**Background:** Yemen has been described as the world's largest humanitarian disaster, with over 24 million people (80% of the population) requiring humanitarian assistance. The civil war, which became internationalized in 2015 with Saudi-led coalition air strikes and Iranian-backed Houthi forces, has destroyed health infrastructure, displaced millions, and pushed nearly 10 million people to the brink of starvation. Despite over five years of war, few attempts had been made to estimate excess mortality due to dysfunctional civil registration and vital statistics systems. This study aimed to determine whether mortality increased significantly during the conflict period (2015–2019) compared to the immediate pre-conflict period (2012–2014).
**Methods:** The researchers aggregated mortality data from 91 small-scale surveys using the Standardized Monitoring and Assessment of Relief and Transitions (SMART) methodology, covering 2,864 geographical clusters across Yemen from 2012–2019. Surveys were sourced from the Complex Emergency Database (CE-DAT) and UN-OCHA, including data from WFP, FAO, the Yemeni Ministry of Health, and UN emergency food security assessments. Twenty-eight surveys were from the pre-conflict period (2012–2014) and 63 from the conflict period (2015–2019). Violent death data were obtained from the Armed Conflict Location & Event Data Project (ACLED) and the Yemen Data Project, covering 2015–2019. A Bayesian Poisson-Gamma model was used to estimate pre-conflict baseline crude death rate (BCDR) and conflict period posterior crude death rate (PCDR), with person-days as an offset variable. The model ran 100,000 iterations with a 50,000 burn-in, and convergence was assessed using Geweke diagnostics. Total excess deaths were calculated by applying the rate difference to the median adjusted population. An insecurity index was constructed using weighted fatalities from six types of violent events (air/drone strikes, battles, explosions/remote violence, protests, riots), classified into quartiles (Insecure, Serious, Severe, Extreme). Association between PCDR and insecurity level was tested using one-way ANOVA.
**Key Results:** The national estimated crude death rate during the conflict period was 0.20 per 10,000 per day (95% CI: 0.17, 0.24), compared to the baseline rate of 0.19 per 10,000 per day (95% CI: 0.17, 0.22). The total excess death rate was 0.03 per 10,000 per day (95% CI: 0.00, 0.07). Applying conflict period rates to the Yemeni population, the researchers estimated 1,115,024 total deaths occurred during the 5-year crisis period, versus 946,812 expected deaths based on baseline rates—a 17.8% increase. This yielded 168,212 excess deaths (approximately 92 excess deaths per day). Of these, 67.2% were attributable to direct combat-related violence, with the remainder due to indirect causes. At the governorate level, PCDR ranged from 0.11 to 0.35 per 10,000 per day. Hajjah, Ibb, Aden, Saada, and Al Jawf governorates presented the highest total excess deaths. The insecurity index was highest along the coastline (Hajjah, Al Hudaydah, Taizz) and in Houthi-controlled areas; Sana'a experienced approximately 4,932 air/drone strikes. However, the one-way ANOVA found no statistically significant association between insecurity level and PCDR across governorates.
**Clinical Implications:** The findings reveal that Yemen's health situation was already poor before the 2015 crisis, with baseline mortality rates higher than regional averages. Unlike many other conflicts where indirect deaths (from disease, malnutrition, health system collapse) vastly outnumber direct violent deaths, Yemen is a rare case where nearly two-thirds of excess deaths were from direct violence—primarily air and ground strikes. The authors suggest that the massive increase in humanitarian aid following international recognition of the crisis may have helped contain indirect mortality, while the sustained barrage of air strikes drove direct fatalities. The lack of association between insecurity level and death rates suggests that the mere threat of violence across all governorates may have disrupted health care and life-sustaining activities nationwide. These results underscore the urgent need for innovative data collection methods in high-insecurity settings and for renewed efforts to understand the human health impacts of the sustained crisis. The study is limited by the exclusion of high-insecurity areas from surveys (over 50% of surveys excluded inaccessible areas), the relatively small number of pre-conflict surveys, and the aggregate analysis that may mask changes in mortality intensity over time.