We constructed a decision analysis model using TreeAge 2022 to assess the cost-effectiveness of ceftazidime-avibactam-based versus meropenem-based HAP/VAP treatment regimens from the perspective of the Chinese healthcare system. Figure 1 provides an overview of the model structure, depicting different pathways for each patient from diagnosis to clinical cure or failure after initiation of empirical therapy.
The decision analysis model was based on the design of the REPROVE clinical trial, which enrolled 879 patients globally, including 56% Asian and 42% Caucasian individuals across 23 countries. Subjects were randomized to receive intravenous ceftazidime-avibactam (2000/500 mg) or meropenem (1000 mg) every 8 hours as empirical therapy for HAP/VAP. The model assumed that each patient entered empirical therapy and continued treatment for the next 48-72 hours (until microbiological results were available). Those without observed resistance continued empirical therapy, while those with any pathogen resistant to the empirical agent switched to second-line treatment with intravenous polymyxin B (1.66 mg) and high-dose meropenem (2000 mg) every 8 hours, which was considered clinical failure if uncured.
Clinical response was determined at end of treatment (EOT, within 24 hours after the last dose). Patients meeting cure criteria were reassessed at first follow-up (21-25 days after randomization) corresponding to the clinical cure assessment (TOC) in the clinical trial to determine if clinical cure was achieved. Those not meeting cure criteria switched to second-line treatment.
Given the lack of evidence on the overall survival impact of clinical cure for HAP/VAP patients, we simply considered the short-term costs and outcomes of both regimens. During the simulation of clinical course, each patient also faced risks of treatment-related adverse events (TRAEs) and in-hospital mortality. Additional medical costs for managing adverse events were considered when Grade 3 or higher severe adverse events occurred. All accumulated costs, life years (LYs), and quality-adjusted life years (QALYs) were calculated over a time horizon or until patient death.
2.1 Treatment Comparisons
The base-case analysis compared two treatment sequences: the first sequence with ceftazidime-avibactam (intervention group) as first-line empirical therapy, followed by second-line therapy based on current conventional clinical practice in China if it failed; the second sequence with meropenem (comparator group) as first-line empirical therapy, also followed by second-line treatment upon failure. The assumed second-line regimen of intravenous polymyxin B combined with high-dose meropenem was applied to both the intervention and comparator groups.
2.2 Model Inputs and Data Sources
Table 1 Baseline distribution and resistance rates of major HAP/VAP pathogens
Pathogens
|
Frequency of baseline pathogensa
|
Resistance rate by pathogensb
|
CAZ-AVI
|
Meropenem
|
Klebsiella pneumoniae
|
32.26%
|
6.5
|
27.1
|
Escherichia coli
|
8.91%
|
4.4
|
1.7
|
Pseudomonas aeruginosa
|
23.17%
|
11.6
|
18.5
|
Proteus mirabilis
|
1.25%
|
4.5
|
1.2
|
Serratia marcescens
|
2.85%
|
10.1
|
5.1
|
Enterobacter cloacae
|
5.53%
|
5.5c
|
6.3c
|
Acinetobacter baumannii
|
26.02%
|
15.9d
|
75.6d
|
Estimated overall resistance rate
|
100%
|
9.96
|
33.37
|
Abbreviations: CAZ-AVI, ceftazidime/avibactam
a Distribution of main bacterial species isolated from 131,097 respiratory specimens (CHINET 2022)
b 2023 resistance data for China (CHINET)
c The data were sourced from the study by Peiyao Jia et al.[22]
d The data were sourced from the study by Jingyuan Xi et al.[23]
The model took antibiotic resistance into account in assessing outcomes. Table 1 summarizes the baseline pathogen distribution and resistance rates to ceftazidime-avibactam and meropenem among major pathogens causing HAP/VAP. The analysis was based on pathogen profiles from 131,097 respiratory isolates reported in the 2022 China Antimicrobial Surveillance Network (CHINET). The input for resistance rates of ceftazidime-avibactam and meropenem was sourced from the 2023 in vitro susceptibility results of CHINET, reflecting real-world levels of risk of resistance among infected patients in China.
2.3 Clinical Inputs
Table 2 summarizes the relevant parameter inputs and data sources used in the model. The criteria for judging treatment efficacy in the model referenced the REPROVE trial design, using cure at end of treatment (EOT) and clinical cure achieved at first follow-up (TOC) as assessment measures.
Table 2 summarizes relevant parameter inputs and data sources used in the model.
inputs
|
value
|
Source
|
Clinical cure, %
|
|
|
CAZ-AVI
|
66
|
REPROVE clinical study data[24]
|
Meropenem
|
74.8
|
REPROVE clinical study data[24]
|
Colistin + high-dose meropenem
|
58
|
Tichy et al [25]
|
AE frequencies, %
|
|
|
CAZ-AVI
|
10.8
|
Table S1
|
Meropenem
|
9.3
|
Table S1
|
Colistin + high-dose meropenem
|
9.3
|
Assumed to be same as meropenem
|
Treatment duration, d
|
|
|
CAZ-AVI
|
6.9
|
QIN et al [26]
|
Meropenem
|
7.3
|
QIN et al [26]
|
Colistin + high-dose meropenem
|
9.5
|
Tichy et al [25]
|
Hospital length of stay, d
|
|
|
With clinical cure
|
16.4
|
REPROVE clinical study data[24]
|
With clinical failure
|
19.1
|
REPROVE clinical study data[24]
|
Percentage of hospitalization days in ICU, %
|
|
With clinical cure
|
43.9
|
REPROVE clinical study data[24]
|
With clinical failure
|
56.02
|
REPROVE clinical study data[24]
|
Daily drug costs (average daily dose)
|
|
|
CAZ-AVI
|
596$
|
Yaozhi Data
|
Meropenem
|
93$
|
Yaozhi Data
|
Colistin + high-dose meropenem
|
225$
|
Yaozhi Data
|
In-hospital death, %
|
|
|
Appropriate empirical treatment
|
14.02
|
WILKE et al [27]
|
Inappropriate empirical therapy
|
26.36
|
WILKE et al [27]
|
Resistant to empirical therapy(CAZ-AVI)
|
5
|
Liu J et al [28]
|
Resistant to empirical therapy(Meropenem)
|
44.6
|
Liu J et al [28]
|
Hospital cost per day General ward
|
83$
|
Shenzhen Basic Medical Service Price Item Catalog
|
ICU
|
854$
|
Shenzhen Basic Medical Service Price Item Catalog
|
SAE cost
|
2195$
|
Guoqiang Liu et al [29]
|
Utility (quality of life)
|
|
|
With clinical response/cure
|
0.87
|
Tichy,SHI et al [25, 30]
|
Without clinical response
|
0.64
|
Tichy,SHI et al [25, 30]
|
Regarding adverse events (AEs), given that only severe AEs (SAEs) would have a substantial impact on associated costs, the model only considered SAEs. The incidence of AEs was derived from a meta-analysis of SAE incidence reported in four clinical trials, results presented in Table S1.[26, 31-33] To simplify the model, the probability of treatment discontinuation due to AEs was set at 0%. Since the REPROVE study did not reflect mortality rates across different pathways in the model, in-hospital mortality rates for HAP/VAP patients were obtained from two other published studies.[27, 28] In-hospital mortality was related to whether patients received appropriate first-line therapy and whether first-line agents developed resistance.
2.4 Economic Inputs
As shown in Table 2, the model utilizes various economic inputs. The treatment duration of ceftazidime-avibactam and meropenem was set with reference to the study by QIN et al. The treatment cycle of second-line therapy referred to the study by Tichy et al. The model assumed that each patient completed the entire treatment process according to the preset treatment duration, unless empirical first-line therapy resulted in resistance or the patient died.
The ICU cost in Table 1 refers to the additional cost of ICU stay, excluding basic ward cost. Total hospitalization cost was calculated as the weighted sum of basic ward cost and additional ICU cost, weighted by the proportion of time patients spent in the ICU. The proportion of time in ICU versus general ward and total hospitalization time were referenced from Tichy et al.’s study, where related data was sourced from cure status at first follow-up in the REPROVE trial.
The model input for SAE costs was a weighted average cost derived from reported costs of different SAEs. Health utilities are a key component for estimating QALYs, which was based on data obtained from published literature, since such data was not captured in the REPROVE study, and no published data on utility for non-responsive HAP/VAP patients was identified. The model input for SAE-related costs was a weighted average calculated from costs reported for different SAE types by Qiang et.al.[29] Health utilities, key for estimating QALYs, were referenced from utility values reported in two published studies.[25, 30]
2.5 Analysis Methods
2.5.1 Base-case Analysis
In the base-case analysis, a 1-year time horizon was chosen to cover relevant events during infection treatment and follow-up. The ceftazidime-avibactam regimen was compared with the meropenem regimen in terms of cure rates, in-hospital mortality, adverse reaction incidence, life years gained, QALYs gained, medication costs, hospitalization costs, adverse reaction management costs, and total costs. The incremental cost-effectiveness ratio (ICER) was calculated. To determine cost-effectiveness, the ICER was compared with China’s willingness-to-pay (WTP) threshold of 3 times per capita GDP.
2.5.2 One-way Sensitivity Analysis
One-way sensitivity analyses were conducted to identify key parameters in the model, including ±20% variation in cost parameters and ±10% variation in probability and utility parameters, while holding all other parameters constant. Results of one-way sensitivity analyses were displayed in tornado diagrams reflecting changes in the ICER.
2.5.3 Probabilistic Sensitivity Analysis
A probabilistic sensitivity analysis (PSA) with 1000 Monte-Carlo simulations was performed to examine the robustness of model outcomes. Medical resource utilization costs were assumed to follow gamma distributions, probability and utility parameters followed beta distributions. Due to lack of precise variability information, the standard error of parameters was set at 10% of the mean. PSA results generated a cost-effectiveness acceptability curve depicting the probability of cost-effectiveness of the two regimens across different WTP thresholds.
2.5.4 Scenario Analysis
Two scenario analyses were conducted to test assumptions in the model. One ignored the impact of resistance. In base-case, we assumed pathogens had resistance against first-line empirical therapy, which may increase in-hospital mortality and hospitalization costs. The other set the efficacy of second-line therapy as 100%, assuming patients switched to appropriate second-line treatment after microbiological testing.