The modelled economic evaluation was undertaken using two paralleled decision-analytic Markov models: comparing SAMe with no treatment (placebo in clinical trials) and UDCA with no treatment. (Figure 1). The utilisation of parallel models stems from the lack of clinical trial data directly comparing SAMe with UDCA. The two modelled outcomes were subsequentially compared to generate the final difference in ICER between SAMe and UDCA. The Markov models involved an initial decision of starting treatment with SAMe, UDCA, or placebo. Patients then transitioned through different health states. Each model cycle represented 2 weeks over a 2-year time horizon and used a discount rate of 3%. Sensitivity analysis was applied to the time horizon of ± 1 year and to the discount rate of ± 2%.
The Markov model has three health states (Figure 2): moderate/severe cholestatic disease (i.e., elevated liver function tests), mild cholestatic disease (i.e., normal liver function tests), and death.
The cost-effectiveness of the intervention was described in terms of improvements in health outcomes and costs associated with the interventions. The net cost per unit of health outcome benefit gained from the intervention is the incremental cost-effectiveness ratio (ICER), which is the main outcome of interest in cost-utility analysis. For example, the ICER of SAMe compared with placebo is defined as:
To estimate QALYs lived, the years of life lived within any health state are multiplied by the utility value relevant to that health state. To estimate costs, annual costs are applied to the relevant years of life lived within health states, similar to the application of utility weights. The formulas implicitly assume that any transition into and out of that health state occurs halfway through any cycle. The same model method was applied to UDCA compared with placebo. The two ICERs were then compared. The intervention is considered cost-effective where the ICER falls below the decision maker’s maximum willingness to pay for the health benefits, also referred to as the decision threshold (8). There is no established ICER threshold in China, and therefore, this study uses an ICER threshold set to 3 x GDP/capita for China, which was CNY 270,000 in 2022.
A broader Chinese societal perspective is taken in this cost-effectiveness analysis.
Costs
The first intervention that is proposed for the treatment of moderate intrahepatic cholestasis is SAMe. SAMe is given once a day at a dose of 1500- 2000 mg, or 3 tablets [11] . A pack of 10 tablets (500mg per tablet) of SAMe costs CNY 171.90 (lowest bidding price, PVBP excluded), which leads to 5 packs of consumption every 2 weeks with the cost of 5 × 171.9 = CNY 859.5.
The second intervention is UDCA. UDCA is given once a day at a dose of 750 mg or 3 tablets. A pack of 25 tablets (250 mg per tablet) cost CNY 210.2 (lowest bidding price, PVBP excluded), which lead to 2 packs of consumption every 2 weeks with the cost of 2 × 210.2 = CNY 420.4.
Since there is no peer-reviewed literature to indicate the usage of healthcare resources by patients diagnosed with intrahepatic cholestasis in China, an expert panel of physicians from China advised the frequency of visiting clinics, hospitals, and specialists for patients who experienced IHC. From the expert panel’s feedback, patients who experienced moderate IHC see a doctor 6 times a year and have blood tested 5 times a year on average. A small group of patients (0.08%) also need to visit specialists 6 times a year. The cost of visiting a doctor is advised to be CNY 830 per time. The cost of visiting a specialist is advised to be an extra CNY 2171 per time on average. The cost of a blood test is CNY 50 per test. Overall, the healthcare cost for patients with severe/moderate IHC is assumed to be (CNY830*6) + (CNY2171*6*0.08%) + (CNY50*5) = CNY 5241.4 per year. Sensitivity analysis varying the cost +/- 10% and a scenario analysis, where the cost of a visit to a doctor is set to zero, were performed.
Efficacy
The effectiveness of the treatment is categorised based on SAMe clinical trials targeting various types of liver diseases, including alcoholic liver disease, non-alcoholic fatty liver disease, non-alcoholic steatohepatitis, drug-induced liver impairment, and intrahepatic cholestasis of pregnancy.
ALP was selected as the biomarker to evaluate IHC severity in the Markov models[12, 13]. The percentage change in ALP for treatment versus placebo derived from three studies was 28.9% versus 7.6% for SAMe and 6.0% versus 5.4% for UDCA, summarised (Table 1). Studies were weighted where necessary using inverse variances.
Table 1. Percentage change in ALP for the indirect comparison of SAMe versus UDCA with placebo.
|
Treatment
|
Baseline ALP;
± change (U/L)
|
% change
|
Study weight
|
Weighted % change
|
Frezza (1990) [14]
|
SAMe
|
268; -77
|
28.9%
|
100%
|
28.9%
|
Placebo
|
284; -22
|
7.6%
|
100%
|
7.6%
|
Summary, total % change
|
SAMe
|
28.9%
|
Placebo
|
7.6%
|
Leuschner (2010) [15]
|
UDCA
|
165; -11
|
6.4%
|
68%
|
4.4%
|
Placebo
|
173; -9
|
5.5%
|
68%
|
3.7%
|
Lindor (2004) [16]
|
UDCA
|
155; -8
|
5.2%
|
32%
|
1.7%
|
Placebo
|
154; -8
|
5.3%
|
32%
|
1.7%
|
Summary, total % change
|
UDCA
|
6.0%
|
Placebo
|
5.4%
|
The diagnostic criteria for cholestatic liver diseases have historically been ALP levels higher than 1.5 × the upper limit of normal (ULN) [1]. This was presented in the European Association for the Study of the Liver guideline for the management of cholestatic liver diseases in 2009 [17], and subsequently endorsed by the Chinese Association of Hepatology in 2015 [1]. The normal ALP range for adults is 30 to 120 IU/L [18]. This corresponds to an ALP level of ≥180 IU/L for IHC patients. A baseline ALP level of IU was calculated from available literature (Table 2).
Table 2. Baseline ALP values of IHC patients at study commencement.
First author (year)
|
Baseline ALP (IU/L)
|
N
|
Choudhuri (2014)[19]
|
276.5
|
67
|
Fiorelli (1999) intramuscular group [20]
|
259.5
|
276
|
Fiorelli (1999) intravenous group [20]
|
276.5
|
253
|
Frezza (1990) treatment group [14]
|
270.0
|
110
|
Frezza (1990) comparison group [14]
|
282.0
|
110
|
Ivashkin (2018) [21]
|
241.2
|
72
|
Leuschner (2010) treatment group [15]
|
165.0
|
95
|
Leuschner (2010) comparison group [15]
|
173.0
|
91
|
Lindor (2004) treatment group [16]
|
154.7
|
79
|
Lindor (2004) comparison group [16]
|
154.4
|
86
|
Manzillo (1992) treatment group [22]
|
289.2
|
99
|
Manzillo (1992) comparison group [22]
|
295.8
|
85
|
Qin (2000) treatment group [23]
|
195.5
|
15
|
Qin (2000) comparison group [23]
|
190.6
|
15
|
Virukalpattigopalratnam (2013) [24]
|
230.6
|
96
|
SUM (ALP × N)
|
354,672
|
SUM (N)
|
1,549
|
Summary, baseline ALP (IU/L)
|
229
|
A time horizon of two years was chosen on the basis that UDCA is a long-term therapy, requiring many months of administration for treatment effectiveness [25].
The proportion of IHC patients reporting the symptoms of fatigue, pruritus and jaundice was 79.7%, 49.5% and 47.0%, respectively at study commencement (considered IHC onset with severe/moderate disease severity) and 33.3%, 11.4% and 24.4% respectively at study conclusion (considered mild disease severity) with study weights calculated using inverse variances in RevMan 5.4 (Table 3).
Table 3. Percentage of IHC patients presenting with symptoms fatigue, pruritus and jaundice.
|
% reporting (n/N)
|
Study weight
|
Weighted % reporting
|
First author (year)
|
Study start
|
Study end
|
Study start
|
Study end
|
FATIGUE
|
Choudhuri (2014) [19]
|
80% (195/243)
|
38% (92/243)
|
25.2%
|
20.2%
|
9.5%
|
Ivashkin (2018) [21]
|
82% (59/72)
|
51% (37/72)
|
24.6%
|
20.2%
|
12.6%
|
Perlamutrov (2014) [26]
|
81% (85/105)
|
11% (12/105)
|
24.9%
|
20.2%
|
2.8%
|
Virukalpattigopalratnam (2013) [24]
|
76% (184/243)
|
33% (79/243)
|
25.3%
|
19.2%
|
8.2%
|
Summary, total % reporting fatigue
|
|
|
|
79.7%
|
33.3%
|
PRURITUS
|
Choudhuri (2014) [19]
|
36% (88/243)
|
10% (24/243)
|
20.1%
|
7.3%
|
2.0%
|
Ivashkin (2018) [21]
|
54% (39/72)
|
11% (8/72)
|
19.8%
|
10.7%
|
2.2%
|
Perlamutrov (2014) [26]
|
81% (85/105)
|
7% (7/105)
|
19.8%
|
16.0%
|
1.3%
|
Virukalpattigopalratnam (2013) [24]
|
39% (95/243)
|
17% (42/243)
|
20.1%
|
7.9%
|
3.5%
|
Manzillo (1992) [22]
|
38% (129/343)
|
12% (42/343)
|
20.1%
|
7.6%
|
2.5%
|
Summary, total % reporting pruritus
|
|
|
|
49.5%
|
11.4%
|
JAUNDICE
|
Choudhuri (2014) [19]
|
81% (197/243)
|
56% (136/243)
|
25.1%
|
20.3%
|
14.0%
|
Ivashkin (2018) [21]
|
39% (28/72)
|
17.5% (13/72)
|
24.8%
|
9.6%
|
4.5%
|
Perlamutrov (2014) [26]
|
20% (21/105)
|
2% (2/105)
|
24.8%
|
5.0%
|
0.5%
|
Virukalpattigopalratnam (2013) [24]
|
48% (116/243)
|
21% (51/243)
|
25.2%
|
12.0%
|
5.4%
|
Summary, total % reporting jaundice
|
|
|
|
47.0%
|
24.4%
|
Quality of life
The quality of life (QoL) used in this model is determined by the primary IHC score and the comorbidities, including jaundice, pruritus, and fatigue, in two different health states. The mean Short Form 6 Dimension (SF-6D) utility score of chronic liver patients is 0.79 [27]. Patients in severe/moderate state have a QoL score of 0.68. Patients with pruritus have the SF-6D QoL score of 0.613-0.706 = -0.093; patients with fatigue have the following utility score: 0.576-0.732 = -0.156 [28]; and patents with jaundice have a utility score of 0.63-0.67 = -0.04 [29].
Total QoL utility is calculated as:
Primary IHC + Jaundice × percentage of Jaundice + Pruritus × percentage of pruritus + Fatigue × percentage of fatigue.
Patients in different treatment groups will have different percentages of comorbidities (Table 4).
Table 4: Quality of life input table.
|
QoL
|
Utility by severity
|
Mild IHC
|
0.79
|
Severe/moderate IHC
|
0.68
|
Pruritus
|
|
IHC without pruritus
|
0.706
|
IHC with pruritus
|
0.613
|
Reduction by pruritus
|
-0.093
|
Fatigue
|
|
IHC without fatigue
|
0.732
|
IHC with fatigue
|
0.576
|
Reduction by fatigue
|
-0.156
|
Jaundice
|
|
IHC without jaundice
|
0.67
|
IHC with jaundice
|
0.63
|
Reduction by jaundice
|
-0.04
|
Sensitivity analysis
One-way and probabilistic sensitivity analyses were performed by varying the input by +/- 10% for cost and efficacy variables. The discount rate was examined by varying it to 1% and 5%, and the time horizon was varied from 1 year to 3 years. Probabilistic sensitivity analysis was performed using @Risk by Palisade.
Microsoft Excel (Microsoft 365 MSO version 2304, Microsoft Corporation) on the Windows 10 platform was used for the economic evaluation, and RevMan 5.4 was used for the meta-analysis.