Structured Hybrid Mechanistic Models for Robust Estimation of Time-Dependent Intervention Outcomes

Estimating intervention effects in dynamical systems is crucial for outcome optimization. In medicine, such interventions arise in physiological regulation (e.g., cardiovascular system under fluid administration) and pharmacokinetics, among others. Propofol administration is an anesthetic intervention, where the challenge is to estimate the optimal dose required to achieve a target brain concentration for anesthesia, given patient characteristics, while avoiding under- or over-dosing. The pharmacokinetic state is characterized by drug concentrations across tissues, and its dynamics are governed by prior states, patient covariates, drug clearance, and drug administration. While data-driven models can capture complex dynamics, they often fail in out-of-distribution (OOD) regimes. Mechanistic models on the other hand are typically robust, but might be oversimplified. We propose a hybrid mechanistic-data-driven approach to estimate time-dependent intervention outcomes. Our approach decomposes the dynamical system’s transition operator into parametric and nonparametric components, further distinguishing between intervention-related and unrelated dynamics. This structure leverages mechanistic anchors while learning residual patterns from data. For scenarios where mechanistic parameters are unknown, we introduce a two-stage procedure: first, pre-training an encoder on simulated data, and subsequently learning corrections from observed data. Two regimes with incomplete mechanistic knowledge are considered: periodic pendulum and Propofol bolus injections. Results demonstrate that our hybrid approach outperforms purely data-driven and mechanistic approaches, particularly OOD. This work highlights the potential of hybrid mechanistic-data-driven models for robust intervention optimization in complex, real-world dynamical systems.

💡 Research Summary

The paper tackles the problem of estimating the effects of time‑varying interventions in dynamical systems, a task that is central to personalized medicine, fluid management, and many engineering applications. Traditional approaches fall into two camps: purely mechanistic ordinary differential equation (ODE) models that encode known physics but are often misspecified, and fully data‑driven models (e.g., neural networks) that can capture complex nonlinearities but tend to fail when the test distribution differs from the training distribution—particularly problematic for causal or counterfactual queries that inherently involve distribution shifts.

To bridge this gap, the authors propose a structured hybrid framework that decomposes the transition operator of the system into four components: (i) parametric + non‑parametric, and (ii) intervention‑dependent + intervention‑independent. Formally, the dynamics are written as

dX/dt = Fψ^p(t,X,β_u) + Fψ^np(t,X,β_u) +