Document resource
Introduction The Brims model classifies patients with pleural mesothelioma (PM) into 4 prognostic groups. However, at diagnosis, clinicians and patients commonly face a binary choice between immediate treatment or ‘watch-and-wait’. The latter is high-risk if patients rapidly progress but reasonable if clinical progression is slow, since immediate treatment may result in unacceptable toxicity. We applied plasma Fibulin-3 data, which in isolation has yet to find a useful clinical role, to design and validate a binary prognostic model that could be used to inform clinical decision-making in this context. The study aspired to create a classifier with prognostic performance that was at least equal to the Brims model using data from the multicentre DIAPHRAGM study (Tsim, JTO 2021) and Scottish National MDT.Methods We first performed a retrospective analysis of prospectively recorded DIAPHRAGM data (n=749, 23 UK centres). Using 100 recruits with a final diagnosis of PM and baseline plasma Fibulin-3 data, we defined a 3-group 1-yr survival model based on hazard ratios generated by multivariable Cox regression. This defined clear high/low-risk groups, plus an intermediate group for stratification using Fibulin-3. Next, we externally validated the performance of the 3-group model in an independent set of 203 Scottish National MDT cases. Finally, to create the intended binary model, we stratified intermediate-risk cases in the validated 3-group model by baseline Fibulin-3, generating the Glasgow Mesothelioma Score (GMS).Results Six independent features contributed to the initial 3-group DIAPHRAGM model ( figure 1a), which delivered superior performance to Brims (C=0.72 v 0.66). The 3-group model was successfully validated in the Scottish MDT dataset, with significant prognostic performance retained (C=0.60, 1-yr survival 38%, 56%, 75% in high/intermediate/low-risk; figure 1b). In the final, binary GMS model, stratification of intermediate-risk cases by Fibulin-3 (dichotomised using an optimal threshold of 16μg/mL), improved performance (C=0.73, 1-yr survival 70% v 9% in high/low risk; figure 1c-d).Abstract S66 Figure 1Conclusion The GMS combines Fibulin-3 with an externally validated 3-group model to create a powerful binary classifier. This model may be useful in identifying patients unsuitable for ‘watch-and-wait’. Ongoing work includes further external validation of the final binary GMS using international cohorts with available Fibulin-3 data.