A storage model with random release rate for modeling exposure to food contaminants

Patrice Bertail, Stéphan Clémençon, Jessica TressouSPIRE 2008

TL;DR

This paper introduces a novel mathematical model for tracking how contaminants accumulate in the human body over time, considering both dietary intake and the body's natural elimination processes. The model, a piecewise-deterministic Markov process, incorporates random variability in elimination rates, which is crucial for realistic risk assessment, and is applied to methyl mercury exposure data.

What Problem Does This Paper Address?

Assessing the health risks associated with food contaminants often involves understanding how these substances build up in the body. Traditional methods often use static models that don't account for the dynamic processes of accumulation and elimination over time. This paper addresses the need for a more sophisticated, dynamic model that captures the temporal evolution of contaminant levels in the body. Specifically, it aims to model the exposure to a contaminant by considering:

  • Repeated dietary intakes: The process of consuming contaminated food at various times.
  • Pharmacokinetics: How the body absorbs, distributes, metabolizes, and excretes the contaminant.
  • Variability in elimination: Recognizing that the rate at which the body eliminates a contaminant can vary due to individual metabolic differences.

This dynamic modeling is particularly important for understanding long-term exposure and assessing risks associated with chronic exposure, such as with methyl mercury.

Methods

The core of the paper is the development and analysis of a piecewise-deterministic Markov process (PDM process). Here's a breakdown of the key methodological aspects:

  • Accumulation Model: Dietary intakes are modeled as a marked point process. Each intake event is characterized by the quantity of food consumed and its contamination level.
  • Elimination Model: Between intakes, the contaminant level in the body decreases according to a linear differential equation. Crucially, the rate of this elimination is random, represented by a parameter $\theta$. This randomness accounts for metabolic variability.
  • Embedded Markov Chain: The analysis of the long-term behavior (ergodicity) relies on studying an embedded Markov chain. This chain captures the contaminant levels immediately after each intake.
  • Statistical Inference: Since the exposure process is often not directly observable, the paper develops a framework for simulation-based statistical methods. This involves using estimated distribution functions to simulate sample paths and then constructing estimators from these simulated trajectories.
  • Coupling Analysis: To establish the validity of simulation-based methods, the authors employ coupling analysis to demonstrate the closeness between the true and approximated laws of the exposure process.

Datasets

The paper uses real-world data for its application:

  • Methyl Mercury Contamination Data: Data on methyl mercury contamination in fish and seafood from French markets, collected between 1994 and 2003. This dataset comprises 2832 analytical data points.
  • National Individual Consumption Survey (INCA): Data from the French national survey on food consumption, providing information on quantities of food consumed and meal types for adults and children.

These datasets are used to estimate the parameters of the intake and inter-intake time distributions.

Key Insights and Ideas

  • Random Elimination is Key: The paper strongly emphasizes that incorporating random variability in the contaminant elimination rate is essential for realistic modeling. This reflects biological reality and significantly impacts long-term exposure predictions.
  • PDM Processes as a Powerful Tool: The use of piecewise-deterministic Markov processes provides a flexible and powerful framework for modeling complex biological systems involving discrete events (intakes) and continuous dynamics (elimination).
  • Simulation for Unobservable Processes: The paper offers a rigorous theoretical foundation for using simulation methods to estimate quantities of interest when the actual exposure process cannot be directly measured. This is a critical step for practical risk assessment.
  • Steady-State Behavior is Achievable: The analysis demonstrates that the exposure process eventually settles into a steady-state regime, allowing for the assessment of long-term average exposure and risk.
  • Time to Steady-State Matters: The paper highlights that the time it takes to reach this steady-state is relevant for practical risk assessment, suggesting that for basic quantities, this can be on the order of several years.

Key Results

  • Model Development: A novel PDM process is proposed and analyzed, extending standard storage models by incorporating random release rates.
  • Ergodicity and Stability: The paper establishes conditions for the positive recurrence and geometric ergodicity of the embedded Markov chain and the continuous-time process.
  • Tail Behavior Analysis: Theoretical results are presented for the tail behavior of the stationary distribution, linking it to the tail behavior of the intake distribution.
  • Simulation Framework: A theoretical framework is provided for the asymptotic validity of simulation estimators, showing that good approximations of input distributions lead to good approximations of exposure process functionals.
  • Application to Methyl Mercury: The model is applied to methyl mercury data, providing estimated parameters for intake and elimination distributions and illustrating the convergence of mean exposure to a steady-state value. For the adult female subpopulation, the estimated steady-state mean exposure is 2.92 µg/kgbw. The time to reach steady-state for average time spent above a reference threshold (Xref,0.7) is estimated to be around 8.5 years.

Limitations and Future Work

  • Model Simplifications: The authors acknowledge that the model makes simplifying assumptions, such as the independence of intake quantities and contamination ratios, and the specific form of the elimination rate. For contaminants present in many foods, a more complex autoregressive structure for dietary behavior might be needed.
  • Data Availability: The study relies on available datasets for methyl mercury. Obtaining more detailed, individual-level data on consumption and pharmacokinetics could further refine the model.
  • Statistical Issues: While the paper provides theoretical grounds for simulation-based inference, further investigation into the statistical issues of validating the mathematical model and exploring nonparametric approaches is suggested.
  • Rare Event Simulation: Estimating probabilities of rare events (e.g., exceeding very high thresholds) using naive Monte-Carlo simulation can be inefficient. The paper points to the need for more sophisticated methods like particle filtering for such scenarios.
  • Sensitivity Analysis: Future work will involve investigating the sensitivity of the model to changes in estimation methods and exploring nonparametric approaches.

Summary updated

Abstract

This paper presents the study of a continuous-time piecewise-deterministic Markov process for describing the temporal evolution of exposure to a given food contaminant. The quantity X of food contaminant present in the body evolves through its accumulation after repeated dietary intakes on the one hand, and the pharmacokinetics behavior of the chemical on the other hand. In the dynamic modeling considered here, the accumulation phenomenon is modeled by a simple marked point process with positive i.i.d. marks, and elimination in between intakes occurs at a random linear rate OX, randomness of the coefficient B accounting for the variability of the elimination process due to metabolic factors. Via embedded chain analysis, ergodic properties of this extension of the standard compound Poisson dam with (deterministic) linear release rate are investigated, the latter being of crucial importance in describing the long-term behavior of the exposure process (X-t)(t >= 0) and assessing values such as the proportion of time the contaminant body burden is over a certain threshold. We also highlight the fact that the exposure process is generally not directly observable in practice and establish a validity framework for simulation-based statistical methods by coupling analysis. Eventually, applications to methyl mercury contamination data are considered.
BibTeX
@inproceedings{bertail2008storage,
  author = {Patrice Bertail and Stéphan Clémençon and Jessica Tressou},
  title = {{A storage model with random release rate for modeling exposure to food contaminants}},
  booktitle = {SPIRE},
  year = {2008},
}
Figure