A predictive, multi-source, attack-level model to quantify and characterize the injury burden and need for reconstructive surgery in Gaza.

BACKGROUND: Since October 2023, the war in Gaza has produced a massive Palestinian civilian injury burden and created a large need for reconstructive surgery. It has also crippled an already fragile healthcare system and reduced its capacity to provide reconstructive care. Effective planning to address this gap requires a consensus on the volume and pattern of injuries, and the ability to forecast future injuries. However, acquiring accurate granular data is challenging in conflict settings. This study computationally estimates and characterizes injuries in Gaza to serve as a comparison to reported figures, forecast future injuries, and aid in planning to meet reconstructive needs. METHODS: A multivariate negative binomial regression model was built using attack data, geospatial mapping, dynamic population density estimates and evolving infrastructure classifications from humanitarian, governmental, and media sources. The model was trained on data from October 2023-March 2024, validated in April 2024, and tested from May 2024-May 2025. The primary outcome was daily injury counts in Gaza, modeled as a function of attack counts, attack types, population density, infrastructure, and geographically moving areas of conflict. Forecasts through May 2026 were generated under varying trajectories of conflict intensity. FINDINGS: For October 7th, 2023-May 1st, 2025, our model predicted 116,020 injuries (10% sensitivity analysis: 108,000-129,000), aligning with the 118,014 reported by the Gaza Ministry of Health (MoH). Injuries were greatest in air and shelling attacks, during periods of actively moving conflict, and in densely populated areas, especially in urban settings before and after they were devastated to rubble. Of those injuries, 29,000-46,000 were predicted to require reconstructive surgery, with over 80% due to explosions, and rising to 34,000-48,000 by May 2026. Model predictions were correlated with observed outcomes, with Spearman's ρ = 0.723 (p = 7.06 × 10-30) in the training set and ρ = 0.487 (p = 3.99 × 10-23) in the testing set. Event rates: 75,982 reported injuries in training period; 2523 injuries in validation period; 41,202 injuries in testing period. INTERPRETATION: To our knowledge this is the first predictive model to forecast the injury burden in Gaza based on attack characteristics using granular multi-source data. Our findings corroborate figures published by the Gaza MoH and demonstrate that the burden will continue to grow without cessation of hostilities, further exacerbating reconstructive surgical need. FUNDING: Bass Connections grant, Duke University.

Duke Scholars

Author Ash Patel Surgery, Plastic, Maxillofacial, and Oral Surgery

Author Frances S. Hasso Gender, Sexuality & Feminist Studies