Cost effectiveness analysis of composite endpoints
Source:    Publish Time: 2012-08-11 07:46   4011 Views   Size:  16px  14px  12px
Author: Xuanqian Xie In general, there are two approaches to model the composite endpoints, the major event, or all eve

Author: Xuanqian Xie

In general, there are two approaches to model the composite endpoints, the major event, or all events.

The first example is based on a HTA project of EZ-IO infusion system (Intraosseous (IO) infusion is the injection directly into the marrow, which generally used when vascular access cannot be achieved on time) (Xie et al, 2008).

A decision analytic model was applied by cost-effectiveness analysis with a theoretical cohort of 1,000 patients (Figure 1). If intravenous (IV) access cannot be established on time, except IO, another alternative is Central Venous Access (CVA), which is the usual practice in XX Health Centre now, particular for adult patient. We compare using IO as the first alternative with using CVA as the first alternative when IV access cannot be achieved on time. In EZ-IO strategy, we assumed that if IO cannot be accessed in 10 minutes, CVA would be successfully replaced. In the other arm, similar as those, IO would be replaced if CVA does not work. We note that the IO diffusion can be used within 24 hours(3;5). Consequently, if survivals still need fluid infusions one day after IO access, they have to switch to CVA or IV access. Patients’ outcomes include well, minor complications, main complications and death. Animal studies show that there are no significant differences in physical responses (blood pressure, heart rate) among central intravenous, peripheral intravenous, and IO routes(2). So, we assumed that IO and CVA strategies do not impact the ultimate clinical result, the mortality rate. Complications directly caused by EZ-IO or CVA are measured in this study. The main consequences considered are pneumothorax, osteomyelitis, other severe non-fatal infections and fracture(2;10-12). Costs of minor complications are counted in our analysis, while the clinical consequences of them are not included.

Branch probabilities and resource costs of this model are based on best available evidence (Table 1). Branch probabilities of EZ-IO are described in “efficacy” and “safety” in results part. Branch probabilities of CVA are mainly based on Hamilton et al. review in Cochrane Database(12) and Ganeshan et al. review(11). We assumed that CVAs are inserted using ultrasound-guidance and possible CVA sites are femoral, subclavian et al. Minor complications of CVA include minor infections, minor bleeding, fibrin sheath formation et al. The mortality rate of target population and the proportion of survivals using CVA after IO treatment are derived from estimates of experts. If patients use both IO and CVA, they will be susceptible to complications of both. The potential complications caused by IV access do not include in this model.


The economic evaluation was performed from the point of view of XXXX, and only the direct health care costs are collected. All costs are expressed as 2007 Canadian Dollar(13;14). Given diverse conditions of patients, health care costs of patients would vary hugely. For simplicity, we assume that the base cost is 20,000 CAD per patient who received emergency services whatever finally they were death or alive. The costs of CVA are included by the base cost, while if patients use EZ-IO, they must pay extra 150 CAD per procedure for purchasing devices (4). When patients suffer main complications or minor complication, they would spend extra resource costs. We assume that all complications occur within one year after IO/CVA infusions.

The main outcomes include incremental cost per main complication averted (ICCA), expected costs and main complication cases of two strategies. Univariate sensitivity analysis and multi-way sensitivity analysis were performed to determine the main factors influencing ICCA. All Data were analyzed using Treeage pro Suite 2007 and Excel 2003.

The major results can be found below.


The 2nd example is based on a project which was not pursued for publication. My role in this project included modelling and computation. All mortality, non-fatal MI and non-fatal stroke were defined as the composite clinical outcomes. See model and results below. (One of other approaches is to model the life-year/QALY using Markov model, which will be discussed in other examples later.) 



Xie X, Brophy JM. EZ-IO for intraosseous access. Montreal (Canada): Technology Assessment Unit (TAU) of the McGill University Health Centre (MUHC); 2008. (Informal brief report). Available from: