We employed a modelling approach to estimate the overall number and prevalence of metastatic/unresectable GIST patients in the UK and those who failed on currently licensed treatments. The model demonstrated that, under a plausible set of assumptions (base-case scenario) of disease incidence and survival, the prevalence of third-line treatment-eligible GIST is 1.0 (95% CI; 0.7–1.3) per 100,000, with a 99.9% likelihood of being below the 2 per 100,000 population threshold for ultra-orphan disease status in the UK. As data values reported in published studies which informed our model parameters are subject to great variation, we examined three alternative sets of assumptions.
When we raised the assumed yearly incidence of GIST to the highest levels reported in Western countries (1.5 per 100,000 persons), the probability that third-line treatment-eligible GIST prevalence remained below the ultra-orphan disease threshold was virtually unchanged (99.6%). This scenario may be plausible given that estimates of the annual incidence of GIST may rise with improved diagnosis of tumours that are tested for c-KIT. Because a new effective treatment line would prolong life in the third-line treatment-eligible GIST and therefore increase prevalence, it was important to assess the longer patient survival impact on the third-line treatment-eligible GIST prevalence. By assuming 1.5 years of survival in the third-line treatment-eligible GIST state, we estimated a 90.6% probability that its prevalence was below the ultra-orphan disease threshold. Only when assuming both higher GIST incidence and longer third-line treatment-eligible GIST survival we obtained a 37.9% probability of remaining below the threshold of interest.
Sensitivity analysis showed that initial GIST incidence and third-line treatment-eligible GIST survival were the most influential variables on the third-line treatment-eligible GIST prevalence estimate, while threshold analysis showed that only under extreme assumptions of these two parameters (GIST incidence greater than 2.2 per 100,000 population or third-line treatment-eligible GIST survival greater than 2.2 years) would the number of third-line treatment-eligible GIST patients surpass the UK ultra-orphan disease threshold. We did not explore further other model parameters in this respect. Varying estimates of post-resection GIST TTT, imatinib-treated GIST TTT, sunitinib-treated GIST TTT and the proportion of resectable GIST did not greatly influence our study findings.
The model predicted the total number of patients living after suffering GIST in the base-case to be 9,365 (95% CI; 6,953–12,325) and the number of GIST patients taking imatinib as 1,422 (95% CI; 838–2,368) or sunitinib as 599 (95% CI: 435–789).
The understanding of the aetiology and management of GIST has evolved since it was first differentiated among other soft tissue sarcomas and two lines of therapy became available for patients with metastatic or unresectable tumours. However, for a proportion of patients, these therapies eventually fail and patients who exhaust their treatments are left with re-challenging with higher imatinib dose or best supportive care. Several potential third-line treatment drugs [23–25] could be candidates for orphan or ultra-orphan disease treatment status. Orphan and ultra-orphan disease status have implications on how public funding supports the provision of treatments, and the speed of access to new treatments for suitable patients. There could be a case for potential treatments after imatinib and sunitinib failure given the low number of patients at third-line treatment-eligible GIST.
The major limitation of the study is the face validity of the model structure and the structure’s inherent assumptions on treatment pathways; however, this study presents a model, understood as a simplified and imperfect description of reality, which estimates the number of subjects at each stage of GIST treatment, based on model parameters available from the literature. Another limitation we encountered was that data needed to inform the model parameters were sparse or unsuitable, which has also been reported in economic modelling studies [26–30]. This limitation is common in modelling studies since it is not always possible to inform all of the model parameters, considering that the samples of the available studies are small , the evidence for the model is obtained from only one study  or data are used from other countries and applied to the country of interest due to lack of local evidence .
Some of the reviewed studies reported on important prognostic variables (i.e., mitotic count, genetic markers) that identify heterogeneous subgroups within the GIST patient population. Our model did not explicitly account for these subgroups as this would have necessitated stratum-specific TTT estimates that were mostly unavailable. As a result, we have given preference to publications with large sample sizes where a pool of GIST patients with a mix of those variables can occur, that is, studies with heterogeneous patient populations (i.e., different treatment histories, stages of GIST) and response criteria. Nonetheless, sensitivity and scenario analysis showed that the importance of data uncertainty mattered mainly for initial incidence of GIST and third-line treatment-eligible GIST survival.
We acknowledge that the model’s structure, where all patients transition through the treatment states sequentially, and assuming that patients cannot skip a particular treatment line, can be questioned. Another modelling study included up to seven plausible treatment pathways for patients with GIST which also depended on limited data for the proportions of patients following each pathway, and required assumptions for death rates and state transition probabilities . We also assumed GIST as not curable, so all subjects diagnosed with GIST could not exit the GIST population. Nevertheless, the post-resection GIST TTT are widely distributed and a long post-resection GIST TTT (longer than the life expectancy) can be assumed as cured GIST. Both of these assumptions are conservative since patients remain in the model for longer, therefore, leading to overestimation of GIST-related state prevalences. Our model did not explicitly include pathways for adjuvant and neoadjuvant use of tyrosine kinase inhibitors. These therapeutic strategies can be considered a combination of the two proposed model pathways.