Data Analysis
Data analysis was performed with Microsoft Excel (Microsoft Office 2010,
Microsoft, Redmond, WA, USA) and IBM SPSS version 21.0 (IBM, Somers, NY,
USA) and in the R statistical software (version 3.1.0. R Development
Core Team, R Foundation for Statistical Computing, Vienna, Austria)
using the metaphor package. Pooled baseline patient characteristics were
calculated with the use of sample-size weighting. Early mortality and
linearized occurrence rates (LOR) of late valve-related complications
were pooled on a logarithmic scale with the use of inverse variance
weighting in a random-effects model. Reported study characteristics and
pre- and peri-operative patient characteristics are presented as mean ±
standard deviation for continuous variables and percentages for discrete
variables. For outcome variables, individual and pooled statistics are
presented as LOR and 95% confidence interval (CI). In studies where
median and ranges instead of mean and variance were reported, the method
described by Hozo et al. (10) was used to
calculate the mean. In case of absence of total number of patient-years,
this was calculated by multiplying the number of patients with the mean
follow-up duration in years. In case a certain event did not occur in an
individual study, we assumed that 0.5 events occurred for that
particular outcome for the purpose of inverse variance weighting. When
an event was not reported, this study was excluded from the analysis of
that event.
For late mortality and reintervention,
subgroup analyses were performed stratifying the root replacement by
prosthesis type (stented vs. stentless), follow-up duration (individual
study mean follow-up less than pooled mean follow-up versus individual
study mean follow-up more than pooled mean follow-up), and age at
surgery. To assess the association of these variables with late
mortality and reintervention rates, linear regression analyses were
performed with weighting the studies according to the inverse variance
of the occurrence rate. Heterogeneity between the studies was assessed
using the I2 test. Funnels plots were used to
investigate publication bias. To investigate the potential influence of
publication bias on pooled outcome, sensitivity analyses were conducted
by temporarily excluding the smallest quartile (by sample size) of
included studies. This systematic review and meta-analysis was conducted
according to the PRISMA guidelines (11).
Microsimulation model: the concept
The microsimulation model is a computer application that simulates the
life of a patient after aortic valve replacement, considering the
morbidity and mortality events that the patient could experience. The
calculated mortality of a patient is composed of the background
mortality of the general population, operative mortality, mortality due
to valve-related events and an additional “excess mortality”. This
so-called excess mortality in the patient compared to a matched person
in the general population reflects mortality associated with the
underlying left ventricular function, valve pathology, and the root
replacement procedure. All pooled and weighted occurrence rates of
(operative) mortality risk, the occurrence rate of valve-related events
together with the risk of mortality and reintervention directly due to
valve-related events were obtained from the meta-analysis. The
occurrence rates of all events were assumed to be linear and non-age
dependent.
For patients aged 61-70 and > 70 years, these “excess
mortality“ hazard ratios were 1.2 and 0.8 for males, and 2.2 and 1.3
for females, respectively. The background mortality of the general
population was obtained from the 2004 United States Life Tables, as 2004
was the pooled median year of intervention, assuming a constant
incidence rate over time in each study
(12).
To obtain age-specific estimates of life expectancy and lifetime risk of
valve-related morbidity, the microsimulation model was run for the ages
of 60, 65 and 70 years for 10,000 iterations each and separately for
males and females. The age-specific outcomes of both genders were then
pooled at the male/female ratio obtained from our meta-analysis. For the
internal validation of the model, we performed an additional run for
10,000 iterations at the pooled mean age (65.5 years) and male/female
ratio (70%) of the meta-analysis. The actuarial survival obtained from
the microsimulation model for these data was then plotted against the
pooled (overall) mortality observed in the meta-analysis. A more
detailed account of the microsimulation and the methodology has been
supplied previously (13).
ResultsStudy and baseline patient characteristicsThe initial literature search exposed 2,106 publications. The selection
procedure is illustrated in Figure 1. Cross-referencing did not result
in additional papers. Thirty-one studies were finally included in this
systematic review with a total number of 5,227 patients, mean follow-up
of 4.1 years (range 1-10 years), and total follow-up of 22,706
patient-years. Root replacements was performed with a stented prosthesis
in 26% of patients and 74% were stentless valve prostheses. In one
study the implantation period was missing, which was provided by the
authors (14). Supplementary file 3 shows
an overview of the included publications and study characteristics.
Pooled pre- and peri-operative characteristics are presented in Table 1.