Context. Managed care plans aggressively seek to contain costs, but few data are available regarding their impact on access to high quality care for their members.
Objective. To assess the impact of managed care health insurance on use of lower-mortality hospitals for children undergoing heart surgery in California.
Design. Retrospective cohort study using state-mandated hospital discharge datasets.
Setting. Pediatric cardiovascular surgical centers in California.
Patients. Five thousand seventy-one children admitted for open cardiac surgical procedures during 1992–1994.
Results. Hospitals were divided into lower- and higher-mortality groups according to adjusted surgical mortality. Using multivariate logistic regression analysis to control for medical, socioeconomic, demographic, and distance factors, children with managed care insurance were less likely to be admitted to a lower-mortality hospital for surgery relative to children with indemnity insurance (odds ratio: .53; 95% confidence interval: .45,.63). Similar findings resulted when the analysis was stratified by race/ethnicity. In addition, length of stay, a correlate of health care costs, was no longer for children admitted to lower-mortality centers than for those at higher-mortality centers (adjusted difference: .54 days shorter at lower-mortality centers; 95% confidence interval: −1.50,.41).
Conclusions. During this study, children with managed care insurance had significantly reduced use of lower-mortality hospitals for pediatric heart surgery in California compared with children with indemnity insurance. Further study is necessary to determine the mechanisms of this apparent insurance-specific inequity.
The delivery of health services in the United States has been characterized increasingly by enhanced competition and efforts to contain costs. Although such efforts have taken a variety of forms, they have primarily relied on limits on patient self-referral, financial incentives for physicians and hospitals to reduce use, and negotiated referral arrangements with hospitals. Managed care systems have grown substantially, with the number of patients enrolled in health maintenance organizations (HMOs) alone rising from 36.5 million in 1990 to 58.2 million in 1995.1,,2
Amid this rapidly changing financial base, the clinical capabilities of health care provided to children have also expanded dramatically. Advances in the care of critically ill children suffering from a variety of underlying causes have markedly reduced mortality.3–5 Among the most important arenas of improved outcome has been the surgical correction of congenital heart defects in children. Strides in the precision of diagnostic tests, surgical techniques, and intensive care have improved the survival and long-term well-being of affected children.6,,7
Given these expanding clinical resources, it is essential that policies and practice patterns ensure that this capability is provided to all those in need. It is possible, however, that shifting patient referral patterns aimed at reducing costs could unintentionally result in reduced access to high quality surgical care for children with managed care insurance.
To explore the impact of managed care insurance status on the provision of highly specialized care, we examined the experience of children requiring surgical correction of congenital heart disease in California for the years 1992 through 1994. Using statewide hospital discharge data, we assessed whether children with managed care insurance were less likely than those with indemnity insurance to undergo cardiac surgery in a facility with a record of low mortality for such procedures. In addition, we examined whether children admitted to hospitals with a record of low mortality had higher hospital lengths of stay, a proxy for inpatient costs.
Abstracts of patient records were drawn from the annual California state-mandated hospital discharge databases8–10 for children <15 years of age and residing in California, who underwent an open-heart operations between 1992 and 1994 and had indemnity, managed care, or Medicaid (Medi-Cal) as their primary health insurance coverage. The information available in these databases includes patient age (in categories: 0–1 year, 1–5 years, 5–15 years of age, etc), race/ethnicity, insurance type, zip code, hospital of admission, hospitalization length of stay, andInternational Classification of Diseases, Ninth Revision-Clinical Modification (ICD-9-CM) diagnosis and procedure codes. The state databases were coded such that children enrolled in Medicaid managed care plans were assigned to the Medicaid insurance group.
Each discharge abstract may include as many as 21 ICD-9-CM procedure codes and 25 ICD-9-CM diagnosis codes. All patients meeting age and residence criteria with any cardiovascular ICD-9-CM procedure code (35.xx–39.xx) were initially considered for inclusion. When more than 1 cardiovascular procedure code was recorded for a patient, a primary procedure code was assigned using a ranking algorithm designed to assign the procedure code with the highest overall complexity to each patient. Patients were then grouped into 38 procedure groups, combining closely related codes and, in some cases, dividing codes into multiple groups based on diagnosis and/or age criteria in an effort to maximize the homogeneity of each group. For example, 3 codes for ventricular septal defect repair were incorporated into a single group (35.53, prosthetic repair; 35.62, graft repair; and 35.72, not elsewhere specified), whereas 2 groups were created from the arterial switch operation code (35.84): 1 for infants and 1 for older children. Finally, we excluded closed procedures, such as patent ductus arteriosus ligation/pulmonary artery band (38.85); ambiguous codes such as enlarge septal defect (35.41); and very low volume procedures groups (<20 cases/year for the entire state).
The procedure groups included in the analyses included aortic valve replacement (35.21 and 35.22), arterial switch operation in infancy (35.84 in children <1 year of age), atrial septal defect repair (35.51, 35.52, 35.61, and 35.71), atrial switch operations (Senning or Mustard, 35.91), cavopulmonary shunt (39.21), endocardial cushion defect repair (35.54, 35.63, and 35.73), Fontan procedure (35.94), mitral valve replacement (35.23 and 35.24), Norwood procedure for hypoplastic left heart syndrome (ICD-9-CM diagnosis code 746.7 with cardiopulmonary bypass in children <1 year of age, excluding procedure codes for cavopulmonary anastomosis [39.21], and the Fontan procedure [35.94]), open valvotomy (35.1, 35.10, 35.11, 35.12, 35.13, and 35.14), right-ventricular to pulmonary artery conduit (35.92), tetralogy of Fallot repair (35.81), total anomalous pulmonary venous connection repair (35.83), transplant (37.5 and 33.6), truncus arteriosus communis repair (35.83), and ventricular septal defect repair (35.53, 35.62, and 35.72).
Because of the difficulty in estimating the mortality rates for hospitals with very low case volumes, children admitted to centers that performed fewer than 5 cases per year were excluded from the study.
Use of Lower-Mortality Centers
The statistical analysis was performed in 2 steps. First, we divided hospitals into lower-mortality and higher-mortality groups based on adjusted hospital-specific mortality rates. Second, we estimated the impact of insurance type on the odds that a child would use a lower-mortality hospital through multivariate logistic regression modeling.
Designation of Lower-Mortality Centers
Individual hospitals were ranked by adjusted odds of inpatient death based on a logistic regression model. All causes of death before discharge were included. Variables were included in this model according to a stepwise, model-building process. In addition to terms for each hospital, variables considered for inclusion in the model were age (dichotomized at 1 year of age), procedure type (entered as a class variable), urgent admission (including infants who underwent surgery during the admission of their birth), Down syndrome, the presence of multiple congenital anomalies, chronic pulmonary hypertension, failure to thrive, prematurity, and the median family income for each patient's zip code of residence (assigned from the 1990 US Census11) as a proxy for neighborhood affluence.
The inclusion of zip code median family income as a variable was based on substantial evidence to suggest that lower socioeconomic class is associated with poorer overall health and increased risk at the time of medical procedures.12–14 The mechanisms are debated, but we believed it was important to allow for the possibility that children from lower socioeconomic settings were at increased surgical risk related to, for example, poor prenatal care, impaired nutrition, or poor medical management before admission for surgery.
Based on the adjusted odds of inpatient death, centers were divided into lower- and higher-mortality groups using a mortality rate cut-point that allocated ∼50% of patients to each group. To ensure that the hospital grouping resulted in a lower-mortality subset for each insurance type, an analysis examining the effect of use of a lower-mortality center on the odds of death was performed separately for each insurance type.
Analysis of Use of Lower-Mortality Centers
Univariate predictors of use of lower-mortality centers were identified using 2 × 2 χ2 tests for categorical variables (eg, gender and age). In the case of categorical variables with more than 2 possible values (race/ethnicity and procedure type), each value was compared with a reference value (eg, white in the case of race/ethnicity). Wilcoxon rank sum tests were used to determine differences among continuous variables (neighborhood affluence, distance to the hospital of admission, and distance to the nearest lower-mortality hospital), each of which was nonnormally distributed.
Multivariate adjusted odds of use of lower-mortality centers were estimated using a logistic regression model. The outcome of interest was death before discharge, because no additional information about the cause of death was available in the databases. Predictors were included in this model according to a stepwise, model-building process. Variables considered for inclusion were insurance type, race/ethnicity, procedure type (entered as a class variable), age, urgency of admission, Down syndrome, the presence of multiple congenital anomalies, chronic pulmonary hypertension, failure to thrive, prematurity, and zip code median family income.
In addition, 2 terms were included to provide measures of the relative geographic accessibility of the nearest lower-mortality center, because the location of a patient's residence may be correlated with both insurance status and the choice of hospital.15–18 Because street addresses were not included in the datasets used, all measures were based on the distances from the geographic centroids of the hospital and patient residential zip codes.19 We evaluated a number of potential geographic measures, including the actual distances, log-transformed distances, exponent-transformed distances, and a measure of relative closeness of the nearest lower-mortality center relative to the nearest higher-mortality center ([DLM − DHM]/DHM). The log-transformed distances most closely fit the relationship between distance and use of a lower-mortality hospital in our dataset and have been used similarly in other studies.20
Given that distance is perhaps the most important factor predicting choice of hospital, we repeated the analysis restricting the sample to patients who lived within 25 miles of both a lower-mortality hospital and a higher-mortality hospital. Our aim was to ensure that any effects noted were not attributable to geographically isolated groups of patients who had access to only 1 type of hospital (lower-mortality or higher-mortality).
The effect of reduced use of lower-mortality centers on mortality was explored by examining the odds of death for each insurance type: 1) without adjustment for hospital, and 2) with adjustment for lower-mortality hospital use. Additional covariates included in this model were the same as those included in the model used to rank hospitals by mortality. The object of this approach was to identify the extent to which any excess mortality associated with an insurance type (relative to indemnity insurance) was eliminated when adjusting for the type of hospital in which a patient was treated, assuming that any excess mortality that was eliminated could be attributed to differences in hospital use patterns.
Analysis of Lengths of Stay
Differences in lengths of stay between lower-mortality and higher-mortality hospitals were estimated using 3 statistical methodologies: a linear regression model estimating the difference in length of stay in days attributable to hospital type; a logistic regression model estimating the odds of being included in a long length of stay group (defined as >90th percentile length of stay for each procedure type); and a Cox regression proportional hazards model estimating the relative risk of discharge at any point in the hospitalization. The latter model was explored with and without censoring of inpatient deaths, where censoring implies that the true length of stay is not known, because the patient died before discharge. Each of these multivariate models was adjusted for age, race/ethnicity, procedure type, insurance type, urgency of admission, and the presence of multiple congenital anomalies.
Analysis of the Relationship Between Case Volume and Mortality
The relationship between hospital case volume (cases/year) and inpatient mortality was explored using a stepwise logistic regression model estimating the odds of death. Variables considered for inclusion were hospital case volume, procedure type (entered as a class variable), age, urgency of admission, Down syndrome, the presence of multiple congenital anomalies, chronic pulmonary hypertension, failure to thrive, prematurity, and zip code median family income.
All testing was performed at the 5% level of significance (2-sided).
The California state databases from 1992 to 1994 contained complete records for 6580 California resident children younger than 15 years of age who underwent open cardiac surgical procedures at 96 California hospitals. When ambiguous and low-volume procedures and low-volume hospitals were excluded, there were 6252 children at 20 hospitals. Of these, 5071 had indemnity, managed care, or Medicaid insurance and were included in the subsequent analyses (Table 1). Children with indemnity and managed care insurance were predominantly white and from higher income zip codes, whereas those with Medicaid insurance were predominantly Hispanic and from less affluent zip codes.
Use of Lower-Mortality Centers
Designation of Lower-Mortality Centers
The covariates included in the hospital ranking model after stepwise selection included age, zip code median family income, each procedure type, urgency of admission, and the presence of prematurity, as well as terms for each individual hospital (Table 2). When hospitals were ranked by adjusted odds of inpatient death, 11 hospitals with the lowest adjusted mortality comprised 48.2% of the total patient population and were designated as the lower-mortality hospital group. The adjusted odds of inpatient death for hospitals in the lower-mortality group (relative to the center with the highest case volume) ranged from .64 to 1.33, whereas for those in the higher-mortality group, which included 9 centers, they ranged from 1.45 to 5.93. The mean case volumes for the lower-mortality and higher-mortality centers were 234 and 281 cases over 3 years, respectively.
To ensure that the hospital grouping was relevant within each insurance type, an analysis was performed examining the effect of use of lower-mortality centers on the odds of death within each insurance group, adjusting for the same covariates included in the model used to rank hospitals. All insurance types were associated with significantly lower odds of death at lower-mortality hospitals (Table 3).
Use of Lower-Mortality Centers—Univariate Effects
Compared with children with indemnity insurance, children with managed care and Medicaid insurance had significantly lower unadjusted rates of use of lower-mortality centers (Table 4; managed care, odds ratio [OR]: .75; 95% confidence interval [CI] .70,.81; Meicaid, OR: .71; 95% CI: .67,.76). Non-white race/ethnicity and atrial septal defect repair were also significant univariate predictors of reduced use of lower-mortality centers, whereas younger age, heart transplant, and urgent admission were predictors of increased use of lower-mortality centers. Children admitted to lower-mortality centers lived farther from their hospital of admission than did children admitted to higher-mortality centers.
The final multivariate model estimating the odds that a patient would use a lower-mortality center included the following covariates: insurance type, race/ethnicity, each of the procedure types, and each of the distance factors.
Compared with children with indemnity insurance (Table 5), children with managed care insurance were significantly less likely to be admitted to a lower-mortality hospital for surgery (OR: .53; 95% CI: .45,.63; P < .001), even after adjusting for the covariates listed above, as were children with Medicaid insurance, albeit to a lesser degree (OR: .71; 95% CI: .61,.84; P < .001). When we restricted the sample to patients living within 25 miles of both a lower-mortality and higher-mortality hospital, the results were similar. Compared with patients with indemnity insurance, patients with managed care insurance were less likely to receive surgery at a lower-mortality hospital (OR: .60; 95% CI: .48,.74; P < .001), as were patients with Medicaid insurance (OR: .68; 95% CI: .55,.83; P< .001).
We explored the sensitivity of our results to the chosen cut-point of 50% inclusion in the lower-mortality group by varying this threshold from 32% to 72% of the total number of patients (Table 6). For each of these supplementary analyses, the odds that a child with managed care insurance would use a lower-mortality center were substantially reduced relative to children with indemnity insurance, ranging from .25 (95% CI: .20,.31;P < .001) to .53 (95% CI: .45,.63; P< .001). For children with Medicaid insurance the impact of insurance type on use of lower-mortality centers was similar, ranging from .61 (95% CI: .52,.73; P < .001) to .71 (95% CI: .61,.84; P < .001).
Although not included in the final model, a significant interaction between insurance type and race/ethnicity was noted during the model-building process. An analysis stratifying by race/ethnicity is presented in Table 6. Again, the adjusted odds that a child with managed care insurance was admitted to a lower-mortality hospital ranged between .40 (95% CI: .27,.60; P < .001) and .62 (95% CI: .50,.77; P < .001) for each of the racial/ethnic groups, in each case, a highly statistically significant result. However, use of lower-mortality centers among patients with Medicaid insurance varied with racial/ethnic groups. Compared with children with indemnity insurance, non-white children with Medicaid insurance were less likely to use a lower-mortality hospital, whereas the odds of using a lower-mortality center was similar between white children with indemnity and Medicaid insurance.
Finally, to explore whether the reduced use of lower-mortality hospitals by children with managed care insurance had a demonstrable effect on mortality, an analysis was performed examining the odds of death for each insurance type with and without adjustment for hospital use patterns (Table 7). Without adjustment for hospital of admission, children with managed care insurance were significantly more likely to die after surgery (OR: 1.54; 95% CI: 1.07,2.22; P = .021) relative to children with indemnity insurance. With adjustment for lower-mortality hospital status, the excess mortality no longer achieved statistical significance (OR: 1.29; 95% CI: .89,1.87; P = not significant [NS]). Children with Medicaid insurance did not have significantly increased adjusted mortality compared with children with indemnity insurance with or without adjustment for lower-mortality hospital status.
Hospital Lengths of Stay
Lengths of stay did not differ significantly between lower-mortality and higher-mortality hospitals using 3 different regression models. The linear regression model revealed a slightly shorter length of stay at lower-mortality hospitals with a difference of −.54 days (95% CI: −1.50,.41; P = NS). The logistic regression model estimated the relative odds of an admission lasting beyond the 90th percentile for the patient's procedure type at .43 for the lower-mortality hospital group relative to the higher-mortality group, although this finding did not reach statistical significance (OR: .43; 95% CI: .16,1.15; P = NS). Finally, a proportional hazards model estimating the relative risk of discharge at any point during a hospitalization revealed a relative odds of .98 for lower-mortality centers relative to higher-mortality centers (95% CI: .92,1.04; P = NS) when patient deaths were censored, and .95 (95% CI: .90,1.00; P = NS) when admissions ending in death were not censored.
Analysis of the Relationship Between Case Volume and Mortality
The final multivariate model estimating the relationship between case volume and mortality after stepwise selection included the following covariates: hospital volume, each of the procedure types, age, urgency, prematurity, and zip code median family income. Increasing hospital case volume was associated with a significant reduction in the odds of death (OR: .75; 95% CI: .62,.90 for each increase in case volume by 100 cases; P = .003).
The present study demonstrates that, during the study, managed care status had an important effect on hospital use patterns among children undergoing heart surgery in California. Compared with children with indemnity insurance, those with managed care insurance were substantially less likely to undergo surgery at a center with a record of low mortality, after adjusting for race/ethnicity, each of the procedure types, and the distance to the nearest lower-mortality and higher-mortality hospital. Children with managed care insurance had a significantly higher mortality after cardiac surgery relative to children with indemnity insurance. At least some of this excess mortality was eliminated when adjusting for use of lower-mortality centers, reducing the differences in mortality between insurance types to nonsignificance. Multivariate analysis suggested that differences in hospital use patterns were at least partly responsible for the increased mortality associated with managed care insurance.
Use of lower-mortality hospitals was also reduced for some, but not all, subgroups of patients with Medicaid insurance. Most notably, among non-white children, Medicaid insurance was associated with significantly reduced use of lower-mortality centers relative to indemnity insurance, whereas for white children, Medicaid insurance was not associated with significantly lower use of lower-mortality centers.
The reasons for such a difference in referral patterns for managed care patients cannot be deduced from these data alone, but payer policies could be involved. Managed care plans seek to provide care at reduced cost, often through controls over referral patterns that aim to reduce use of hospitals with higher costs.21 Although such policies might have minimal impact on outcomes for common or low-risk procedures, it is possible that uncommon or more high-risk disorders, such as pediatric heart surgery, could be adversely affected.22,,23 In addition, the impact of such referral policies on outcomes would be difficult to identify without specific programs dedicated to that purpose.24 Industry-wide monitoring of referral patterns and outcomes could, however, provide the information necessary for more appropriate use of resources for various patient populations.
An issue that warrants discussion is whether higher-mortality hospitals might have higher mortality because they have a disproportionate number of managed care patients, who are higher risk for some reason to begin with. Although the increased surgical mortality among children with managed care insurance theoretically could have been caused by referral of a sicker population, the criteria for surgical repair of most congenital heart lesions are relatively uniform among the small community of pediatric cardiologists. In addition, during this study, few nonsurgical options existed. For example, there were no catheter-delivered atrial septal defect closures in California during the study (E. Sideris, G. Das, and K. Jenkins, personal communication, April 1997).
The present study confirms previous reports relating increased case volume to reduced mortality for pediatric heart surgery.25,,26 This analysis was included to offset concerns, based on the average case volumes of the 2 hospital groups, that the reverse might be the case. Although it is true that the lower-mortality hospital group had a lower average case volume than the higher-mortality group, the actual relationship between case volume and mortality is obscured when the data are aggregated at the hospital level and dichotomized.
Our study inferences should be viewed in light of the limitations of the datasets used. The use of large administrative hospital discharge data added certain strengths to the study, such as the option to restrict the study population to patients with a homogenous group of procedures, as well as avoidance of selection biases, because submission to the dataset is mandated by state legislation. Such datasets, however, present greater challenges in controlling for medical comorbidities than do prospectively collected data. For example, preoperative risk factors, such as hypoxia and acidosis included in more specialized datasets,27 were not included in the California administrative datasets. In contrast, we were able to investigate many other important potential predictors of mortality, including age, urgency of admission, a proxy for socioeconomic status, and medical risk factors such as Down syndrome, the presence of multiple congenital anomalies, chronic pulmonary artery hypertension, failure to thrive, and prematurity. In addition, we controlled for procedure types individually, in contrast to previous studies, which have adjusted for procedure risk by grouping procedures into categories.27,,28 One advantage to analyzing procedures individually is that much of the variability in preoperative comorbidities that were not available in the California administrative datasets may be included in the variability of individual procedures (ie, cyanosis and acidosis are common in transposition of the great arteries, but rare in ventricular septal defects).
Linked aggregate Census data were used to adjust for zip code-specific median family income, a proxy for socioeconomic status. Such data are ecologic in that they assign values from a group to the individuals within the group and may be misleading when used to determine the significance of correlations among individuals.29–31Consequently, care must be taken in the interpretation of the results related to this variable, particularly with respect to the levels of significance.
Another limitation includes the use of length of stay as a proxy for hospitalization costs. Of the 2 cost measures available in the datasets, lengths of stay and hospital charges, only the former was available for patients admitted to hospitals operated by managed care organizations. Length of stay does not reflect all factors contributing to cost, such as variability in the use of expensive resources over the course of a hospitalization. Short of an actual measure of costs, however, numerous studies have demonstrated that length of stay is an excellent correlate of hospital costs.32,,33 Indeed, the recognition that length of stay is a major determinant of costs has driven an increasing interest in methods to predict34–37and reduce32 the duration of hospitalizations.
In summary, between 1992 and 1994, children in the state of California with managed care health insurance were significantly less likely to use lower-mortality cardiac surgical centers than were children with indemnity insurance. In addition, lower-mortality hospitals were not associated with increased lengths of stay, a proxy for health care costs, suggesting that lower-mortality hospitals were no more expensive than higher-mortality centers. Further research is necessary to identify the specific mechanisms and generalizability of our findings to other uncommon, technically complex procedures and to other states.
This research was supported in part by the Kobren Fund, Burton G. Bettingen Corporation, and Mattina Proctor Fund.
- Received March 1, 1999.
- Accepted November 15, 1999.
Reprint requests to (L.C.E.) Department of Cardiology, Children's Hospital, 300 Longwood Ave, Boston, MA 02115. E-mail:
- HMO =
- health maintenance organization •
- ICD-9-CM =
- International Classification of Diseases, Ninth Revision-Clinical Modification •
- OR =
- odds ratio •
- CI =
- confidence interval •
- NS =
- not significant
- ↵Group Health Association of America. Patterns in HMO Enrollment. Washington, DC: Group Health Association of America; 1994
- ↵American Association of Health Plans. Managed Health Care Overview. Washington, DC: American Association of Health Plans; 1996
- ↵Orenstein DM. Cystic Fibrosis. In: Rudolph AM, Hoffman JIE, Rudolph CD, eds. Rudolph's Pediatrics. Stamford, CT: Appleton and Lange; 1996:1640–1650
- Rizzoli G,
- Mazzucco A,
- Fracasso A,
- et al.
- ↵Office of Statewide Health Planning and Development. 1992 Discharge Data. Sacramento, CA: Office of Statewide Health Planning and Development; 1993
- ↵Office of Statewide Health Planning and Development. 1993 Discharge Data. Sacramento, CA: Office of Statewide Health Planning and Development; 1996
- ↵Office of Statewide Health Planning and Development. 1994 Discharge Data. Sacramento, CA: Office of Statewide Health Planning and Development; 1996
- ↵Bureau of the Census. Census of Population and Housing, 1990: Summary Tape File 3 on CD-ROM. Washington, DC: Bureau of the Census; 1992
- Marshall SW,
- Kawachi I,
- Pearce N,
- Borman B
- ↵Bureau of the Census. Zip Code Centers With Longitude and Latitude Values. Washington, DC: Bureau of the Census; 1995. Available at: www.census.gov/tiger/pts/zip cent.zip
- ↵Prospective Payment Assessment Commission. Medicare and the American Health Care System: Report to Congress, June 1994. Washington, DC: Prospective Payment Assessment Commission; 1994
- Newacheck PW,
- Hughes DC,
- Stoddard JJ,
- Halfon N
- Schlesinger M,
- Mechanic D
- ↵Scanlon WJ. Medicaid Manage Care: Challenge of Holding Plans Accountable Requires Greater State Effort. Washington, DC: United States General Accounting Office; 1997. General Accounting Office/Health, Education, and Human Services Division-97-86
- Jenkins KJ,
- Newburger JW,
- Lock JE,
- et al.
- Hannan EL,
- Racz M,
- Kavey RE,
- Quaegebeur JM,
- Williams R
- Hannan EL,
- Racz M,
- Kavey RE,
- Quaegebeur JM,
- Williams R
- Jenkins KJ,
- Newburger JW,
- Lock JE,
- et al.
- Tu JV,
- Jaglal SB,
- Naylor CD
- Tu JV,
- Mazer CD,
- Levinton C,
- Armstrong PW,
- Naylor CD
- Copyright © 2000 American Academy of Pediatrics