WA2 - Santé / Healthcare 1

Systematic review synthesizing 33 studies applying process mining to healthcare logistics. We map logistics domains, PM methods, KPIs, and gaps, highlighting predominance of single-site retrospective studies, inconsistent event-log reporting, and limited economic/environmental evaluations. We propose optimization-focused research directions for scalable, data-governed logistics improvements and stakeholder engagement across diverse hospital settings.

In autoclave sterilization, medical instruments must be loaded into non-stacking, height-adjustable racks while preserving similarity to validated historical configurations. We formulate this as a 3D bin-packing problem with non-overlap constraints, solved via CP-SAT. A geometric heuristic and MIP-based initialization framework leverage historical templates to improve feasibility, runtime, and layout similarity.

Emergency Departments (EDs) must deliver timely care under highly variable demand while ensuring fair and sustainable working conditions for clinicians and nursing staff. Traditional rostering models primarily emphasize coverage and fairness, whereas the service quality delivered during a shift may also depend on the capability profile of the on-duty team (e.g., competencies, seniority, and experience). This study explores the integration of service quality considerations into ED staff rostering with the objective of levelling the expected capability across shifts while maintaining equity in assignments.
We formulate an integrated staff rostering problem that assigns physicians and nurses to shifts over a finite planning horizon. The model incorporates legal and organisational constraints (e.g., working-time rules, minimum rest periods, and shift succession), coverage and competency requirements, and preference-related constraints. Service quality levelling is modelled through indicators derived from staff capability profiles, with the objective of reducing variability in shift-level capability while balancing workload distribution and undesirable shifts. The proposed framework is designed to be evaluated using data from a Canadian hospital Emergency Department.
Current work focuses on implementing the optimization model and defining capability indicators reflecting team composition during each shift. The forthcoming computational experiments will explore how incorporating service-quality levelling influences roster feasibility, workload equity, and shift composition under realistic operational constraints. The analysis will also examine potential trade-offs between traditional rostering objectives and service capability balancing.
This study proposes a modelling framework that explicitly incorporates service quality considerations into ED staff rostering. By accounting for clinician and nursing profiles in shift composition, the approach aims to support more balanced and robust staffing plans. The model provides a foundation for future empirical evaluation, including analyses of downstream operational performance (e.g., waiting times, length of stay, left-without-being-seen) and, where data linkage is feasible, exploratory assessment of patient outcome proxies.

Protein complexes play a crucial role in cellular processes, and their accurate identification is essential for understanding biological systems. Most existing computational approaches for detecting protein complexes in Protein-Protein Interaction (PPI) networks rely on topological features and assume non-overlapping structures, assigning each protein to a single complex. However, this assumption does not reflect biological reality, where proteins often participate in multiple complexes.

In this work, we propose a novel optimization-based model for detecting overlapping protein complexes by leveraging connectivity-driven measures derived from PPI networks. To enhance the exploration capability of the search process, we introduce a new heuristic mutation operator that improves solution diversity and avoids premature convergence.

The proposed method is evaluated on three benchmark PPI datasets (Yeast-D1, Yeast-D2, and Collins) using two reference standards (MIPS and CYC2008). Experimental results demonstrate that our approach outperforms several state-of-the-art methods in terms of accuracy and quality of detected complexes.

These results highlight the effectiveness of integrating tailored heuristic operators within evolutionary optimization frameworks for uncovering biologically meaningful overlapping structures in PPI networks.