Female Sex as a Risk Factor for In-Hospital Mortality Among Children Undergoing Cardiac Surgery
Background— The purpose of this study was to investigate whether sex disparity in cardiovascular outcomes exists in children who undergo cardiac surgery.
Methods and Results— Statewide hospital discharge data from California from 1995 to 1997 were used. Children <21 years old who had a procedure code (by ICD9-CM) that indicated cardiac surgery were selected. The outcome variable was binary, in-hospital death versus alive at discharge. Twenty-three surgical procedures were selected and adjusted for risk by procedure type. We used logistic regression analysis to evaluate the effect of sex on in-hospital mortality, controlling for age, race and ethnicity, type of insurance, home income, type of admission, date and month of surgery, hospital case volume, and type of procedure. There were 6593 cases of cardiac surgery, with 345 in-hospital deaths (mortality rate 5.23%). Crude mortality rates for males (4.98%) and females (5.54%) were not significantly different. However, fewer females were neonates, and females had more low-risk procedures than males. Multivariate logistic regression showed that females had a higher odds ratio (OR) for mortality than males (OR 1.51, P<0.01). The OR for mortality was 3.86 for neonates and 2.98 for infants compared with children aged ≥1 year. Low-volume hospitals had higher mortality rates than high-volume hospitals (OR 1.67, P<0.01). The risk-adjusted length of hospital stay and charges were similar between females and males.
Conclusions— For children undergoing cardiac surgery, female sex was associated with 51% higher odds of death than male sex. The mechanism by which female sex acts as a risk factor requires further investigation.
Received May 6, 2002; revision received June 19, 2002; accepted June 25, 2002.
In adults undergoing CABG surgery, many studies have reported higher mortality rates in women.1–3⇓⇓ The risk-adjusted odds ratio (OR) for mortality for females compared with males is generally reported to be 1.5 to 2.5.4–6⇓⇓ Weintraub et al7 reported 13 368 patients undergoing CABG at a single institution and found an OR for risk-adjusted mortality of 1.79 for females compared with males. O’Connor et al8 reported an OR of 2.23 for females. Although the finding of higher mortality for females undergoing cardiac surgery has been reported repeatedly, the mechanism by which sex acts as an independent risk factor for adult coronary artery disease remains unclear. Investigators have postulated that outcome differences by sex are due to biological factors, referral bias, and differences in utilization of services.9–11⇓⇓ It is unknown whether a similar sex disparity in cardiovascular outcomes exists among children undergoing surgery for congenital heart disease (CHD).
We hypothesized that sex is an important determinant for surgical outcomes of children with CHD. We further hypothesized that the causes for the sex difference in outcomes include biological factors and factors related to healthcare access and utilization. The objectives of the present study were to determine whether the risk-adjusted in-hospital mortality rate is higher in female pediatric patients than in male patients undergoing cardiac surgery and to explore the possible causes for sex difference in cardiac outcomes in children.
Abstracted hospital discharge data from the California Office of Statewide Health Planning and Development (OSHPD) were used to conduct the analysis. The OSHPD database includes all discharges from >500 acute-care hospitals in California. In the present study, we used OSHPD data on hospital discharges from January 1995 to December 1997. The OSHPD database contains International Classification of Diseases, Ninth Revision, Clinical Modification (ICD9-CM) discharge diagnosis and procedure codes assigned by California hospitals to each individual discharge. Fields are provided for up to 24 diagnoses and 20 procedures. Routine demographic and administrative information such as age, sex, race, admission type and source, discharge status, length of hospitalization, and total hospital charges are listed in the OSHPD database.
We selected pediatric patients (<21 years of age) with procedure codes in the database that indicated 1 or more cardiac surgical procedures listed in Table 1. Patients undergoing closure of patent ductus arteriosus were excluded from the present study.
Study Variables and Risk Adjustment
In-hospital death, identified by the discharge status in the OSHPD database, was used as the main outcome variable. To compare differences in mortality between males and females, we used a multivariate regression analysis that included patient-related, healthcare system-related, and medical variables as independent variables. Each independent variable was divided into mutually exclusive subgroups to characterize each patient undergoing surgery. In addition to sex, the patient-related variables were age, race and ethnicity, type of insurance, and family income (by median household income of the home zip code). Selected healthcare system-related variables were month of surgery, day of surgery, type of admission, source of admission, and hospital case volume of pediatric cardiac surgery. The subgroups for each of these variables are shown in Table 2.
Medical variables included in the study included both the type of surgical procedure performed and the presence or absence of 1 or more of 4 comorbid conditions. Pediatric cardiac surgery comprises a variety of distinctly different procedures, and many procedures have a small number of cases, insufficient to allow multivariate analysis to be conducted on each surgical group individually. Previous investigators used different strategies to group common pediatric cardiac procedures into 4 risk categories primarily defined by expert opinion.12,13⇓ However, these risk-adjustment strategies may not adequately account for the complexity and variety of cardiac surgical procedures in children. Recently, Erickson et al14 proposed a risk-adjustment strategy that includes the majority of pediatric cardiac surgical procedures (16 procedure groups) as independent variables in a multivariate analysis. This approach may be appealing, because the medical risk for patients in each procedure group is more homogeneous. We modified Erickson’s risk-adjustment strategy by expanding to 23 mutually exclusive procedure groups to further improve the homogeneity of the cases within each group while maintaining an adequate number of cases in each group to allow for a meaningful multivariate analysis. These groups are listed in Table 2.
To determine the type of procedure each patient underwent, we examined the principal procedure and other procedures listed in each patient’s discharge data. Although in some cases there were many other procedures listed in the database, we examined the first 4 other procedures to standardize the selection process. In addition, we examined the principal diagnosis and other diagnoses to further verify the type of procedure each patient underwent and to eliminate cases with ambiguous coding and coding errors. Each patient was assigned to 1 of the 23 procedure groups. If a patient had a procedure code for atrial septal defect (ASD) closure and a procedure code for any of the other procedures listed in Table 1, the patient was assigned to the non-ASD procedure group. Among patients with ventricular septal defect (VSD) closure, if a procedure code was found for excision of subaortic membrane, infundibulectomy, and coarctation repair, the patient was listed in the complex VSD group. Patients undergoing orthotopic heart transplantation were separated into 2 groups: CHD as an underlying heart condition before transplantation and non-CHD (eg, cardiomyopathy). Patients undergoing aortopulmonary shunt who also had atrial septectomy for single-ventricle type of CHD were separated from the aortopulmonary shunt group. We also included 4 comorbid conditions as independent, nonprocedural medical variables: Down syndrome (ICD9-CM code 758.0), pulmonary hypertension (416.0), failure to thrive (783.4), and prematurity (765.0 and 765.1).
To determine whether sex is an independent risk factor for pediatric cardiac surgery mortality, logistic regression analysis was conducted. The sole dependent variable was a binary outcome variable: alive at discharge versus in-hospital death. Within each independent variable, 1 subgroup was selected as the reference (OR=1) to calculate OR for mortality for the other subgroups relative to the reference subgroups. The reference subgroups were as follows: male for “sex,” child (age ≥1 year) for “age,” white for “race/ethnicity,” public insurance for “type of insurance,” annual family income >$60 000 for “family income,” weekday for “day of surgery,” October through December for “month of surgery,” elective admission for “type of admission,” non-emergency room admission for “source of admission,” high-volume hospital (annual case volume >100) for “cardiac center,” VSD closure for “type of surgical procedure,” and absence of comorbidity for the 4 comorbidity variables. Data for each group of variables were presented as ORs comparing each subgroup with its respective reference subgroup. For each OR, the 95% CI of the OR was calculated.
In the logistic regression, risk adjustment was accomplished by treating each procedure as an independent variable. If a patient had that particular procedure performed, a value of “1” was listed; if not, value “0” was listed. Each patient had only 1 value of “1” listed among the 23 procedure groups. The Hosmer and Lemeshow goodness-of-fit test was conducted to determine how the regression model fit observed data.15 The regression was conducted with and without adjustment for clustering of cases in surgical centers by use of STATA 6.0 software (Stata Corp), and the results with and without adjustment were compared.
Additional analyses were conducted to test the hypothesis that sex differences in mortality result from differences in utilization of services between males and females. We compared the length of hospital stay and hospital charges between males and females. The analysis of length of hospital stay and hospital charges was also adjusted for the patient, healthcare system, and medical variables by multiple regression. A probability value <0.05 was considered statistically significant.
We identified 6972 children undergoing 1 of the 23 types of surgical procedures for CHD in 65 hospitals in California in the period 1995 to 1997. Although the surgeries were performed in 65 hospitals, 30 hospitals had only 1 pediatric cardiac surgery case over the 3-year period. After cases in hospitals with <10 cases over the study period and cases with incomplete or miscoded data were excluded, 6593 cases of surgery from 20 hospitals were analyzed. In this group, there were 345 in-hospital deaths, yielding an overall mortality rate of 5.23%.
Table 2 summarizes the sociodemographic and clinical variables of the study subjects in the univariate (not risk-adjusted) analysis. The in-hospital mortality rate was 4.98% for males and 5.54% for females. In this analysis, females did not appear to have a significantly higher mortality rate than males (OR 1.11, CI 0.90 to 1.38, P=0.33). Mortality rates were 16.50% for neonates, 5.57% for infants, and 1.75% for children (≥1 year). No significant difference in mortality was found among white, black, Hispanic, and Asian groups, although “other races” tended to have higher mortality. Patients with various types of insurance had similar mortality rates. Among patients of different home income levels, a trend toward higher mortality rate was noted with decreasing home income, although the differences were not statistically significant.
With regard to the variable “month of operation,” we found the mortality rate in January through March was significantly higher than that in October through December (P<0.01). Surgeries performed on weekends and nonelective surgeries had significantly higher mortality rates than those performed on weekdays or those performed after elective admissions. Similarly, patients admitted through the emergency room had higher mortality rates. Seven hospitals with an average annual case volume >100 were defined as high-volume hospitals, and the remaining 13 hospitals, with annual case volume ≤100, were considered low-volume hospitals. The high-volume hospitals had significantly lower overall mortality rates (5.00%) than the low-volume hospitals (5.74%).
The numbers of cases and mortality rates for the 23 procedure groups used to account for medical risk are listed in Table 2. Closure of ASD was the most commonly performed procedure, accounting for 15.74% of all cases. This was followed by VSD closure and tetralogy of Fallot repair. The mortality rates for the procedures ranged widely, fom 0.19% for ASD closure to 41.13% for the Norwood operation.
The results of comparisons of the patient, healthcare system, and medical characteristics between female and male patients are listed in Table 3. There were 3597 males and 2996 females, with a male-to-female ratio of 1.20. The age distributions of females and males were significantly different (P<0.01). Males represented a higher proportion of neonates than females. The composition of race/ethnicity, type of insurance, and household income level were similar between females and males. Although there were no differences between males and females in day or month of admission, admission through emergency room, or surgery at low- versus high-volume hospitals, males had a higher proportion of nonelective surgeries than females (30.39% versus 24.80%, P<0.01). Among the 4 comorbid conditions selected, we found that females had a higher incidence of Down syndrome, pulmonary hypertension, and failure to thrive than males.
Of particular interest, there were significant differences in the overall distribution of the procedures between males and females (P<0.01). On examination of the specific procedure groups, females had a much higher proportion of ASD closure than males (20.69% versus 11.62%), whereas males had more heart transplantation, aortic valve replacement, and arterial switch operation (with and without VSD closure) than females. In addition, procedure groups with high morality rates, such as aortopulmonary shunt with atrial septectomy and the Norwood operation, were more common in males than in females.
The results of multivariate logistic regression analysis of all patients, controlling for all nonsex variables, including medical risk as represented by the 23 procedure variables, are listed in Table 4. The Hosmer and Lemeshow goodness-of-fit test for the logistic regression model showed that the regression model fit the observed data well, with a χ2 value of 4.21 (P=0.84).
When all other patient, system, and medical variables were controlled, females had a significantly higher OR for mortality than males (OR 1.51, P<0.01). Neonates had an OR for mortality of 3.86 and infants had an OR of 2.98 compared with children ≥1 year old (P<0.01). Black, Asian, and Hispanic patients had mortality rates similar to those for white patients; however, “other races” had significantly higher mortality (OR 1.59). A weakly positive increase in OR for mortality was noted for private insurance patients (OR 1.64) compared with children covered under managed care (P=0.04). The median income by home zip code did not appear to have a significant effect on surgical mortality, although the OR for mortality appeared to increase slightly as median home income decreased.
Surgeries performed on weekends carried higher mortality than weekday surgeries in the univariate analysis (11.73% versus 5.03%, respectively). However, with risk adjustment, the mortality OR for surgery performed on weekends was 1.18, which was not significantly different from that for surgeries performed on weekdays. Even after adjustment for risk, the mortality rate for surgeries performed in January, February, or March was higher than that for surgeries performed in October, November, or December (OR 1.52, P=0.02). Surgeries performed in April, May, or June and July, August, or September did not appear to have greater mortality rates than those performed in October, November, or December. Admission through the emergency department did not have a significant effect on mortality. Nonelective surgeries carried a 2-fold higher mortality than elective surgeries (P<0.01). Finally, the OR for mortality for low-volume hospitals was significantly higher than that for high-volume hospitals (OR 1.67, P<0.01).
Four comorbid conditions were added to the risk adjustment: Down syndrome, pulmonary hypertension, failure to thrive, and prematurity. The OR for mortality increased with the presence of pulmonary hypertension (OR 2.67, P<0.01) or if the patient was premature (OR 3.00, P<0.01). The presence of Down syndrome or failure to thrive did not produce a significantly increased difference in the OR for mortality.
Length of Hospital Stay and Hospital Charges
The median length of hospital stay was 6 days for males and 5 days for females (mean length of stay 11.1 days and 10.1 days, respectively; P<0.01). The median hospital charge was $60 046 for males and $52 558 for females (mean hospital charge $95 886 and $87 478, respectively; P<0.01). There was no statistical difference in hospital length of stay or charges between males and females among patients who died in the hospital.
When multiple regression was used to control for medical, patient, and healthcare system variables, we found males had only 0.1 day longer hospital stay than females when everything else was equal (P=0.77). In the multiple regression analysis, males had $338 less in hospital charges than females, and this difference was not statistically significant (P=0.90).
The major finding of this study is that female sex was associated with higher in-hospital mortality among children undergoing cardiac surgery in California from 1995 to 1997. This difference is probably not due to variations in service utilization during hospitalization, because the risk-adjusted hospital length of stay and hospital charges were similar between males and females. We have also found age of the patient and case volume of the hospital at which the surgery is performed are important factors in determining outcomes, whereas race and ethnicity, type of insurance, and home income level are not.
Significant differences in mortality between males and females were seen after adjustment for the variations that exist in the patients, healthcare system-related variables, and the risks of the various types of cardiac surgery. Although the crude mortality rate (without risk adjustment) was slightly higher in females than in males, with no statistically significant difference, fewer females were neonates, and females underwent a larger number of low-risk procedures (such as ASD closure) and a lower number of high-risk procedures (such as the Norwood operation). After adjustment for these differences between females and males, the odds of dying for females were actually 50% higher than for males. Although risk-adjustment strategies have been proposed in studies of pediatric cardiac surgery,12–14⇓⇓ we have found that the approach of including patient, system, and detailed procedure variables in the risk-adjustment scheme is useful in distinguishing patients with different levels of risk.
To the best of our knowledge, a difference in cardiac surgical mortality between male and female has not been reported in pediatric patients. In a study by Jenkins et al,12 sex was not found to be a significant predictor for in-hospital death by a multivariate analysis. In a study by Hannan et al,13 the risk-adjusted mortality rates for males and females were not reported; however, the unadjusted mortality rate appeared to be slightly higher in males than in females (7.33% versus 6.10%). The possible explanations for sex effect found in the present study but not by other authors are as follows: (1) case selection—specifically, patent ductus arteriosus closure, a low-risk, high-prevalence surgery, which was included in other studies, was not included in the present study; (2) risk-adjustment methodologies—previous studies used 4 risk categories to account for the variation in surgical risk compared with the 23 procedure groups used in the present study; and (3) different geographic areas and time frames from which data were derived.
Many researchers have proposed possible explanations for sex disparities in cardiovascular outcomes in adults. The postulated causes for sex differences can be broadly summarized as issues related to (1) biological differences between females and males, such as a higher prevalence of diabetes and hypertension,7 a lower functional class at the time of surgery,16 and smaller body surface area in females,8 and (2) differences in access to and use of health services, including a lower rate of utilization for thrombolytic therapy and cardiac catheterization in females10,11⇓ and a sex-derived referral bias for CABG.6,9⇓ In the present study, we were able to assess in-hospital healthcare utilization by analyzing risk-adjusted hospital charges, which we found to be similar between males and females. To the extent such information is available, it is unlikely that the sex difference in mortality is due to differences in service utilization during hospitalization. Another possible explanation is that males are discharged earlier than females, and therefore fewer male deaths occur in the hospital. However, we found the risk-adjusted length of hospital stay to be comparable between females and males. We also found that females have a higher incidence of comorbid conditions, such as Down syndrome, pulmonary hypertension, and failure to thrive. Thus, we speculate that the sex difference in surgical outcomes discovered in the present study may be due to other biological differences between young males and females that were not characterized in this administrative data set. Further studies are clearly needed to provide more evidence to support our speculations.
It is not surprising that neonates and infants had higher mortality rates for cardiac surgery. It is apparent from Table 2 that neonates accounted for only 16.37% of all surgeries, but >50% of all in-hospital deaths occurred in neonates. Although the predominant types of cardiac surgery in neonates and infants differ from those in older children, risk adjustment for various types of procedures did not eliminate age-related differences in mortality. When other variables were controlled, infants still had 2.98 times, and neonates 3.86 times, higher mortality than children ≥1 year old. These ORs are similar to the OR of 2.32 for <90-days-old and 1.96 for 90-day-old to 1-year-old infants compared with children aged ≥1 year reported by Hannan et al.13
Recent studies have shown that children’s race, ethnicity, and insurance coverage may affect the age at which they are referred to a cardiologist and the age at which surgery for definitive repair of CHD is performed.17,18⇓ In the present study, we did not find significant differences in mortality among whites, Hispanics, and blacks. This result is consistent with the findings reported by Jenkins et al12 that race is not a predictor of in-hospital mortality. We had previously reported that Asians tended to be older at the time of complete surgical repair for CHD, which suggests racial or cultural differences in access to and use of healthcare services.17 However, in the present study, any potential differences in access or use in Asians did not appear to lead to an increased risk for mortality when undergoing cardiac surgery. In contrast, in adults undergoing CABG surgery, race has been found to be an independent risk factor for surgical mortality.19,20⇓ It is interesting that race and ethnicity appear to affect the outcomes of adult but not pediatric cardiac surgery. Whether the differences between adults and children with regard to the effect of race and ethnicity on cardiac surgery outcomes are due to biological or sociocultural factors remains a topic for further investigation.
Bell and Redelmeier21 reported that patients with acute, serious medical conditions who are admitted to hospitals on weekends have higher mortality rates than patients admitted on weekdays. In the present study, we found the crude in-hospital mortality rate to be more than twice as high for surgeries performed on weekends than for those performed on weekdays. However, risk adjustment eliminated this difference. After risk adjustment, surgery on weekends had only slightly and nonsignificantly higher mortality (OR 1.18) than surgeries performed on weekdays. The “weekend effect” reported by Bell and Redelmeier is probably less significant in the present study population because most pediatric cardiac surgery admissions were elective (scheduled), and only a small number of surgeries were performed on weekends (Table 2). In addition, surgeries performed on weekends were generally high-risk procedures, which therefore resulted in a higher crude (unadjusted) mortality rate.
The relation between hospital case volume and favorable patient outcomes has been demonstrated by several studies for a variety of surgical procedures.22–24⇓⇓ Specifically, a correlation between case volume and outcomes for pediatric cardiac surgery has been reported by Jenkins et al12 and Hannan et al,13 respectively. Both studies demonstrated that risk-adjusted, in-hospital mortality rates are lower in hospitals with higher volumes of pediatric cardiac surgery. The study by Jenkins et al12 found the OR for mortality was 7.7 for cases at hospitals with <10 cases/year, 2.9 for hospitals with 10 to 100 cases/year, and 3.0 for hospitals with 101 to 300 cases/year compared with hospitals with >300 cases/year (reference group). In the study by Hannan et al,13 the OR for mortality for hospitals performing <100 cases/year was 1.42 compared with hospitals with ≥100 cases/year. Similarly, in the present study, we found an OR of 1.67 for low-volume hospitals (≤100 cases/year) compared with high-volume hospitals.
A major limitation of this study results from errors related to miscoding and missing data in an administrative database.25–27⇓⇓ In the present study, we attempted to minimize the effect of errors from miscoding by restricting our case selections to hospitals with ≥10 cases/year of pediatric cardiac surgery. In addition, we carefully examined the procedure and diagnosis codes for each patient. Despite these safeguards, missing data and miscoding of patients selected for the present study may exist and have the potential to bias our findings.
Sex appears to be an important determinant of surgical outcome among children undergoing cardiac surgery. In this series of cases, females had 51% higher mortality than males when other medical and nonmedical variables were held equal. The sex difference in outcomes does not appear to be related to in-hospital health services utilization. Future studies to investigate the mechanisms by which sex influences surgical outcomes may help to improve outcomes of children undergoing cardiac surgery.
Dr Chang was a postdoctoral fellow of the Agency for Healthcare Research and Quality. This study was supported in part by an institutional grant from the Harbor-UCLA Research and Education Institute and grants (RR 00425 and K23 RR17041-01) from the National Center for Research Resources, National Institutes of Health.
Presented in abstract form at the meeting of the American Academy of Pediatrics, Chicago, Ill, October 18, 2000.
- ↵Greenland P, Reicher-Reiss H, Goldbourt U, et al. In-hospital and 1-year mortality in 1,524 women after myocardial infarction: comparison with 4,315 men. Circulation. 1991; 83: 484–491.
- ↵O’Connor GT, Morton JR, Diehl MJ, et al, for the Northern New England Cardiovascular Disease Study Group. Differences between men and women in hospital mortality associated with coronary artery bypass graft surgery. Circulation. 1993; 88: 2104–2110.
- ↵Jenkins KJ, Newburger JW, Lock JE, 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 EL, Racz M, Kavey RE, et al. Pediatric cardiac surgery: the effect of hospital and surgeon volume on in-hospital mortality. Pediatrics. 1998; 101: 963–969.
- ↵Erickson LC, Wise PH, Cook EF, et al. The impact of managed care insurance on use of lower-mortality hospitals by children undergoing cardiac surgery in California. Pediatrics. 2000; 105: 1271–1278.
- ↵Hosmer DW, Lemeshow S. Assessing the fit of the model.In: Hosmer DW, Lemeshow S, eds. Applied Logistic Regression. New York, NY: John Wiley & Sons; 1989: 140–145.
- ↵Chang R-KR, Chen AY, Klitzner TS. Factors associated with age at operation for children with congenital heart disease. Pediatrics. 2000; 105: 1073–1081.