Research Works & Publications

Standard cardiovascular risk calculators estimate the conditional probability P(CHD | SysBP = s) rather than the interventional quantity P(CHD | do(SysBP = s)). We applied Pearl's do-calculus to the Framingham Heart Study Offspring Cohort (n = 4,240). A structurally corrected DAG incorporating four biologically motivated corrections was subjected to conditional independence testing. G-computation yielded an ACE of 3.40% absolute risk reduction (95% CI: 2.64%–4.14%), compared with a naive observational estimate of 4.14% — a relative overestimation of ~21.8% (+35.7% on the log-odds scale). Significant treatment-effect heterogeneity was detected across age strata and diabetes status. These findings suggest standard risk tools overestimate the absolute benefit of blood pressure reduction, with implications for clinical risk stratification and prescribing thresholds.

Glioblastoma multiforme (GBM), designated as WHO Grade IV IDH-wildtype adult-type diffuse astrocytic tumour, represents the most prevalent and therapeutically challenging primary brain malignancy in adults. A critical clinical challenge is the reliable differentiation of true tumour progression (TP) from pseudoprogression (PsP). This study presents a comprehensive radiomics and machine learning framework leveraging DCE-MRI and pharmacokinetic modelling. Radiomic features were extracted from five pharmacokinetic parameter maps derived via a novel parsimonious modelling framework. A cohort of 82 histologically confirmed GBM patients was evaluated, with MGMT promoter methylation status incorporated as a molecular covariate. The Random Forest classifier integrating parsimonious DCE-derived radiomic features with MGMT status achieved test AUC 0.89 ± 0.07, sensitivity 0.93 ± 0.09, specificity 0.76 ± 0.19, and F1-score 0.90 ± 0.07, improving AUC by 26 pp over the conventional Extended Tofts-only approach.