Optimising the selection of welfare indicators in farm animals

Jon E. L. Day^1,2*Mohamed Ben Haddou³Rita Kylling⁴Guro Vasdal⁵Heleen Ariane van de Weerd^1,2

• ¹Cerebrus Advies BV, Dinxperlo, Netherlands
• ²Chronos Sustainability Ltd, Chichester, United Kingdom
• ³Mentis SA, Brussels, Belgium
• ⁴Matprat, Oslo, Norway
• ⁵Animalia AS, Oslo, Norway

Introduction: Risk assessment (RA) frameworks are increasingly applied to improve the welfare of farmed animals. At their core is a logic chain linking welfare hazards (risks) with welfare consequences, each measured by one or more indicators. Effective monitoring often requires selecting a subset of indicators from a large pool. Selecting “iceberg indicators,” associated with multiple consequences, can be advantageous, but no standardised, data-driven method exists to select optimal combinations under practical constraints. This study addresses this gap by creating an algorithmic approach to optimise indicator selection. Methods: The work followed six phases: 1) construction of a structured database of welfare indicators; 2) a proof-of-concept study; 3) design of a greedy selection algorithm; 4) enhancement using branch-and-bound and backtracking; 5) performance and sensitivity testing; and 6) development of case studies. A dataset of 382 indicators across seven farm species was compiled from European Food Safety Authority (EFSA) scientific opinions and published literature. EFSA opinions link indicators with associated hazards and consequences through systematic reviews and expert elicitation. For algorithm development, Coverage was determined as the number of unique consequences linked to each indicator. Metadata on Impact of consequence, Ease of hazard mitigation, and Ease of indicator use were generated via expert elicitation. Data were standardised using max–min normalisation, and an objective function was defined to allow selection according to user-defined criteria. Optimisation was performed using both greedy and enhanced algorithms. Performance and robustness were assessed via sensitivity analyses, scenario testing, and benchmarking. Results: The greedy algorithm was computationally efficient but plateaued in Coverage as more indicators were added. The enhanced algorithm identified globally optimal combinations within 0.2 seconds across all species. In a broiler chicken case study, it excluded moderately difficult indicators. A pig case study showed overlap with greedy outputs but demonstrated added value by identifying high-impact, easy to implement indicators suitable for certification. Discussion: The enhanced algorithm incorporates multiple selection criteria, moving beyond iceberg indicators to provide data-agnostic, flexible optimisation for research, industry, and policy contexts. Future work should refine weighting processes, scenario testing, and stakeholder engagement to maximise both relevance and practicality.