The prevalence of AEs in patients taking OCS is widespread; studies report that up to 93% of patients using long-term OCS experience at least one condition linked to their use [3, 13, 39]. OCS continue to be used globally despite the well-demonstrated associations with severe adverse effects. This is in part due to their widespread availability and affordability, making these treatments particularly favoured in low- and middle-income countries [40,41,42]. However, in the long-term, the AEs associated with OCS treatment lead to significant healthcare costs that outweigh the initial lower treatment costs [43, 44]. As such, reducing the need for OCS treatment and offering alternative anti-inflammatories is vital. Current HTAs do not fully capture the potential benefits of OCS sparing, and therefore risk undervaluing new therapies [23]. In this study, we describe a modelling approach that allows more accurate predictions of the value of emerging treatments with OCS-sparing capabilities, which is crucial to inform future decision making. Using this model, we demonstrate that avoidance of OCS and OCS-related AEs led to improvements in patient quality of life and reductions in AE-related costs.

In the model base-case, assuming a patient population taking an average dose of 10 mg of OCS per day and an OCS-sparing agent capable of sparing 50% of that dose resulted in lifetime incremental cost savings of £2403 (£2203–£2668) and 0.071 (0.064–0.077) QALYs gained. These cost savings increased significantly when considering a population with an average daily OCS dose of 15 mg at baseline, with cost savings of £10,627 (deterministic) or £19,501 (−£51,836 to −£748, probabilistic) and QALY gains of 0.116 (deterministic) or 0.356 (−0.022 to 2.404, probabilistic) QALYs. Therefore, the introduction of an OCS-sparing treatment has the potential to reduce the number of transient and chronic AEs a patient experiences, resulting in cost savings and QALY and life-year gains. While these estimated gains may appear small, they would be incremental to the additional effects of a new treatment and would otherwise have been missed using previous modelling approaches. Therefore, they should be considered in the context of the benefits of the treatment itself. Likewise, considering all disease indications where OCS treatment is approved, almost 1% of adults in the UK are prescribed OCS, and therefore small individual cost savings have the potential to accumulate into substantial cost savings for the NHS when considering the large patient population they affect [25, 45].

Our results estimate significant additional long-term costs and QALYs incurred due to OCS exposure resulting from OCS-related AEs, especially for patients taking higher doses. This is particularly impactful for conditions such as rheumatoid arthritis or ulcerative colitis, where higher average daily doses of over 7.5 mg/day and up to 40 mg/day, respectively, are commonly administered to control symptoms [46, 47].

The cost savings modelled here are aligned with similar modelling studies considering the costs of OCS treatment. In the UK, healthcare costs for patients with asthma have been demonstrated to increase according to OCS exposure, and patients receiving maintenance OCS treatment are estimated to accrue 39% and 51% higher healthcare costs for patients with mild-moderate and severe asthma, respectively, compared with patients not receiving OCS [43, 48]. In these studies, patients receiving maintenance OCS had increased prescription costs as well as increased healthcare utilisation costs [43, 48]. This translated to an additional annual OCS-related cost per year of £224 for mild asthma and £1310 for severe asthma [43].

In addition, Asaria et al. estimated the lifetime treatment costs for the general population, defined as the expected costs of hospital admission over the average life expectancy, to be between £50,908 and £69,671 (inflated from 2011/2012 costs) [49]. Our total deterministic lifetime cost estimates due to OCS exposure (£30,277 in the OCS-only arm) represent 41.6–56.8% of these costs, demonstrating the high additional burden that OCS related-AEs represent. This additional burden is well demonstrated; in the study by Barry et al. [43], OCS-related AEs were found to increase lifetime prescription costs by 39–51% in patients with severe asthma compared with patients with moderate or no asthma.

Compared with existing models of OCS-related AEs, our model has a significant advantage in that it is the only model to include, in detail, three OCS-related chronic AEs; T2DM, eCVD, and osteoporosis, while also including other relevant AEs such as renal impairment, glaucoma, cataract, peptic ulcers, pneumonia, and adrenal insufficiency. The use of chronic health states addresses a significant weakness of prior approaches such as the previous tezepelumab and benralizumab NICE TAs [21, 23], where chronic events were considered as transient events with one-off costs and disutilities applied, resulting in an underestimation of the total cost and QALY impact from OCS-related AEs.

A direct comparison of outcomes between outcomes herein and the tezepelumab and benralizumab NICE TAs is not possible as results are redacted due to a commercial arrangement [21, 23]. The approach applied in TA278, for omalizumab for the treatment of severe persistent allergic asthma, applied the costs and health losses incurred with the excess relative risk associated with each OCS-related AE into an annual OCS cost or disability-adjusted life-year (DALY) loss per patient taking OCS per year [50]. A reanalysis by Norman et al. found incorporation of OCS AEs reduced the ICER from £50,181/QALY to £46,634/QALY [51]. Similarly, in NICE TA278, the Assessment Group found that including the adverse effects of OCS substantially reduced the potential ICER of omalizumab [50]. The outcomes from our model suggest that including the impact of avoiding OCS complications could likewise be expected to offset a significant proportion of OCS-sparing agent treatment costs (which are not considered in this model and hypothetical scenario), contribute to additional QALY gain, and therefore may have a significant impact on the cost effectiveness of OCS-sparing agents for future HTA assessments.

While this model accounts for several acute and chronic OCS-related transient events, it also relies on several assumptions that may be considered limitations to the analysis. For instance, mortality hazards and interactions are based on populations observed in the general literature, and therefore they may not reflect a specific OCS-eligible population. Furthermore, the cost estimates here do not take into account the treatment costs of OCS-sparing agents; in the UK, the net prices of such treatments are currently confidential and therefore could not be included in the analysis. Rather than use estimated prices, which may not reflect reality, the results are presented as the costs associated with the reduction of-related AEs, and this limitation should be taken into account when considering the cost effectiveness of OCS-sparing agents in the UK.

The population data used in this study were aligned to the OPRI RWE study, which used data from the OPCRD and CPRD databases, with patient characteristics defined based on an alternative published UK RWE study [1, 27, 28]. As such, these analyses may not be generalisable beyond the UK, and caution should be taken when comparing these results across countries. However, in a related study conducted in Italy, the costs of OCS-related AEs were predicted and were applied to epidemiological data from the Severe Asthma Network in Italy (SANI) registry [44]. In this study, the annual cost per patient for OCS-related AEs was reported at €1957.50. Averaging our costs across the life-years included, we estimate average per patient per year costs of £2183, excluding discounting. These costs align with the results reported by Canonica et al., with minor differences explained by differences in the unit costs of treating complications across countries. Together, these results demonstrate the burden of OCS-related AEs, but highlight the need for country-specific considerations for exact costings.

While the baseline characteristics of the model are likely to align well with the OCS-taking population in the UK, the OCS-related AE event incidence rates and additional risk applied to those events following OCS exposure were sourced from the OPRI study [1], which considered a younger patient population with more males. This limitation is expected to have an uncertain impact on model outcomes, as a younger population will be expected to experience a lower baseline incidence of OCS-related AEs, and may also experience higher OCS-related hazard ratios due to this. The model may therefore be underestimating the general population incidence rate of OCS-related AEs, while overestimating the impact of increasing OCS exposure.

In addition, prior cumulative dose for patients upon entering the OPRI study were unavailable [1], therefore prior OCS dose was assumed to be zero. This is not reflective of the true OCS-eligible population, as many patients would have prior OCS exposure. Similarly, in the model, patients may discontinue from the OCS-sparing scenario. Upon discontinuation, the OCS dose returns to that of the ‘OCS only’ arm immediately, whereas in reality, the OCS dose would increase more slowly. These are conservative assumptions that are likely to result in underestimation of the benefits of OCS sparing.

In the model, the OCS dose is captured using cumulative OCS dose over time, reflecting the progressive and cumulative nature of OCS-related AEs, which do not resolve upon discontinuation. However, as the risk of some AEs may decrease with OCS discontinuation, the model may overestimate the OCS-related risk for patients who have discontinued from OCS, yet are still assumed to experience the OCS-related risk driven by their cumulative OCS exposure.

The model assumes that the relationship between OCS dose and beta values (informing hazard ratios) is linear, however it may not be truly linear and this may result in an overestimation of OCS-related AEs, particularly at high cumulative doses (> 15 g), as this would represent an extrapolation beyond the OPRI study population [1]. This is an issue particularly for adrenal insufficiency, for which there is a very strong relationship between incidence and OCS exposure [31], which when applied with the linear exponential relationship and high cumulative doses results in clinically implausible incidence estimates. This limitation, along with the decision to use a skew log-normal distribution (to reflect skewed dosing observed in clinical practice) with a standard error of 20% of the mean value, explain the right-hand side skew observed in the 15 mg probabilistic scenario analyses, with higher incremental costs and QALYs observed driven by the small number of patients receiving very high OCS doses. Unfortunately, the relationship between the incidence of adrenal insufficiency and cumulative OCS exposure is unknown. However, to address this limitation, conservative assumptions have been applied where possible, and for the 15 mg scenario, both deterministic and probabilistic results are presented. These include removing patients receiving inhaled corticosteroids from the baseline incidence rate and applying costs due to adrenal insufficiency only for the duration of the model cycle (28 days) to avoid double counting where patients experience adrenal insufficiency multiple times in a single year.

A limitation of the current approach compared with previous models is the data used to inform this model, which does not always align with the OCS-receiving population in the UK. While both the tezepelumab technology appraisal and this model use the same source for OCS-related AE incidence, the modelling approach used was significantly different, with the tezepelumab model applying incidence rates for OCS-related AEs by dosage categories to estimate the incidence of transient events, whereas the model presented here used cumulative dose-estimated hazard ratios per additional gram of OCS. The approach for transient events in both models is similar. However, for chronic events, significantly more QALYs and costs are captured in our model, as the lifetime impact of incident events of chronic events are captured.