The decision problem facing health care commissioners in this context in the UK or internationally is whether to invest in and provide to PAD patients FlowOx™ therapy in order to help manage their condition. Our objective was to use an economic model to calculate the relative cost-effectiveness of the FlowOx™ therapy as part of a pathway of care as compared with standard NHS care (without the FlowOx™ therapy) in a hypothetical cohort of patients with mild to severe PAD. Although FlowOxTM therapy is intended for use by patients who present with reduced blood flow (ABPI <0.8) or symptoms of claudication in advance of surgical intervention, the device may also provide some preventative options, thus we conducted a cost analysis for all disease states in order to evaluate the most appropriate and cost-effective stage at which to begin therapy.

In constructing our model, we opted for a Markov model, with three FlowOx™ therapy scenarios:

The FlowOx™ therapy plus nominal care (S0)

The FlowOx™ therapy only (S1)

The FlowOx™ therapy plus standard care (S2).

We chose the base case analysis as FlowOx™ therapy plus nominal care, and standard care as the comparator. We chose FlowOx™ therapy plus nominal care as the base case to examine the impact of substituting FlowOx™ therapy for standard care. The inclusion of nominal care was to acknowledge that patients would still need regular monitoring and clinical input. We defined nominal care is a reduced standard care, comprising fewer nurse visits and reduced medication use. Costs associated with different care options can be found in Table A of S1 Appendix; while costs associated with different health states can be found in Tables B and E of S1 Appendix. See Table 1 for an overview of study intervention scenarios.

Our model was designed to mimic the progression of the PAD to lower limb ulcers and CLI. To allow for this flexibility, a Markov model was developed. Each state in the Markov model represents a plausible, discrete health state. We used transition probabilities to describe the causal relationships and the probability of a patient moving from one state to another. Sources of data used in the model are shown in Tables A, B and E in S1 Appendix.

Markov modelling allows complex real-life events to be represented in a simplified health state form [24]. In our Markov model we followed a cohort of 1000 patients moving through defined Markov states and time periods. The modelled patients can remain in their current health state, move to another health state or reach the ‘absorbed’ state (a health state with no cost or health benefit e.g. dead in this model) according to certain transition probabilities [25]. The Markov model had finite and independent states with each state being assigned a health utility and cost value to allow cost per QALY estimates to be calculated. Parameters such as clinical outcomes, healthcare costs incurred, time duration of the intervention and quantity of intervention were included in the model construction.

Our aim was to create a realistic but simplified model, which took account of the complexity of the issue, whilst also allowing ease of modelling. The challenge of modelling the cost-effectiveness of diseases such as PAD is that they have asymptotic and symptomatic levels, and each level has sub-levels. In clinical practice, the Rutherford and Fontaine classification are used to distinguish between these different levels [26,27].

Our Markov model states, and transition probabilities describe, to the best of our ability, the biological progression of disease severity of PAD, LLI, lower limb ulcers and CLI. The Fontaine and Rutherford classifications, as shown in Table 2, were applied in the criteria and boundary requirements for the Markov state classification. We used these classifications to investigate progression through health states (see Fig 2 for a pictorial representation of the model structure). The model consists of seven-disease states: mild, progressive, severe, minor amputation, major amputation, recovered and a dead state (which acts as the absorbed state). The recovered state is dependent on whether the patient has undergone at least one amputation or not; this state has a health index less than the ‘mild’ state due to the presence of disability. See Table C in S1 Appendix for utilities for each health state. The structure of the Markov model includes the amputation states as different severity stages due to the impact on patient outcomes [2830].

*Ankle Brachial (Pressure) Index (ABPI/ABI): The ratio of blood pressure at the ankle to the pressure at the upper arm [31].

For a robust and valid comparison of the cost-effectiveness of FlowOx™ therapy, the cycle length in this Markov model was three months, this was constructed to match one complete dosage of FlowOx™ therapy. The cycle length was intended to be short enough to capture patient transition and long enough to include all health benefits observed.

A time horizon of five years was adopted for this health economic model analysis to allow for ample time to reflect the trajectory towards costs and outcomes between patients receiving FlowOx™ therapy and patients receiving standard care. The time horizon allows time for many of the patients in more severe stages of PAD, for example patients experiencing CLI, to progress through the model structure with rates of major amputation within the first year estimated to be up to 40% [2] and mortality estimated as high as 35% within one year and 50% within five years [8].

The two FlowOx™ therapy trials described earlier had a follow up of three months, the observed probabilities from the clinical follow up and published evidence was used to populate the transition probabilities. Where there was insufficient data, point estimates were derived from published literature [3,32]. The relative risk and effects of disease progression were derived from the study data and are presented with the transition probabilities in Table D in S1 Appendix.

The model assumed all participants were at the symptomatic phase of PAD (progressive state) at the beginning of the model and then progressed through the model using the given transition probabilities. Cost and health outcomes were computed from the distributed simulation.

The sample sizes of the primary data sources were limited, hence point estimates from similar studies were combined to give a more robust model. Cost data associated with PAD have a wide variance, which has a direct impact on average cost estimates. Costs were based on the average of unit cost in the UK NHS, for 2017/18, and were expressed in pounds Sterling (£GBP). The cost estimates for the Markov states and patient cohort were pooled from the national schedule of NHS reference costs [33] and other published literature [11,34,35]. These costs were inflated in accordance to the Hospital and Community Health Services pay and price index (HCHS) [36]. Resource use costs included community, inpatient and outpatient costs for healthcare service use. The range of unit costs used for each model state are presented in Table E in S1 Appendix. An NHS perspective was adopted in this study, as there was insufficient evidence to allow a wider, societal perspective. The Markov state costs by level of severity are summarised in Table 3.

Note: Costs are reported in 2017/2018 financial year and were inflated in the computation to 2018/19. Recover state does not have PAD treatment costs.

These annual costs were rescaled to match the cycle length (i.e. three months) of the model (see Table E in S1 Appendix). Where these costs had different prices, the mean was adopted and these costs were discussed and agreed upon by expert advisors. Expert advisors comprised a vascular surgeon and nurses working within secondary care vascular medicine. This model focused on FlowOx™ therapy using a rental-based pricing structure, at a cost of £15 per day to NHS services, as advised by the manufacturer. No overheads were added to the FlowOx™ therapy due to the predominant focus on patient-led home use.

Note: The content above has been extracted from a research article, so it may not display correctly.

Please log in to submit your questions online.
Your question will be posted on the Bio-101 website. We will send your questions to the authors of this protocol and Bio-protocol community members who are experienced with this method. you will be informed using the email address associated with your Bio-protocol account.

We use cookies on this site to enhance your user experience. By using our website, you are agreeing to allow the storage of cookies on your computer.