Authors: Geofrey Nyabuto, Charles Kibet Ng'etich, Edwin Seno, George Kihara Mburu, Marion Jeptoo, Joyadams Munene, Muriithi Alex Karani, John kimani Muragu, Nyairo Charles Magati
Abstract: Association rule mining (ARM) is a data mining approach used to discover frequent co-occurrence patterns and conditional relationships in large datasets. In healthcare, ARM has been applied to electronic health records, claims databases, laboratory data, prescription data, disease registries, and public health datasets to reveal clinically meaningful patterns that may support diagnosis, medication safety, risk stratification, and service planning. Objective: This review synthesizes how ARM has been applied in healthcare, focusing on methods, clinical application areas, implementation challenges, and future research directions. A systematic review design guided by PRISMA 2020 was used to structure the manuscript. Literature was organized around peer-reviewed ARM studies in healthcare, including clinical decision support, diagnostic test ordering, disease-medication association mining, adverse drug reaction signal detection, risk factor discovery, hospital readmission analysis, privacy-preserving mining, and emerging causal or hybrid ARM approaches. The literature shows that Apriori remains the most frequently used ARM algorithm, although FP-Growth, weighted Apriori, class association rules, negative association mining, privacy-preserving ARM, and causal irredundant ARM are increasingly used to address computational, interpretability, privacy, and clinical validity limitations. ARM is valuable because it produces transparent IF-THEN rules that clinicians can inspect, but uncontrolled rule generation, weak validation, data quality limitations, and spurious associations remain major barriers. ARM has clear potential in healthcare knowledge discovery and decision support, particularly where interpretability is required. Future research should prioritize external validation, clinician-centered rule evaluation, integration with electronic medical records, explainable hybrid models, privacy-preserving analytics, and evidence from low- and middle-income healthcare settings.
DOI: https://doi.org/10.5281/zenodo.20199703
