Upcoming GRBIO seminars

Online, si voleu assistir-hi, contacteu amb grbio@grbio.eu

Longitudinal Modeling of Non-invasive Biomarker GDF15 in Oropharyngeal Squamous Cell Carcinoma Using Linear Mixed-Effects Models

Growth differentiation factor 15 (GDF15) is a stress-responsive cytokine implicated in systemic inflammation and cancer-related wasting. This study aimed to characterize the longitudinally evolution of circulating GDF15 evolution in patients with oropharyngeal squamous cell carcinoma (OPSCC) and to quantify the independent effects of HPV status, treatment response, and relevant clinical, nutritional and body composition variables on its temporal trajectory. We analyzed data from a prospective cohort of 138 OPSCC patients. Baseline associations between GDF15 and HPV status, as well as nutritional and body composition parameters, were evaluated.
Longitudinal GDF15 measurements (baseline and end of treatment) were modeled using linear mixed-effects models with a patient-specific random intercept to account for within-subject correlation. Fixed effects included time, HPV status, treatment modality and treatment response, and selected interaction terms (time-by-HPV and time-by-treatment). Regression coefficients (β) represent adjusted mean differences in GDF15 levels (pg/mL). This approach allows separation of between-subject and within-subject variability, providing robust inference on temporal change and effect modification in a clinically heterogeneous cohort.

Biosketch

Sara Tous holds degrees in Statistics and Statistical Sciences from UPC (2001, 2008). Since 2005, she has been lead statistician at the Unit of Molecular Epidemiology and Genetics within the Cancer Epidemiology Research Programme at the Catalan Institute of Oncology. Her work focuses on HPV-related anogenital and head and neck cancers. She has co-authored more than 30 peer-reviewed publications and is an active member of the Catalan Statistical Society, contributing to initiatives that promote statistical literacy.

Online, si voleu assistir-hi, contacteu amb grbio@grbio.eu

Spatial Bayesian distributed lag non-linear models (SB-DLNM) for small-area exposure-lag-response epidemiological modelling

Distributed lag non-linear models (DLNMs) are the standard approach for modelling lagged non-linear associations but are limited in small-area analyses. Recent spatial Bayesian DLNMs (SB-DLNMs) address these limitations by jointly modelling spatial dependence and non-linear lagged effects in a one-stage framework. This approach enables robust estimation of geographically varying risks in small areas by borrowing strength across space. The presentation introduces SB-DLNMs, demonstrates their application in epidemiological studies, and discusses key gaps and future methodological directions.

Biosketch

Marcos Quijal Zamorano is a postdoctoral researcher at the Climate Epidemiology & Public Health research group at University of Bern, supervised by Dr. Ana Maria Vicedo Cabrera. His research focuses on methodological developments in environmental epidemiology, extending spatial and flexible exposure–response models for small-area analyses, developed during his PhD at ISGlobal. He participated in studies that assessed temperature-related health impacts across Europe under current and future climate conditions and informed heat–health early warning systems.

Online, si voleu assistir-hi, contacteu amb grbio@grbio.eu

Estimands and Estimation Strategies for Platform Trials with Time Trends

"In platform trials, model-based approaches have been proposed to mitigate bias from time trends. However, these methods typically condition on time, yielding treatment effect estimates specific to particular calendar periods. While such conditional estimates are unbiased for their respective time-specific estimands, investigators typically seek an unconditional treatment effect - one that does not depend on calendar time and reflects the effect in the overall trial population. This raises fundamental questions: when combining data from multiple periods, what is the appropriate target estimand? What defines the overall trial population if arms continuously enter and leave? Which estimator should be used?
In this work, we examine both conditional and marginal estimands in platform trials with time trends, clarifying the target populations of inferential interest. We evaluate model-based approaches, G-computation and augmented inverse probability weighting estimators, comparing their properties, including bias and variance. We discuss how the choice of estimand and the target population, as well as which trial data to use for estimation, affects the performance of estimators."