Four health states were defined: GINA Step 3, GINA Step 2, asthma without medications (complete withdrawn of all asthma medications) and any-cause death (absorbing Markov state). GINA Step 3 and GINA Step 2 states were defined as medium and low dose of ICS with salbutamol, respectively. An asthma-related death state was not included in the model due to the low frequency of this event in this level of disease severity. All-cause mortality probabilities were calculated based on vital statistics data of the Colombian National Administrative Department of Statistics (20). A Markov cycle of three months was used in the model as it is the minimum period of time in which asthma medications can be modified according to the clinical response of patients as recommended by the GINA report (18). A decision tree was embedded in the model to include exacerbations as events (Figure 1).
Within the model the following transitions were possible: medication step-down (from GINA Step 3 to GINA Step 2), remission (from GINA Step 2 to asthma without medications) and death (from any health state). The following assumptions were considered: 1) patients start in the GINA Step 3 state and end in the asthma remission state, 2) transition from GINA Step 3 state to the asthma remission state was not allowed, 3) hospitalizations and ED visits were included in GINA Step 3 state, and only ED visits in the GINA Step 2 state, 4) transition probabilities of medication step-down and remission generated by the SCIT+ICS would be constant during the first five years, and a reduction in exacerbations during the complete time horizon, 5) the ICS strategy would generate medication-step-down and remission in the first two years of the time horizon, 7) no adverse events were considered in the model due to their relative low frequency and impact in terms of HCRU, costs and quality of life, 8) an exacerbation would require medical attention through ED or hospitalization.
Our model is based on the assumption that SCIT+ICS therapy generates higher disease remission and reduction of exacerbations compared to ICS without SCIT (21,22). As patients move among health states (i.e. as disease severity is reduced) they accumulate health-state and event-specific costs, and gains in utilities that are translated into QALYs. The negative effect of asthma exacerbations on quality of life was expressed through disutilities.

Inputs and data sources

Transition probabilities

Table 1 shows the parameters used in the decision model. A literature review was conducted to obtain effectiveness parameters and were subsequently validated with clinical experts. The probability of asthma remission for the ICS strategy was defined as the proportion of patients that discontinued all asthma medications in the control group of an observational study by Sánchez et al. in Colombia that included patients with moderate persistent asthma (23). Although authors do not specify the ICS used by patients, we assumed that the majority of patients were treated with Beclomethasone dipropionate (BDP) as it is the only IC covered by the Colombian health care system (24). The probability of medication step-down was obtained from the control arm of an RCT conducted by Zielen et al. in which the steroid-sparing effect of SCIT was evaluated and the proportion of patients achieving a controller medication step reduction was reported (25). This was the only study retrieved in our search that used the reduction in medication steps as an outcome to evaluate the steroid-sparing effect of SCIT + ICS and ICS. As authors used fluticasone propionate as the controller treatment, we assumed similar effectiveness between BDP and fluticasone propionate (26). Probabilities of medication step-down and remission for the SCIT+ICS cohort were obtained from the study by Sánchez et al in which a sample of 122 patients received SCIT + ICS over a 3-year period and the proportion of patients with a reduction/complete withdrawn of asthma medications was reported (23). All initial probabilities were converted as rates using their periodicity and then re-expressed as probabilities with a 3-month interval (cycle length) (27)