Abstract Association rule mining (ARM) offers an interpretable alternative to black-box models for extracting actionable patterns from Intensive Care Unit (ICU) data; however, its application to cardiac surgery ICUs and sex-stratified clinical decision support remains limited. In the perioperative healthcare supply chain, the cardiac surgery ICU is a capacity-constrained downstream node in which ventilator availability and workflow coordination must be managed under uncertainty. To address this gap, we applied the Cross-Industry Standard Process for Data Mining (CRISP-DM) framework to a cardiac surgery ICU dataset (1092 records; 22 retained attributes after cleaning) to mine clinically interpretable rules associated with the target outcome, “onₚump”, a binary attribute indicating whether a ventilator is connected to the patient. Continuous variables were discretized into clinically meaningful grades and categorical variables were one-hot encoded to form patient transactions. Frequent itemsets were extracted using the Apriori algorithm and converted into association rules, retaining only rules with consequent onₚump under minimum support = 0. 05 and minimum lift = 1. 0. Rule mining was performed for the full dataset and separately within female and male subgroups. Across sex strata, 62 unique antecedent→onₚump rules were identified (31 female-only, 29 shared, 2 male-only), corresponding to 60 rules in females and 31 in males. After Fisher’s exact testing with Benjamini–Hochberg False Discovery Rate (FDR) adjustment, 52 of 60 female rules and 25 of 31 male rules remained statistically significant. High-strength rules were dominated by procedural combinations involving CrossClamp and CABGValve (confidence ≈ 0. 96–0. 97; lift ≈ 2. 4). These interpretable rules can be operationalized as a lightweight decision-support layer for perioperative resource planning (e. g. , ventilator and ICU capacity) within the healthcare supply chain. Overall, the findings support ARM as a transparent, clinician-auditable approach for decision support and hypothesis generation in the cardiac surgery ICU.
Bahrami et al. (Sat,) studied this question.