## Journal of the American College of Cardiology

# Survival analysis and risk factors for mortality in transplantation and staged surgery for hypoplastic left heart syndrome

## Author + information

- Received September 9, 1999
- Revision received March 15, 2000
- Accepted May 24, 2000
- Published online October 1, 2000.

## Author Information

- Pamela C Jenkins, MD, PhD∗,* (pcj{at}hitchcock.org),
- Michael F Flanagan, MD, FACC∗,
- Kathy J Jenkins, MD, MPH†,
- James D Sargent, MD∗,
- Charles E Canter, MD, FACC‡,
- Richard E Chinnock, MD§,
- Robert N Vincent, MD, FACC∥,
- Anna N.A Tosteson, ScD¶ and
- Gerald T O’Connor, PhD, DSc, FACC¶

- ↵*Reprint requests and correspondence: Dr. Pamela C. Jenkins, Department of Pediatrics, Dartmouth Medical School, Hanover, New Hampshire 03755

## Abstract

OBJECTIVES

We compared survival in treatment strategies and determined risk factors for one-year mortality for hypoplastic left heart syndrome (HLHS) using intention-to-treat analysis.

BACKGROUND

Staged revision of the native heart and transplantation as treatments for HLHS have been compared in treatment-received analyses, which can bias results.

METHODS

Data on 231 infants with HLHS, born between 1989 and 1994 and intended for surgery, were collected from four pediatric cardiac surgical centers. Status at last contact for survival analysis and mortality at one year for risk factor analysis were the outcome measures.

RESULTS

Survival curves showed improved survival for patients intended for transplantation over patients intended for staged surgery. One-year survival was 61% for transplantation and 42% for staged surgery (p < 0.01); five-year survival was 55% and 38%, respectively (p < 0.01). Survival curves adjusted for preoperative differences were also significantly different (p < 0.001). Waiting-list mortality accounted for 63% of first-year deaths in the transplantation group. Mortality with stage 1 surgery accounted for 86% of that strategy’s first-year mortality. Birth weight <3 kg (odds ratio [OR] 2.4), highest creatinine ≥2 mg/dL (OR 4.7), restrictive atrial septal defect (OR 2.7) and, in staged surgery, atresia of one (OR 4.2) or both (OR 11.0) left-sided valves produced a higher risk for one-year mortality.

CONCLUSIONS

Transplantation produced significantly higher survival at all ages up to seven years. Patients with atresia of one or both valves do poorly in staged surgery and have significantly higher survival with transplantation. This information may be useful in directing patients to the better strategy for them.

Hypoplastic left heart syndrome (HLHS) is a lethal congenital heart malformation, resulting in a 95% mortality in the first month of life when untreated (1) and accounting for 22% of deaths from congenital heart disease in the first year of life (2). Infant heart transplantation and a staged revision of the native heart currently offer hope of survival for children with this condition. Staged surgery is more commonly available, partly because of the limited number of donor hearts of an appropriate size.

Survival analyses and risk factor analyses have been performed for each strategy. Razzouk et al. (3) found a seven-year survival of 70% for transplanted patients and 61% for all listed patients. For staged surgery, a five-year survival of 30% to 71% has been found (4–7). The two procedures have been compared by Bando (8) for 50 neonates with HLHS, by Starnes (9) for 35, and by Jacobs et al. (10) for 323 patients with aortic atresia. Two studies found transplantation to be superior, and one found staged surgery to have better outcomes. However, two of these (8,9) analyzed crossover patients with a treatment-received analysis rather than with intention-to-treat analysis, a more conservative analytic approach (11,12). Jacob’s study (10) found some advantage of transplantation over staged surgery in three-year follow-up by intention-to-treat analysis. Reports vary regarding which preoperative variables increase mortality risk for infants with HLHS (5–8,10,13–18). The choice that offers the better chance of survival to these infants remains unclear.

Four pediatric cardiac surgical centers have collaborated to collect preprocedure and outcome data on consecutive patients with HLHS born between 1989 and 1994. These infants were admitted with the intention to perform either heart transplantation or staged surgical reconstruction. We used intention-to-treat analysis to understand how babies intended for a strategy fare. Intention-to-treat analysis informs the dilemma of which treatment to undertake, as opposed to reporting the outcomes of surgery. We compared long-term survival between the transplantation strategy and the staged surgical strategy, determined risk factors for one-year mortality in infants with HLHS and developed a prediction equation to help direct patients to the optimal strategy based on individual characteristics. Tailoring the treatment to minimize the risk of mortality could improve survival for these children.

## Methods

### Selection criteria

Hypoplastic left heart syndrome was defined as normal segmental anatomy with mitral and/or aortic atresia or stenosis and a left ventricle too small to sustain the systemic circulation. Babies with HLHS born between January 1, 1989, and December 31, 1993, of birth weight ≥2 kg, 37 to 42 weeks gestation, and admitted to the surgical center with the intention to perform surgery were included. Hypoplastic left heart syndrome babies with other cardiac malformations, including any anomalous pulmonary venous return (n = 9), transposition of the great arteries (n = 2), ventricular septal defect (n = 23) or double outlet right ventricle (RV) (n = 18) were excluded. Patients with noncardiac malformations and chromosomal abnormalities (n = 16) were also excluded.

### Participating sites

These sites are Egleston Children’s Hospital in Atlanta, Children’s Hospital in Boston, Loma Linda University Children’s Hospital and St. Louis Children’s Hospital. The Internal Review Board of each hospital approved this study.

### Subjects

All patients with HLHS were identified at each site by hospital database search. Medical records of eligible patients were reviewed by the principal investigator. Of 242 eligible patients (Fig. 1), 11 charts were not located (5%). Survival analysis included 231 cases: 109 in the staged surgical group and 122 in the transplantation group. Risk factor analysis included 221 cases; 10 patients died in other hospitals while waiting on a participating center’s transplant list.

### Statistical methods

#### Intention-to-treat groups

Treatment path was determined as the original intent for surgery. Infants listed for transplantation remained in that strategy; patients who received stage 1, then transplantation, remained in the staged surgical group. Patients were followed to the last known physician encounter. Seven patients listed for transplantation received staged surgery; six patients received a transplant after stage 1 (Fig. 1). Three patients received retransplantation.

#### Censoring

Patients lost to follow-up or known to be still alive were censored. The Kaplan–Meier method of survival analysis (19) was used to generate and adjust survival curves using preoperative variables that differed between the treatment groups. The Mantel–Haenszel log-rank test (11) tested the equality of the survivor functions. The software program used was STATA 5.0 (Stata Corp., College Station, Texas).

### Risk factor analysis methods

#### Univariate analysis

One-year survival was the dependent variable to assess risk factors for mortality. Variables tested by univariate logistic regression (20) are presented in Tables 1 and 2⇓. ⇓ All variables were tested for first-order interactions. Correlation between explanatory variables was tested. P-values for tests of trend were generated by the Pearson chi-square statistic (11).

#### Missing data

Missing data were dealt with in two ways (21,22): 1) for variables not entered into the logistic regression model, missing data were not substituted; 2) sensitivity analysis was performed on three variables missing ≥20% of data tested in the regression: aortic diameter, moderate-to-severe tricuspid regurgitation (TR) and moderate-to-poor RV function. Sensitivity analysis determines whether the logistic regression model is sensitive to the way missing values are included. Patients were divided into treatment groups and missing variables were assigned normal or abnormal values by group. Each permutation of normal/abnormal assignment by treatment group for missing data was tested for univariate association with mortality.

#### Multiple regression analysis

A logistic regression model (20) for the 221 admitted patients characterized the preoperative factors that placed infants with HLHS at risk for death at one year. The model included variables with univariate p-values of ≤0.25 and was reduced using the likelihood ratio test (20). The relative contribution of each variable to the explained mortality risk was calculated using z-scores.

#### Model assessment techniques

We assessed the model for its discrimination, using the area under the relative operating characteristic (ROC) curve; internal validity, using bootstrap resampling; and goodness-of-fit between the model and the data, using the Hosmer–Lemeshow test. These techniques are described in the Appendix.

## Results

Of 231 patients with HLHS, 111 (48%) had died by one year. Forty-three percent of the deaths occurred in the transplantation strategy and 57% in the staged surgical strategy.

When preoperative characteristics of patients in the treatment groups were compared (Table 1), three characteristics were found to be significantly different. The rate of moderate-to-severe TR on the initial echocardiogram (p = 0.01) and preoperative atrial septostomy rate (p = 0.002) were higher for transplantation than for staged surgery. Waiting time prior to surgery differed, with 7% in transplantation and 68% in staged surgery receiving an operation within a week of life (p < 0.001). All other characteristics were similar.

### Survival analysis

Transplantation generated a higher survival at all time points through seven years of follow-up (Fig. 2) (p < 0.001). The transplantation strategy had an unadjusted one-year survival of 61% and a five-year survival of 55% (p < 0.01) (Table 3). Staged surgery had an unadjusted one-year survival of 42% and a five-year survival of 38% (p < 0.01). For staged surgery, median survival was 2.6 months; for transplantation, median survival was two years. After adjusting the survival curves for the different incidence of moderate-to-severe TR on the initial echocardiogram, one-year survival in the staged surgical arm remained at 42%, while survival in the transplantation arm improved to 70% (p = 0.01). The rates of restrictive atrial septal defect (ASD) and age at surgery were consequences of the wait and were not included in the adjustment.

Predischarge mortality following stage 1 surgery represented 78% (49/63) of the one-year mortality for this strategy (Fig. 3). Ninety-three percent of the five-year mortality for staged surgery occurred in the first year. For transplantation, preoperative mortality accounted for 63% (30/48) of the one-year mortality, and 87% of the five-year mortality occurred in the first year. Longer listing time did not increase five-year mortality (p = 0.16). All surviving infants on the transplant list had received a donor heart by six months of age.

Treatment-received analysis was also performed on the survival data (Table 3). Treatment-received analysis produced a larger difference between treatments, evidenced by smaller p-values.

Survival curves stratified by center showed no difference in outcomes for staged surgery (p = 0.86). For transplantation, a significant difference in mortality by center was found (p = 0.02). Institutions listing fewer patients tended to have higher mortality rates.

### Risk factor analysis

Univariate associations between pre-operative characteristics and one-year mortality (Tables 1 and 2) show that valvar atresia was predictive of higher one-year mortality in the staged surgical strategy but not in transplantation (Fig. 4). One-year mortality for mitral and aortic stenosis was 32%, for atresia of one valve 62% (OR 3.4, p = 0.02) and for atresia of both valves 76% (OR 6.7, p ≤ 0.001). Aortic diameter ≤2 mm was associated with higher one-year mortality in staged surgery. In the transplant arm, a maximum preoperative creatinine of ≥2 mg/dL or restrictive ASD requiring septostomy yielded a higher one-year mortality. Use of epinephrine and age at surgery ≤7 days were also significant in univariate analysis.

Sensitivity analysis on missing data showed that the model was sensitive only to missing data for birth weight; replacing missing birth weight as ≥3 kg did not change the model, but replacement as <3 kg caused birth weight to drop from the model. The model with missing birth weight as ≥3 kg is presented. Missing data substitutions for TR, RV function and aortic diameter produced nonsignificant results.

Multiple logistic regression (Table 4)showed that a birth weight <3 kg (OR 2.4, p = 0.01), creatinine ≥2 mg/dL (OR 4.7, p < 0.001) and atrial septostomy (OR 2.7, p = 0.02) were significant risk factors for one-year mortality in all patients (n = 221). Atresia of one (OR 4.2, p = 0.001) or both valves (OR 11.0, p < 0.001) were significant risk factors for staged surgical mortality but not for transplantation. Figure 5 shows the relative contribution of each risk factor to the explained risk. This model produced an area under the ROC curve of 0.77 (95% CI 0.71–0.83) (Appendix). The model generated an area under the ROC curve of 0.73 (95% CI 0.57–0.78) for transplantation and 0.75 (95% CI 0.60–0.80) for staged surgery.

Table 4 includes the probability of mortality based on individual patient characteristics using the logistic regression model. Any patient with HLHS having no risk factors (birth weight 4 kg, creatinine <2 mg/dL, no septostomy and mitral stenosis and aortic stenosis) has an OR of 0.22, and a probability of one-year mortality of 18%. A patient with HLHS having a creatinine < 2 mg/dL who received staged surgery with mitral atresia and aortic atresia, a birth weight of 4 kg and no septostomy has an OR of one-year mortality of 2.39 and a probability of 71%. A similar transplanted patient would have an OR of 0.22, and a probability of one-year mortality of 18%.

## Discussion

This analysis compares the staged surgical strategy to the transplantation strategy, by survival curves and risk factors for mortality, for 231 infants with HLHS at four pediatric cardiac surgical centers from 1989 to 1994. Babies listed for transplantation had a significantly higher survival than those babies intended for staged surgery at one and five years (p < 0.01). Four risk factors for one-year mortality were important for infants with HLHS: birth weight <3 kg, maximum creatinine ≥2 mg/dL, atrial septostomy and atresia of one or both left-sided valves in the staged surgical approach.

The difference in survival is not due to systematic, case-mix differences or selection bias. Children with HLHS placed in each strategy were similar, except for the initial rate of moderate-to-severe TR and the consequences of the surgical wait (atrial septostomy rate, preoperative mortality rate) (Table 1). Adjusting for moderate-to-severe TR increased the survival difference. The difference in survival is not due to bias introduced by missing data. When variables with large amounts of missing data were assigned normal or abnormal values, no univariate relationship between one-year mortality and TR, RV function, or aortic diameter existed.

Nearly all of the mortality for both paths occurred in the first year of life (Fig. 3). For patients who survived the first year, approximately 90% in each strategy survived to age five. Mortality after stage 1 tended to occur in-hospital. Waiting-list mortality dominated the transplantation arm’s first-year mortality.

These survival outcomes in this cohort of patients are not the best reported outcomes from 1989 to 1993 for either transplantation or staged surgery. The results are likely to be more representative of the outcomes most children with HLHS experienced. Such information is valuable for physicians discussing options and preparing families for the survival potential of their child with HLHS.

### Comparison to risk factor analyses in the literature

Prior studies of preoperative risk factors for mortality in each treatment strategy have used case numbers ranging from 20 to over 200. Restrictive ASD or failed balloon septostomy, nonidentical blood type and age at transplantation have been important for the transplant group (13–15). For the staged surgical approach, some centers, but not all, have found aortic diameter ≤2 mm, atresia of both mitral and aortic valves, preoperative acidosis, age >1 months at operation, earlier year of operation, obstruction to pulmonary venous return, noncardiac congenital conditions and weight <3 kg to predict a higher risk of death (5–8,16–18). Infants with aortic atresia had increased three-year mortality risk for lower birth weight, noncardiac anomalies, intention not to treat and staged surgical path (10).

A high preoperative creatinine level and a restrictive ASD increase mortality in transplantation by univariate analysis (Table 1). Creatinine’s effect on transplantation survival conceivably reflects the nephrotoxicity of some immunosuppressives. The need for atrial septostomy could be a proxy measure for increased waiting time and may be subject to practice variations. This study found no effect of nonidentical blood type or age over one month at transplantation for the transplant group, and no effect of preoperative acidosis, aortic diameter or age at surgery for the staged group. Obstruction to pulmonary venous return or noncardiac congenital conditions were exclusions from the study. Of note, prenatal diagnosis did not improve survival in this population.

### Study limitations

The limitations of this study include the possibility of “center effects.” Center effects would apply if centers had varying outcomes for each strategy and large-volume centers overwhelmed the analysis. Only widely disparate outcomes in survival are detectable in a four-center analysis. Institutions listing fewer patients tended to have higher mortality rates, possibly due to differences in listing practices, to regional organ donation rates or to volume effects. This effect can be better understood in a study involving many more transplant centers. However, the disparity in survival between centers for listed patients supports our conclusion that survival for listed patients is higher than survival for staged surgical patients. A center with a high waiting-list mortality will lower the overall survival for the transplantation strategy and so decrease the difference in survival between transplantation and staged surgery. If significance is achieved even when the curves are conservatively weighted, the difference in survival is probably real and not due to statistical manipulations.

Although surgical mortality did not show a significant difference in survival by time period, one-year mortality rates may have improved for these strategies. Both strategies have the potential for significant late mortality in adolescence and adulthood.

### Intention-to-treat analysis

Treatment-received analysis (8,9) can, in the worst case, make one strategy look better than it actually is and make the other strategy look worse. Bando et al.’s report (8) of 50 babies with HLHS found an 18% transplant mortality and a 50% staged surgical mortality, but if all seven of the listed patients who received a stage 1 surgery died, patients seeking transplantation could have expected only a 45% survival (10/22); patients intended for the staged procedure would have a 64% survival (18/28) by intention-to-treat analysis. An intention-to-treat approach, performed here and by Jacobs et al. (10), yields a conservative comparison of survival. In this study, intention-to-treat methods gave transplantation poorer results than the treatment-received analysis did, and gave staged surgery better results (Table 3). Treatment-received analysis widened the difference between treatments. Although the survival difference between intention-to-treat and treatment-received analysis in this study was not statistically significant (p > 0.10), the higher crossover rate in other studies could significantly bias results.

Intention-to-treat analysis also seeks to address a different question than does treatment-received analysis. Treatment-received analysis usually evaluates the outcomes of procedures. With intention-to-treat methods, researchers seek to understand outcomes for patients who choose a particular treatment. It informs the initial treatment decision.

### Conclusions

From 1989 to 1994, transplantation produced a significantly higher survival than staged surgery in the treatment of HLHS at the participating centers. Babies with HLHS having birth weights ≥3 kg, normal creatinine levels, no need for atrial septostomy and mitral stenosis/aortic stenosis tend to do relatively well in both treatment paths. Patients with atresia of one or both valves are at increased risk for one-year mortality in the staged surgery strategy.

The multiple logistic regression model in Table 4 is a useful tool for determining one-year mortality risk for particular patients with HLHS. It can help inform the choice between strategies and allow the comparison of survival results over time or across centers by risk adjustment of caseloads. Once risk adjustment can be securely achieved, centers can undertake initiatives to improve outcomes for babies with HLHS.

### Implications

The low first-year survival for both arms has two major implications: 1) if more donor hearts were available and waiting-list mortality could be reduced, five-year survival in the transplantation arm for HLHS could approach 80% at major surgical centers; 2) improvements in stage 1 mortality could make this strategy equally successful as transplantation.

The higher transplantation survival implies that more babies with HLHS should be listed for transplantation. The outcome of such a policy on the local or national level must be considered. Increasing the number of potential recipients could create greater public awareness of the need for organ donation and hence increase donations. Alternatively, if more babies were listed with the current shortage of infant donor hearts, proportionately fewer babies would receive a transplant and waiting-list mortality would increase. A practical solution to this dilemma is to recommend staged surgery for babies likely to do well in that path and list for transplantation babies at higher risk in staged surgery.

In our dataset, valvar atresia was the only differentiating factor in choosing between transplantation and staged surgery for HLHS. The higher one-year mortality in staged surgery was due primarily to poorer survival in patients with one- or two-valve atresia. Transplantation may improve survival in these babies if a successful transplant program with adequate organ donation is available. Attention to risk factors for first-year mortality, perioperative care and strategies for improving donor availability offer the potential for improving survival in children with HLHS.

## Appendix

### Risk factor model assessment

It is important to assess how well the multivariable model produced by a particular dataset actually works. The ideal assessment of a model should answer several questions: 1) How well does the model discriminate between patients who lived and patients who died? 2) How consistent is that discrimination at different levels of sensitivity and specificity? 3) Are there patients for whom the risk factor model predicts poorly? and 4) How well does it discriminate between patients who lived or died in other datasets? We addressed three of these four questions, using the area under the ROC curve, bootstrap resampling and the Hosmer-Lemeshow goodness-of-fit test.

The area under the ROC curve measures the ability of the logistic regression model to discriminate on the basis of survival (23). The graph plots the model’s ability to correctly predict outcome at different levels of true positive rate (sensitivity) and false positive rate (1-specificity). A value of 0.5 represents a worthless model that predicts outcome as well as a coin toss would. A value of 1.0 represents a model that will perfectly predict outcome every time. Our model yielded an area under the ROC curve of 0.77, similar to values for prediction equations in other fields.

Bootstrap resampling was performed for the area under the ROC curve. Bootstrapping is a nearly unbiased method of internal validation (24). Bootstrapping avoids the loss of power inherent in the training- and testing-set method. Bootstrap resampling derives an area under the ROC curve for a subset, replaces that subset and chooses another subset; 1,000 iterations produce the mean and 95% confidence interval.

The Hosmer-Lemeshow goodness-of-fit test (25) assessed how well the model fit the data. A significant p-value indicates an unsatisfactory fit of the model to the data. The test calculated expected mortality for three categories of predicted risk for each regression model and compared expected to observed mortality in each category. A well-parameterized model would predict equally well for each risk subgroup. Predicted risk was close to observed risk for low- and high-risk patients in our model.

A complete assessment of a logistic regression model also requires external validation. The risk factors must be pertinent in other datasets as well. Performing this next step will allow the secure application of this information to newborns with HLHS.

## Footnotes

☆ This work was supported by a National Research Service Award grant from the National Heart, Lung, and Blood Institute, number HI09488-03.

- Abbreviations
- ASD
- atrial septal defect
- HLHS
- hypoplastic left heart syndrome
- OR
- odds ratio
- ROC
- relative operating characteristic
- RV
- right ventricle, right ventricular
- TR
- tricuspid regurgitation

- Received September 9, 1999.
- Revision received March 15, 2000.
- Accepted May 24, 2000.

- American College of Cardiology

## References

- ↵
- Fyler D,
- Buckley L,
- Hellenbrand W

- ↵
- ↵
- ↵
- ↵
- ↵
- ↵
- ↵
- ↵
- Rothman K,
- Greenland S

- Newell D

- ↵
- Canter C,
- Moorhead S,
- Huddleston C,
- Spray T

- Canter C,
- Naftel D,
- Caldwell R,
- et al.

- ↵
- Kleinbaum D

- ↵
- Kleinbaum D,
- Kupper L,
- Muller K,
- Nizam A

- ↵
- Greenland S,
- Finkle W

- Little R,
- Rubin D

- ↵
- Swets J

- ↵
- ↵
- Hosmer D,
- Lemeshow S