The complex relationship between pediatric cardiac surgical case volumes and mortality rates in a national clinical database

      Objective

      We sought to determine the association between pediatric cardiac surgical volume and mortality using sophisticated case-mix adjustment and a national clinical database.

      Methods

      Patients 18 years of age or less who had a cardiac operation between 2002 and 2006 were identified in the Society of Thoracic Surgeons Congenital Heart Surgery Database (32,413 patients from 48 programs). Programs were grouped by yearly pediatric cardiac surgical volume (small, <150; medium, 150–249; large, 250–349; and very large, ≥350 cases per year). Logistic regression was used to adjust mortality rates for volume, surgical case mix (Aristotle Basic Complexity and Risk Adjustment for Congenital Heart Surgery, Version 1 categories), patient risk factors, and year of operation.

      Results

      With adjustment for patient-level risk factors and surgical case mix, there was an inverse relationship between overall surgical volume as a continuous variable and mortality (P = .002). When the data were displayed graphically, there appeared to be an inflection point between 200 and 300 cases per year. When volume was analyzed as a categorical variable, the relationship was most apparent for difficult operations (Aristotle technical difficulty component score, >3.0), for which mortality decreased from 14.8% (60/406) at small programs to 8.4% (157/1858) at very large programs (P = .02). The same was true for the subgroup of patients who underwent Norwood procedures (36.5% [23/63] vs 16.9% [81/479], P < .0001). After risk adjustment, all groups performed similarly for low-difficulty operations. Conversely, for difficult procedures, small programs performed significantly worse. For Norwood procedures, very large programs outperformed all other groups.

      Conclusion

      There was an inverse association between pediatric cardiac surgical volume and mortality that became increasingly important as case complexity increased. Although volume was not associated with mortality for low-complexity cases, lower-volume programs underperformed larger programs as case complexity increased.

      CTSNet classification

      Abbreviations and Acronyms:

      ABC (Aristotle Basic Complexity), O/E (observed mortality/expected mortality), OR (odds ratio), RACHS-1 (Risk Adjustment for Congenital Heart Surgery, Version 1), STS (Society of Thoracic Surgeons)
      Earn CME credits at http://cme.ctsnetjournals.org
      The drive to quantify and publicly report hospital quality has intensified the search for comparable quality measures. Because of its simplicity and ready availability, surgical volume is one of the most often cited metrics. The relationship between hospital surgical volume and in-hospital mortality for coronary artery bypass grafting surgery has been one of the most studied. Importantly, the strength of the relationship has depended on the database used. The inverse association between hospital volume and mortality is more apparent in administrative data than in clinical data.
      • Birkmeyer J.D.
      • Siewers A.E.
      • Findlayson E.V.
      • et al.
      Hospital volume and surgical volume in the United States.
      • Peterson E.D.
      • Coombs L.P.
      • DeLong E.R.
      • Haan C.K.
      • Ferguson T.B.
      Procedural volume as a marker of quality for CABG surgery.
      This is in part due to the more sophisticated risk adjustment possible with clinical data and differences in the cohorts of hospitals.
      • Welke K.F.
      • Peterson E.D.
      • Vaughn-Sarrazin M.S.
      • et al.
      Comparison of cardiac surgical volumes and mortality rates between the Society of Thoracic Surgeons and Medicare databases from 1993 through 2001.
      Although less studied, the relationship between hospitals' pediatric cardiac surgical volumes and mortality rates has been the subject of several previous investigations.
      • Jenkins K.J.
      • Newburger J.W.
      • Lock J.E.
      • et al.
      In-hospital mortality for surgical repair of congenital heart defects: preliminary observations of variation by hospital caseload.
      • Hannan E.L.
      • Racz M.
      • Kavey R.E.
      • Quagebeur J.M.
      • Williams R.
      Pediatric cardiac surgery: the effect of hospital and surgeon volume on in-hospital mortality.
      • Sollano J.A.
      • Gelijns A.C.
      • Moskowitz A.J.
      • et al.
      Volume-outcome relationships in cardiovascular operations: New York State, 1990-1995.
      • Chang R.K.
      • Klitzner T.S.
      Can regionalization decrease the number of deaths for children who undergo cardiac surgery? A theoretical analysis.
      • Bazzani L.G.
      • Marcin J.P.
      Case volume and mortality in pediatric cardiac surgery patients in California, 1998-2003.
      • Welke K.F.
      • Diggs B.S.
      • Karamlou T.
      • Ungerleider R.M.
      The relationship between hospital surgical case volumes and mortality rates in pediatric cardiac surgery: a national sample 1988-2005.
      These studies have used either administrative data or single-state clinical data. The largest and most recent of these studies demonstrated an inverse and nonlinear association between volume and mortality after adjustment for patient age and surgical case mix.
      • Welke K.F.
      • Diggs B.S.
      • Karamlou T.
      • Ungerleider R.M.
      The relationship between hospital surgical case volumes and mortality rates in pediatric cardiac surgery: a national sample 1988-2005.
      Although this study benefited from a national database, the reliance on administrative data limited the ability to adjust for patient-level risk factors. Surgical case-mix designation was also limited by administrative coding.
      The purpose of our study was to determine the relationship between program surgical volume and mortality after pediatric cardiac surgery. For the investigation, we used clinical data from the Society of Thoracic Surgeons (STS) Congenital Heart Surgery Database. We first examined the raw association between overall surgical volume and mortality. We then investigated how this association is affected by adjustment for both patient-level risk factors and surgical case mix.

      Materials and Methods

       Database

      This study was designed as a retrospective cohort analysis. We obtained data from the STS Congenital Heart Surgery Database.

      The Society of Thoracic Surgeons Congenital Heart Surgery Database 2007 Congenital Report Executive Summary. Available at: http://www.sts.org/documents/pdf/2007_Congenital_Report_Executive_Summary_-_All_Patients.pdf. Accessed May 8, 2008.

      The STS maintains the provider-led voluntary cardiac surgical clinical database as a means of supporting national quality-improvement efforts. The database grew from 6790 operations from 20 participants during calendar year 2002 to 19,853 operations from 56 participants during calendar year 2006. Database participants are either hospitals or surgical groups that are affiliated with hospitals at which cardiac surgery is performed. Each record corresponds to a primary cardiac surgical procedure. Data elements include basic patient demographic information, comorbidities and preoperative risk factors, diagnoses, type of operation, and outcomes, including in-hospital mortality, 30-day mortality, major morbidity, and postoperative length of stay.

       Study Population

      The study population consisted of patients age 18 years or less who underwent a cardiovascular operation at an STS-participating program between January 1, 2002, and December 31, 2006. Programs were excluded if they had more than 10% missing data on key study variables, including preoperative risk factors, noncardiac abnormalities, discharge mortality, and length of stay. Ten programs were deleted because of data quality issues, leaving 48 programs in the study cohort. Seven of the 48 programs performed operations at 2 hospitals. In each case the vast majority of operations were performed at one of the hospitals. Patients were initially included if they underwent one of the cardiovascular procedures for which the Aristotle Basic Complexity (ABC) Score and/or the Risk Adjustment for Congenital Heart Surgery, Version 1 (RACHS-1) complexity level is defined.
      • Jenkins K.J.
      • Gauvreau K.
      • Newburger J.W.
      • Spray T.L.
      • Moller J.H.
      • Iezzoni L.I.
      Consensus-based method for risk adjustment for surgery for congenital heart disease.
      • Lacour-Gayet F.
      • Clark D.
      • Jacobs J.
      • et al.
      The Aristotle score: a complexity-adjusted method to evaluate surgical results.
      In addition to cardiovascular procedures, the ABC score is also defined for 13 noncardiovascular procedures that were excluded from the analysis. Patients undergoing one of these 13 noncardiovascular procedures were only included if the noncardiovascular procedure was performed concomitantly with a cardiovascular procedure and was not the primary procedure of the operation. In addition, patients weighing less than or equal to 2500 g undergoing patent ductus arteriosus ligation as their primary procedure were excluded from the analysis. In cases in which patients had multiple operations during the same hospital admission, only the first operation was analyzed. Finally, patients were excluded if they were missing data on 2 key risk-adjustment variables: age and weight.

       Definition of Surgical Volumes

      Average annual program volumes were based on operations submitted to the STS database between January 1, 2002, and December 31, 2006. Case volumes were defined by counting the number of admissions for which the index operation (first operation of the hospitalization) was cardiovascular. The average annual program volume was calculated by dividing the number of admissions during the study period by the number of months that the program participated in the database during the study period and then multiplying by 12. For tabular presentation, volumes were categorized into 4 groups: less than 150, 150 to 249, 250 to 349, and 350 or more cases per year. These categories were chosen to ensure an adequate sample size (number of events) in each category when analyzed in the overall population, as well as subgroups. The association between volume and mortality was also explored without categorizing volume, as described below.
      We repeated the calculation of surgical volumes by counting the total number of cardiovascular operations (instead of the number of admissions). When these operation-based volumes were compared with the original admission-based volumes, the Pearson correlation was 0.99. Hence there is little risk that a program's volume status was misclassified based on the decision to define volume in terms of admissions or operations.

       End Point

      Our primary end point was in-hospital mortality, which was defined as death during the same hospitalization as the operation, regardless of timing.

       Analysis of Mortality

      Unadjusted mortality rates were compared across volume categories by using logistic regression with generalized estimating equations to account for correlation of outcomes within hospitals. A linear trend test was conducted by assigning integer scores to the volume categories and testing whether the coefficient for these integer scores was zero. Next, a generalized estimating equation logistic regression model was used to assess the association between volume and mortality while adjusting for patient-level risk factors and surgical case mix.

       Patient-level risk factors

      The following variables were included in the adjusted mortality model: age (modeled as a piecewise linear variable with knots at 30 days and 1 year); age-for-weight-and-sex z score; interaction between age and age-for-weight-and-sex z score; preoperative stay for more than 2 days; number of prior operations (0, 1, and ≥2); renal failure or dialysis; acidosis, circulatory support, or shock; preoperative ventilatory support or tracheostomy; asplenia, polysplenia, or a22q11 deletion; DiGeorge syndrome; Down syndrome; procedure or procedure group; and operation date (modeled as a linear trend).

       Modeling of volume

      Volume was initially modeled as a categorical variable, with categories corresponding to less than 150, 150 to 249, 250 to 349, and 350 or more cases per year. Because categorization can reduce statistical power, the analysis was also performed by entering volume in the model as a single continuous linear variable. This approach was used for testing the overall null hypothesis of no association between volume and mortality. Finally, to explore possible nonlinear volume effects, volume was analyzed by using restricted cubic splines. Knots for the spline function were placed at 150, 250, and 350 cases per year. Because the shape of a spline function might depend on the choice of arbitrary knot locations, the model was subsequently repeated by using 2 alternative specifications: knots at 100, 300, and 500 cases and knots at 200, 300, and 450 cases.

       Adjustment for type of procedure

      Because of the large number of procedures with small sample sizes, it was not feasible to adjust for procedure identity by including a separate term for each individual procedure. Instead, a coarser covariate adjustment was performed by grouping procedures of similar complexity. Any procedure with at least 20 deaths recorded in the database was considered to be its own stratum. All remaining procedures were stratified by ABC level and RACHS-1 category. The possible values of the ABC level were 1, 2, 3, 4, and “unassigned.” The possible values of the RACHS-1 level were 1, 2, 3, 4, 6, and “unassigned.” “Unassigned” indicates that the procedure type was not included by the developers of the ABC or RACHS-1 methodology. To perform the covariate adjustment, we grouped procedures according to the 30 unique combinations of ABC and RACHS-1 (5 levels of ABC [including “unassigned”] × 6 levels of RACHS-1 [including “unassigned”] = 30 unique combinations). Strata with zero deaths do not provide information for studying the volume-outcome association and were therefore excluded. The remaining 28 combinations were entered into the logistic regression model as a set of category indicator variables.

       Subgroup analyses

      In addition to studying the overall volume–outcome association, we also assessed the volume–outcome association separately for low-difficulty and high-difficulty procedures. We did this by including an interaction between volume and difficulty in the regression models. Procedures were classified by difficulty based on the difficulty component of the Aristotle score. The categories were Aristotle difficulty of 3.0 or less versus 3.0 or greater. Procedures that were not classified by using the Aristotle system were excluded from this analysis. We also performed a separate subgroup analysis to assess the volume-outcome association specifically for the Norwood operation. The Norwood operation was the only procedure with enough events to make a standalone analysis informative.

       Sensitivity analysis

      To determine whether the observed volume–outcome association was an artifact of unusually high performance at just 1 or 2 high-volume hospitals, we repeated the analysis 3 ways: (1) after excluding the single participant with the largest average annualized number of index operations, (2) after excluding the 2 participants with the largest average annualized number of index operations, and (3) after excluding the participant in the highest volume category with the lowest observed mortality to expected mortality (O/E) ratio. The first program participated for 3 years, contributed 2080 cases to the regression analysis, had an average annualized volume of 850 cases per year, and had the third lowest O/E ratio among participants in the highest volume category. The second program participated for 5 years, contributed 3279 cases to the regression analysis, had an average annualized volume of 683 cases per year, and had the second lowest O/E ratio among participants in the highest volume count. This program contributed more cases to the regression analysis than any other participant. (Note: the average annualized volume does not correspond to the number of cases in the regression analysis because of differing inclusion/exclusion criteria for calculating participant volumes and analyzing outcomes.) The third program participated for 5 years, contributed 2769 cases to the analysis, had an average annualized volume of 600.6, and had the lowest O/E ratio among participants in the analysis.
      If the observed volume–outcome association disappeared after excluding these hospitals, this might indicate that the apparent volume–outcome association was an artifact of unusual results at just 1 or 2 high-volume hospitals.

      Results

      We identified 32,413 operations in the STS Congenital Heart Surgery Database from 2002 to 2006 that met our inclusion criteria. These operations took place at 48 programs. The average annual pediatric cardiac surgical volume at these programs ranged from 66 to 850 cases (Figure 1).
      Figure thumbnail gr1
      Figure 1Distribution of hospital average annual case volumes.
      Patient-level risk factors and surgical case mix varied by volume category (Table 1). The case mix at larger programs included a higher percentage of younger patients. As a result, operations done at larger-volume programs more frequently involved lower-weight patients. The case mix at larger-volume programs also included a greater percentage of more complex operations. Patient demographics and preoperative risk factors did not differ by volume category.
      Table 1Baseline characteristics of the study cohort by volume category
      Volume category
      All<150150–249250–349≥350
      VariableNFrequency (%)NFrequency (%)NFrequency (%)NFrequency (%)NFrequency (%)
      Participants481514127
      Patients32,4133715608812,00710,603
      Demographics
       Age (d)<30686321.267418.1132621.8236919.7249423.5
      ≥30 and <36511,72636.2129334.8226837.3426335.5390236.8
      ≥36514,82442.7174847.1249441.0537544.8420739.7
       Weight (kg)<2.510923.4952.62293.83683.14003.8
      ≥2.5 and <5.010,28331.7110229.7196432.3360830.1360934.1
      ≥5.021,03864.9251867.8389564.0803166.9659462.2
      z Score (weight for age)<−53791.2371.0741.21361.11321.2
      ≥−5 and <−331229.63509.462010.210759.0107710.2
      ≥−3 and <−113,26240.9151140.7250641.2474839.5449742.4
      ≥−1 and <113,10040.4149040.1242439.8497141.4421539.8
      >=125507.93278.84647.610779.06826.4
      Female sex14,60845.1170145.8278645.8535244.6476945.0
      Abnormalities
       22q11 Deletion1390.4250.7360.6610.5170.2
       DiGeorge syndrome7102.2691.91512.52562.12342.2
       Down syndrome29049.03649.862610.310288.68868.4
       Asplenia5001.5521.4821.41871.61791.7
       Polysplenia1290.490.2320.5570.5310.3
      Preoperative risk factors
       Preoperative length of stay≥0 and <119,61360.5228661.6377562.0733761.1621558.6
      ≥1 and <5690521.375020.2110518.2224018.7281026.5
      ≥5 and <10335210.33429.269711.5137711.59368.8
      ≥10 and <3018585.72326.33936.57736.44604.3
      ≥306752.11032.81161.92772.31791.7
       Previous cardiothoracic operations≥0 and <122,47770.7279575.4427570.4784768.7756071.4
      ≥1 and <2488615.454414.7101016.6172615.1160615.2
      ≥2444014.036910.079213.0184916.2143013.5
       Acidosis6942.1742.01722.83132.61351.3
       Mechanical circulatory support720.280.280.1400.3160.2
       Renal failure with or without dialysis3231.0340.9631.01301.1960.9
       Shock4011.2742.0911.51631.4730.7
       Tracheostomy1050.3110.3340.6460.4140.1
       Ventilatory support349010.83449.379713.1140411.79458.9
      Operative case mix
       Aristotle difficulty≤3.028,48688.8332991.4537089.31063289.3815587.0
      >3.0359811.23128.664710.8127010.7136913.0
       Aristotle basic complexity level1493215.469019.088314.7198016.6137913.1
      214,12344.0170446.8287847.8514243.2439941.8
      3868527.188824.4146524.4328527.6304729.0
      4434413.53599.979113.2149512.6169916.1
       RACHS-1 category1413114.661018.478814.7143714.1129613.8
      211,94942.3150345.4231243.1444243.6369239.3
      3856730.391027.5160830.0301629.6303332.3
      423008.12156.54298.08238.18338.9
      560.0200.0010.0210.0140.04
      613094.6722.22284.34644.65455.8
      The overall unadjusted mortality rate for the cohort was 3.7%. When programs were grouped into 4 volume categories, the overall unadjusted mortality rate at small programs was lower than that at medium-volume programs (4.0% vs 4.1%) and slightly higher than those at large- and very large-volume programs (3.8% and 3.3%, respectively; Table 2). When mortality risk was modeled as a function of program volume categories, the c statistic was low (0.53), indicating that volume alone was a poor predictor of mortality. Adjustment for patient risk factors and surgical case mix improved the discrimination of the model substantially (c statistic = 0.84). After adjustment, the mortality rates at medium and large programs were similar to the mortality rates at very large programs (odds ratios [OR], 1.05 and 1.14, respectively), but compared with very large programs, mortality at small programs was significantly higher (OR, 1.51; P = .0005; Table 3).
      Table 2Raw mortality rates (percentage mortality and 95% confidence interval) by volume category
      Volume category
      Procedure subgroupAll<150150–249250–349≥350P value
      P value for linear trend across volume groups.
      All procedures3.7 (3.3–4.1)4.0 (2.9–5.0)4.1 (3.1–5.2)3.8 (3.3–4.2)3.3 (2.7–3.8).101
      Aristotle difficulty ≤3.02.6 (2.3–2.9)2.7 (1.9–3.4)2.9 (2.2–3.6)2.9 (2.5–3.2)2.2 (1.9–2.5).086
      Aristotle difficulty >3.09.8 (8.4–11.1)14.8 (11.4–18.2)11.1 (7.7–14.5)9.3 (7.6–11.0)8.4 (6.2–10.7).020
      Norwood procedures21.3 (18.4–24.3)36.5 (27.4–45.7)24.2 (17.7–30.7)22.7 (18.9–26.6)16.9 (14.0-19.9)<.0001
      P value for linear trend across volume groups.
      Table 3Association between annual case volume and mortality
      The smaller numbers of operations in this table compared with those in Table 1 reflect the exclusion of patients with missing risk adjustment data. Although Table 1 includes the entire cohort, Table 3 excludes 89 patients who were in a procedure stratum with zero deaths (as mentioned in the methods section) and 44 patients with missing mortality status.
      Volume categoryNDeathsAdjusted odds ratio (95% confidence interval)P value
      All procedures
       ≥35010,570346Reference.004
      P for linear trend.
       250–34911,9784501.05 (0.86–1.29).63
       150–24960512501.14 (0.84–1.55).41
       <15036811481.51 (1.19–1.90).0005
      Low-difficulty procedures
       ≥3508663188Reference.29
      P for linear trend.
       250–34910,2522951.16 (0.87–1.53).31
       150–24951041481.08 (0.76–1.52).68
       <1503229861.21 (0.87–1.69).26
      High-difficulty procedures
       ≥3501855135Reference.0008
      P for linear trend.
       250–34916361380.89 (0.69–1.15).38
       150–249894791.22 (0.81–1.84).35
       <150406542.41 (1.89–3.06)<.0001
      Norwood procedures
       ≥35047981Reference<.0001
      P for linear trend.
       250–349418951.43 (1.06–1.95).020
       150–249194471.59 (1.09–2.32).016
       <15063232.91 (1.98–4.28)<.0001
      The smaller numbers of operations in this table compared with those in Table 1 reflect the exclusion of patients with missing risk adjustment data. Although Table 1 includes the entire cohort, Table 3 excludes 89 patients who were in a procedure stratum with zero deaths (as mentioned in the methods section) and 44 patients with missing mortality status.
      P for linear trend.
      We then subdivided the case mix by difficulty to see whether there was a differential effect of volume (Table 3). For low-difficulty operations (defined as Aristotle difficulty ≤3.0), all 4 volume groups performed similarly (P = .29, test of trend). However, for high-difficulty operations (defined as Aristotle difficulty >3.0), small programs had substantially higher adjusted mortality relative to very high-volume programs (OR, 2.41; P < .0001). Mortality rates for high-difficulty operations performed at medium and large programs were similar to the mortality rates at very large programs. We then examined the volume–mortality relationship for the Norwood procedure. This operation was chosen because of the high level of system knowledge and coordination needed to achieve success. For the Norwood procedure, very high-volume programs outperformed all other volume groups.
      To further investigate the volume–mortality relationship, we analyzed volume as a continuous variable and used logistic regression to adjust for patient-level risk factors and surgical case mix. We found an inverse relationship between overall surgical volume as a continuous variable and mortality (P = .002). We then plotted the ORs for risk-adjusted mortality by using a volume of 800 cases per year as the arbitrary reference (OR, 1.0). For overall surgical volume, the slope of the volume–mortality curve was steepest below an infection point, which occurred between 200 and 300 cases per year (Figure 2). This nonlinear effect was not significant for low-complexity cases (P = .06, test of no volume–mortality association; Figure 3) but was consistent for high-complexity cases (P = .007, test of no volume–mortality association; Figure 3) and the Norwood operation (P < .001, test of no volume–mortality association; Figure 4). Importantly, the reliability of the inflection point cannot be assured because as a result of the number of programs in the cohort, the 95% confidence intervals were not sufficiently narrow. When the analysis was repeated by using 2 alternative specifications for the knot locations, the shape of the curves was similar, but the location of the inflection point increased to between 300 and 400 cases per year.
      Figure thumbnail gr2
      Figure 2Association between overall annual volume and risk-adjusted mortality (P = .002, test of no volume-mortality association).
      Figure thumbnail gr3
      Figure 3Association between surgical volume and risk-adjusted mortality by Aristotle difficulty (A, low difficulty, ≤3 [P = .059, test of no volume-mortality association]; B, high difficulty, >3 [P = .007, test of no volume-mortality association]).
      Figure thumbnail gr4
      Figure 4Association between hospital volume and risk-adjusted mortality for Norwood operations (P < .001, test of no volume-mortality association).
      To further assess the stability of our results, we conducted a series of sensitivity analyses. After removal of the largest and 2 largest programs from the cohort, there was no substantial difference in our findings. Similarly, removal of the lowest-mortality program did not appreciably alter the results.

      Discussion

      After risk adjustment for patient-level variables and surgical case mix, there was an inverse relationship between overall surgical volume as a continuous variable and mortality. When analyzed as a categorical variable, the relationship was most apparent for difficult operations. For one of the most complex procedures (the Norwood procedure), the largest programs had results that were significantly better than those of all other groups. Although volume alone was an unreliable discriminator of mortality, mortality rates adjusted for patient risk factors and surgical case mix suggest that in aggregate higher-volume programs achieve lower mortality rates for complex operations.
      Although we did find a relationship between case volume and mortality, this finding must be interpreted with caution. The aggregation of programs into volume groups gave us sufficient statistical power to analyze important relationships but disguised individual programs. One should not conclude that all larger programs perform better than all smaller programs. Although on average this was true, there were low-volume programs that had low mortality rates and those that had volumes too low for any mortality rate difference to be observed. Importantly, our analysis highlights the unsuitability of volume alone, without adjustment for patient risk factors and surgical case mix, as a marker of quality. In general, a patient's own risk characteristics and level of disease burden account for the vast majority of his or her mortality risk, and the effect of program volume on the mortality risk of an individual patient is small.
      The results of this study complement those of the largest and most recent previous study of the volume–mortality relationship in pediatric cardiac surgery.
      • Welke K.F.
      • Diggs B.S.
      • Karamlou T.
      • Ungerleider R.M.
      The relationship between hospital surgical case volumes and mortality rates in pediatric cardiac surgery: a national sample 1988-2005.
      Although the 2 investigations were done with different types of data, both found that larger programs performed better than smaller programs. The previous study used the Nationwide Inpatient Sample, which is an involuntary administrative database, which is a stratified cross-sectional sample that includes approximately 20% of all community (nonfederal) hospital discharges in the United States selected from a sampling frame that comprises approximately 90% of all hospital discharges in the United States. The STS Congenital Heart Surgery Database, as used in the present study, is a voluntary clinical database. As such, the participants are more likely to be higher-volume programs specializing in congenital cardiac surgery. Consequently, the volume categorizations differed between the 2 studies. Despite this population difference and the greater ability for risk adjustment with clinical data, the findings of the 2 studies were similar.
      The limitations of this study are a result of the choice of the data source. The STS Congenital Heart Surgery Database was designed primarily for quality improvement. The database contains a self-selected group of centers that voluntarily chose to submit data. The data were collected by individuals at the institutions, many of whom have a stake in the outcomes of the program. Although some have questioned the validity of such data because of this involvement, the contributions of clinical personnel to the data collection process and the design of the database specifically for congenital cardiac surgery might result in higher-quality data and superior risk adjustment. Unfortunately, at the time of the data collection for this study, there was no data audit process to ensure accuracy. The consistency of the aggregate morality rates and the volume–mortality relationships with previous reports supports the quality of the database.
      Because we did not have information on preoperative decision making, we were not able to assess the appropriateness of the operations performed. For example, for a certain diagnosis, a patient might receive an operation with a lower in-hospital mortality rate but lower long-term survival rather than a more appropriate operation with a higher up-front mortality and better long-term survival. In addition, the choice of the low-mortality operation might necessitate a second operation, resulting in a higher combined mortality rate. The lack of long-term follow-up limited our ability to compare the “true” outcomes important to congenital heart surgery, including not only mortality but also morbidity, functional status, and neurologic status. Because the overall mortality rate for our cohort was 3.7%, these outcomes, rather than in-hospital mortality, are of importance to more than 96% of our patients. Future efforts need to investigate these outcomes. Notwithstanding these limitations, our study was conducted by using a national multicenter database with adequate power to generate current, stable mortality rates.

      Conclusion

      We found overall unadjusted volume to be a poor discriminator of mortality. However, after adjustment for patient risk factors and surgical case mix, larger programs achieved superior results for more complex operations. Many factors contribute to the mortality risk of a patient undergoing pediatric cardiac surgery. The relationship between volume and mortality is complex, making volume a difficult choice as a quality measure for pediatric cardiac surgery. Rather than accepting an imperfect proxy, the process measures and system characteristics for which volume is a surrogate need to be identified. The widespread implementation of these factors is likely to lead to substantial improvement in the outcomes of our operations.

      References

        • Birkmeyer J.D.
        • Siewers A.E.
        • Findlayson E.V.
        • et al.
        Hospital volume and surgical volume in the United States.
        N Engl J Med. 2002; 346: 1128-1137
        • Peterson E.D.
        • Coombs L.P.
        • DeLong E.R.
        • Haan C.K.
        • Ferguson T.B.
        Procedural volume as a marker of quality for CABG surgery.
        JAMA. 2004; 291: 195-201
        • Welke K.F.
        • Peterson E.D.
        • Vaughn-Sarrazin M.S.
        • et al.
        Comparison of cardiac surgical volumes and mortality rates between the Society of Thoracic Surgeons and Medicare databases from 1993 through 2001.
        Ann Thorac Surg. 2007; 84: 1538-1546
        • Jenkins K.J.
        • Newburger J.W.
        • Lock J.E.
        • et al.
        In-hospital mortality for surgical repair of congenital heart defects: preliminary observations of variation by hospital caseload.
        Pediatrics. 1995; 95: 323-330
        • Hannan E.L.
        • Racz M.
        • Kavey R.E.
        • Quagebeur J.M.
        • Williams R.
        Pediatric cardiac surgery: the effect of hospital and surgeon volume on in-hospital mortality.
        Pediatrics. 1998; 101: 963-969
        • Sollano J.A.
        • Gelijns A.C.
        • Moskowitz A.J.
        • et al.
        Volume-outcome relationships in cardiovascular operations: New York State, 1990-1995.
        J Thorac Cardiovasc Surg. 1999; 117: 419-428
        • Chang R.K.
        • Klitzner T.S.
        Can regionalization decrease the number of deaths for children who undergo cardiac surgery? A theoretical analysis.
        Pediatrics. 2002; 109: 173-181
        • Bazzani L.G.
        • Marcin J.P.
        Case volume and mortality in pediatric cardiac surgery patients in California, 1998-2003.
        Circulation. 2007; 115: 2652-2659
        • Welke K.F.
        • Diggs B.S.
        • Karamlou T.
        • Ungerleider R.M.
        The relationship between hospital surgical case volumes and mortality rates in pediatric cardiac surgery: a national sample 1988-2005.
        Ann Thorac Surg. 2008; 86: 889-896
      1. The Society of Thoracic Surgeons Congenital Heart Surgery Database 2007 Congenital Report Executive Summary. Available at: http://www.sts.org/documents/pdf/2007_Congenital_Report_Executive_Summary_-_All_Patients.pdf. Accessed May 8, 2008.

        • Jenkins K.J.
        • Gauvreau K.
        • Newburger J.W.
        • Spray T.L.
        • Moller J.H.
        • Iezzoni L.I.
        Consensus-based method for risk adjustment for surgery for congenital heart disease.
        J Thorac Cardiovasc Surg. 2002; 123: 110-118
        • Lacour-Gayet F.
        • Clark D.
        • Jacobs J.
        • et al.
        The Aristotle score: a complexity-adjusted method to evaluate surgical results.
        Eur J Cardiothorac Surg. 2004; 25: 911-924