Ruey-Kang R. Chang, MD, MPH; Sandra Rodriguez, MS; Maggie Lee; MPH, Thomas S. Klitzner, MD, PhD 

Am Heart J.  2006;152(2):386-393.  ?2006 Mosby, Inc.

Abstract and Introduction

Abstract

Background: Little is known regarding the risk factors for early and late death after hospital discharge among pediatric patients undergoing cardiac surgery.
Methods: Statewide hospital discharge data from California in 1989 to 1999 were used to study outcomes of children <18 years old who had a procedure code (by International Classification of Diseases, Ninth Revision, Clinical Modification) indicating cardiac surgery. The outcome variable was death occurring after hospital discharge. Postdischarge deaths were identified by linking hospital discharge data to statewide death registry data. Cardiac surgical procedures were grouped into 23 categories to adjust for risk involved with the procedures. We used logistic regression to evaluate risk factors for postdischarge mortality, including variables age, sex, race and ethnicity, type of insurance, home income, date and month of surgery, type of admission, hospitals case volume, and the various types of procedures.
Results: There were 25402 cardiac surgery cases with 1505 inhospital deaths. Of 23897 hospital discharges, 148 deaths (0.62%) occurred within 365 days after discharge, including 37 deaths within 30 days; 44 deaths at 31 to 90 days; and 67 deaths at 91 to 365 days. Logistic regression showed young age was an important risk factor for postdischarge death with an odds ratio of 4.8 for neonates and 3.5 for infants, compared with children >1 year old. Another significant risk factor was the type of procedure. For death <30 days after discharge, Norwood operation (odds ratio 8.4 compared with closure of ventricular septal defect) was a risk factor. For death that occurred between 31 and 365 days, significant risk factors were truncus arteriosus repair, total anomalous pulmonary vein repair, aortopulmonary shunt, and open valvotomy. Sex, race/ethnicity, home income, and hospital case volume were not significant predictors of postdischarge deaths.
Conclusions: Many demographic and socioeconomic variables affecting inhospital death were not significant predictors for postdischarge death. Important risk factors for postdischarge death were young age and the type of surgery performed.

Introduction

Outcomes of pediatric cardiac surgery have improved dramatically over the past 20 years.^[1-4] The surgical mortality for congenital heart disease in the current era is in the range of 5% to 6% in most large reported series. Most studies, however, define surgical outcome as "inhospital" mortality or the status of patient at the time of hospital discharge. A more appropriate measure of immediate outcome after surgery would be "30-day mortality"; however, it is generally more difficult to ascertain this information because many centers do not track mortality after hospital discharge.^[5] Very limited information has been reported on deaths occurring after hospital discharge in children undergoing cardiac surgery.

Deaths occurring in the community setting among patients with congenital heart disease (CHD) have been examined by some investigators. Most of the postdischarge deaths in these patients are characterized as "sudden death." In a population-based study evaluating patients <19 years old who underwent surgical treatment of CHD in Oregon from 1958 to 1996, Silka et al^[6] reported that most deaths occurred in patients with aortic stenosis, coarctation, transposition of the great arteries, and tetralogy of Fallot. The common causes of death are arrhythmia and circulatory (embolic or aneurysm rupture) and acute heart failure.^[6]

Although it has been known to clinicians that some deaths after pediatric cardiac surgery occur after hospital discharge, the incidence and characteristics of these occurrences are unknown. Identifying children who are at risk for postdischarge death after CHD surgery will allow clinicians to arrange proper surveillance and follow-up, which have the potential to improve outcomes of these patients. Therefore, we conducted this study with 2 objectives: to characterize the epidemiology of postdischarge death among infants and children undergoing cardiac surgery and (2) to identify risk factors for early and late postdischarge death after pediatric cardiac surgery.

Methods

Data Sources

Abstracted hospital discharge data from the California Office of Statewide Health Planning and Development (OSHPD) were used to conduct the analysis.^[7] The OSHPD database includes all discharges from >500 acute care hospitals in California. In the current study, we used the OSHPD data on hospital discharges from January 1989 to December 1999. The OSHPD data contain International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) discharge diagnosis and procedure codes assigned by California hospitals to each individual discharge during the year.

To capture postdischarge death, we also used 1989 to 1999 statewide death data from the Death Statistics Master File from Vital Statistics compiled by the California Department of Health Services, Center for Health Statistics.^[8] The Death Statistics Master File data include state-mandated reports of all deaths to the California Department of Health Services. Data fields available in the public release of Death Statistics data include date of death, date of birth, sex, race/ethnicity, operation performed before death, underlying cause of death (in International Classification of Diseases, Ninth Revision codes), and 5-digit home zip code.

Case Selection from OSHPD

We selected pediatric patients (<18 years of age) with procedure codes in the database indicating cardiac surgery. The procedure and diagnosis codes in the OSHPD data are based on the ICD-9-CM. The list of cardiac procedures and the corresponding ICD-9-CM codes used to select study samples in the current study has been reported in a previous article.^[9]

Identifying Postdischarge Death

We identified postdischarge deaths by linking the OSHPD hospital discharge data with Death Statistics data. From the Death Statistics data, we first selected cases based on age at death <18 years, operation before death, and underlying cause of death for cardiovascular disease. Cases of inhospital death from the OSHPD data were matched with the list from Death Statistics data by linking the age, sex, cardiac diagnosis, and 5-digit home zip code. Matched cases were eliminated for the postdischarge death analysis.

From the list of selected Death Statistics cases after eliminating those who died in hospital, we matched reported deaths with OSHPD cases that indicated alive at discharge. The matching algorithm used patient age (Death Statistics minus OSHPD age difference <365 days), sex, 5-digit home zip code, and diagnosis (OSHPD diagnosis matched with underlying cause of death). Because the assignment of principal diagnosis may differ between OSHPD and Death Statistics databases, we also matched the Death Statistics principal diagnosis to the OSHPD diagnostic fields. If 1 case in Death Statistics data matched to multiple cases in the OSHPD data, we used additional data fields including date of discharge, race, and ethnicity to determine the matched case in OSHPD. If a patient had subsequent hospitalizations and cardiac procedures and died in hospital, the patient was not included in the postdischarge deaths because the death was likely attributable to the subsequent procedure.

In the OSHPD database, ages of patients were listed in days and years for those <4 years of age and only in years for those ≥4 years of age. From both OSHPD and Death Statistics databases, we calculated the age of each patient in days. For patients <4 years of age at admission, the age of patient at hospital discharge was calculated by "age in days at admission" plus "length of hospital stay." Among the patients matched as postdischarge deaths, days from discharge to death were determined by the difference between age at death and age at discharge.

For patients ≥4 years of age, we used the data on the month of admission available in OSHPD data and assumed the date of admission to be the 15th of the month of admission. The date of hospital discharge was calculated by date of admission plus length of hospital stay. The number of days from discharge to death was calculated by the difference between the date of discharge calculated from OSHPD data and the date of death listed in Death Statistics data.

Study Variables

Postoperative death was used as the main outcome variable. Postoperative death included inhospital death, which was identified by the "discharge status" in the OSHPD database, and postdischarge death. Postdischarge deaths were identified by linking the OSHPD data with death registry data, as detailed in the previous section.

To determine the risk for postdischarge death, we selected the following patient-related variables and healthcare system?related variables. The patient variables were age, sex, race and ethnicity, type of insurance, and family income. For the age variable, we divided patients into 3 age groups: neonate (<1 month), infant (1 month to 1 year), and child (>1 year). Race/ethnicity included white, black, nonwhite and nonblack Hispanic, Asian and Pacific Islander, and others (including Native Americans). Type of insurance included private insurance (traditional indemnity insurance), managed care (including health maintenance organization [HMO] and preferred provider organization [PPO]), public insurance (Medicaid and other government-sponsored programs), and others. Family income, defined by median household income by home zip code, was grouped to >$60000, $40000 to $60,000, $20000 to $40000, and <$20,000. Information on median income of zip codes was obtained from 1990 US Census data.

Selected healthcare system variables were month of surgery, day of surgery, type of admission, source of admission, and hospital surgical volume. Month of surgery was grouped by quarter: January to March, April to June, July to September, and October to December. Day of surgery was dichotomized to weekday (Monday through Friday) and weekend (Saturday and Sunday). Type of admission comprised elective and nonelective. Source of admission was defined as admission through emergency department or scheduled admission. Hospital average annual case volume was used to define the hospitals as low volume (≤100 cases per year) and high volume (>100 cases per year).

Adjustment for Medical Risk

Pediatric cardiac surgery comprises a variety of distinctly different procedures, and many procedures have small numbers of cases insufficient for multivariate analysis. Recently, a consensus-based risk adjustment method for congenital heart surgery, The Risk Adjustment for Congenital Heart Surgery, was developed.^[10,11] The Risk Adjustment for Congenital Heart Surgery, which stratifies congenital heart surgery into 6 risk categories, is useful for risk adjustment of small number of cases but may be inadequate for analyzing large sample size database. We have previously used a risk adjustment strategy, which expands the number of procedure categories to 23 to improve the homogeneity of the procedures within each group while maintaining adequate number of cases in each of the groups to allow multivariate analysis.^[9] We used this set of 23 surgical procedure groups to adjust for risks involved in the various types of procedures in the current study.

To determine the type of procedure each patient underwent, we examined the Principal Procedure variable and 3 Other Procedure variables listed in each patient's discharge data. In addition, we examined the Principal Diagnosis and Other Diagnosis variables to further determine the type of procedure each patient underwent and to eliminate cases with coding errors. Each patient was assigned to one of the following 23 procedure groups (listed in "Type of procedure" in Table I ). The designation of procedure groups has been detailed previously.^[9]

In this study, we included 4 comorbidity conditions listed in the OSHPD database as independent medical variables to account for some noncardiac medical variables: Down syndrome (code 758.0), pulmonary hypertension (code 416.0), failure to thrive (code 783.4), and prematurity (codes 765.0 and 765.1).

Data Analysis

Logistic regression models were used to determine risk factors for mortality and calculate odds ratios. The dependent variable was binary: alive or dead. Regression models were created using the following outcomes as dependent variables, respectively: inhospital mortality, 30-day postdischarge mortality, late (31-365 days) postdischarge mortality, and overall (0-365 days) post discharge mortality. The following independent variables were used to identify risk factors for mortality. Within each independent variable group, 1 subgroup was selected as the reference to calculate odds ratio (OR) for mortality for the other subgroups relative to the reference group: male for "sex," child (age>1) for "age," white for "race/ethnicity," public insurance for "type of insurance," annual family income >$60000 for "family income," weekday for "day of surgery," October to December for "month of surgery," elective for "type of admission," non?emergency room for "source of admission," high volume hospital (annual case volume >100) for "cardiac center," ventricular septal defect (VSD) closure for "type of surgical procedure," and absence of comorbidity for the comorbidity variables. Data for each group of variables are presented as odds ratios comparing each subgroup to its respective reference subgroup.

Results

There were 25402 cardiac surgery cases with 1505 inhospital deaths. Of 23897 "alive" hospital discharges, 148 deaths (0.62%) occurred within 365 days after discharge, including 37 deaths, which occurred within 30 days; 44 deaths at 31 to 90 days; and 67 deaths at 91 to 365 days. Of the 148 deaths within 365 days from discharge, there were 76 neonates, 50 infants, and 22 children (>1 year old). These deaths represent 2.1% of neonates who were alive at discharge, 0.7% infants, and 0.2% children. Twelve patients (8.1% of all postdischarge deaths) were >4 years of age. Most postdischarge deaths (95.5%) occurred in hospitals. There were 5 deaths that occurred at home and in additional 2 patients who were dead on arrival to a hospital.

Table I summarizes the demographic and clinical information of the study population. There were 95 deaths among 13068 boys who were alive at discharge (0.73%). This is significantly >53 deaths among 10827 girls who were alive at discharge (0.49%, P = .02). Postdischarge deaths were similar among race/ethnic groups: white 0.58%, black 0.65%, and Hispanic 0.73%. The postdischarge death rate for nonelective surgeries was 4 times that for elective surgeries (1.57% vs 0.39%, P < .01). Of various surgical procedures, Norwood operation was associated with the highest postdischarge death rate (6.77%), followed by aortopulmonary shunt with atrial septostomy (3.45%), repair of total anomalous pulmonary venous return (2.41%), truncus arteriosus repair (2.38%), and aortopulmonary shunt (2.13%).

In the univariate analysis on the risk for postdischarge mortality, we found the mortality in boys was higher than girls (relative risk [RR] 1.34, P = .02), neonates and infants were higher than children >1 year old (RR 13.0 and 14.6, P < .01, respectively), surgeries performed on weekends were higher than surgeries performed on weekdays (RR 1.7, P = .01), and nonelective surgeries were higher than elective surgeries (RR 4.6, P < .01). Median income is also an important factor for postdischarge death?lower median income was related to higher risk ratio for postdischarge death. Compared with VSD closure, the relative risk for postdischarge mortality was 33 for Norwood operation (RR, 12 for truncus arteriosus and total anomalous pulmonary venous return [TAPVR] repair; RR, 11 for aortopulmonary shunt procedure).

Results from the multivariate logistic regressions are presented in Table II . The logistic regression model using all postdischarge deaths (0-365 days post discharge) as the dependent variable showed OR for death was significantly higher in neonates and infants when compared with children >1 year old (OR 4.76 and 3.46, respectively, P < .001). There was no sex difference in postdischarge death after risk adjustment. Comparing procedure type to closure of VSD, the Norwood operation (OR 8.39), truncus arteriosus repair (OR ,6.01), repair of TAPVR (OR 5.84), aortopulmonary shunt (OR 5.36), right ventricle to pulmonary artery conduit (OR 4.14), and thoracic vessel procedures such as coarctation repair and pulmonary artery banding (OR 3.07) had significantly higher risk. In Table II , the results from regression using the combined inhospital and postdischarge deaths as the dependent variable are presented for comparison.

To identify risk factors for deaths within 30 days from hospital discharge, we conducted logistic regression using the postdischarge deaths that occurred ≤30 days post discharge as the dependent variable. There was no sex difference in 30-day postdischarge death after risk adjustment. The only significant predictors for 30-day postdischarge death were age groups of neonates and infants and Norwood operation (OR 8.39).

For deaths occurred between 31 and 365 days, significant predictors were newborns and infants and procedures including truncus arteriosus repair, TAPVR repair, open valvotomy, aortopulmonary shunt, right ventricle to pulmonary artery conduit, thoracic vessel procedures, and aortopulmonary shunt (all procedures compared with VSD closure). Of the above logistic regression models for overall, early, and late postdischarge deaths, we found that sex, race/ethnicity, home income, and hospital surgical volumes were not significant predictors for postdischarge deaths.

Discussion

The literature contains very limited information on postdischarge death among infants and children who undergo cardiac surgery. Because most surgical centers do not track mortality after hospital discharge, most studies on outcomes of CHD surgery from the United States have relied on inhospital mortality.^[5] In the current study, we were able to link hospital discharge record to death registry data and identify postdischarge deaths. In addition, we were able to identify the day when postdischarge death occurred and further categorized the deaths to <30 days, 31 to 90 days, and 91 to 365 days post hospital discharge. The numbers of days between hospital discharge and death in patients ≥4 years old were less precise because of the assumption of date of admission on the 15th of the month. However, patients ≥4 years old accounted for a very small portion (only 8%) of all postdischarge deaths.

In this study, we excluded patients who, after initial hospital discharge, were readmitted to the hospital and died in hospital after subsequent cardiac procedures. We found that postdischarge deaths within 1 year of surgery were not very common, occurring in only 0.62% of all discharges. The low incidence of postdischarge death may be related to the exclusion of subsequent inhospital deaths after another cardiac procedure. It is also possible that our linking algorithm might be too stringent that some cases might have been missed. Our database-linking algorithm was designed to identify cases accurately and to be relatively certain that the linked cases referred to the same patient. If there were discrepancies in the manner that data were reported between the 2 databases, some cases might be missed using this algorithm.

Many variables have been shown to be risk factors for inhospital mortality in previous studies.^[9] These variables include female sex, type of insurance, nonelective surgeries, surgeries on weekends, and case volume of the surgical center; we did not find statistical significance in the logistic regression model in predicting postdischarge death for these variables. One possible explanation is that postdischarge deaths are much less frequent than inhospital deaths; therefore, even with a sample size of 25402 cases, the statistical power was inadequate.

Age is an important independent risk factor for both inhospital and postdischarge death. Neonates (<30 days old) are at particularly high risk for death after surgery, followed by infants <1 year of age. This is consistently seen in inhospital and postdischarge deaths. In fact, age is a more significant risk factor for postdischarge deaths than inhospital death (comparing the OR in Table II ).

Infants with hypoplastic left heart syndrome who have undergone the Norwood procedure are a group known to be at risk for late mortality.^[12-14] These deaths typically occur between Norwood procedure and Glenn shunt, which is normally performed at 3 to 6 months of age. In the present study, we found that Norwood procedure is a risk factor for 30-day postdischarge mortality, consistent with the findings of previous studies. Norwood procedure was not a significant risk factor for late mortality probably because most HLHS patients who had undergone Norwood had been operated for a Glenn shunt by 3 to 6 months of age.

Aortopulmonary shunt is a significant risk for death beyond the early postoperative period. Fenton et al reported that 21 (14%) of 146 infants who underwent aortopulmonary shunt died after discharge and before further planned surgery.^[15] At autopsy, cause of death were attributed to shunt thrombosis in one third of cases.^[15] TAPVR repair was also a risk factor for late death. This may be related to pulmonary venous stenosis, which occurs in 11% of TAPVR repair cases, and results in pulmonary hypertension.^[16]

Sudden death is known to occur in patients who have CHD with or without surgical correction.^[17-19] Common causes of sudden cardiac death in patients with CHD include from arrhythmia, left ventricular outflow tract obstruction, pulmonary vascular disease, heart failure, and coronary ischemia. In the present study, we were not able to identify the cause of death and, therefore, cannot determine weather the death was directly or indirectly related to the surgery.

The current study identified risk factors for early and late death after hospital discharge for CHD surgery in children. Our findings indicate that neonates and infants are at increased risk and require frequent follow-up and investigative studies. Patients who underwent Norwood operation, aortopulmonary shunt, TAPVR repair, and truncus arteriosus repair are also at risk for early postdischarge death and, therefore, need more frequent periodic follow-up surveillance.

Limitations

A major limitation of this study is the data accuracy in an administrative database. Missing data and miscoding of patients selected for our study may exist and potentially bias our findings. Another limitation inherent to an administrative database is the limited clinical information available in data collected primarily for administrative purposes.

We found the overall rate of postdischarge deaths to be 0.62%. Although there is no previous study for comparison, this rate is quite low. As stated previously, this low rate of postdischarge death could be due to the more stringent data matching algorithm we used in the current study. Further validation of data matching may be needed.

Conclusion

Many demographic and socioeconomic variables affecting inhospital death were not significant predictors for postdischarge death. Important risk factors for postdischarge death were young age and the type of surgery performed. Neonates and infants who undergo Norwood procedure, aortopulmonary shunt, TAPVR, and truncus arteriosus repair are at high risk for postdischarge death therefore need close follow-up.