Overview of the model:
We constructed a Markov model using proprietary software (TreeAge Pro 2011 Software, Williamstown, MA) concerning a hypothetical reference case, which was similar to the approach adopted in previously published studies[24, 26–29]. A hypothetical cancer population of 64-year-old, 70 kg, and with VTE event receiving treatment with NOACs or LMWHs was considered for the model. A 1-month cycle length with a 6-month and 5-year time horizon was used. The 6-month time horizon was chosen based on the applied data period from the randomized controlled trials (RCTs) and the 5-year time horizon was chosen as it is commonly used to reflect important clinical and economic impacts of NOACs for CAT and general cancer survival [26, 30–32].
The Markov model, consisting of 10 health states, as depicted in Fig. 1, included on anticoagulant treatment, off anticoagulant treatment, recurrent pulmonary embolism (rPE), recurrent deep vein thrombosis (rDVT), intracranial hemorrhage (ICH), non-ICH major bleeding (MB), clinically relevant non-major bleeding (CRNMB), PE-related death, MB-related death, and death by any case. Specifically, patients entered the model following a VTE event, at the beginning of the decision tree, the patients would receive one of the four following agents for the treatment of VTE in cancer patients: apixaban, rivaroxaban, edoxaban, or LMWHs that dosages for agents were based on their respective trials, and then they either remained in their current on-treatment state, moved to an event state, transitioned to an off-treatment state or died, during the course of 1-monthly cycles. Each state was associated with a cost and utility weighting to calculate the total costs and QALYs of patients simulated in the model.
Parameters of the model input:
The clinical inputs for various event probabilities used in the model are summarized in Table 1. The probabilities of events of NOACs and LMWHs during the first 6 months were obtained from 4 good quality RCTs including the Hokusai VTE Cancer trial[33], Select-D[34], the Caravaggio study[35], and ADAM VTE trial[36] in which each of the NOACs were directly compared with LMWHs for the treatment of VTE focused on patients with active cancer. All studies had the primary efficacy outcome (recurrent VTE) and the primary safety outcome (major bleeding). In each study, patients were followed for at least 6 months. NOACs were shown to be noninferior to dalteparin for recurrent VTE and major bleeding. Bleeding was more common in patients with GI malignancies taking edoxaban and rivaroxaban compared with dalteparin[33, 34]. In contrast, apixaban was not associated with an increased risk of bleeding compared with dalteparin in the ADAM and Caravaggio trials[35, 36]. Transition probabilities (TP) for 7–12 months were derived directly from the Hokusai-VTE study and the same estimates were extrapolated to the time horizon beyond 12 months (Supplemental Table 1). The time-varying TP for recurrent VTE when off anticoagulant treatment was estimated from a large population study of cancer patients [37]. The probability of events of bleeding seen in patients with gastrointestinal malignancy was derived from the randomized controlled trials as above, including the Hokusai-VTE[33], Select-D[34], Caravaggio[35] trials. The event rates were translated into monthly transition probabilities with the following formula: Tp = 1−(1 − p)^(1/n)(with Tp = monthly probability of events, p = event probability as reported in the literature, and n = number of months).
The cost analysis was evaluated from the healthcare system perspective setting in China. In analysis, it includes patients' direct medical costs related to drugs and complications, without considering indirect costs and direct non-medical costs. The daily drug acquisition costs of NOACs and LMWHs were collected from public databases[38]. An average of the edoxaban, rivaroxaban, and apixaban total drug costs was used for the NOACs arm. The wording NOACs refer to apixaban, edoxaban, or rivaroxaban, where did not include dabigatran as dabigatran was not used in any study. Costs for enoxaparin were used for this model due to its widespread use in China, although the clinical trials in cancer patients have used LMWHs such as dalteparin. Monthly costs (each cycle) were derived from 30-day prescriptions of the drugs at the labeled dosing frequency. The cost of symptomatic DVT or PE considered both the diagnosis and hospitalization costs incurred for such events[39]. The resource use in managing a major bleeding event was based on a Chinese study analyzing the costs for inpatient admissions due to major bleeding events[40]. Costs for other states were also based on values in the previously published literature. All the costs were calculated and reported in US dollars (USD) with the average exchange rate in 2020 (Ұ= $0.14). Also, a discount rate of 5% was used, as recommended by Chinese guidelines for pharmacoeconomic evaluations[41] each year. All costs are reported in Table 1.
The quality adjusted life years (QALYs; duration times utility) was incorporated in the model by using the values of utility. Evidence from previously published literature was used to determine the various utility values. As the literature on the utility of various events in cancer patients with VTE events is scarce, thus, most data were obtained from VTE patients without cancer [42, 43]. The base utility was considered to be 0. 95 and oral anticoagulant treatments were assumed not to change the utility value [42]. The utility inputs of the direct effects of the treatment drugs and the series of clinical events are reported in Table 1.
Table 1
Parameters for model input with a cycle length of 1 month
Parameters
|
Base case |
Ranges |
Distribution |
Source |
1-6month transition probabilities (tp) a, %
|
|
|
|
|
NOACs
|
|
|
|
|
Recurrent DVT
|
0.33
|
0.26–0.40
|
β
|
[33–36]
|
Recurrent PE
|
0.54
|
0.43–0.65
|
β
|
[33–36]
|
Fatal PE
|
4.45
|
3.56–5.34
|
β
|
[33–36]
|
MB
|
0.70
|
0.56–0.84
|
β
|
[33–36]
|
Fatal MB
|
0.28
|
0.22–0.34
|
β
|
[33–36]
|
ICH
|
0.02
|
0.016–0.024
|
β
|
[33–36]
|
Fatal ICH
|
0
|
0
|
β
|
[33–36]
|
Major GI bleeding
|
1.17
|
0.94–1.40
|
β
|
[33–36]
|
CRNMB
|
1.81
|
1.45–2.17
|
β
|
[33–36]
|
Death of any case
|
4.46
|
3.57–5.35
|
β
|
[33–36]
|
Treatment discontinuation
|
2.91
|
2.33–3.49
|
β
|
[33–36]
|
LMWHs
|
|
|
|
|
Recurrent DVT
|
0.61
|
0.49–0.73
|
β
|
[33–36]
|
Recurrent PE
|
0.74
|
0.59–0.89
|
β
|
[33–36]
|
Fatal PE
|
2.24
|
1.79–2.69
|
β
|
[33–36]
|
MB
|
0.47
|
0.38–0.56
|
β
|
[33–36]
|
Fatal MB
|
1.26
|
1.01–1.51
|
β
|
[33–36]
|
ICH
|
0.08
|
0.06–0.10
|
β
|
[33–36]
|
Fatal ICH
|
5.45
|
4.36–6.54
|
β
|
[33–36]
|
Major GI bleeding
|
0.56
|
0.45–0.67
|
β
|
[33–36]
|
CRNMB
|
1.09
|
0.87–1.31
|
β
|
[33–36]
|
Death of any case
|
4.52
|
3.62–5.42
|
β
|
[33–36]
|
Treatment discontinuation
|
4.73
|
3.78–5.68
|
β
|
[33–36]
|
OffDVT
|
1.90
|
1.52–2.28
|
β
|
[37]
|
OffPE
|
2.03
|
1.63–2.44
|
β
|
[37]
|
Costs, $
|
|
|
|
|
NOACs 1st month
|
207.34
|
165.87-248.81
|
γ
|
[38]
|
NOACs 2nd month onwards
|
825.89
|
660.71-991.07
|
γ
|
[38]
|
LMWHs (enoxaparin) 1st month
|
149.91
|
119.93-179.89
|
γ
|
[38]
|
LMWH(enoxaparin)2nd month onwards
|
412.94
|
365.30-547.94
|
γ
|
[38]
|
DVT
|
693
|
329–941
|
γ
|
[39]
|
PE
|
1121
|
448–1793
|
γ
|
[39]
|
MB
|
654
|
603–704
|
γ
|
[40]
|
ICH
|
3834
|
2684–4984
|
γ
|
[44]
|
CRNMB
|
8.25
|
5.77–10.72
|
γ
|
[44]
|
Utilities
|
|
|
β
|
|
Base
|
0.95
|
0.76-1.00
|
β
|
[42]
|
NOACs
|
0.95
|
0.76-1.00
|
β
|
Assumedb
|
LMWHs (enoxaparin)a
|
0.85
|
0.68-1.00
|
β
|
[42]
|
DVT
|
0.84
|
0.67-1.00
|
β
|
[45]
|
PE
|
0.63
|
0.50–0.76
|
β
|
[45]
|
MB
|
0.65
|
0.52–0.78
|
β
|
[45]
|
ICH
|
0.33
|
0.26–0.40
|
β
|
[45]
|
CRNMBa
|
0.91
|
0.73-1.00
|
β
|
[24]
|
DVT, deep vein thrombosis; PE, pulmonary embolism; MB, major bleeding; ICH, intracranial hemorrhage; CRNMB, clinically relevant non-major bleeding; NOACs, new oral anticoagulants; LMWHs, low molecular weight heparins; offDVT, risk of deep-vein thrombosis while off-treatment; offPE: risk of pulmonary embolism while off-treatment. |
a: Upper and lower bounds estimated to vary ± 20% of the mean value for these input parameters estimates. |
b: Utility associated with NOACs treatment was assumed to be same as base. |
Analysis
We assessed the cost-effectiveness of treatment with NOACs compared to LMWHs among patients with CAT. In addition, given the increased rate of bleeding seen in patients with gastrointestinal malignancy on edoxaban[33] and rivaroxaban[47], a subgroup cost-effectiveness analysis was performed on this patient population. The primary outcome measure of this study is the incremental cost-effectiveness ratio (ICER) in costs per QALY. According to the world health organization (WHO) recommendation, When the ICER was less than three times the gross domestic product (GDP) per capita, cost-effectiveness would be considered[48]. We used three times the per-capita GDP of China in 2020 ($10142.58) with willingness to pay (WTP) thresholds of US $30427.74 per QALY as the cost-effectiveness threshold.
To explore the effect of parameter uncertainty in the model, we performed one-way sensitivity analysis (OWSA) and probabilistic sensitivity analysis (PSA). In OWSA, the value range of each parameter was either based on the reported or estimated 95% CIs in the referenced studies or determined by assuming a 20% change from the point estimate in the base-case analysis. The 10 most influencing parameters were presented in a tornado diagram. PSA was performed using a Monte Carlo simulation with 1000 iterations. the distributions assumed for the input parameters were gamma (cost), beta (utility weights and TP), and log-normal (RR). All the analyses were performed in TreeAge Pro 2011.