Machine Learning Approaches for Predicting NSAID-Induced Multi-System Adverse Drug Reactions in Orthopaedic Patients: A Review

Abstract

Adverse Drug Reactions (ADRs) associated with Non-Steroidal Anti-Inflammatory Drugs (NSAIDs) remain a clinical concern in orthopaedic care due to their association with complications affecting multiple organ systems, including gastrointestinal, renal, cardiovascular, hepatic, and neurological functions. The major concern is that the traditional ap-proaches fail to recognize the disorders and ADRs, which get worse for the patients; hence, with the modern approach of Machine Learning, Artificial Intelligence, Graph Neural Networks and Deep Learning, the predictions have become easier with all the possessed clinical data. This review explores how computational methods are used to predict adverse drug reactions, keeping in mind the techniques applied, datasets used, evaluation measures, and the predictive performance from various carried on studies. Although many studies have worked on predicting adverse drug reactions, there are still gaps. The big issue is that not much importance is given to orthopaedic patients who use NSAIDs for a prolonged duration. Also, most existing models are not good at predicting multiple system-related problems. To make predictions still better here, more attention should be given to each patient’s treatment pattern and models that clinicians can interpret. At last, this framework and model discussed in this survey aim at helping clinicians prescribe NSAIDs more safely purely based on analysed patients’ patterns and personal nature, spot risks before itself, and make the full treatment process a very useful and effective one thereby.

Introduction

This review paper focuses on the prediction of Adverse Drug Reactions (ADRs) caused by prolonged use of Non-Steroidal Anti-Inflammatory Drugs (NSAIDs) in orthopaedic patients. NSAIDs such as Ibuprofen, Diclofenac, Naproxen, and Celecoxib are widely prescribed to manage pain, inflammation, fractures, arthritis, and other musculoskeletal disorders. However, long-term use can lead to serious ADRs affecting multiple organ systems, including the gastrointestinal, renal, cardiovascular, hepatic, and neurological systems. Early detection of these reactions remains challenging due to delayed symptoms, complex patient histories, drug interactions, and limitations in clinical monitoring.

The paper reviews various machine learning (ML), deep learning (DL), natural language processing (NLP), and graph neural network (GNN) approaches used for ADR prediction. Traditional ML models such as Support Vector Machines, Random Forests, Naïve Bayes, Logistic Regression, and XGBoost have shown good performance in identifying ADRs from clinical and pharmacovigilance datasets but struggle with imbalanced data and complex biological relationships. NLP-based systems analyze social media, healthcare forums, and patient reviews to detect ADRs from unstructured text, enabling early signal detection but facing challenges such as informal language and noisy data.

Advanced DL models, including CNNs, Bi-LSTMs, multimodal neural networks, and GNN-based frameworks like DruGNN and AutoDDI, improve ADR prediction by modeling drug-drug, drug-gene, and gene-gene interactions. Although these methods achieve higher accuracy, they often suffer from high computational complexity, lack of explainability, and difficulties in handling unseen drugs or clinical deployment.

The study identifies a significant research gap: most existing ADR prediction systems focus on general drug safety, cancer, or cardiovascular conditions, while orthopaedic patients using NSAIDs are largely overlooked despite being at high risk for multi-system adverse reactions. Challenges include non-specific datasets, data imbalance, scalability issues, and limited clinician trust in black-box AI models.

To address these issues, the paper proposes an orthopaedic-specific ADR prediction framework. The framework uses clinical data such as age, treatment duration, dosage patterns, medical history, and concurrent medications. After data preprocessing and normalization, multiple models—including Logistic Regression, SVM, Random Forest, Gradient Boosting, and Deep Learning algorithms—are trained and compared. The best-performing model is then selected to predict multi-system NSAID-induced ADR risks, enabling earlier detection, safer prescribing decisions, and improved patient outcomes in orthopaedic care.

Conclusion

The reviewed studies demonstrate the effectiveness of using machine learning and deep learning models in predicting ADRs across clinical settings. Still, challenges including data imbalance, limited model interpretability, real-world implementation, and insufficient representation of orthopaedic patients on prolonged medications persist. Future research should focus on di-verse, balanced datasets and interpretable models to develop reliable ADR prediction systems that increase patient safety and support clinical decision-making.

References

The reviewed studies demonstrate the effectiveness of using machine learning and deep learning models in predicting ADRs across clinical settings. Still, challenges including data imbalance, limited model interpretability, real-world implementation, and insufficient representation of orthopaedic patients on prolonged medications persist. Future research should focus on di-verse, balanced datasets and interpretable models to develop reliable ADR prediction systems that increase patient safety and support clinical decision-making.

Copyright

Copyright © 2026 Ankith H S, Bimba Prasad, K Aprameya Hebbar, Santhosh J R, Yashas S. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.