Article Text

Microsimulation and clinical outcomes analysis support a lower age threshold for use of biological valves
  1. Serban Stoica1,
  2. Kimberley Goldsmith2,
  3. Nikolaos Demiris2,
  4. Prakash Punjabi3,
  5. Geoffrey Berg4,
  6. Linda Sharples2,
  7. Stephen Large5
  1. 1Bristol Royal Children's Hospital, Bristol, UK
  2. 2Medical Research Council Biostatistics Unit, Cambridge, UK
  3. 3Hammersmith Hospital, London, UK
  4. 4Golden Jubilee National Hospital, Glasgow, UK
  5. 5Papworth Hospital, Cambridge, UK
  1. Correspondence to Stephen Large, Papworth Hospital, Cambridge CB3 8RE, UK; stephen.large{at}


Objective To characterise contemporary results of aortic valve replacement in relation to type of prosthesis and subsequent competing hazards.

Methods 5470 procedures in 5433 consecutive patients with aortic valve replacement ± coronary artery bypass grafting (CABG) were studied. Microsimulation of survival and valve-related outcomes was performed based on meta-analysis and patient data inputs, with separate models for age, gender and CABG. Survival was validated against the UK Heart Valve Registry.

Results Patient survival at 1, 5 and 10 years was 90%, 78% and 57%, respectively. The crossover points at which bioprostheses and mechanical prostheses conferred similar life expectancy (LE) was 59 years for men and women (no significant difference between prosthesis types between the ages of 56 and 69 for men, and 58 an 63 for women). The improvement in event-free LE for mechanical valves was greater at younger ages with a crossover point of 66 years for men and 67 years for women. Long-term survival was independently influenced by age, male gender and concomitant CABG, but not by type of prosthesis. In bioprostheses the most common long-term occurrence was structural deterioration. For men aged 55, 65 and 75 at initial operation it had a lifetime incidence of 50%, 30% and 13%, respectively. The simulation output showed excellent agreement with registry data.

Conclusion Bioprostheses can be implanted selectively in patients as young as 56 without significant adverse effects on life expectancy, although event-free life expectancy remains significantly lower with bioprostheses up to age of implant of 63.

  • Aortic valve, replacement
  • heart valve, bioprostheses
  • heart valve, mechanical
  • surgery-valve
  • aortic valve disease

Statistics from

Prosthesis choices for aortic valve replacement (AVR) between the ages of 50 and 70 years are not standardised, even in the absence of other factors that reduce life expectancy. Of the two trials comparing mechanical and biological valves, one found better survival with mechanical valves.1 In practice the trend is towards using more bioprostheses but the evidence for that is not strong. In the series reported here, for example, the ratio of mechanical to biological valves changed from 1:1 in the first half of the study period to 1:1.8 in the second half. Thus it remains unclear how the bleeding hazard for mechanical valves compares long-term with the risk of structural degeneration of bioprostheses. Randomised trials for such few and remote outcomes, as well as longitudinal follow-up, are logistically difficult, particularly when the valve industry innovates constantly. Use of bioprostheses below the age of 65 is supported by some publications,2–4 discouraged by others,5 whereas another view is that patient factors are more important in determining long-term outcomes.6 Some studies have used microsimulation of the patients' course with point estimate inputs not allowing for variability (deterministic inputs).2 3 The aim of this study was to analyse life expectancy and complication rates for UK patients undergoing AVR with bioprostheses and mechanical prostheses. Age, gender, concomitant coronary artery bypass grafting (CABG) and other patient factors are examined using simulation of the lifetime after AVR surgery with non-deterministic inputs incorporating variability.

Patients and methods

Local research ethics committee approval was not required for this retrospective study. Five thousand four hundred and seventy consecutive AVR ± CABG procedures (5433 patients) at three centres between 1993 and 2006 were studied. Patient data were used to study perioperative mortality. A microsimulation model was constructed to predict life histories for the study patients, using patient data and published event rates for inputs (see online supplementary data). Since detailed follow-up information was not available for study patients this was supplemented with results from a meta-analysis of valve-related morbidity and mortality. Validation of the simulation results, including overall survival, was carried out against data from the UK Heart Valve Registry (UKHVR). UKHVR contains reoperation and survival statistics for individual patients but no follow-up on valve-related events.

Patients and valves

Patient information from the three centres was collected prospectively. Patients data were de-identified to maintain confidentiality. Table 1 shows a list of valves used in the study period (1993–2006) and table 2 shows patient characteristics. Homografts and valve repairs were excluded.

Table 1

Types of valves inserted

Table 2

Patient characteristics by prosthesis type

Perioperative survival

Estimates of operative death from first and subsequent AVR were taken from the multiple variable analysis of perioperative mortality from the three participating centres. The effects of age, gender, prosthesis type, concomitant CABG, creatinine and reoperation on perioperative mortality were studied using logistic regression. Variables such as the New York Heart Association score for breathlessness and left ventricular function were not recorded consistently in the databases and so did not appear in the final model. Manual stepwise selection based on the likelihood ratio statistic was used to detect variables for the multivariable models. For each age group, different survival probabilities were estimated for four subgroups defined by sex (male/female) and concomitant CABG surgery (yes/no), according to results from the logistic regression.


In the absence of long-term follow-up of valve-related events for UK patients, a systematic review of the literature and random effects meta-analysis were performed,7 while for other event rates we had similar inputs and model assumptions as Puvimanasinghe (2004)3 (supplementary table E1). PubMed was searched between 1990 and 2009 with the Boolean string ‘aorta OR aortic AND valve replacement’. The search was limited to articles on humans published in English in core clinical journals and resulted in 2085 hits. Inclusion criteria were: (1) outcomes of AVR listed in table E1 with prostheses listed in table 1, size ≥19 mm; (2) concomitant procedure, if present, CABG or mitral valve replacement (all mitral patients in combined series, with or without concomitant AVR, were subsequently excluded); (3) guideline compliant, or in the spirit of published guidelines8; (4) follow-up available in ≥95% of patients. Exclusion criteria were: (1) absence of any morbidity data; (2) linearised occurrence rate of events unpublished or unobtainable from the published data; (3) specialised series (redo, small aortic root, extremes of age); (4) overlapping or previously published series, in which case the last report was chosen.

Forty-seven articles met the criteria, containing 28 623 patients and 152 075 patient-years follow-up (supplementary table E2). Where necessary, linearised occurrence rates and their SEs were calculated with the available data (see online supplementary data). For embolism, valve thrombosis and non-structural dysfunction, we used a random rate, simulated from a distribution obtained by meta-analysis. For endocarditis, we assumed a constant rate for the first 6 months and a smaller constant rate thereafter. For structural valve deterioration, we assumed an increasing rate over time for bioprostheses3 and a zero event rate for mechanical valves. Finally, for haemorrhage, we employed a constant rate over time for bioprostheses and an exponentially increasing rate for mechanical valves.9 Estimates of reoperation and death rates associated with these events are from the literature.3


Conditional on perioperative survival, the microsimulation for studying valve outcomes was similar to that described in detail by the Rotterdam group.2 3 10 Briefly, after AVR a number of factors influence the trajectory between the states alive and deceased: background, additional and operative mortality, and valve-related events. Background mortality was taken from government life tables.11 Additional mortality reflects the increased hazard of death for a patient undergoing AVR as compared with the general population.3 Heart rhythm and function or type and severity of valve disease can contribute to this hazard. We used an accepted simulation method10 but this was subject to probabilistic sensitivity analysis.

For each of the parameters in the model that are known imprecisely (eg, event rates, covariate effects), a value was simulated for the distribution of the parameter (Poisson for event rates, binomial for adverse event probabilities) and, based on these estimates, the life histories of 50 000 individuals for each age group were simulated using the following steps:

  1. Random event rates were drawn for valve-related morbidity and its sequelae of reoperation and subsequent death.

  2. A random probability of death from operation/reoperation was drawn from its distribution. This was used to decide if a simulated individual survived the initial operation. If so, the main body of the simulation continued.

  3. Based on UK mortality tables, adjusted for the additional mortality related to AVR, a random age was sampled that became the age-at-death of the individual should no valve-related event occur.

  4. Using the occurrence rates in table E1 and the event rates from step 1, a random age at which each event may occur was simulated. If all these times were greater than the age-at-death in step 3, the simulation ended and that was the age-at-death. Otherwise, the event that corresponded to the shortest time was assumed to occur first. The mortality rate associated with each event was used to determine whether the individual died owing to the event or not. In the former case, the age-at-death was (re)set to the time at which the event occurred. If the individual survived the event but required surgery because of the event, any additional surgical mortality was also simulated. Thereafter, the time to the next event was determined, possibly altering some event probabilities conditional on the simulated individual history. This procedure continued until the simulated individual died or reached the random age-at-death of step 3.

  5. Steps 2–4, were repeated 50 000 times for each valve type, and life expectancy (LE), event-free life expectancy (EFLE) and the probabilities of valve-related events were calculated from the simulated individual histories. In order to estimate the distributions of these outputs, steps 1–4 were repeated for 100 draws from the event rate distributions.


Using UKHVR data from the three centres an observed (empirical) survival curve for the patients was constructed using Kaplan–Meier methods. Predicted survival curves from the simulation were compared with UKHVR observed survival curves. Estimates of uncertainty for the model resulted from probabilistic sensitivity analysis, in which all model parameters were varied simultaneously, and are represented graphically in the output. The effects of age, gender, prosthesis type, concomitant CABG and creatinine levels on long-term survival were studied using Cox proportional hazards models applied to the observed data. In addition to patient survival, prosthesis lifetimes were calculated. Patients were considered to have had the event (first prosthesis failure or operative death) if they had a second AVR operation, died perioperatively at initial surgery, or died of any valve-related or cardiovascular cause of death. Otherwise, patients were censored. For second prosthesis lifetime only patients having had a second AVR operation were included.


Patients and valves

Between 1993 and 2006 5470 consecutive AVR operations were performed in 5433 patients (2239 mechanical and 3231 biological valves) (table 2). In univariate analysis patients receiving bioprostheses had 1.6 times the odds of dying perioperatively as compared with patients receiving a mechanical valve (table 3). In multiple variable analysis, type of prosthesis and gender were not associated with perioperative mortality when other characteristics were analysed. Concomitant CABG was associated with 1.4 higher odds of death (p=0.05). Each 10 year increment in age had a 1.5 times increase in the odds of operative mortality (p<0.001). A 10 unit increase in creatinine levels was associated with a 5% increase in the risk of death soon after the valve operation (p<0.001). The Hosmer and Lemeshow test (p=0.53) and the c-statistic for the final model (0.67, 95% CI 0.63 to 0.71) showed no evidence of poor fit. These results were taken to the next stage.

Table 3

Relationship between perioperative mortality and patient characteristics


Figures 1 and 2 show differences in LE and EFLE after AVR between bioprosthetic and mechanical valves for men and women with and without CABG. Men had a small but consistent survival benefit in receiving a mechanical valve for patients up to about age 59. Although this was the crossover point, there was no significant difference between bioprostheses and mechanical valves between the ages of 56 and 69 for men. The improvement in EFLE was greater at younger ages and bioprostheses were favoured after the age of 66, although there was no significant difference between the two types of valves between the ages of 62 and 68. For women the crossover points at which the valve types conferred the same LE and EFLE were approximately the same as for men (59 and 67, respectively). Patients who had AVR and CABG had only slightly shorter LE and EFLE than those having AVR alone. The only difference in model inputs for the combined procedure was a small but significant increase in operative death rate.

Figure 1

Mean difference (95% confidence limits) in life expectancy between bioprostheses and mechanical valves by age. Values above 0 favour bioprostheses. CABG, coronary artery bypass grafting.

Figure 2

Mean difference (95% confidence limits) in event-free life expectancy between bioprostheses and mechanical valves by age. Values above 0 favour bioprostheses. CABG, coronary artery bypass grafting.

Figure 3 illustrates valve-related complications for men. Supplementary table E3 summarises valve-related complications for both sexes. The plots and tables for men show that prosthetic endocarditis had an estimated lifetime probability of 6–13% depending on age at first procedure. Although it was more common for mechanical valves, the differences were small. Non-valve systemic thromboembolism had a slightly higher estimated incidence in recipients of bioprostheses compared with mechanical valves, in agreement with inputs from table E1. Haemorrhage, as expected, was significantly more common among patients with a mechanical valve. Structural valve disease was common for bioprostheses with a lifetime probability of 58%, 50%, 40%, 30%, 20% and 13% for men aged 50, 55, 60, 65, 70 and 75, respectively, at initial operation. Valve thrombosis was fairly uncommon, occurring with a lifetime probability of <8% in male patients with a mechanical valve and <0.2% in male patients with a bioprosthesis, in all age groups. Non-structural valve dysfunction was also uncommon, with a marginally higher occurrence for the mechanical group, with a lifetime probability of 3.6% and 4.3% for 50-year olds, decreasing to 1.8% and 2.0% for 75-year olds for the bioprosthesis and mechanical groups, respectively. Event rates were slightly higher in general for women but had a similar distribution (table E3).

Figure 3

Lifetime probability of valve-related complications for men, aortic valve replacement only. (A) Endocarditis. (B) Thromboembolism. (C) Haemorrhage.

Validation and long-term survival

Model-predicted patient survival at 1, 5 and 10 years was 90%, 78% and 57% (figure 4), whereas first valve survival was 91%, 85% and 71%, respectively. These figures agreed well with registry data. In multiple variable analysis, long-term survival was adversely affected by increasing age, male gender and CABG but not tissue prosthesis, although there were clearly age-specific differences between valve types (table 4). Redo AVR operations were captured via the UKHVR: there were 74 during the study period (68 first time and six second time). The most common causes of redo operations were intrinsic valve failure (22%), prosthetic endocarditis (12%) and dehiscence (9%). Figure 4 shows that the UKHVR actuarial survival curves for mechanical and bioprosthetic valves compared with predicted survival from the simulation for a 65-year-old man. The registry curves lie almost entirely within the confidence bounds for the model.

Figure 4

Kaplan–Meier survival estimates from combined analysis of UK patients with aortic valve replacement compared with simulation predicted survival for men aged 60, 65 and 70. (A) The first curve from the top shows the UK general life expectancy for 60-year-old men. The model prediction is shown as dots and the solid lines are 95% confidence bands. All these hypothetical survival curves are extended to 100 years of age. The two shorter curves are Kaplan–Meier estimates for 60-year-old men from the UK Heart Valve Registry data. (B, C) Same as above for men aged 65 and 70.

Table 4

Relationship between survival and patient characteristics


This is a clinical study reporting long-term contemporary outcomes of AVR and, in particular, satisfactory results with bioprostheses below the age of 65. The microsimulation methodology builds on the pioneering work of the group in Rotterdam10 and combines patient data with mathematical modelling to estimate life histories after AVR. Regression models were used to evaluate short- and long-term mortality. We have attempted to achieve greater accuracy of estimates through the following design measures: (1) a larger contribution of actual patient data to the model; (2) detailed analysis of CABG and gender for estimating prognosis, in addition to age only as reported previously3; (3) a large meta-analysis of bioprostheses and mechanical prostheses as reported in contemporary practice; (4) probabilistic sensitivity analysis, which also provides CIs rather than point estimates and (5) model validation based on individual survival data from a national valve registry.

The results show that, for men who do not have concomitant CABG, the two valves have equivalent LE and EFLE at age 59 and 65, respectively, and that differences between the two valves are not statistically significant for ages as low as 56 and 62 (figures 1 and 2). The crossover ages for women are 58 and 64, respectively. The slightly steeper curves seen in women are probably related to longer LE overall. These results, partly reflecting the 95% CI around the estimate, suggest that there is flexibility in choosing prosthesis type around this age. Taking other patient characteristics into consideration as well as patient preference seems appropriate. A higher risk of endocarditis for mechanical valves (figure 3A) is probably related to the inputs we used and the fact that we used large datasets resulting in narrow CIs. As detailed in ‘Patients and methods’ we preferred to use a two-phase hazard for endocarditis3 rather than the constant rate derived from meta-analysis. In clinical terms the differences remain small. A jump in the risk of haemorrhage at age 60 (figure 3C) reflects an exponential increase in this complication with age, modelled with age as a categorical covariate as described by van de Meer et al.9 By incorporating CABG the crossover points do not change significantly. This is in keeping with other data from Rotterdam and is probably a reflection of the fact that a shorter life expectancy in patients with coronary heart disease is also associated with fewer valve-related complications.12

The role of renal function, as described by creatinine levels, seems to be small for short- and long-term survival. Creatinine clearance is related to age and so we did not use these two correlated covariates together in the analysis to avoid making the model statistically unstable. The agreement between the simulation estimated survival and the observed Kaplan–Meier survival, demonstrating excellent calibration of the model, is remarkable since with such a large dataset we would be able to detect small differences (figure 4). There is some disagreement at the end of the observed data at age 75, but this is likely to reflect smaller numbers at risk.

Our study has a number of limitations. Its retrospective nature means that bias in valve selection is introduced by both surgeons and patients. Valve-related morbidity and mortality are underestimated retrospectively, therefore segments of the model are heavily reliant on meta-analysis of published data; the retrospective papers included are not immune either from this shortcoming. Additionally only PubMed was searched systematically, along with reference lists from identified papers, so that it is possible that some relevant studies were not represented in the meta-analyses. Other factors that we did not study could certainly have a role. With accurate, prospectively collected data, this analysis can be extended to include the influence of preoperative rhythm and degree of heart failure as well as patient–prosthesis mismatch. Moreover, results should not be extrapolated uncritically to populations with substantially different underlying life expectancy, although the methods are generalisable. Our conclusions are based on a variety of prostheses. However, contemporary valves have comparable short- and long-term results and so this amalgamation should be acceptable. Intuitively, the lower anticoagulation targets for modern mechanical valves may be paralleled by improved long-term outcomes by concomitant improvements in tissue valve technologies. Finally, microsimulation itself, much like any mathematical modelling, is as good as its inputs. Oversimplifications are inevitable in places–for example, by assuming that valve-related complications are constant over time and independent of patient age and time elapsed from surgery.

An analysis of quality-of-life outcomes in middle-aged patients undergoing AVR highlighted some significant disadvantages of mechanical valves.13 Increasingly, with better operative outcomes, the decision-making process in selecting a valve substitute shifts from mortality to morbidity. Patients may show different attitudes in their attitude to risk from haemorrhagic complications versus the possibility of reoperation for a failed bioposthesis.14 Our results support the notion that quality-of-life dividends are not associated with a decrease in quantity of life by implanting bioprostheses selectively from age 56 upwards.


Maria Benedicta Edwards, from the United Kingdom Heart Valve Registry, assisted in linking our data to the registry.


Supplementary materials


  • Linked articles 207050.

  • Competing interests None.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

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.

Linked Articles