Noise Is Not Error: Detecting Parametric Heterogeneity Between Epidemiologic Time Series.

Area of Science:

• Epidemiology
• Mathematical Biology
• Computational Science

Background:

• Mathematical models, particularly ordinary differential equations (ODEs), are crucial in epidemiology for data integration, experimental design, and hypothesis generation.
• Deterministic ODE models assume noise arises solely from error or parametric heterogeneity, potentially misinterpreting random fluctuations as true biological variation.
• This misclassification can lead to unstable predictions and flawed public health policies or research directions.

Purpose of the Study:

• To quantify the capacity of ODE models under different hypotheses (fixed vs. random effects) to accurately represent individual data variations.
• To compare the performance of state-of-the-art stochastic fitting methods against traditional least squares approximations for noisy time-series data.
• To develop a framework for assessing the limitations and risks associated with conventional ODE fitting methodologies.

Main Methods:

• Utilized an exactly solvable mathematical model exhibiting initial exponential growth to compare fitting approaches.
• Employed and evaluated advanced stochastic fitting techniques alongside traditional least squares approximations.
• Analyzed the impact of noise on parameter estimation and model prediction accuracy.

Main Results:

• Stochastic fitting methods demonstrate superior ability in distinguishing random noise from genuine parametric heterogeneity compared to traditional least squares.
• Traditional ODE fitting methods applied to noisy data risk misattributing random variations to biological differences, leading to inaccurate parameter estimates.
• The study highlights specific scenarios where noise significantly impacts model stability and predictive power.

Conclusions:

• Accurate epidemiological modeling, especially with noisy data, necessitates the use of stochastic approaches over traditional deterministic methods.
• Understanding the limitations of conventional fitting techniques is vital for reliable data interpretation and evidence-based decision-making in public health.
• The findings have direct implications for interpreting epidemic data, such as that from the 2014-2015 Ebola outbreak, and refining future modeling strategies.