Data-driven logistics can cut medical risk in disaster zones

A new peer-reviewed study published in the journal Algorithms signals a major shift in how humanitarian logistics can be planned under extreme uncertainty, introducing a data-driven model that blends machine learning with advanced optimization techniques to improve casualty evacuation, medical treatment, and shelter allocation after disasters.

Titled Data-Driven Planning for Casualty Evacuation and Treatment in Sustainable Humanitarian Logistics, the study proposes a comprehensive framework designed to support emergency planners faced with volatile conditions, damaged infrastructure, and incomplete information.

A new model for managing chaos after disasters

The study introduces a bi-objective planning model that seeks to balance two competing priorities that define disaster response: minimizing health risks to affected populations and controlling the operational costs of emergency logistics. Rather than focusing on a single outcome, the model simultaneously evaluates how delays in treatment increase injury severity while also accounting for the financial and operational limits faced by humanitarian agencies.

Unlike many previous models, the framework explicitly distinguishes between severely injured casualties, lightly injured casualties, and homeless individuals. Each group follows different evacuation and care pathways. Severely injured people may require helicopter transport directly to hospitals or staged transfers through temporary clinics. Lightly injured individuals are typically routed to clinics first, then to shelters. Homeless populations require rapid shelter allocation but may not need immediate medical intervention. By embedding these differences into the planning structure, the model reflects real-world disaster dynamics rather than simplified averages.

The logistics network designed in the study includes hospitals, temporary clinics, shelters, and warehouses supplying medical goods. Decisions about where to locate these facilities, how many to activate, and how to route vehicles and helicopters are made simultaneously. This integrated approach contrasts with earlier research that treated facility location, evacuation, and supply distribution as separate problems, often leading to fragmented or inefficient outcomes.

A critical innovation lies in how the model handles uncertainty. Disaster environments are characterized by unknown casualty numbers, fluctuating travel times, uncertain medical demand, and the risk of facility disruptions due to aftershocks or infrastructure failure. Instead of applying a single uncertainty method across all parameters, the authors adopt a hybrid robust optimization strategy. Scenario-based robustness is used for facility disruption risks, box-type uncertainty addresses bounded variations such as travel time and casualty estimates, and polyhedral uncertainty sets manage medical supply demand where flexibility is essential. This tailored approach allows the model to remain realistic without becoming overly conservative.

Machine learning enters humanitarian logistics

The study integrates machine learning into humanitarian logistics planning. The researchers employ unsupervised clustering techniques to group casualties and homeless individuals based on geographic location and injury severity. These clusters serve as demand nodes within the optimization model, replacing the impractical task of modeling every affected individual separately.

This data-driven clustering improves both accuracy and speed. By identifying spatial concentrations of need, the model can allocate resources more efficiently and reduce unnecessary transport times. The approach also allows planners to estimate demand even when detailed individual-level data is unavailable, a common limitation in the immediate aftermath of disasters.

The clustering process is not arbitrary. The researchers use objective validation methods to determine the optimal number of clusters, ensuring that each group represents a meaningful concentration of affected individuals. This step is critical because poorly defined clusters could distort demand estimates and lead to misallocation of resources.

Once clusters are established, the model uses geographic distances and assumed vehicle speeds to calculate travel times for ambulances and helicopters. These times feed directly into the health risk calculations, linking spatial decisions to medical outcomes. In effect, the model translates geography into survival probabilities, making distance and access central to emergency planning.

Machine learning is not a replacement for human judgment but a tool that strengthens decision-making under pressure, the study asserts. By combining data-driven insights with robust optimization, the framework allows planners to explore trade-offs between speed, cost, and coverage before deploying resources in the field.

Evidence from a real earthquake case

To test the model’s real-world applicability, the researchers apply it to the 2017 Kermanshah earthquake in Iran, one of the country’s most devastating recent disasters. The event resulted in hundreds of deaths, thousands of injuries, and tens of thousands of people left homeless, while damaging hospitals and transport infrastructure across the region.

Using realistic assumptions derived from the disaster context, the model evaluates multiple scenarios reflecting different levels of uncertainty and facility disruption. The results show that the proposed hybrid robust model consistently produces feasible and stable solutions across scenarios, even when conditions worsen.

Comparative analysis reveals that deterministic models, which rely on fixed estimates, frequently fail when applied to realized disaster conditions. As uncertainty increases, these models either underestimate demand or allocate insufficient capacity, leading to infeasible solutions. In contrast, the hybrid robust approach maintains functionality by anticipating variation and planning for worst-case deviations within controlled limits.

The study also compares the proposed framework with combined robust-stochastic methods. While stochastic models perform well when reliable probability distributions are available, such data is rarely accessible in disaster settings. The findings show that the new approach achieves more consistent outcomes across scenarios, with only marginal increases in cost relative to less robust alternatives.

The results highlight critical policy insights too. Increases in uncertainty around severely injured casualties have a disproportionately large impact on overall health risk, underscoring the need to prioritize rapid evacuation and treatment for this group. Sensitivity analysis further shows that expanding clinic capacity can significantly reduce health risk, while modest increases in budget can eliminate the number of unevacuated casualties entirely.

The model also demonstrates the importance of multimodal transportation. Helicopter deployment, while costly, sharply reduces deterioration for critically injured patients and improves survival outcomes. The findings suggest that investing in air evacuation capacity can yield outsized benefits in high-impact disasters.