Cécile Proust-Lima, University of Bordeaux and Inserm

The study of neurodegenerative diseases, such as the Multiple System Atrophy (MSA: a rare alpha-synucleinopathy with poor prognosis) using data from epidemiological cohorts presents various statistical challenges:

• There is a large heterogeneity in the typical profiles of progression;
• Multiple dimensions/processes that evolve over time are involved;
• Most dimensions are measured by scales made of continuous or ordinal items (e.g., cognitive functions, motor function, quality of life).
• Repeated measurements are collected at highly variable times across participants, with missing data.
• Events (e.g., diagnosis, death, dropout) truncate the observation process.

Through the analysis of a large cohort of MSA patients, I describe several methodologies based on latent variables (random effects, latent processes, latent classes) to model disease progression in continuous time while addressing these challenges. Specifically, I introduce a continuous-time item response model (for continuous and/or ordinal items) to handle the irregular and individual-specific timings of observation in cohorts (1,2) and its extension to the joint modeling of a clinical event (3). Then, I describe an approach to study the progression of health-related Quality-of-Life over time along the clinical progression (4). Finally, I investigate the phenotypic heterogeneity via a latent class model for multiple longitudinal processes and time-to-death (5). These approaches are available in open-source R software, notably in the lcmm R package (6).

References:

1. Proust-Lima, C., Amieva, H., & Jacqmin-Gadda, H. (2013). Analysis of multivariate mixed longitudinal data: A flexible latent process approach. British Journal of Mathematical and Statistical Psychology, 66(3), 470–487. https://doi.org/10.1111/bmsp.12000
2. Proust-Lima, C., Philipps, V., Perrot, B., Blanchin, M., & Sébille, V. (2022). Modeling repeated self-reported outcome data: A continuous-time longitudinal Item Response Theory model. Methods, 204, 386–395. https://doi.org/10.1016/j.ymeth.2022.01.005
3. Saulnier, T., Philipps, V., Meissner, W. G., Rascol, O., Pavy-Le Traon, A., Foubert-Samier, A., & Proust-Lima, C. (2022). Joint models for the longitudinal analysis of measurement scales in the presence of informative dropout. Methods, 203, 142–151. https://doi.org/10.1016/j.ymeth.2022.03.003
4. Saulnier, T., Fabbri, M., Le Goff, M., Helmer, C., Pavy-Le Traon, A., Meissner, W. G., Rascol, O., Proust-Lima, C., & Foubert-Samier, A. (2024). Patient-perceived progression in multiple system atrophy: Natural history of quality of life. Journal of Neurology, Neurosurgery & Psychiatry. https://doi.org/10.1136/jnnp-2023-332733
5. Proust‐Lima, C., Saulnier, T., Philipps, V., Traon, A. P. L., Péran, P., Rascol, O., Meissner, W. G., & Foubert‐Samier, A. (2023). Describing complex disease progression using joint latent class models for multivariate longitudinal markers and clinical endpoints. Statistics in Medicine, 42(22), 3996–4014. https://doi.org/10.1002/sim.9844

6. Proust-Lima, C., Philipps, V., & Liquet, B. (2017). Estimation of Extended Mixed Models Using Latent Classes and Latent Processes: The R Package lcmm. Journal of Statistical Software, 78(2), 1–56. https://doi.org/10.18637/jss.v078.i02

About the Speaker

Dr. Cécile Proust-Lima is a Director of Research in Biostatistics at the Bordeaux Population Health Research Center for the Inserm (French Institute of Health and Medical Research) and the University of Bordeaux in France. Her research mainly focuses on the development of dynamic statistical models to describe, explain and predict chronic disease progression. She has specialized over the years in the joint analysis of multivariate repeated markers and event time history with applications mainly in cerebral aging and neurodegenerative diseases (Multiple System Atrophy, Alzheimer’s Disease and related dementias). Her works often involve latent processes to translate dynamic health phenomena measured by repeated noisy markers, and latent classes to translate the heterogeneity of disease progression. Her research is highly motivated by epidemiological and clinical questions thanks to strong collaborations with epidemiologists and clinicians, and access to large cohort studies. It covers a wide range of statistical topics, from causal inference to dynamic prediction techniques, always with an emphasis on modeling and data challenges. The developments of her research group are made available in R packages (e.g., lcmm) with constant maintenance and upgrades.