Overview
This chapter describes the structure of the economic model, the scenarios evaluated and the probabilistic sensitivity analysis. The main assumptions of the model are also presented in this section. The underlying model is based on the study by Sharples et al.30 which has been adapted for our decision problem and updated with new data. For more detailed information, readers are referred to Chapter 7 and 8 in the Sharples et al.30 report.
Model structure
An economic model was developed based on a multistate model of patient experience from the UK during the period April 2005 to November 2011. The aim of the economic model was to compare BTT with MM treatment for patients eligible for HT.
The model is a semi-Markov, multistate model as shown in . In the model each patient can be in one of three mutually exclusive health states, namely alive with VAD (or MM) support (state 1), alive after HT (state 2) or dead (state 3). Each individual may move between health states or remain in the same state. State 3 (dead) is an absorbing health state. The transition between each of these health states, referred to as the TP, is represented by the quantities p12, p13 and p23. Transition probabilities are not fixed but depend on time t since the VAD was implanted (p12, p13) or the time t* since transplantation (p23).30 For patients on MM support, a precisely similar model was constructed, with different estimates of pre-transplantation transition probabilities (p12, p13), but the same estimates of post-transplantation transition probabilities (p22 and p23).
Discrete-time, semi-Markov, multistate model of health states for VAD patients. For patients on MM instead of a VAD, a precisely similar model was constructed, with different estimates of the pre-transplantation transition probabilities (p12, p13), but (more...)
Cycle length was set at 1 month and transition between each health state occurs at the end of each cycle.
The model was evaluated over several time horizons. For the base-case scenario, a lifetime horizon, spanning approximately 50 years, was used. The model was also run for shorter time horizons of 3 and 10 years. The model evaluates costs from the perspective of the NHS. Thus, only direct costs related to VAD implants have been included and indirect costs are excluded. All costs are at 2010/11 UK prices in pounds sterling (£). Health outcomes were measured in QALYs. In accordance with current UK guidelines,101 an annual discount rate of 3.5% was applied to both costs and health outcomes. Both deterministic and probabilistic approaches were used to estimate the cost-effectiveness of VADs. The probabilistic approach was used to account for uncertainty in the various variables within the model.
Base-case analysis
For the base-case analysis we used observed survival data from the BTDB. We hypothesised that although survival rates are different for each group (patients who received a second- or third-generation approved VAD as a BTT or patients who received MM support to transplant), they would have common post-transplant survival rates, with a constant death rate for months 3–12. In the base-case analysis survival up to 3 years from VAD implantation/listing were estimated using data from the BTDB based on constant death rates beyond 6 months post transplantation. Several assumptions were made when estimating longer-term survival rates after 42 months (see Chapter 7).
Structural assumptions
Disease state/pathways
Two pathways were modelled for this economic evaluation of VADs. In the base case, patients with more severe HF (based on inotrope medication) either followed the VAD pathway or were allocated to the MM pathway. In both pathways patients received a HT after a certain period of time (which was varied according to different sensitivity analyses). Some of the patients died before receiving a HT.
Strategies/comparators
For the two research questions we compared:
use of VADs as a BTT with MM using the inotrope subgroup of patients as the comparator group
use of VADs as an ATT with use of VADs as a BTT. For an ATT, transition probabilities were kept the same as for patients in the BTT base-case arm, except that the probability of receiving a donor heart was set to zero.
For the sensitivity analyses we included comparisons of:
use of the HW only, as a BTT with MM using the inotrope subgroup of patients as the comparator group as in the base case
use of VADs as a BTT with all MM patients (both inotrope and non-inotrope)
use of VADs as a BTT with an artificially constructed MM group using the VAD patients as their own controls. (Based on predicted survival of the VADs group – had they been treated with MM not VADs. Predictions were made using the SHFM; see
Chapter 7, Selection of comparator group and sensitivity analyses.)
Cost-effectiveness summaries
Incremental costs and QALYs gained were estimated and summarised as the ICER, the additional cost per QALY gained. Specifically, given mean costs CA, CB, CC and CD and mean benefits (QALYs) QA, QB, QC and QD for the groups, the ICER for group A relative to group B, say, is:
The mean costs and benefits for each group were estimated from the economic model using data from the BTDB.
The joint distribution of incremental mean costs and benefits was plotted on the cost-effectiveness plane and used to estimate both the incremental net benefit (INB), for example:
and the cost-effectiveness acceptability curve (CEAC), for example:
where λ represents the maximum acceptable cost for one unit of benefit, in this case one QALY.
Estimation of model parameters
Three types of input were considered for the economic analysis: transition probabilities estimated from the BTDB, utilities derived from the published literature, and costs computed from UK data. The following assumptions were made in the base-case analysis and subsequent scenarios.
Transitions to the HT state were assumed to occur at monthly intervals and a whole month of pre-transplantation survival and costs were included. However, in practice, a transplant may take place at any time during the month and, on average, at the mid-point of the relevant month. Also, costs and utilities associated with death were assigned zero. A half-cycle correction was added to reflect the fact that a death could occur at any time during the month, although transitions were assumed to occur at monthly intervals. Thus, a transition to death would result in a reduction in survival time of 0.5 months. For the month in which death occurred no reduction in costs was required, as only costs up to death were included in these months.
In summary, for the economic model, a simple discrete-time, discrete-state model was constructed. Cost-effectiveness summaries of interest were estimated by weighting time in each state of the model by the utility and cost associated with that state. Transition probabilities, costs and utilities have been estimated using data from the NHS BTDB (see Chapter 7).
Quality of life and utilities
Health-related quality of life remains relatively static in HF patients who are medically managed,102 and improves after receiving a VAD53,103,104 or HT,104–106 with improvements maintained for several years.103,105,107–109 Recipients of HT report better HRQoL than recipients of VAD.104 In the model, health outcomes were measured in QALYs, in accordance with current UK guidelines.101 The EuroQoL EQ-5D110 is the preferred measure of decision-making bodies such as the National Institute for Health and Care Excellence (NICE).101 The literature revealed two applicable sources of EQ-5D utility scores derived from patients suffering from chronic HF: Sharples et al.30 and Gohler et al.111
Sharples et al.30 derived EQ-5D utility scores from UK patients suffering from chronic HF who were either implanted with a VAD or medically managed while waiting for HT. A subset of the group were reassessed post HT. shows the extracted data.
European Quality of Life-5 Dimensions utility scores derived from Sharples et al. (adapted from Sharples et al.)
Gohler et al.111 collected EQ-5D data on a subsample of the Eplerenone Post-AMI Heart Failure Efficacy and Survival Trial (EPHESUS) trial participants. EPHESUS was a multinational RCT which investigated the effect of the aldosterone antagonist eplerenone (Inspra®, Pharmacia) in patients with chronic HF after acute myocardial infarction. Responses to the EQ-5D descriptive system were used to generate an EQ-5D utility score by applying the appropriate tariff based on participant's country of origin. Univariate and multivariate analyses were used to investigate the association of EQ-5D utility scores with NYHA class. The findings highlight the utility loss associated with worsening NYHA class, with excellent model fit found in the multivariate models. The association between NYHA class and HRQoL, including EQ-5D utility scores, is supported in the literature.25,112,113
shows the relationship between NYHA class and utility.
European Quality of Life-5 Dimensions utility scores by NYHA class (adapted from Gohler et al.)
For the purposes of this analysis we used the data provided by the BTBD to determine EQ-5D utility scores for health states in the model. The HT data set recorded NYHA class for 1011 patients who received a HT (from 2002 till the end of 2011). NYHA class was entered at initial registration and 3 months after VAD implant (for the 83 of 235 patients who subsequently received a HT). For those who received a HT, NYHA class was recorded post transplant at their 3, 12 and 24 months outpatient visits. The BTDB suggests that there is some improvement in NYHA class after HT; however, this translates into very minor changes in the weighted EQ-5D utility scores over time. summarises the data.
New York Heart Association class of patients post-VAD implantation and post HT
For the model, the weighted derived EQ-5D utility score was based on the proportions of patients for each NYHA class for VAD patients (3 months post implant) and HT patients (3 months post transplant) (see ). Thereafter, utility was assumed to remain constant.
A weighted EQ-5D utility score for all MM patients and for those MM patients receiving inotropes was similarly determined using NYHA data recorded at registration () and, as with previous analysis, EQ-5D utility score was assumed to remain constant thereafter (). shows the EQ-5D utility scores used for the base-case analysis. For the sensitivity analysis, data reported by Sharples et al.30 were used (see ).
New York Heart Association class of inotrope MM patients and all MM patients
European Quality of Life-5 Dimensions utility scores for base-case analysis
In the next chapter we describe derivation of transition probabilities between model states in more detail.
