Context: Mustering (gathering) feral horses (Equus ferus caballus) often cause mortalities, yet cause-specific details are lacking.

Aims: Given the need to optimise horse welfare, we analysed public horse muster data from the USA to understand specific causes of mortalities.

Methods: We coded 393 individual horse mortality reports for 92 cause-specific mortality terms (keywords informing the deciphering of specific causes of mortality classified as anatomical, causal or conditional) and demographic details (age, sex, and body condition). Data were derived from 50 musters across seven states with at least one horse mortality. Musters were coded for type (helicopter or bait), emergency or regular planned efforts, and number of horses mustered and shipped daily.

Key Results: More horses were euthanased than died naturally (330 (84.0%) and 39 (9.9%) respectively), and more horses had chronic than acute conditions (317 (80.7%) and 76 (19.3%) respectively), with both trends holding for both sexes and across ages. Body condition scores (BCS) for female horses were skewed low, whereas male horse BCS data were more normally distributed. Female horses had lower BCS than did male horses (P < 0.001). On average, each horse mortality had two cause-specific mortality terms, ranging from 1 to 7. Only 57 horses (14.5%) had terms describing anatomy, cause and condition, concurrently. Phi coefficients (φ; indicators of fidelity and constancy) for cause-specific terms were related to demographic or muster attributes and were analysed with post hoc ANOVA tests of estimated marginal means to allow for ranking. Female horses were most often described as emaciated, weak, and starving, whereas male horses were described as lame, arthritic, blind or dangerous. Bait trapping and emergency musters included horses that were starving, dehydrated and weak.

Conclusions: Generally, disorders associated with legs and feet, eyes, necks and nutrition were the most prevalent cause-specific mortality issues. Using a machine learning approach, validation and test accuracy were high for predicting euthanasia versus natural mortalities, but low for predicting acute versus chronic mortalities. Individual horse demographics or daily muster features had a greater relative influence than did capture type or emergency status in both comparisons.

Implications: These results provide practical insight for potential cause-specific mortalities relative to demographics and muster techniques.

背景：集结（聚集）野马（Equus ferus caballus）通常会导致死亡，但缺乏具体原因的细节。

目标：鉴于需要优化马匹福利，我们分析了来自美国的公共马匹集合数据以了解死亡的具体原因。

方法：我们对 393 份马匹死亡率报告进行了编码，涉及 92 个特定原因的死亡率术语（关键字，用于解释被分类为解剖、因果或条件的特定死亡原因）和人口统计信息（年龄、性别和身体状况）。数据来自七个州的 50 次集合，其中至少有一匹马死亡。集合被编码为类型（直升机或诱饵），紧急或定期计划的努力，以及每天集合和运送的马匹数量。

主要结果：安乐死的马比自然死亡的马多（分别为 330 (84.0%) 和 39 (9.9%)），患有慢性疾病的马多于急性疾病（分别为 317 (80.7%) 和 76 (19.3%)），两者均男女和不同年龄段的趋势。雌性马的身体状况评分 (BCS) 偏低，而雄性马的 BCS 数据更呈正态分布。雌性马的 BCS 低于雄性马（P < 0.001)。平均而言，每匹马的死亡率都有两个特定原因的死亡率术语，范围从 1 到 7。只有 57 匹马 (14.5%) 同时具有描述解剖结构、原因和状况的术语。特定原因术语的 Phi 系数 (φ；保真度和稳定性指标) 与人口统计或集合属性相关，并使用估计边际均值的事后 ANOVA 检验进行分析以允许排名。雌性马最常被描述为消瘦、虚弱和饥饿，而雄性马则被描述为跛足、关节炎、失明或危险。诱饵诱捕和紧急召集包括饥饿、脱水和虚弱的马。

结论：通常，与腿和脚、眼睛、颈部和营养相关的疾病是最普遍的特定原因死亡率问题。使用机器学习方法，在预测安乐死与自然死亡方面的验证和测试准确性很高，但在预测急性与慢性死亡方面的准确性较低。在这两种比较中，与捕获类型或紧急状态相比，单个马的人口统计数据或每日集合特征具有更大的相对影响。

含义：这些结果为与人口统计和集合技术相关的潜在原因特定死亡率提供了实用的见解。