Home / Current Issue / Paper 1720286
Responsible Artificial Intelligence for Polypharmacy and Medication-Safety Risk in Older Adults: An Explainable FHIR-Based Clinical Decision-Support Framework
Subject area: Biological & Medical Sciences · Area of research: Health Data Science
DOI: https://doi.org/10.64388/IREV10I1-1720286
Abstract
Medication-related harm in older adults is driven by the interaction of multimorbidity, age-related pharmacokinetic change, polypharmacy, potentially inappropriate medications, drug-drug interactions, fall-risk-increasing drugs and impaired renal function. Existing electronic prescribing alerts often treat these risks as isolated rules, generating high alert burden without ranking residents according to the combined probability and clinical significance of harm. This study develops an explainable, FHIR-based and fairness-aware clinical decision-support framework for identifying potentially inappropriate medications, high-risk combinations, excessive medication burden, medication-related fall risk, renal- and age-sensitive prescribing risks, and residents requiring pharmacist or clinical review. Because no single openly accessible Kaggle dataset contains all medication, laboratory, renal, fall and long-term-care variables needed for the full framework, a reproducible synthetic cohort of 10,000 older residents across 30 facilities was calibrated to the Kaggle Drug Adverse Event Detection Dataset, public adverse-drug-event datasets and relationships reported in geriatric pharmacotherapy literature. Logistic regression, random forest and gradient boosting models were compared using facility-level holdout validation. Performance was assessed with AUROC, area under the precision-recall curve, Brier score, sensitivity, specificity, positive predictive value and F1 score. Fairness was evaluated by sex, race and chronic-kidney-disease status at a capacity-constrained top-quintile review threshold. Logistic regression produced the most transportable discrimination across medication harm, medication-related falls and pharmacist-review need, while calibration and subgroup analysis showed that average performance alone was insufficient for safe deployment. The framework maps medication and clinical inputs to FHIR resources and produces an explanation packet that identifies risk level, contributing medicines, interaction mechanisms, renal-dose concerns, data quality and recommended human review. The study offers a transparent foundation for prospective validation in nursing homes, ambulatory geriatric practice and post-acute care.
Keywords
Polypharmacy, Medication Safety, Older Adults, Potentially Inappropriate Medications, Adverse Drug Events, Falls, Renal Dosing, Explainable Artificial Intelligence, FHIR, Clinical Decision Support.
How to cite this paper
@article{1720286,
author = {Kwame Ofori Boakye, Grace Mupa, Rumbidzai Lyn Kasinamunda, Rudorwashe Tsitsi Karuma, Munashe Naphtali Mupa},
title = {Responsible Artificial Intelligence for Polypharmacy and Medication-Safety Risk in Older Adults: An Explainable FHIR-Based Clinical Decision-Support Framework},
journal = {Iconic Research And Engineering Journals},
year = {2026},
volume = {10},
number = {1},
pages = {4031-4044},
issn = {2456-8880},
url = {https://www.irejournals.com/formatedpaper/1720286.pdf},
abstract = {Medication-related harm in older adults is driven by the interaction of multimorbidity, age-related pharmacokinetic change, polypharmacy, potentially inappropriate medications, drug-drug interactions, fall-risk-increasing drugs and impaired renal function. Existing electronic prescribing alerts often treat these risks as isolated rules, generating high alert burden without ranking residents according to the combined probability and clinical significance of harm. This study develops an explainable, FHIR-based and fairness-aware clinical decision-support framework for identifying potentially inappropriate medications, high-risk combinations, excessive medication burden, medication-related fall risk, renal- and age-sensitive prescribing risks, and residents requiring pharmacist or clinical review. Because no single openly accessible Kaggle dataset contains all medication, laboratory, renal, fall and long-term-care variables needed for the full framework, a reproducible synthetic cohort of 10,000 older residents across 30 facilities was calibrated to the Kaggle Drug Adverse Event Detection Dataset, public adverse-drug-event datasets and relationships reported in geriatric pharmacotherapy literature. Logistic regression, random forest and gradient boosting models were compared using facility-level holdout validation. Performance was assessed with AUROC, area under the precision-recall curve, Brier score, sensitivity, specificity, positive predictive value and F1 score. Fairness was evaluated by sex, race and chronic-kidney-disease status at a capacity-constrained top-quintile review threshold. Logistic regression produced the most transportable discrimination across medication harm, medication-related falls and pharmacist-review need, while calibration and subgroup analysis showed that average performance alone was insufficient for safe deployment. The framework maps medication and clinical inputs to FHIR resources and produces an explanation packet that identifies risk level, contributing medicines, interaction mechanisms, renal-dose concerns, data quality and recommended human review. The study offers a transparent foundation for prospective validation in nursing homes, ambulatory geriatric practice and post-acute care.},
keywords = {Polypharmacy, Medication Safety, Older Adults, Potentially Inappropriate Medications, Adverse Drug Events, Falls, Renal Dosing, Explainable Artificial Intelligence, FHIR, Clinical Decision Support.},
month = {July},
doi = {https://doi.org/10.64388/IREV10I1-1720286}
}