Bus Jacinta D, de Mol Rudi M, Webb Laura E, Bokkers Eddie A M, Boumans Iris J M M
Animal Production Systems group, Wageningen University & Research, Wageningen, 6700AH, The Netherlands.
Wageningen Livestock Research, Wageningen, 6708WD, The Netherlands.
Porcine Health Manag. 2025 Aug 7;11(1):44. doi: 10.1186/s40813-025-00456-3.
Modern sensor technologies and algorithms have the potential to continuously monitor indicators of individual animal welfare, but in growing-finishing pigs the validity of such welfare monitoring remains low for unclear reasons. This study explored how sudden deviations in individual pig feeding behaviour, detected as alerts by a dynamic linear model with Kalman filter, relate to the onset of welfare issues.
Alerts frequencies varied across feeding behaviour components, with higher occurrences for feed intake, feeding duration, feeding rate and night intake (approximately 14.5% of days with an alert) than for feeding frequency and circadian rhythm strength (approximately 7.7% of days with an alert). Limited temporal overlap was observed between feeding components (mean co-occurrence: 18 ± 2%, range 8-47%). Alert prevalence was lower in the first month of the growing-finishing phase for all feeding components except frequency and night intake, which showed opposing patterns. Substantial inter-individual variation in alert prevalence was observed (range: 1.1 - 22.2% alert days). The co-occurrence of alerts and welfare issue onsets, i.e. the sensitivity, was poor and not better than expected by chance (health issues: 2.0 - 48.7%, heat stress: 6.4 - 23.2%). Highest sensitivities were obtained for combinations of coughing, ear tip damage, lameness, rectal prolapse, or tail damage with feed intake, feeding duration or night intake. Sensitivities dropped further when only positive (range: 0.0 - 30.8%) or negative (range: 1.0 - 33.3%) alerts were considered. Sensitivities beyond chance expectations were obtained for feed intake, feeding duration and/or night intake in lame or tail-bitten pigs with specific feeding strategies (range: 4.6 - 66.7%).
These results suggest that sudden deviations in feeding behaviour co-occur poorly with onsets of health issues and heat stress, and that current achievements may be largely based on statistical probabilities. However, mild sensitivities were identified for certain health issues and feeding components, especially for pigs with specific feeding strategies. In addition, the results imply that different types of deviations exist, which differ in suddenness and persistence across the welfare issues' duration. Considering the importance of individual feeding strategies and basal feeding behaviour, stimulating more consistent basal behaviour by improving pigs' housing conditions may reveal new avenues for continuous welfare monitoring.
现代传感器技术和算法有潜力持续监测个体动物福利指标，但在生长育肥猪中，此类福利监测的有效性仍较低，原因不明。本研究探讨了通过带有卡尔曼滤波器的动态线性模型检测到的个体猪采食行为的突然偏差与福利问题发作之间的关系。
警报频率在采食行为各组成部分中有所不同，采食量、采食持续时间、采食速率和夜间采食量（约14.5%的天数有警报）的警报发生率高于采食频率和昼夜节律强度（约7.7%的天数有警报）。在采食各组成部分之间观察到有限的时间重叠（平均同时发生：18±2%，范围8 - 47%）。除采食频率和夜间采食量呈现相反模式外，在生长育肥阶段的第一个月，所有采食组成部分的警报发生率均较低。观察到警报发生率存在显著的个体间差异（范围：1.1 - 22.2%的警报天数）。警报与福利问题发作的同时发生情况，即敏感性较差，并不比偶然预期的更好（健康问题：2.0 - 48.7%，热应激：6.4 - 23.2%）。对于咳嗽、耳尖损伤、跛行、直肠脱垂或尾巴损伤与采食量、采食持续时间或夜间采食量的组合，获得了最高的敏感性。当仅考虑阳性（范围：0.0 - 30.8%）或阴性（范围：1.0 - 33.3%）警报时，敏感性进一步下降。对于具有特定采食策略的跛行或咬尾猪，采食量、采食持续时间和/或夜间采食量获得了超出偶然预期的敏感性（范围：4.6 - 66.7%）。
这些结果表明，采食行为的突然偏差与健康问题和热应激发作的同时发生情况较差，当前的成果可能很大程度上基于统计概率。然而，对于某些健康问题和采食组成部分，尤其是具有特定采食策略的猪，发现了轻度的敏感性。此外，结果表明存在不同类型的偏差，这些偏差在福利问题持续期间的突然性和持续性方面存在差异。考虑到个体采食策略和基础采食行为的重要性，通过改善猪的饲养条件来刺激更一致的基础行为可能会为持续福利监测开辟新途径。
