Seminar Series

Past Seminars

You can find the recordings of some of our past seminars here.

17 September 2026

Title: Causal inference with unmeasured exposures in birth cohort studies: aligning questions with algorithms

Speaker: Izzuddin Aris (Harvard Pilgrim Health Care Institute and Harvard Medical School)

Summary: Observational studies have shown strong associations between BMI metrics in childhood (e.g., early age at BMI rebound, overweight/obesity) and subsequent cardiometabolic outcomes including type 2 diabetes and metabolic syndrome in adolescence and adulthood. Despite adjustment for covariates, these studies have not explicitly considered causal inference, including causal effect definitions and assumptions sufficient for unbiasedly estimating that effect with available data. In this presentation, we consider the problem of using existing observational data from a pre-birth cohort in USA (Project Viva) to estimate effects of stochastic interventions on a chosen subset of modifiable (but unmeasured) exposures, expected to be associated with BMI metrics (a proxy exposure), on adolescent cardiometabolic and respiratory outcomes in early adolescence.

22 January 2026

Title: Synthetic cohorts for life course etiologic research

Speaker: Katrina Kezios (Boston University School of Public Health)

Summary: Because no single U.S. cohort spans the entire life course, researchers increasingly rely on data combination approaches (e.g., pooling, integration, fusion) that leverage multiple, age-overlapping cohorts covering different life stages to construct life course data. When the analytic goal in the combined cohort is exposure–disease effect estimation, such approaches require attention to causal assumptions, and principles from data fusion theory can guide cohort construction. This seminar reviews conditions under which such cohort fusion is possible (Kezios et al. 2024. Epidemiology) and presents an applied example that creates a longitudinal synthetic cohort by fusing the National Longitudinal Survey of Youth 1979 and Health and Retirement Study cohorts to estimate the effect of income volatility across adulthood on later memory decline.

26 September 2025

Title: Functional Principal Component Analysis (FPCA) to describe longitudinal data

Speaker: Corentin Segalas (Bordeaux Population Health)

Summary: In this talk, the functional data analysis framework will be introduced and we will see how FPCA - one of its central tools - may be used to describe biomedical longitudinal data. Finally, a simulation study that investigates FPCA robustness to various missing data scenarios will be discussed.

06 February 2025

Title: SITAR enhancements to support state-of-the-art analysis of individual growth curves and their correlates

Speaker: Ahmed Elhakeem (University of Bristol)

Summary: I will present the aims and objectives for my New Investigator Research Grant, which are to develop and test methodology for a) SITAR to model post-peak growth, b) a Bayesian SITAR model, c) joint SITAR and outcome/exposure models, and to develop and test R (& DataSHIELD) packages to implement these methods and develop resources and guidance for researchers.

Title: Bayesian Implementation of Superimposition by Translation and Rotation (SITAR) Growth Model: An Introduction to the ‘bsitar’ R Package

Speaker: Satpal Sandhu (University of Bristol)

Summary: To enhance the utility of the SITAR model, we have developed the ‘bsitar’ R package, which implements the Bayesian version of the SITAR model. The objective of this presentation is to provide an overview of the ‘bsitar’ package and demonstrate some of its capabilities.

31 October 2024

Title: Joint modelling of non-time to event longitudinal data: a systematic review and case-study on height and weight

Speaker: Rehema Ouko (London School of Hygiene & Tropical Medicine)

Summary: Joint Bayesian modelling approach is feasible for modelling growth data and incorporates correlation between measurements. In longitudinal studies, it can be used in the case of missing data for two or more correlated variables of interest. Therefore, there is a potential benefit of using joint modelling of repeated measurements leveraging its advantages.

06 March 2024

Title: Derivative curve estimation in longitudinal studies using P-splines

Speaker: Marìa Alejandra Hernàndez Velandia (Basque Center for Applied Mathematics)

Summary: In this presentation, we will explore the use of P-spline models for estimating derivatives within the framework of longitudinal data analysis. The main contribution lies in adding an extra penalty on the coefficients of both the population curves and the individual deviations of the model. This integration focuses on issues of under-smoothing and boundary problems in derivative estimation. The problem is interesting because derivative estimation is valuable for describing individuals’ biological features or identifying change regions of interest. As an illustration, we will showcase an application involving longitudinal height measurements in teenagers to analyze their growth and maturity patterns over time.

31 January 2024

Title: Weight trajectories through adulthood and prostate cancer: a two-stage approach to the analysis of longitudinal and survival data

Speaker: Marisa da Silva (Lund University)

Summary: The Obesity and Disease Development Sweden (ODDS) study is a pooled cohort including over 4 million individuals with 9 million observations. We sampled ~250 000 men with repeated weight observations to estimate life and age span weight trajectories and used the parameters in a survival model. I will present our model building process and some of the discussions behind the steps.