Presentation (CTI)

OBJECTIVES: Indirect treatment comparisons frequently require reconstruction of individual patient data from published Kaplan-Meier curves and selection of appropriate parametric survival models for long-term extrapolation. These analyses often rely on multiple software packages and manual coding, making them time-consuming and difficult to reproduce. This study aimed to develop and validate an AI-augmented interactive R Shiny tool that automates KM reconstruction, parametric survival modelling, and economic evaluation within a reproducible workflow.
METHODS: An R Shiny tool was developed to support survival analyses using published KM curves or raw IPD. The tool reconstructs IPD, fits seven parametric survival distributions (Exponential, Weibull, Log-Normal, Log-Logistic, Gompertz, Gamma, and Generalised Gamma) using the “flexsurvreg” package, and ranks models using Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and log-likelihood. Additional functionalities include KM estimation, Cox proportional hazards modelling, restricted mean survival time, subgroup analyses, diagnostic plots, and long-term extrapolation. Treatment costs and health utility inputs generate quality-adjusted life years (QALYs) and enable real-time cost-effectiveness comparisons. Validation compared survival, cost, and QALY estimates with published oncology results.
RESULTS: Using the tool, IPD was successfully reconstructed from published KM curves, generating parametric survival estimates across all evaluated distributions. Automated model ranking identified the optimal survival model, while diagnostic visualisations assessed distributional assumptions and extrapolation suitability. The integrated workflow reduced manual programming by combining IPD reconstruction, survival modelling, diagnostics, and extrapolation within a single interface. Validation demonstrated high concordance with published results, with relative deviations below 2% across survival, cost, and QALY estimates. Reproducible analyses and downloadable survival traces supported transparent evidence generation for HTA.
CONCLUSIONS: The AI-augmented R Shiny tool provides an integrated solution for automated survival analysis and evidence synthesis, improving transparency and reproducibility. The tool has the potential to support researchers, HTA agencies, and pharmaceutical sponsors by accelerating survival modelling and informing health economic evaluations and reimbursement submissions.