BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Aurore Zhe LI - Neural Ordinary Differential Equations enhanced Li
 near Mixed-effects Models
DTSTART:20260917T110000Z
DTEND:20260917T120000Z
DTSTAMP:20260710T025500Z
UID:indico-event-16572@indico.math.cnrs.fr
CONTACT:denis.rustand@u-bordeaux.fr
DESCRIPTION:Speakers: Aurore Zhe LI (BPH)\n\nSpeaker: Aurore Zhe LI from B
 PH\nTitle: Neural Ordinary Differential Equations enhanced Linear Mixed-ef
 fects Models\nAbstract: \nIntroductionLongitudinal cohort studies generat
 e irregularly spaced\, partially missing repeated measurements where outco
 mes may depend on the full trajectory of time-varying exposures rather tha
 n their instantaneous values. Classical linear mixed-effects models (LMMs)
  can incorporate path-dependent or cumulative exposures through hand-engin
 eered functionals like lags\, weighted cumulative integrals\, or basis exp
 ansions of covariate history\, whose functional form\, time window\, and i
 nteraction structure must be pre-specified by the analyst\, and so cannot 
 be discovered from data. We propose Neural Ordinary Differential Equation 
 enhanced LMMs (NODE-LMMs)\, a unified framework embedding continuous-time 
 latent dynamics within a mixed-effects model\, which learns the form of th
 e path-dependent effect jointly with the rest of the model rather than fix
 ing it a priori.\nMethodsBaseline and static covariates are encoded into a
 n initial latent state\, which evolves over time via a Neural ODE with aut
 onomous dynamics. The latent process is decoded through a LMMs observation
  model\, enabling estimation of population-level fixed effects and subject
 -specific random effects\, with parameters learned by maximizing the margi
 nal likelihood. To interpret covariate effects\, we compute Partial Depend
 ence Plots (PDPs) under counterfactual interventions and define the ∆PDP
  as the difference in population-averaged predictions between exposure lev
 els with uncertainty quantified via a full-parameter delta method. The app
 roach is evaluated in the simulation studies under diverse data-generating
  processes\, and compared to the classical LMMs. It is also applied to the
  real population-based aging Trois-Cités (3C) cohort (N=7\,324) to assess
  the association between cardiometabolic health (e.g.\,BMI\, glycemia) and
  cognitive decline over 15 years of follow-up.\nResultsUnder an instantane
 ous association (X(t) --> Y(t))\, our approach achieves ∆PDP estimation 
 performance comparable to the correctly specified classical LMM. Under a c
 umulative burden process\, the NODE-LMMs correctly distinguish sustained e
 arly exposure from a late spike to the same level\, a discrimination the c
 lassical LMMs cannot achieve without committing in advance to a cumulative
 -burden specification. In the 3C cohort\, our method delivers interpretabl
 e marginal associations between trajectories of cardiometabolic factors (B
 MI and glycemia) and cognitive decline via the PDP.\nDiscussionThe NODE-LM
 Ms extend classical LMMs with continuous-time neural dynamics while preser
 ving their statistical interpretability. Simulation and real-data results 
 demonstrate that the framework matches LMMs performance when covariate eff
 ects are instantaneous and outperforms it when effects are path-dependent 
 or cumulative\, a setting highly relevant inlife-course epidemiology.\nKey
 words: Irregular Longitudinal Data\, Neural ODE\, mixed-effects models\nCa
 lendar subscription link for the complete seminar series:https://indico.ma
 th.cnrs.fr/category/711/events.ics\nProgram of the Biostatistics seminars:
 https://indico.math.cnrs.fr/category/711/\nSubscribe to the seminar mailin
 g list:https://diff.u-bordeaux.fr/sympa/subscribe/seminaire.biostat.bph\nF
 ormer e-seminars on our YouTube channel (mostly in French): https://www.yo
 utube.com/channel/UCURp-hEQL7k23UzGfqgEurA/videos\n \nBiostatistics semin
 ar series from the Department of Public Health from the University of Bord
 eaux and the Bordeaux Population Health UMR 1219 research center\n \n\nht
 tps://indico.math.cnrs.fr/event/16572/
LOCATION:amphi Louis (ISPED)
URL:https://indico.math.cnrs.fr/event/16572/
END:VEVENT
END:VCALENDAR