Objectives To investigate survival in patients with Eisenmenger syndrome based on a systematic review of the literature and reanalysis of data. We specifically tested the hypothesis that previous publications have been subject to immortal time bias, confounding survival analyses.
Methods A systematic review of the literature was performed to evaluate survival in treatment naïve patients with Eisenmenger syndrome and standardised mortality ratios were calculated. Furthermore, we used a contemporary cohort of 219 treatment naïve patients with Eisenmenger syndrome from the own institution as a comparison group.
Results Overall, 12 studies (published 1971–2013) were identified, including a total of 1131 patients. Only one study seemed to deal appropriately with immortal time bias in this setting. All other studies did not account for this effect, thus overestimating survival prospects of patients with Eisenmenger syndrome by up to 20 years. After accounting for this effect we found high standardised mortality ratios, a 10-year mortality rate approaching 30–40% and no evidence of superior survival prospects of current era patients compared with those seen in the 1970s, 1980s and 1990s. Only, a historical Eisenmenger-cohort from the 1950s/1960s had worse survival.
Conclusions The current analysis challenges the traditional view of benign survival prospects of patients with Eisenmenger syndrome. In addition, survival prospects do not seem to have considerably improved over the last decades in untreated patients. These results support a proactive treatment strategy including a more aggressive approach trying to avoid the development of the condition.
- CONGENITAL HEART DISEASE
Statistics from Altmetric.com
In 1897, the Austrian physician Victor Eisenmenger reported the case of a man with cyanotic heart disease and severe pulmonary hypertension, a condition that was later to carry his name in the form of Eisenmenger syndrome.1 The patient died at the age of 32 years. Approximately, half a century later Paul Wood described a cohort of 127 patients with Eisenmenger syndrome: the average age at death was 33 years in those with post-tricuspid cardiac defects and 36 years in those with atrial septal defects.2 Other publications from the same era confirmed poor survival prospects in Eisenmenger syndrome with the vast majority of patients dying during childhood and young adulthood (with the possible exception of some pretricuspid defects).3 ,4
More recently it has been increasingly appreciated that Eisenmenger syndrome is a multisystem disorder affecting various organ systems and leading to a multitude of complications.5 ,6 To add to morbidity, patients with Eisenmenger syndrome have the highest prevalence of renal dysfunction and the worst exercise capacity among all patients with congenital heart disease—both risk factors linked to poor outcome.6 ,7 Paradoxically, however, survival prospects of patients with Eisenmenger syndrome are generally believed to be relatively benign especially considering the degree of pulmonary hypertension, the detrimental long-term effects of cyanosis and the complexity of the underlying cardiac condition in some patients. This interpretation is largely based on the results of studies using age as the time scale variable for Kaplan-Meier survival analyses.8–10 These studies have suggested that a sizeable proportion of patients with Eisenmenger syndrome survive to mid if not old age. However, using age as time to event variable is only appropriate if all patients are observed from birth to end of follow-up or death, an assumption that is unrealistic in this setting. In order to be enrolled in studies, patients have to survive to an age where they can enter clinical follow-up at research institutions. This introduces bias into the statistical analysis, potentially leading to overoptimistic estimates of survival prospects.
Interest in survival prospects has recently re-emerged as pulmonary vasoactive disease targeting therapies (DTTs) have become available and have been demonstrated to improve symptoms and exercise capacity in patients with Eisenmenger syndrome. A study from our own institution, investigating the association between DTT and outcome in Eisenmenger syndrome, reported a high mortality rate (approximately 25% at 5 years of follow-up) in treatment naïve patients with Eisenmenger syndrome.11
The current study was designed to shed light into these apparently discrepant findings and investigate survival prospects based on a systematic review of the literature and reanalysis of own data (from the Royal Brompton Hospital, London). Specifically we aimed to test the hypothesis that previous publications may have been subject to left truncation and immortal time bias, confounding survival analyses.
We used a two pronged approach to clarify survival prospects of patients with Eisenmenger syndrome: First, we performed a systematic review of the literature using a broad PubMed search for the term ‘Eisenmenger’. After excluding publications with the search term appearing only in the authorship field as well as case reports, reviews and publications in languages other than English, French or German, 492 articles published between 1947 and 2013 remained and underwent further manual screening. After excluding short-term studies, not providing survival data (or only composite end points) and studies reporting on a combination of patients with Eisenmenger syndrome with and without DTT, 12 papers published between 1971 and 2013 remained and were included in the current analysis. Of these, one provided individual patient data in table format that could be extracted and allowed us to use it directly for survival analysis. The remaining publications provided either Kaplan-Meier survival curves or information on the proportion of patients surviving at 1 year, 3 years, 5 years or 10 years of follow-up. If only a graphical representation of the data was available the data was extracted using appropriate computer software (Plot Digitizer; http://sourceforge.net/projects/plotdigitizer/, last accessed 16 September 2013) by digitising data points after calibration for x axis and y axis ranges.
To estimate standardised mortality ratios (SMRs) compared with an age and gender matched sample of the general population the method reported by Finkelstein et al12 was used wherever raw data was available. When only the overall survival function of the population was available from a previous publication, this was compared with life table data from the same era and country of origin using custom written software. Specifically, we used life table data available online from the United Nations, Department of Economic and Social Affairs (http://esa.un.org/wpp/Excel-Data/mortality_1.htm, last accessed 16 September 2013). Using this data we generated a hypothetical cohort of 20-year-old individuals with a gender distribution matching the respective study. The annual risk of death of this hypothetical cohort was subsequently adjusted (by multiplication with a constant amount, thus being conceptually similar to SMRs) to yield a survival function as close as possible to the published one. This was achieved by minimising the sum of squared differences between the curves (details on the procedure can be found in the online supplementary appendix section).
Second, where raw data was available, we used standard survival analysis, using age as time to event variable, with and without accounting for left truncation and immortal time bias. Left truncation was accounted for by specifying age at start and end of follow-up (age at censor or age at event time) within the survival model.
The methods used for generating survival curves that account for left truncation are implemented in various statistical packages. The software package used here (R with the survival library), allows to account for left truncation by specifying start and end times (or age) specifically within the survival function in addition to the usual indicator variable (0 or 1 for (right)-censored or deceased individuals, respectively) as follows:
survfit(Surv(start_age, end_age, indicatorvariable ∼1)).
Here start_age indicates the age at start of follow-up (first contact with the centre), end_age is the age at last contact (death, loss to follow-up or end of study). An overview over corresponding commands for other statistical packages is provided in the appendix section of the publication of Lamarca et al.13
A meta-analytical summary of the survival curve data from various studies was produced, as described in detail by Arends et al14 This method is based on fitting a multivariate, random mixed effects model to Ln-minus-Ln transformed survival proportions at multiple time points, using the R lme function. R-package version 2.13.0 was employed for all analyses (http://cran.r-project.org/).
The corresponding author has the right to grant on behalf of all authors and does grant on behalf of all authors, an exclusive licence (or non-exclusive for government employees) on a worldwide basis to the BMJ Publishing Group Ltd and its Licensees to permit this article (if accepted) to be published in Heart editions and any other BMJPGL products to exploit all subsidiary rights.
Details on the included studies with year of publication and the number of patients included are presented in table 1. In addition, table 2 presents details on the cohort of treatment naïve patients from our own institution included for analysis. This cohort was chosen to ensure comparability with a previous study from our institution11 and to allow potential use of the results for further modelling studies and future health economic evaluations.
Figure 1 illustrates the problem of immortal time bias based on data published by Young et al in 1971.3 Relevant patient data was available in the form of a table in the original publication and could be extracted. This data included gender, age at start and end of follow-up as well as information on mortality. Individual patients are plotted, and sorted by age at the end of follow-up. The figure illustrates the significantly different mortality estimates obtained using an inappropriate approach for performing survival analysis using age as time to event variable but ignoring left-truncation (red survival curve, time-fixed analysis), as opposed to correcting for immortal time by accounting for age at entry to the study (blue survival curve). As illustrated, the different results are the consequence of a discrepant number of patients in the denominator used as part of the Kaplan-Meier estimator. For example, at point A, four patients died. The instantaneous risk of death at point A is estimated as the number of deaths (ie, four) divided by the number of patients at risk. Depending on the approach used, this population at risk is either 55 (ie 57 total population minus 2 patients censored before this point) if left truncation is ignored (yielding a ‘mortality’ of 7.3%), or just 15 after accounting for late start of the observation period (yielding a ‘mortality’ of 26.7%). A similar effect could be shown when analysing our own raw data. As illustrated in figure 2 the survival curve ‘shifts’ rightwards by almost 20 years (measured at median survival time) when ignoring immortal time in the data, that is, when assuming than all patients are followed from birth to their age of event. This also affects the calculated SMRs, which ‘increase’ from 9.0 to 48.2 after accounting for this bias. Table 3 shows the SMRs calculated for the different studies included. Ignoring immortal time results in an estimated SMR for a hypothetical 20-year-old patient with Eisenmenger syndrome ranging between 9 and 14.6 in various studies and our own data. When accounting for immortal years (left truncation), the SMR is calculated between 33.4 and 76, thus suggesting a much higher relative annual mortality rate. As a consequence of these widely discrepant results, survival estimates from different studies can only be compared when accounting for the statistical approach employed.
To test whether survival prospects of treatment naïve patients with Eisenmenger syndrome have improved over time we compared survival in the current era (based on our own data) with the results of older studies.3 ,8 ,9 ,15 In figure 3A, survival curves from studies ignoring left truncation are compared with the survival curve generated by analysing our own data using a similar approach. Survival curves from studies performed after the 1970s appear to overlap, demonstrating no clear improvement over time.
Survival prospects for patients with Eisenmenger syndrome appear to have been somewhat worse in the 1950s and 1960 s, based on the study by Young et al, compared with more contemporary cohorts.3
Data from the only study accounting for left truncation (Corone et al,15 published in 1992) compared with survival data from our own cohort (1999–2007) analysed using a similar analysis approach (figure 3B), also suggest no improvement in the survival of treatment naïve patients with Eisenmenger syndrome over the last two to three decades. However, by accounting for immortal time, survival estimates are much worse compared with studies which failed to do so (figure 3A).
Various studies have provided survival data for patients with Eisenmenger syndrome during short-term follow-up, with time of entry to the study as time zero for the survival analysis, rather than using patient age. This approach is—by definition—not subject to immortal time bias. We combined the survival rates provided at 1 year, 3 years, 5 years and 10 years of follow-up to obtain average survival rates. Figure 4 illustrates the aggregated survival function based on the meta-analytical approach proposed by Arends et al, as well as the simple weighted means of the survival (proportions) at 1 year, 3 years, 5 years and 10 years of follow-up, respectively. The weights were based on the number of patients included in the different studies. Figure 4 illustrates that the predicted 10-year survival of treatment naïve patients with Eisenmenger syndrome is in the range of 60–70%.
Using a systematic review of the literature with analysis of available data from previous studies and raw data from our own institution we demonstrate that patients with Eisenmenger syndrome not treated with DTT have relatively poor survival prospects, with high standardised mortality rates and 10 year mortality rates ranging between 30–40%. No obvious improvement in the survival of treatment naïve patients was evident after the 1960s. Accounting for left truncation and immortal time bias is essential in order to avoid overestimation of survival prospects when using age as the time variable in survival analysis, especially for chronic, long-standing conditions such as Eisenmenger syndrome.
Patients with Eisenmenger syndrome are afflicted by a high morbidity and mortality, driven by the combination of the underlying heart defect, chronic cyanosis and systemic levels of pulmonary arterial hypertension.5 ,6 On the other hand, it is generally accepted that survival prospects of patients with Eisenmenger syndrome are significantly better in comparison with other forms of pulmonary hypertension. This notion is based on the results of large retrospective studies, suggesting low annual mortality rates in adults with the condition.8 ,9 ,16 However, the results of these studies are in stark contrast with older publications and a recent paper from our institution, which reported a high mortality rate in adults with Eisenmenger (mean age 34.5 years) syndrome. The results of the current analysis indicate that the optimistic survival estimates reported in some of the previous studies could be—at least in part—attributable to the presence of immortal time bias, inherent in the study approach used. It is important to emphasise that all studies in question did report the methodology employed appropriately. Moreover, greater awareness of the phenomenon of immortal time bias and the effects of left-truncation has only been achieved in recent years, after the publication of such studies.
The problem of immortal time bias entered prime time with illustrious publications, such as those comparing the survival prospects of Academy award winners with those of Oscar nominees.17
By not accounting for left truncation and immortal time bias, an Oscar winner would be considered a winner throughout their life, rather than after obtaining the award. Moreover, actors have to be alive to obtain the award (or star in the film for which they are awarded), thus, biasing survival estimates towards a spuriously better survival for Oscar winners.18 ,19 Similarly, patients with Eisenmenger syndrome entering a study cannot have had an event (death) between the time of development of the condition and the time of enrolment. Therefore, the time from diagnosis to start of follow-up should be removed from the analysis, by accounting for the age at entry to the study (figure 1). Ignoring this is likely to substantially underestimate mortality, resulting in the better survival prospects reported by some studies. After accounting for immortal time bias, a much more dire picture emerged, which is consistent with the poor survival prospects suggested by Corone et al15 and early studies.2–4 The prevalence of Eisenmenger syndrome in the community, however, is not well defined and it is possible that mainly symptomatic patients and those representing the worse end of the spectrum of Eisenmenger syndrome are referred to tertiary centres. Therefore this could provide a counterbalance to immortal time bias when assessing the entire Eisenmenger population.
Accurately estimating the survival prospects of patients with Eisenmenger syndrome is clinically relevant for at least two reasons. First, appreciating the high mortality of patients with Eisenmenger syndrome and the apparent lack of progress over preceding decades provides a rationale for a more proactive treatment strategy in this chronic progressive form of pulmonary hypertension. In fact, data from our institution suggest that treatment with DTTs may improve outcome and this potential survival benefit should be taken into account in the management of treatment naïve patients. Second, the results presented here call for a reconsideration of the survival prospects of adult patients with Eisenmenger syndrome and for an even more proactive approach to early diagnosis and repair, with the scope of avoiding the development of the condition.
The likely lack of a sizeable improvement in survival prospects over time in treatment naïve patients with Eisenmenger syndrome is of interest, and partly surprising. Although no specific therapy for the condition has been available until approximately one to one and a half decades ago, when intravenous and later oral DTT were introduced,20 ,21 one may have expected that the pathophysiological insights into the condition and especially the recognition of the potentially detrimental impact of phlebotomy and iron deficiency,5 ,22 as well as endocarditis prophylaxis and heart lung transplantation may have improved outcome. This finding suggests that any survival benefit of DTT shown as part of retrospective observational studies cannot be attributed to improvements in supportive management (including the avoidance of pitfalls and of previous detrimental practices) which have occurred over the few decades, alone. This supports a proactive approach with DTT or heart lung transplant in this challenging group of patients.
The SMR estimation procedure used here is based on the implicit assumption of a constant SMR (ie, independent of age). This assumption may be violated especially in older age but it is used to provide an estimate of the average excess mortality in patients with Eisenmenger syndrome and the possible magnitude of the error introduced by failing to account for left truncation. Like standard survival analysis methods under conditions of censoring, the methods employed here are based on a number of assumptions: These include the assumption of uninformative censoring and left-truncation and the assumption that the data represent a set of independent and identically distributed observations.13 ,23 In addition, the meta-analytical method proposed by Arends et al14 assumes that the estimated cumulative hazards are independent over different intervals. In addition, the chosen transformation of the data corresponds to a Weibull distribution assumption. While we have not formally tested these assumptions, figure 4 illustrates that the survival estimates obtained by using this method are consistent with those observed in our own data set. We accept that the studies included in this analysis may have been subject to referral bias, favouring patients with more advanced disease and worse survival prospects than those expected in the community. This could lead to an overestimate of the mortality risk in patients with Eisenmenger syndrome. Consistent with the majority of studies, we focused on adult patients and thus the analysis presented here ignores mortality in paediatric patients with Eisenmenger syndrome. In addition, the included studies were heterogeneous in terms of the underlying diagnosis, average patient age, publication era and country of origin. This by and large reflects the nature of the disease covering a broad spectrum of underlying heart conditions. Furthermore, the effect of treatment era and country of origin is accounted for (at least in part) by using appropriate country and era-specific life tables for statistical analysis. We cannot prove that no progress in terms of survival prospects for treatment naïve patients with Eisenmenger syndrome has been made during the last four decades. Rather this conclusion has to be made based on the visual inspection of the survival curves. However, this conclusion is supported by the consistent results of different statistical approaches.
Patients with Eisenmenger syndrome not treated with DTT have relatively poor survival prospects, with no clear evidence of improvement in survival seen after the 1960s. The overoptimistic estimate of the long-term outcome of patients with Eisenmenger syndrome reported in previous studies is likely attributable—at least in part—to the method used for analysing left-truncated data, which did not account for immortal time bias. Treatment with DTTs may be considered for adult patients with Eisenmenger syndrome, given the potential for improving their survival prospects.
What is already known about this subject?
Patients with Eisenmenger syndrome represent the extreme end of the spectrum of pulmonary hypertension in congenital heart disease. Despite the well appreciated morbidity and functional limitation of these patients, survival prospects are generally considered to be relatively benign compared with other forms of pulmonary hypertension.
What does this study add?
The current study highlights the issues of immortal time bias in the setting of patients with Eisenmenger syndrome. It also highlights that survival prospects in untreated patients are less benign than previously thought and may provide a rationale for a more proactive treatment strategy.
How might this impact on clinical practice?
The overoptimistic estimate of the long-term outcome of patients with Eisenmenger syndrome reported in previous studies is likely—at least partly—attributable to the method used for analysing left-truncated data, which did not account for immortal time bias. Treatment with disease targeting therapies may be considered for adult patients with Eisenmenger syndrome, given the potential for improving their survival prospects.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
- Data supplement 1 - Online appendix
Contributors G-PD: design of the study, data analysis, writing of the manuscript. AK: data analysis, critical review of the manuscript. RI: data acquisition and statistical analysis. RR, SJW, HB, MAG: study design, critical review of the manuscript. KD: study design, writing of the manuscript, critical review of the manuscript.
Funding This study was supported by a research grant from the EMAH Stiftung Karla Voellm, Krefeld, Germany. G-PD, AK, KD, MAG and the Adult Congenital Heart Centre and Centre for Pulmonary Hypertension, Royal Brompton Hospital, London, UK have received support by Actelion UK, Pfizer UK, GSK UK, the British Heart Foundation and the NIHR Cardiovascular and Respiratory Biomedical Research Units.
Competing interests G-PD, AK, KD, MAG and the Adult Congenital Heart Centre and Centre for Pulmonary Hypertension, Royal Brompton Hospital, London, UK have received support by Actelion UK, Pfizer UK and GSK UK.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.