3.1 Patient Population
The base case economic model generated results for seriously ill patients with a confirmed CR infection in which cefiderocol is expected to be indicated (described as ‘confirmed CR infection’ henceforth). A scenario analysis was also run for a separate population of seriously ill patients who were judged to be at high risk of a carbapenem-resistant infection (described as ‘suspected CR infection’ henceforth). The modelled patient population is consistent with the criteria for prescribing cefiderocol indicated by the Italian Medicines Agency in the health care context: “Treatment of adult patients hospitalized with infections caused by carbapenem-resistant Gram-negatives in whom there are limited treatment options or with invasive infections of strongly suspected aetiology due to carbapenem-resistant Gram-negative bacteria.” [11].
Patients could have experienced one of the three following infection types: complicated urinary tract infection (cUTI), pneumonia or bloodstream infection (BSI)/sepsis. Additionally, patients were limited to those whose infections were caused by the following pathogen types: Acinetobacter baumannii, Pseudomonas aeruginosa, Enterobacteriaceae and Stenotrophomonas maltophilia. The analysis was based on a weighted average of infection and pathogen types, as described further in Section 3.4.
3.2 Model Structure
A decision tree was developed in Microsoft Excel to estimate the cost-effectiveness of cefiderocol compared with colistin and colistin based regimens and ceftazidime/avibactam for the treatment of seriously ill patients with a confirmed or suspected CR infection respectively (Fig. 1). Only a small section of the decision tree (as outlined in red in Fig. 1 and described further within the following section) was relevant to the confirmed CR population, which was used in the base case analysis. With exception of the amendments described below, and the differing treatment effectiveness (as explained within Section 3.3.1), the modelling approach and data sources remained equivalent between the two populations.
Patients with a suspected CR infection should receive a microbiological test to determine the pathogen strain causing the underlying infection (e.g. Enterobacterales) and whether the pathogen was carbapenem-resistant or susceptible. Healthcare professionals use the results of the test to inform the optimal treatment that patients should receive. Within the base case analysis, a hypothetical cohort of patients entered the model at the point in which the microbiological test had confirmed that the infection was caused by a CR resistant pathogen. Patients received either cefiderocol or colistin upon entry to the model and remained in the decision until they experienced one of two possible endpoints: sustained response (cure) or death.
All patients who experienced a sustained clinical response (i.e. no relapse), were classified as cured and it was assumed no more clinical events were incurred by these patients. The probability of sustained response was based upon clinical cure rates obtained from cefiderocol and comparator clinical trials (as described further in Section 3.3.1). Patients that did not experience a sustained response could receive salvage therapy (the next optimal intervention based on the pathogen group identified), of which they could respond to, or die. It was assumed that patients in Italy would receive colistin as salvage therapy within both treatment arms of the model.
Due to a lack of data to quantify the reduction in efficacy for patients with prior exposure to treatment, it was assumed that treatment efficacy was equivalent if administered as a first or second-line treatment option. Secondly, to avoid overcomplicating the model it was assumed that patients could only receive two rounds of treatment within the model time horizon. Therefore, any patients who did not have a sustained response to salvage therapy were assumed to die - this assumption was considered to be acceptable since these patients are severely ill upon entry to the model.
Whilst a short time horizon (i.e. less than one year) was considered sufficient to capture the costs and health-related quality of life impact of infection-related events, a lifetime time horizon was used to capture the long-term impact for those that were cured. All patients that died during the initial decision tree element of the model were assumed to incur no further quality-adjusted life years (QALYs) nor costs. Due to a lack of data to inform future events, it was assumed that all patients with a sustained response would not experience a relapse of their original infection and, therefore, no long-term costs were included in the model. Therefore, these patients were expected to have the same long-term morbidity and mortality as the Italian general population and the long-term QALYs accumulated by all patients with a sustained response were estimated using background mortality rates and age-adjusted utility values for the Italian general population.
3.2.1 Scenario Analysis: Suspected CR Infection
It takes approximately three days for the results of the aforementioned microbiological test to become available. To minimise the likelihood of mortality, an antimicrobial agent must be administered at the first sign of infection. A health care professional can determine whether an optimal treatment was administered, after three days, following identification of the pathogen strain. Therefore, a scenario analysis was run with a hypothetical cohort of suspected CR infection patients within a decision that was expanded to capture this short period in which the pathogen status tests are completed. Within this scenario, all suspected CR infection patients received a microbiological test upon entry to the model and were prescribed either cefiderocol or ceftazidime/avibactam whilst awaiting the test results. More details on the suspected cohort are provided below.
Appropriate Treatment
Firstly, it was understood that a proportion of patients with a suspected CR infection will be carbapenem-susceptible (CS). Therefore, the first chance node in the decision tree separated patients based on the confirmed susceptibility status.
Patients were separated into two distinct groups following the diagnosis of a confirmed CR infection through a microbiological test: appropriate treatment and inappropriate treatment. The susceptibility of the infection pathogen strain was used to determine whether the treatment administered initially was appropriate or inappropriate. For example, if an Enterobacterales infection was diagnosed, and this had a 76.1% susceptibility to cefiderocol, then 76.1% of patients would enter the ‘appropriate’ pathway. An infection was considered susceptible to a treatment if the patient experienced an observed clinical response (i.e. the intervention was effective).
It is assumed that all microbiological tests were 100% accurate (the tests had a sensitivity and specificity of 100%) to avoid over-complication of the model structure because the accuracy of microbiological tests used within Italian clinical practice is not homogenous. Furthermore, insufficient data were available to model changes in patient outcomes due to different tests.
It was possible for any patients identified as carbapenem susceptible following microbiological testing to be switched onto a cheaper antimicrobial agent (assumed to be meropenem) without any loss in efficacy (also known as treatment de-escalation). Not all patients received treatment de-escalation because doctors may be reluctant to alter the treatment regimen of a critically ill patient. Therefore, patients were separated into two separate branches following confirmation of appropriate treatment: treatment de-escalation and treatment maintenance. The clinical outcomes were assumed to be equivalent across these two branches and, therefore, the efficacy levels for the original intervention were still applied. However, the treatment costs associated with the cheaper antimicrobial agent were applied to patients within the treatment de-escalation branch.
Inappropriate Treatment
Evidence from the wider literature indicates that there are several unintended consequences following the administration of an inappropriate intervention. This includes a higher rate of mortality and a prolonged stay in hospital [12–14]. Therefore, it was assumed that patients receiving inappropriate treatment would not experience any form of clinical response leading to an increased risk of death. Patients who did not die, but showed no signs of response, received salvage treatment with colistin. All patients who experienced a sustained clinical response (i.e. no relapse), were classified as cured and it was assumed no more clinical events were incurred by these patients. The probability of sustained response was based upon clinical cure rates obtained from cefiderocol and comparator clinical trials. Patients that did not experience a sustained clinical response were able to receive a second round of salvage treatment with colistin of which they could respond to, or die.
3.3 Data Sources and Model Parameters
3.3.1 Clinical Efficacy
All clinical efficacy inputs used to inform the movement of patients throughout the decision tree in the base case analysis are presented in Table 1. The probability of sustained response associated with cefiderocol and colistin within the confirmed CR population was obtained from the clinical cure at the end of treatment within the CREDIBLE Phase III randomised controlled trial [15]. The time to initial response (as initial and salvage therapy) and time to death associated with cefiderocol and colistin were identified from relevant clinical trials (further information is presented within the supplementary appendix (Table S.1)). These treatment-dependent inputs were used to inform the period over which patients experienced a reduction in quality of life due to their infection. The efficacy and susceptibility data used to inform the movement of patients through the decision tree within the suspected CR population are presented in Supplementary Appendix A (Table S.2).
3.3.2 Adverse events
Safety data related to all treatments within the model were examined to identify serious adverse events that occurred in ≥ 3% of patients. The two adverse events which met this criterion are particularly prominent within the CR infection population, clostridium difficile infection and renal impairment, and were included in the model to capture their impact on overall costs and patient quality of life. The probability of a patient experiencing each adverse event is presented in Table 1.
3.3.3 Mortality
As aforementioned, it was assumed that all patients with a sustained response had no further events related to their original infection. Therefore, an all-cause mortality risk, sourced from Italian age-related population norms, was applied to all patients that achieved a sustained response [16].
Table 1
Clinical parameters and safety inputs used in model
Infection type
|
Distribution of infections
|
Distribution of pathogens
|
Source
|
Enterobacterales
|
Acinetobacter Baumanni
|
Stenotrophomonas maltophilia
|
Pseudomonas aeruginosa
|
cUTI
|
28.4%
|
83.5%
|
3.6%
|
0.0%
|
12.9%
|
[17]
|
Pneumonia
|
32.0%
|
37.5%
|
14.0%
|
12.5%
|
36.0%
|
BSI/sepsis
|
39.7%
|
76.4%
|
7.9%
|
2.7%
|
13.0%
|
Infection type
|
Treatment
|
Probability of sustained response from initial treatment in confirmed CR population and salvage therapy in suspected CR population
|
Source
|
cUTI
|
Cefiderocol
|
76.5%
|
[15]
|
Colistin
|
60.0%
|
Pneumonia
|
Cefiderocol
|
60.0%
|
Colistin
|
63.2%
|
BSI/sepsis
|
Cefiderocol
|
69.6%
|
Colistin
|
50.0%
|
Weighted infection
|
Cefiderocol
|
68.5%
|
Colistin
|
57.0%
|
Treatment
|
Probability of an adverse event
|
Renal impairment
|
Clostridium difficile infection
|
Source
|
Cefiderocol
|
0.7%
|
0.2%
|
Data on file
|
Colistin
|
27.7%
|
0.0%
|
[18, 19]
|
BSI: Bloodstream infection; CR: carbapenem-resistant; cUTI: Complicated urinary tract infection |
3.3.4 Utility
Infection-specific utilities were applied during the decision tree element of the model to capture the negative impact of each infection on patient quality of life (as presented in Table 2). Patients who died due to the infection were assigned a utility of zero for the remainder of the model time horizon. Patients experiencing a sustained response were assumed to have the same long-term morbidity as the general population and population norms were used to estimate the long-term QALYs accumulated [20]. The disutilities incurred by patients following each adverse event are also presented in Table 2.
Table 2
Cost and utility inputs used in model
Treatment
|
Unit cost
|
Units per pack
|
Units per day
|
Cost per day
|
Source
|
Cefiderocol
|
€1,500.00
|
10
|
6
|
€900.00
|
[11]
|
Colistin
|
€10.80
|
1
|
9
|
€71.93
|
[21]
|
Infection
|
Hospital cost per day
|
Length of stay (days)
|
Cost per hospital stay
|
Source
|
cUTI
|
€443.29
|
6.9
|
€3,058.70
|
SDO 2019 [22]
|
Pneumonia
|
€477.29
|
8.5
|
€4,056.97
|
BSI/sepsis
|
€556.33
|
13.3
|
€7,399.19
|
Weighted infection
|
€499.99
|
9.45
|
€4,964.84
|
Calculation
|
Adverse event
|
Overall treatment cost
|
Source
|
Renal impairment
|
€3,734
|
[23]
|
Clostridium difficile infection
|
€10,224
|
[24]
|
Infection
|
Utility value
|
Source
|
cUTI
|
0.782
|
[25]
|
Pneumonia (hospitalisation)
|
0.730
|
[26]
|
Pneumonia (post-hospitalisation)
|
Home (recovered)
|
44%
|
0.840
|
Long-term care
|
56%
|
0.650
|
Weighted average
|
0.728
|
BSI/sepsis
|
0.530
|
[27]
|
Adverse event
|
Disutility value
|
Length of time applied
|
Source
|
Renal impairment
|
-0.263
|
15 days
|
[28, 29]
|
Clostridium difficile infection
|
-0.360
|
10 days
|
[30]
|
BSI: Bloodstream infection; cUTI: Complicated urinary tract infection |
3.3.5 Unit costs and Resource use
Three types of costs were captured in the model: treatment acquisition, hospitalisation and adverse event costs (as presented in Table 2). The unit costs for all interventions were based upon simulations of real prices which take into account confidential discounts applied to hospitals in Italy [21]. The unit cost per day associated with hospitalisation and average length per stay associated with patients that received appropriate treatment for their infection were obtained from the NISAN database [22]. The unit costs associated with the treatment of renal impairment and Clostridium difficile infection were also obtained from published literature [23, 24]. Further cost inputs are presented in Supplementary Appendix A (Table S.3).