📄 medicine

The Quantified Epidemic Energy (QEE) Model: A Stochastic Framework for Climate- Sensitive Cholera Forecasting in Yemen

This study introduces the Quantified Epidemic Energy (QEE) model, a novel stochastic framework that integrates climate risk and reporting delays to accurately forecast cholera incidence in Yemen's Al-Hodeidah Governorate, offering a robust, uncertainty-quantified tool for proactive early-warning systems in data-scarce, conflict-affected settings.

Original authors: Hussein Bakery Hussein Dedy, Ali Bannawi ALZubaidy

Published 2026-06-26✓ Author reviewed ⓘ

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Hussein Bakery Hussein Dedy, Ali Bannawi ALZubaidy

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: Predicting a Storm Before It Hits

Imagine trying to predict a massive flood in a city where the roads are broken, the weather is crazy, and the people keeping track of the water levels are often too busy or scared to send reports on time. That is the situation in Yemen regarding Cholera, a waterborne disease.

The authors of this paper, Hussein Dedy and Ali Bannawi ALZubaidy, created a new tool called the Quantified Epidemic Energy (QEE) Model. Think of this model not as a standard calculator, but as a "Weather Vane for Disease." Instead of just counting sick people, it measures the invisible "energy" or force driving the outbreak, combining how the disease spreads with how the weather (rain and heat) makes it worse.

The Problem: Why Old Maps Didn't Work

The paper explains that old ways of predicting cholera were like using a flat, static map to navigate a stormy ocean.

Old Models (Deterministic SIR): These assumed everything was smooth and predictable. They thought if you knew the number of sick people yesterday, you could perfectly guess today. But in Yemen, the situation is chaotic. The population is moving (displaced by conflict), the water is contaminated, and the weather is extreme.
The Reality: The old models failed to capture the "wobbly" nature of real life. They couldn't handle the sudden spikes in cases caused by heavy rains or the delays in reporting data.

The Solution: The "Epidemic Energy" Engine

The new QEE model treats an outbreak like a boiling pot of water.

The Heat (Transmission): This is how fast the disease spreads from person to person.
The Lid (Recovery): This is how fast people get better and stop spreading it.
The External Fire (Climate): This is the "Climate Risk Index." The model found that heat, heavy rain, and humidity act like a blowtorch under the pot. When the weather gets extreme, it adds "energy" to the pot, making the water boil over (an outbreak) much faster.

The model uses a special math formula (called a Stochastic Differential Equation) that adds a little bit of "randomness" to the mix. This is like acknowledging that in a war zone, you can't predict every single event perfectly, so the model accounts for that uncertainty.

How They Built the Model (The "Time Machine")

The researchers faced a big hurdle: Missing Data. In Yemen, records for every single week between 2016 and 2024 weren't available.

The Fix: They acted like forensic detectives. They took the official numbers they did have and used smart math to "fill in the blanks" for the missing weeks, creating a continuous timeline of what likely happened.
The Time Lag: They also realized that when people get sick, it takes about 7 days on average for the case to be officially reported. The model "rewinds" the clock by about a week to see what was actually happening before the report came in, making the predictions more accurate.

The Results: A Better Crystal Ball

The team tested their new "Energy Model" against the real data from the Al-Hodeidah region in Yemen.

Accuracy: The model was incredibly good at guessing what would happen next. It got the right answer about 83% to 92% of the time (a score called R²).
Comparison:
- The old "flat map" models (SIR/SEIR) were off by a lot (about 1,800 cases per week error).
- The new QEE model was off by only about 500 cases per week.
- It beat even complex computer learning programs (Machine Learning) because it actually understood the rules of the disease, rather than just guessing patterns.
Reliability: They ran the simulation 1,000 times with slightly different settings (like rolling dice 1,000 times). In almost every run, the real-world data fell inside the "safe zone" the model predicted.

The "Early Warning" Switch

The paper claims the model can act as a traffic light system for health officials:

Green (Watch): The "energy" is low. Just keep an eye on things.
Yellow (Alert): The "energy" is rising (often because of rain). Get supplies ready.
Red (Action): The "energy" is high. The model predicts an outbreak is coming in about 3 weeks. This gives officials time to send doctors and clean water before the sickness explodes.

What the Model Doesn't Do (Limitations)

The authors are honest about the model's limits:

It's not magic: It relies on the data they could find. If the data was smoothed out to fill gaps, the model might look slightly too perfect.
It simplifies war: It doesn't explicitly track military battles or sudden changes in where people are hiding, because that data is hard to get.
It assumes a steady delay: It assumes it always takes about 7 days to report a case, but during intense fighting, that delay might get longer, which could throw off the prediction slightly.

The Bottom Line

The paper concludes that by turning the chaos of a cholera outbreak into a measurable "energy" function driven by the weather, they have built a scientifically grounded, uncertainty-aware tool. It doesn't just count the sick; it explains why the sickness is surging (the climate) and gives health workers in fragile places a reliable 3-week heads-up to prepare.

Technical Summary: The Quantified Epidemic Energy (QEE) Model for Cholera Forecasting in Yemen

1. Problem Statement

Cholera remains a critical public health threat in fragile, conflict-affected states like Yemen, where transmission is driven by complex interactions between hydro-climatic variability, population displacement, and compromised water, sanitation, and hygiene (WASH) infrastructure. Traditional deterministic compartmental models (e.g., standard SIR/SEIR) often fail in these contexts because they assume population homogeneity, fixed parameters, and continuous transitions. They struggle to capture the systemic volatility, threshold behaviors, and multi-faceted uncertainties inherent in humanitarian crises. Furthermore, predictive accuracy is severely compromised by reporting delays, data truncation, and the lack of high-frequency surveillance data. There is a methodological gap in unifying mechanical transmission dynamics, longitudinal population mobility, and non-linear hydro-climatic drivers into a single, cohesive stochastic framework capable of operating under profound data scarcity.

2. Methodology

2.1 Data Sources and Reconstruction

The study utilized epidemiological data from the World Health Organization (WHO) Yemen Cholera Situation Reports and the Electronic Disease Early Warning System (eDEWS) for Al-Hodeidah Governorate (2016–2024). Hydro-climatic variables (precipitation, temperature, humidity) were sourced from CHIRPS and ERA5 reanalysis data.

Weekly Series Reconstruction: Due to missing weekly surveillance records, a constrained reconstruction framework was applied. This involved zero-incidence constraints, linear interpolation, exponential growth reconstruction, and cumulative-to-weekly differencing, all smoothed via a centered moving average to ensure epidemiological consistency with official annual totals.
Reporting Delay Adjustment: A gamma-distributed delay kernel (mean lag ≈ 7.1 days) was fitted to deconvolve observed incidence, correcting for systematic right-truncation in conflict-affected reporting systems.

2.2 The Climate Risk Index (CRI)

A data-driven Climate Risk Index was developed to quantify environmental forcing. Weights for temperature, rainfall, and humidity were derived directly from observed cross-correlations with cholera incidence (lagged 2–4 weeks).

Weights: Temperature (0.40), Rainfall (0.35), Relative Humidity (0.25).
Function: $CRI_t = 0.4T + 0.35R + 0.25H$ , where anomalies are relative to the 1979–2025 baseline.

2.3 The Quantified Epidemic Energy (QEE) Framework

The core innovation is the conceptualization of epidemic intensity as a dynamic "energy" function ( $E_t$ ) derived from a modified, climate-forced Susceptible-Infectious-Recovered (SIR) system.

Deterministic Forcing: The model defines epidemic energy as the net force driving changes in active cases ($dI/dt$). It couples transmission potential (kinetic energy analogue), recovery-driven dissipation, and environmental potential energy (scaled by CRI and mobility).
Stochastic Formulation: To account for demographic stochasticity and unmodeled socio-behavioral dynamics, the framework employs an Itô Stochastic Differential Equation (SDE):
$dE_t = \left[ \beta_0 \frac{S_t I_t}{N} - \alpha_3 \gamma I_t + \delta (CRI_t I_t) + \alpha_2 (M_t I_t) \right] dt + \sigma E_t dW(t)$
Where $M(t)$ is a normalized mobility index, $\sigma$ is noise intensity, and $W(t)$ is a Wiener process.
Numerical Solution: Due to non-linear coupling, the SDE is solved using the Euler-Maruyama discrete numerical scheme with reflective boundary conditions to prevent unphysical negative values.

2.4 Validation Architecture

Holdout Strategy: The dataset (2016–2024) was partitioned temporally. 2018 served as a structural calibration anchor; all other years were treated as out-of-sample forecasts.
Metrics: Performance was evaluated using Root Mean Square Error (RMSE) and the coefficient of determination ( $R^2$ ).
Uncertainty Quantification: 1,000 Monte Carlo simulations were run to generate 95% probabilistic confidence bands.
Sensitivity Analysis: Parameters ( $\beta, \gamma, \delta, \sigma$ ) were perturbed by $\pm10\%$ to test structural robustness.

3. Key Results

3.1 Predictive Accuracy

The QEE model demonstrated high predictive fidelity across distinct epidemic waves:

$R^2$ Range: 0.83 to 0.92 across validation years.
RMSE: Averaged approximately 500 cases/week (ranging from ~147 to ~947 depending on epidemic intensity).
Comparative Performance: QEE significantly outperformed deterministic SIR ( $R^2=0.65$ ), SEIR ( $R^2=0.72$ ), Bayesian Hierarchical models ( $R^2=0.83$ ), and Gradient Boosting Machines ( $R^2=0.85$ ). The improvement over deterministic models was statistically significant ( $p < 0.05$ ).

3.2 Robustness and Generalizability

Cross-Validation: K-fold cross-validation across five chronological phases (Emergence, Peak, Decline, Low Transmission, Resurgence) showed RMSE variation below 10%, confirming strong generalizability.
Sensitivity: Perturbing core parameters by $\pm10\%$ resulted in less than a 5% shift in performance metrics, indicating structural stability.
Probabilistic Reliability: Observed incidence consistently fell within the 95% confidence bands generated by Monte Carlo ensembles (92–96% coverage).

3.3 Mechanistic Insights

Lag Analysis: After correcting for reporting delays, the optimal forecasting window was identified as a 3-week lag between Climate Risk Index (CRI) anomalies and case surges.
Energy Thresholds: The model successfully identified "threshold breaches" where transmission pressure and climate forcing overcame clinical dissipation, corresponding to major outbreak waves in 2017, 2019, and 2024.

4. Significance and Claims

The paper claims that the QEE framework provides a scientifically grounded, uncertainty-quantified decision-support instrument for proactive public health planning in data-scarce, conflict-affected settings.

Mechanistic Interpretability: Unlike "black box" machine learning models, QEE maintains epidemiological interpretability by linking specific energy components (transmission, recovery, environmental forcing) to outbreak dynamics.
Early Warning Capability: By establishing a 3-week lead time between hydro-climatic anomalies and epidemic energy acceleration, the model offers a reliable window for early warning.
Operational Utility: The authors propose three strategic response thresholds based on normalized epidemic energy:
1. Watch Domain (0.50–0.70): Localized monitoring and water testing.
2. Alert Domain (0.70–0.85): Pre-positioning of oral rehydration points and chlorination.
3. Action Domain (>0.85): Emergency deployment of mobile medical teams.

The study concludes that by mapping epidemiological dynamics onto an environmentally forced energy function, the QEE model establishes a robust baseline for climate-resilient resource allocation in vulnerable regions, effectively bridging the gap between classical deterministic theory and the stochastic reality of humanitarian crises.

5. Limitations Acknowledged by the Authors

The authors explicitly note several limitations:

Data Imputation: The reconstruction of missing weekly data may introduce smoothing bias.
Simplified Indices: The CRI aggregates complex environmental interactions into a linear index.
Conflict Dynamics: Active military frontlines and sudden displacement changes were not explicitly modeled as independent forcing functions due to data constraints.
Stationary Assumptions: The reporting delay model assumes a stationary delay distribution, which may be violated during acute conflict escalations (e.g., the 2018 Hodeidah offensive), potentially leading to slight overestimation of epidemic energy during spikes.
Calibration: The use of a single calibration anchor year (2018) reduces statistical power for parameter inference, though leave-one-year-out tests suggested comparable performance.