5,12,13 Cumulative volume may be a closer proxy for the learning by doing phenomenon. We sought to identify whether either cumulative volume or current volume was associated with better risk-adjusted outcomes in a recent 9 year period in one US state. We used administrative discharge data with very limited additional chart data items. We hasten to acknowledge the very well-known limitations of such data compared with registry or electronic medical record data.14

Methods

Based on the available literature, we prespecified three hypotheses that guided our data collection and study design:

  1. [State-level mortality] Over time, there would be a reduction in overall risk-adjusted mortality in the state because of either a learning or scale effects;
  2. [Hospital experience] For individual facilities, that facility’s cumulative risk-adjusted mortality would be inversely related to that facility’s cumulative experience;
  3. [Hospital current scale] For individual facilities, in any calendar year, that facility’s risk-adjusted mortality would be inversely related to that facility’s volume in that year.

We obtained administrative data from the Pennsylvanian state agency Pennsylvania Health Care Cost Containment Council (PHC4), which is the state regulator for all hospitals that are obliged to report to the state of Pennsylvania. These data, thus, do not include the federal Veterans Affairs hospitals or federal prison hospitals.

Data Used

Our data comprised all 2,948 consecutive adult ECMO admissions billed between January 1, 2007, and December 31, 2015, using diagnosis related grouper (DRG) 3 and with procedural codes 39.65 international classification of diseases, 9th revision (ICD-9) and 5A15223 (ICD-10) for ECMO. The unit of analysis was the hospitalization. This data contained admission type (emergent, urgent, and routine); whether the admission was through the emergency department (ED), referral, or transferred in; ethnicity; race; sex; age; county and state of patient; hospital type (bed size, specialty, or general acute care); admission time and day of week; insurance status; and discharge disposition.

PHC4 requires hospitals to report a limited number of laboratory tests early in the admission. For patients admitted before 6 pm, the highest or lowest test result on that calendar day must be reported. For patients admitted after 6 pm, highest or lowest test result on that calendar day and the next calendar day must be reported. These test results, therefore, span a variable period of 24 hours through 30 hours.

Early admission laboratory tests included maximum and minimum values of white blood cells, hemoglobin, band neutrophil percentage, sodium, potassium, glucose, calcium, base units, pH, and pCO2; maximum values of prothrombin time, international normalized ratio, partial thromboplastin time, blood urea nitrogen (BUN), creatinine, bilirubin, aspartate, alkaline phosphatase, creatinine kinase, creatinine kinase MB, troponin I, troponin T, brain natriureteric peptide (BNP), and bicarbonate; and minimum values of platelets, albumin, O2 saturation and pO2.

Where variables contained missing values, we constructed a separate missing indicator and set the missing laboratory value to zero. Missingness was least for hemoglobin (76% of patients had data for this) and most for BNP (91% of patients had no data for this). Our case-mix variables also did not contain any relevant vital signs such as Glasgow coma score, weight, height, body mass index, temperature, blood pressure, heart rate, and respiratory rate. Importantly, this dataset also did not capture lactate values that have been shown in the prediction of cardiogenic shock outcome for AMI patients salvaged by VA-ECMO (ENCOURAGE) score model15 and by others to be associated with mortality.16

Patient-Level Mortality Model

Given the data constraints, we were unable to construct risk models incorporating the breadth and depth of clinical data as others have generally been able to.17–22 We conservatively performed a backwards step-wise single-level logistic regression on in-hospital mortality. The significance criterion for retention in the logistic model was a p value of 0.10 for the variable under consideration for discarding. We used a non-hierarchical model as this has been found to be more conservative in modeling surgical outcomes,23 compared with more complex multilevel hierarchical models.24

The final model had 31 covariates (see Table 1, Supplemental Digital Content, https://links.lww.com/ASAIO/A197). Interpretation of the impact of these predictors is not the main focus of the article, but these generally make sense. For example, the higher the early admission maximum value of pH, the lower the odds of in-hospital mortality. If a hospital tests and reports band neutrophils, making these available, that raises the odds of death. Some predictors do not immediately make sense. For example, being admitted on an emergent or urgent basis appear protective with lower odds for in-hospital death, compared with being a routine admission. We speculate that the use of ECMO later in a routine admission that “went wrong” may be driving that apparent increase in risk. Finally, some predictors, like uninsured status, seem unusually strongly associated with mortality. These may be correlated with unobserved, mediating variables such as care before admission or multiple comorbidities.

Despite these concerns and despite the acknowledged inadequacy, in general, of administrative data predictors, the model was nevertheless reasonably well fitting with a pseudo R2 of 11.7%. The risk model discriminated acceptably with an area under the receiver operating characteristic (ROC) curve of 0.72. To put this into perspective, this comports well with five recent survival risk models with generally excellent clinical chart data. Schmidt et al.’s17 respiratory ECMO survival prediction (RESP) score had an area under the ROC curve of 0.74, Klinzing et al.18 had 0.67 amalgamating cannulation modes, Huang et al.19 had a range of 0.76–0.84, Chen et al.20 has 0.65 for sequential organ failure assessment (SOFA) without lactate (our model also did not include lactate), and Kim et al.21 had 0.83 for the simplified acute physiology score II (SAPS II) model.

The risk model was also well calibrated with an insignificant Hosmer–Lemeshow test and good graphical calibration (Figure 1). A perfect model would lie on the 45 degree line. In such a model, patients estimated to have a higher likelihood of dying would actually be observed to die at that higher rate. Our model approaches this and generally lies on or close to the 45 degree line.

Figure 1.
Figure 1.:
Calibration of risk model showing predictions generally matching with actual mortality. DRG, diagnosis related grouper.

The risk model’s classification was adequate with a sensitivity of 68% for in-hospital mortality, a specificity of 64%, and overall accuracy of 66% at a threshold of 0.50, close to the average mortality. These model performance statistics suggest that the model was able to reasonably well predict mortality and adjust for differential patient baseline risk.

State-Level Measurements

We used the patient-level model to calculate individual predicted probabilities of in-hospital death for every patient, i.e., expected risk. We aggregated these at the state level for each calendar quarter in the dataset. We also summed up total observed mortality at the state level for each calendar quarter. We divided these two measurements to obtain a mortality index (i.e., an observed/expected ratio (O/E ratio)) in each quarter, to understand evolution in risk-adjusted mortality over time in the whole state under hypothesis 1.

Hospital-Level Measurements

We aggregated the same patient-level individual predicted probabilities of in-hospital death to the hospital level. For each hospital, we aggregated their patients’ predicted probabilities, their patients’ actual observed mortality, and their total number of adult ECMO cases. For hypothesis 2, we considered all patients seen in a hospital from the time first observed to perform ECMO in our dataset (mid to end of 2007) to the time last observed to perform ECMO (for larger hospitals this was end 2015, but some hospitals exited the ECMO business earlier). For this hypothesis, we calculated cumulative volume of adult ECMO and cumulative risk-adjusted mortality for each hospital.

For hypothesis 3, we considered all patients in a hospital receiving ECMO throughout a calendar year and calculated total cases performed in that year and total risk-adjusted mortality in that year. We performed this calculation separately for each observed year of operation to a maximum of 9 years.

We used Pearson’s correlation to test for linear associations between hospital aggregate outcomes and cumulative volume, annual volume, or time. We graphed all relationships using overlaid fractional polynomial regression predictions to account for any nonlinearity. This study was approved by our Institution’s College of Medicine’s Institutional Review Board.

Results

Selected characteristics of the patients show that a typical patient was a middle-aged male staying 3 weeks in hospital with a slightly greater than even chance (51.7%) of dying in hospital (Table 1). In this state in general, and in this clinical setting in particular, there was less than average representation by Asian or Hispanic minorities.

Table 1.
Table 1.:
Selected Baseline Characteristics of ECMO Patients in Pennsylvania, 2007–2015

The number of facilities performing adult ECMO tripled from 10 in the last quarter of 2007 to 31 in the last quarter of 2015 (see Figure 1, Supplemental Digital Content, https://links.lww.com/ASAIO/A199). Entry and withdrawal of facilities from the market led to a total of 60 distinct facilities billing for ECMO over the 36-quarter period. There was a total of 608 facility*quarters and a total of 278 facility*years in the data.

Overall, statewide quarterly adult ECMO case volume sextupled from 25 to 186 per calendar quarter in the 9 year study period, in line with the substantial national increases in the use of this technology (see Figure 2, Supplemental Digital Content, https://links.lww.com/ASAIO/A200).

Hypothesized Reduction in Statewide Aggregate Risk-Adjusted Mortality over Time

Over the 36 calendar quarters observed, the in-hospital observed mortality declined from 72.0% to 47.9% (Figure 2A). However, expected mortality declined similarly from 68.9% to 49.2% (Figure 2B). Both of these reductions were significant over time with Pearson correlation −0.65 and −0.83, both p < 0.001.

Figure 2.
Figure 2.:
Statewide trends over time. (A) In unadjusted observed mortality showing systematic decrease. B: Statewide trend in expected mortality showing systematic decrease over time. C: Statewide trend in risk-adjusted mortality showing no systematic change over time.

However, their quotient failed to show a significant decline (Figure 2C) with risk-adjusted mortality falling from 1.05 to 0.97 statewide over the quarters of the data and an insignificant Pearson correlation of −0.02 (p = 0.62). Fitting a fractional polynomial regression to this time series to account for nonlinearity revealed essentially a best fit line around one.

Hypothesized Reduction in Facility Risk-Adjusted Mortality with Greater Experience

Across the 60 facilities observed over 9 years or 36 possible quarters, the median number of quarters observed was 7, with interquartile range of 3–13. There were four facilities in operation for 32 quarters or more, which saw the majority of patients (86.2%). There were substantial differences in risk-adjusted mortality by experience. Among patients whose hospitals were in the first quartile, the O/E ratio was 0.66; in the interquartile range, this was 0.67; and in the highest quartile, this was 1.02.

Across all facilities (Figure 3), there was a significant positive association between cumulative volume and risk-adjusted mortality (Pearson correlation for linear trend 0.37; p < 0.001). However, as the figure shows, this trend was nonlinear. Smaller institutions with less than 50 total cases had highly variable performance that tended to be below the aggregate O/E ratio of one. Above 50 cumulative cases, risk-adjusted mortality was essentially flat.

Figure 3.
Figure 3.:
Facility cumulative volume and cumulative risk-adjusted mortality showing little improvement in outcomes beyond 50 cumulative procedures. CI, confidence interval; ECMO, extracorporeal membrane oxygenation.

Analysis of the components of the O/E ratio (see Figures 3 and 4, Supplemental Digital Content, https://links.lww.com/ASAIO/A201) suggests that observed mortality differences were driving the O/E ratio differences, with expected mortality steady over the range of cumulative volume through 600 ECMO procedures.

Hypothesized Reduction in Facility Risk-Adjusted Mortality with Higher Annual Volume

Considering calendar year annual volume, there were a total of 278 facility*calendar units. Among these, a similarly significant curvilinear relationship (Figure 4) was found between annual volume and annual risk-adjusted mortality. As shown above, above a threshold, this relationship failed to continue. Beyond 10 annual cases, risk-adjusted mortality was essentially flat. Analysis of the components of the O/E ratio (see Figures 5 and 6, Supplemental Digital Content, https://links.lww.com/ASAIO/A202) suggests that again observed mortality differences were driving the O/E ratio differences, with expected mortality steady over the range of annual volume through 100 ECMO procedures.

Figure 4.
Figure 4.:
Facility annual volume and risk-adjusted mortality showing little improvement in outcomes beyond 10 annual procedures. CI, confidence interval; ECMO, extracorporeal membrane oxygenation.

Post Hoc Analyses of Patients at High and Low Cumulative Experience Facilities

Given the unexpected finding of no systematic inverse relationships between scale or experience and risk-adjusted outcomes, we performed several post hoc analyses. These analyses sought to understand any systematic patient-level differences that might account for the lower expected risk among patients receiving an ECMO run in a smaller facility or one with less experience.

In Table 2, we show that across some important laboratory test values from the initial 24 to 30 hours of a patient’s admission, patients in hospitals with more than 50 cumulative cases appear somewhat sicker than patients in hospitals with fewer than 50 cumulative cases. Hemoglobin and glucose values are significantly lower, for both minimum and maximum values, and maximum BUN, creatinine, and clotting times are significantly higher for the higher experience hospitals. We also find that oxygen saturation and partial pressures are better among the higher experience hospitals, perhaps reflecting greater supplemental oxygen support.

Table 2.
Table 2.:
Patient Characteristics by Facility Volume, Less Than or More Than 50 Cumulative Procedures

There is a strong association between cumulative volume and cumulative quarters in existence. Therefore table 2 (see Table 2, Supplemental Digital Content, https://links.lww.com/ASAIO/A198) shows a very similar result: patients in the four hospitals in operation 32 or more calendar quarter appear somewhat sicker early in their admission than the many in operation for less time.

Separately, we considered admitting diagnoses by cumulative volume to understand whether the etiology of ECMO initiation differed at less experienced hospitals. The combination of lower observed mortality and more potential respiratory indications, which generally have better mortality, would be consistent with different etiologies of ECMO initiation in different types of hospitals. We found that among patients seen at hospitals with fewer than 50 ECMO cases ever performed, there were significantly fewer respiratory admitting diagnoses (20.0% vs. 29.8%; p < 0.001) and significantly more cardiac admitting diagnoses (45.9% vs. 38.9%; p < 0.001), compared with among patients seen at more experienced hospitals.

In this same line of inquiry, we found that although patients with congestive heart failure or cardiogenic shock were similarly distributed across lower and higher cumulative volume facilities (44.9% vs. 43.6%; p = 0.57), there were substantially more postcardiotomy patients in the lower volume facilities (159 of 521 patients, 30.5%), proportionately, than in the higher cumulative volume facilities (464 of 2427, 19.1%; p < 0.001). Because postcardiotomy patients tend to do better after ECMO, this may help to explain the paradoxically lower risk-adjusted mortality among patients in smaller facilities.

Finally, we examined length of stay (LOS) as a proxy for initiation of ECMO and early transfer to a larger center. The combination of lower mortality through discharge and lower length of stay would be consistent with live transfer out to a larger center. Among patients seen at hospitals with fewer than 50 ECMO cases ever performed, median LOS was 5 and average LOS was 11.0, significantly lower when compared with 16 and 23.5, respectively, among patients seen at hospitals with greater ECMO cumulative volume (p values for both median test and t-test < 0.001).

Discussion

ECMO is a well-established and relatively safe rescue technology,25 approved by use by peer societies and rapidly and widely adopted in the last decade. Despite the high mortality, this technology offers hope to patients with respiratory distress (e.g., from influenza), cardiogenic shock (e.g., from a heart attack or postoperatively after cardiac surgery or from a cardiac arrest), or both (e.g., patients with sepsis), until their native organ function improves or until artificial or transplant organs become available.

In this setting of rapid growth in volumes, we expected to observe volume–outcome relationships driven by learning or scale effects at the individual hospital level or at the statewide level. We hypothesized that improvements in risk-adjusted mortality would ensue because of one or more of these mechanisms. These prespecified hypotheses are buttressed by decades of research on the volume–outcome relationship,4 which generally finds some improvement in outcomes with more volume, or deteriorations in outcomes with less volume or after breaks.26

However, none of these hypotheses were supported in this study. Aggregated across all facilities and over the 9 year period, risk-adjusted mortality did not decrease significantly. While observed mortality fell, expected mortality fell in line with this.

This could represent expansion of ECMO within patient subsets at lower risk as physicians become more comfortable using a rescue technology more broadly. It could also represent intensivists working down an appropriateness curve toward patients that may, objectively, not need ECMO support. Our data with its limited information on patient acuity does not allow further exploration of these potential effects. Further research is needed, using clinical chart data to compare risk-adjusted outcomes over time in homogenous subsets of patients.

No previous study in ECMO has looked at cumulative experience as a driver of mortality, but our results found no role for this. This was unexpected given the ramping up “from zero” and the wide variation in total experience observed in this dataset across 60 different facilities. Unlike the major international registry–based study by Barbaro et al.,10 we found no systematic benefit to scale. We are unable to explain whether this is because of vintage of patients captured in the extracorporeal life support organization (ELSO) registry or international differences in indications or infrastructure.

However, as McCarthy et al.27 have also recently found, we paradoxically observed lower risk-adjusted mortality among low-volume facilities. We speculate that this could represent successful initiation of ECMO in a smaller hospital before transfer of a live patient to a larger center. We found some support for this by considering length of stay: patients in lower volume centers with less than 50 ECMO cases ever performed have less than half the stay of patients in higher volume centers. Early transfer out of a live patient to another, presumably larger regional center, could systematically bias outcomes. The transferring center would record better than expected mortality; the receiving center would record as expected mortality.

It is also possible that lower volume facilities tend to overuse ECMO among “healthier” patients who do better, with resulting better risk-adjusted mortality. We found some support for this latter alternative explanation in examining common intensive care laboratory test values: patients in lower volume centers with less than 50 ECMO cases ever performed appeared somewhat sicker early in their admission than patients in higher volume centers. Additionally, patients in lower volume centers appeared to have systematically fewer respiratory and more cardiac admitting diagnoses and, in particular, significantly more cardiotomy procedures. Given the lower mortality of postcardiotomy ECMO, this is again consistent with a selection bias toward healthier patients among the lower experience facilities.

Limitations

As with all administrative discharge data studies, our study is greatly limited by lack of clinically relevant data. In this setting, e.g., cannulation mode (venovenous or venoarterial) and duration of ECMO are both relevant but missing. Although the administrative data included some laboratory test values that allowed some rudimentary case-mix adjustment, many important variables were missing such as lactate. It is, therefore, possible that volume–outcome relationships are actually present but were confounded in this administrative dataset.

Mitigating these data concerns somewhat are two key points. First is the reasonable fit and predictive validity of the risk model. Had the patient-level model had a smaller R2, worse calibration, discrimination, or classification, then the lack of additional risk adjusters would signal an inadequate risk adjustment. Especially in comparison with five recent risk models relying on generally excellent clinical chart data,17–21 this study’s risk model had similar area under the ROC curve.

Second, when cases are aggregated at the state level, risk adjustment becomes less important. This is because adjustable differences in patient risk tend to become smaller, the larger the area of aggregation. For example, individual patients differ greatly in risk, but aggregated patient risk across different attending physicians in the same hospital will differ less so and, in turn, aggregated risk across different hospitals will differ still less. Finally, although using centralized database data such as that of the ELSO registry is nearly always preferable,28 there are some aspects of selection that may make such registries less than complete. In this study, real-world data on the entire market in one state was available to provide an unselected view of quality and outcomes.

Conclusions

This study suggests several policy-relevant conclusions. First, as others have pointed out in related fields such as cardiac surgery,29 if volume is not a precise and reliable proxy for quality, then we should focus on quality measures (e.g., risk-adjusted mortality, status on discharge) rather than on imprecise and unreliable proxies such as volume or experience.

Second, attention must be paid to selection biases. Our results suggest that smaller volume centers differ systematically from larger centers in that their use of ECMO is disproportionately related to cardiac surgery, their patients appear systematically somewhat healthier even early in the admission, and they appear to spend substantially fewer days in hospital despite surviving ECMO. Such differences could confound measurement of quality in lower volume centers.

Third, centralization of ECMO would be warranted if either scale or cumulative experience significantly affected patient outcomes. However, in our study, once institutions have as few as 50 admitted cases’ total experience or perform just 10 cases annually, there was little if any systematic improvement in outcomes. In a setting of very fast learning, decentralized operations are not handicapped by slow movement down a cumulative experience curve. In a setting of little returns to scale, even modest scale facilities can obtain good outcomes.

Lastly, some legitimate concerns surround patient selection and the fast adoption of this rescue technology. Attention to the development of more stringent appropriate use criteria may be indicated, as may be more attention to the clinical benefits obtained by very high risk (who may not survive because of their underlying organ dysfunction) or very low-risk patients (who may not need this technology).

Acknowledgment

The Pennsylvania Health Care Cost Containment Council (PHC4) is an independent state agency responsible for addressing the problem of escalating health costs, ensuring the quality of health care, and increasing access to health care for all citizens regardless of ability to pay. PHC4 has provided data to Pennsylvania State University (PSU) in an effort to further PHC4’s mission of educating the public and containing healthcare costs in Pennsylvania. PHC4, its agents, and staff have made no representation, guarantee, or warranty, express or implied, that the data—financial, patient, payor, and physician-specific information—provided to PSU is error free or that the use of the data will avoid differences of opinion or interpretation. This analysis was not prepared by PHC4. This analysis was done by PSU. PHC4, its agents, and staff bear no responsibility or liability for the results of the analysis, which are solely the opinion of PSU.

References

1. Rihal CS, Naidu SS, Givertz MM, et al.Society for Cardiovascular Angiography and Interventions (SCAI); Heart Failure Society of America (HFSA); Society for Thoracic Surgeons (STS); American Heart Association (AHA); American College of Cardiology (ACC): 2015 SCAI/ACC/HFSA/STS Clinical Expert Consensus Statement on the Use of Percutaneous Mechanical Circulatory Support Devices in Cardiovascular Care (Endorsed by the American Heart Association, the Cardiological Society of India, and Sociedad Latino Americana de Cardiologia Intervencion; Affirmation of Value by the Canadian Association of Interventional Cardiology-Association Canadienne de Cardiologie d’intervention). J Card Fail 2015.21: 499518,
2. Extracorporeal Life Support Organization. International Summary, 2016. Available at: https://www.elso.org/Registry/Statistics/InternationalSummary.aspx.
3. Dyga BStarting an ECMO program. UPMC Presentation, 2016. Available at: http://docplayer.net/20511261-Starting-an-ecmo-program-2015-asaio-journal.html.
4. Birkmeyer JD, Stukel TA, Siewers AE, Goodney PP, Wennberg DE, Lucas FLSurgeon volume and operative mortality in the United States. N Engl J Med 2003.349: 21172127,
5. David G, Brachet TRetention, learning by doing, and performance in emergency medical services. Health Serv Res 2009.44: 902925,
6. Sheetz KH, Norton EC, Birkmeyer JD, Dimick JBProvider experience and the comparative safety of laparoscopic and open colectomy. Health Serv Res 2017.52: 5673,
7. Gaynor M, Seider H, Vogt WBThe volume–outcome effect, scale economies and learning-by-doing. Am Econ Rev 2005.2: 243247,
8. Ho VLearning and the evolution of medical technologies: The diffusion of coronary angioplasty. J Health Econ 2002.21: 873885,
9. Seymour CW, Iwashyna TI, Ehlenbach WJ, Wunsch H, Cooke CRHospital level variation in the use of intensive care. Health Serv Res 2012.47: 20602080,
10. Barbaro RP, Odetola FO, Kidwell KM, et al.Association of hospital-level volume of extracorporeal membrane oxygenation cases and mortality. Analysis of the extracorporeal life support organization registry. Am J Respir Crit Care Med 2015.191: 894901,
11. McCarthy FH, McDermott KM, Kini V, et al.Trends in U.S. extracorporeal membrane oxygenation use and outcomes: 2002–2012. Semin Thorac Cardiovasc Surg 2015.27: 8188,
12. Huesch MD, Sakakibara MForgetting the learning curve for a moment: How much performance is unrelated to own experience? Health Econ 2009.18: 855862,
13. Huesch MDLearning by doing, scale effects, or neither? Cardiac surgeons after residency. Health Serv Res 2009.44: 19601982,
14. MacLaren G, Fortenberry JD, Dalton HJLies, statistics, and ECMO data mining: Digging dirt or striking gold? Pediatr Crit Care Med 2016.17: 799802,
15. Muller G, Flecher E, Lebreton G, et al.The ENCOURAGE mortality risk score and analysis of long-term outcomes after VA-ECMO for acute myocardial infarction with cardiogenic shock. Intensive Care Med 2016.42: 370378,
16. Li CL, Wang H, Jia M, Ma N, Meng X, Hou XTThe early dynamic behavior of lactate is linked to mortality in postcardiotomy patients with extracorporeal membrane oxygenation support: A retrospective observational study. J Thorac Cardiovasc Surg 2015.149: 14451450,
17. Schmidt M, Bailey M, Sheldrake J, et al.Predicting survival after extracorporeal membrane oxygenation for severe acute respiratory failure. The Respiratory Extracorporeal Membrane Oxygenation Survival Prediction (RESP) score. Am J Respir Crit Care Med 2014.189: 13741382,
18. Klinzing S, Wenger U, Steiger P, et al.External validation of scores proposed for estimation of survival probability of patients with severe adult respiratory distress syndrome undergoing extracorporeal membrane oxygenation therapy: A retrospective study. Crit Care 2015.19: 142,
19. Huang L, Li T, Xu L, et al.Performance of multiple risk assessment tools to predict mortality for adult respiratory distress syndrome with extracorporeal membrane oxygenation therapy: An external validation study based on Chinese single-center data. Chin Med J (Engl) 2016.129: 16881695,
20. Chen WC, Huang KY, Yao CW, et al.The modified SAVE score: Predicting survival using urgent veno-arterial extracorporeal membrane oxygenation within 24 hours of arrival at the emergency department. Crit Care 2016.20: 336,
21. Kim KI, Lee HS, Kim HS, et al.The pre-ECMO simplified acute physiology score II as a predictor for mortality in patients with initiation ECMO support at the emergency department for acute circulatory and/or respiratory failure: A retrospective study. Scand J Trauma Resusc Emerg Med 2015.23: 59,
22. Formica F, Avalli L, Colagrande L, et al.Extracorporeal membrane oxygenation to support adult patients with cardiac failure: Predictive factors of 30-day mortality. Interact Cardiovasc Thorac Surg 2010.10: 721726,
23. Livingston EH, Cao JProcedure volume as a predictor of surgical outcomes. JAMA. 2010.304: 9597,
24. Huesch MD‘Fit’ between cardiac surgeons and hospitals. Health Serv Res. 2011.1: 126,
25. Vaja R, Chauhan I, Joshi V, et al.Five-year experience with mobile adult extracorporeal membrane oxygenation in a tertiary referral center. J Crit Care 2015.30: 11951198,
26. Huesch MDForgetting? Impact of breaks in surgery on Californian cardiac surgeons. J Healthc Qual. 2014.36: 4249,
27. McCarthy FH, McDermott KM, Spragan D, Hoedt A, Kini V, Atluri P, et al.Unconventional volume-outcome associations in adult extracorporeal membrane oxygenation in the united states. Ann Thorac Surg. 2016.102: 489495,
28. Huesch MD, Brehm CThe challenges in predicting ECMO survival, and a path forward. ASAIO J 2017. Epub ahead of print.
29. Shahian DM, Edwards FH, Ferraris VA, et al.Society of Thoracic Surgeons Quality Measurement Task Force: Quality measurement in adult cardiac surgery: Part 1–Conceptual framework and measure selection. Ann Thorac Surg 2007.83 (suppl 4): S3S12,
Keywords:

learning; practice makes perfect; scale economies; extracorporeal membrane oxygenation; mortality

Supplemental Digital Content

Copyright © 2017 The Author(s). Published by Wolters Kluwer Health, Inc. on behalf of the ASAIO.