Abstract

Introduction

It was widely reported that adverse drug reactions remained one of the most significant causes of morbidity and mortality among elderly patients globally, with disproportionate impact in low resource healthcare systems. Elderly individuals reportedly experienced higher risk of adverse drug reactions due to age related physiological changes, polypharmacy, and increased prevalence of chronic diseases requiring multiple medications (Mangoni & Jackson, 2004). These physiological changes reportedly affected drug absorption, distribution, metabolism, and elimination, thereby increasing susceptibility to toxicity and unpredictable drug responses. In Nigeria and similar healthcare contexts, limited pharmacovigilance infrastructure and inadequate monitoring systems reportedly compounded the risk, resulting in under detection and under reporting of adverse drug reactions among elderly populations (Fadare et al., 2015). The central goal of this study was reportedly defined as evaluating the performance of machine learning models in predicting adverse drug reactions among elderly patients using simulated clinical and pharmacological datasets relevant to low resource healthcare environments. Machine learning was reportedly described as a branch of artificial intelligence that enabled computer systems to learn patterns from data and generate predictive models without explicit programming instructions (Obermeyer & Emanuel, 2016). Scholars reportedly argued that machine learning models possessed the capacity to analyze complex multidimensional clinical datasets, identify hidden patterns, and predict adverse drug reactions with greater accuracy than traditional statistical methods.

It was reported that traditional pharmacovigilance systems relied heavily on spontaneous reporting, which suffered from significant limitations including under reporting, reporting bias, and delayed detection of adverse drug reactions. Hazell and Shakir (2006) reportedly estimated that only approximately 6 percent of adverse drug reactions were formally reported, indicating substantial gaps in surveillance. These limitations reportedly highlighted the need for predictive approaches capable of identifying high risk patients before adverse reactions occurred. Machine learning reportedly offered such predictive capability by analyzing patient demographics, medication profiles, laboratory values, and comorbidity patterns to generate individualized risk predictions.

The theoretical framework of this study reportedly integrated Predictive Analytics Theory and the Clinical Risk Prediction Model Framework. Predictive Analytics Theory reportedly posited that data driven models could identify patterns and relationships that were not immediately apparent through conventional statistical analysis, thereby improving predictive accuracy (Shmueli & Koppius, 2011). This theory reportedly provided the conceptual foundation for applying machine learning algorithms to pharmacovigilance data. The Clinical Risk Prediction Model Framework reportedly emphasized the importance of identifying risk factors, quantifying their contribution to adverse outcomes, and generating risk scores to guide clinical decision making (Steyerberg, 2019). Together, these frameworks reportedly supported the use of machine learning models to enhance adverse drug reaction prediction and prevention.

Machine learning models such as logistic regression, decision trees, random forests, and support vector machines were reportedly increasingly applied in healthcare predictive modeling. Logistic regression reportedly provided interpretable probability based predictions, while decision trees and random forests reportedly enabled identification of nonlinear relationships between variables. Random forest models reportedly demonstrated particularly strong performance in clinical prediction tasks due to their ability to handle complex interactions and reduce overfitting (Breiman, 2001). Support vector machines reportedly offered robust classification performance, especially in high dimensional datasets.

It was further reported that elderly patients represented an ideal population for machine learning based adverse drug reaction prediction due to the presence of multiple risk factors. Polypharmacy reportedly emerged as one of the strongest predictors of adverse drug reactions, with studies indicating that patients taking five or more medications experienced significantly increased risk (Maher et al., 2014). Additionally, comorbid conditions such as renal impairment and hepatic dysfunction reportedly altered drug metabolism, further increasing adverse reaction risk.

In low resource healthcare systems, implementation of machine learning models reportedly faced several challenges, including limited electronic health records, inadequate data infrastructure, and lack of technical expertise. However, scholars reportedly emphasized that even relatively simple machine learning models could significantly improve risk prediction compared to traditional methods when applied appropriately (Rajkomar et al., 2019). These findings reportedly suggested that machine learning could provide valuable support for clinical decision making even in resource constrained environments.

It was also reported that predictive machine learning models could improve patient safety by enabling early identification of high risk patients and facilitating preventive interventions such as medication adjustment, enhanced monitoring, or alternative therapy selection. This proactive approach reportedly represented a shift from reactive pharmacovigilance to predictive pharmacovigilance, with potential to significantly reduce adverse drug reaction incidence.

The introduction reportedly emphasized that while machine learning models demonstrated promising predictive performance in developed healthcare systems, limited evidence existed regarding their applicability in low resource settings. Differences in patient demographics, disease patterns, medication use, and healthcare infrastructure reportedly necessitated context specific evaluation. Therefore, this study reportedly aimed to evaluate and compare the predictive performance of multiple machine learning models in identifying elderly patients at risk of adverse drug reactions using simulated datasets reflective of low resource healthcare environments.

Ultimately, the study reportedly sought to contribute to improved pharmacovigilance, patient safety, and clinical decision support by demonstrating the potential of machine learning based predictive models to enhance adverse drug reaction detection and prevention among elderly patients in resource limited healthcare systems