Explainable Artificial Intelligence (XAI) has emerged as a promising approach for understanding the importance of features in Machine Learning (ML) models. In this study, researchers developed an XAI-based technique for pharmacovigilance monitoring by predicting adverse outcomes using individual health information. The study focused on predicting the probability of Acute Coronary Syndrome (ACS) adverse outcomes based on an individual's drug history and comorbidities. To achieve this, linked Western Australian datasets were utilized to extract relevant data, including drug histories, comorbidities, and other key features of individuals aged over 65 who were prescribed Musculo-skeletal system or Cardiovascular system drugs between 1993 and 2009. Multiple ML models were trained to predict ACS-related adverse outcomes such as death or hospitalization with a discharge diagnosis of ACS. To determine the contribution of specific drugs to these predictions, various ML and XAI techniques were employed. Notably, XAI libraries LIME and SHAP were used to identify important and unimportant features, with SHAP slightly outperforming LIME in this context. The analysis revealed that the drug dispensing features for rofecoxib and celecoxib had a significant positive contribution to ACS-related adverse outcome predictions on average. Moreover, the study achieved a prediction accuracy rate of 72% for ACS-related adverse outcomes. These findings demonstrate that ML models trained on linked administrative health datasets in conjunction with XAI algorithms can effectively quantify feature importance. This research lays the foundation for potential applications in pharmacovigilance monitoring techniques. With further refinement and development, these methods could contribute to improved understanding and detection of adverse outcomes associated with specific medications.
Explainable Artificial Intelligence (XAI) is a way to understand how important certain things are in computer programs that learn. Researchers made a new technique using XAI to predict bad outcomes from medicine using people's health information. They focused on predicting the chance of bad heart problems based on what medicines people took and other health issues they had. They used information from Western Australia about people over 65 who took certain medicines between 1993 and 2009. They trained many computer models to predict if someone would have a bad heart problem, and found that some medicines like rofecoxib and celecoxib were more likely to cause these problems. The models were able to predict with 72% accuracy if someone would have a bad heart problem. This research can help make better ways to monitor medicine safety."
Definitions
- Explainable Artificial Intelligence (XAI): A method for understanding how important certain things are in computer programs that learn.
- Pharmacovigilance: Monitoring the safety of medicines.
- Adverse outcomes: Bad results or side effects.
- Acute Coronary Syndrome (ACS): Serious heart problems.
- Comorbidities: Other health issues someone has at the same time as another condition.
- Musculo-skeletal system: The bones, muscles, and joints in the body.
- Cardiovascular system: The heart and blood vessels in the body.
- ML models: Computer programs that learn from data to make predictions or decisions.
- SHAP and LIME:
Introduction
Artificial Intelligence (AI) has revolutionized many industries, including healthcare. In recent years, there has been a growing interest in Explainable Artificial Intelligence (XAI), which aims to provide transparency and interpretability in AI systems. XAI allows us to understand the decision-making process of AI models, making them more trustworthy and reliable for critical applications such as healthcare.
One area where XAI is gaining traction is pharmacovigilance monitoring - the process of monitoring and evaluating adverse drug reactions (ADRs). ADRs are a significant public health concern, with an estimated 5% of hospital admissions being due to ADRs. Therefore, it is crucial to develop effective methods for detecting and predicting ADRs.
In this study, researchers utilized XAI techniques to predict Acute Coronary Syndrome (ACS)-related adverse outcomes using individual health information. The goal was to identify the contribution of specific drugs in predicting these outcomes, providing valuable insights for pharmacovigilance monitoring.
Methodology
The study used linked administrative health datasets from Western Australia between 1993 and 2009. These datasets included drug histories, comorbidities, and other key features of individuals aged over 65 who were prescribed Musculo-skeletal system or Cardiovascular system drugs during this period.
Multiple Machine Learning (ML) models were trained on this data to predict ACS-related adverse outcomes such as death or hospitalization with a discharge diagnosis of ACS. To determine the importance of specific drugs in these predictions, various ML and XAI techniques were employed.
Two popular XAI libraries - LIME (Local Interpretable Model-Agnostic Explanations) and SHAP (SHapley Additive exPlanations) - were used to identify important and unimportant features in the ML models' predictions.
LIME
LIME works by creating a local linear approximation of the ML model's predictions for each instance. It then weights the features based on their contribution to the prediction, providing an explanation for that particular instance.
SHAP
SHAP uses game theory to assign a value to each feature in a prediction. It considers all possible combinations of features and calculates their contributions, resulting in more accurate explanations compared to LIME.
Results
The analysis revealed that the drug dispensing features for rofecoxib and celecoxib had a significant positive contribution to ACS-related adverse outcome predictions on average. This finding is consistent with previous studies that have linked these drugs to increased risk of cardiovascular events.
Moreover, the study achieved a prediction accuracy rate of 72% for ACS-related adverse outcomes, demonstrating the effectiveness of ML models trained on linked administrative health datasets in conjunction with XAI algorithms.
Implications
This research has several implications for pharmacovigilance monitoring techniques. Firstly, it highlights the potential of using XAI techniques in understanding and quantifying feature importance in predicting ADRs. This can aid healthcare professionals in identifying high-risk medications and implementing appropriate measures to mitigate potential harm.
Secondly, this study lays the foundation for further development and refinement of XAI-based methods for pharmacovigilance monitoring. With advancements in AI technology and access to larger datasets, we can expect even more accurate predictions and insights into ADRs associated with specific medications.
Lastly, this research also emphasizes the importance of transparency and interpretability in AI systems used in critical applications such as healthcare. By providing explanations for its decisions, XAI can increase trustworthiness and acceptance of AI systems among healthcare professionals and patients alike.
Conclusion
In conclusion, this study demonstrates how XAI techniques can be effectively applied in pharmacovigilance monitoring. By identifying the contribution of specific drugs in predicting ACS-related adverse outcomes, this research provides valuable insights for healthcare professionals and lays the foundation for potential applications in ADR detection and prevention.
With further refinement and development, XAI-based methods can play a crucial role in improving our understanding of ADRs and their associated risk factors. This has the potential to significantly impact patient safety and contribute to better overall health outcomes.
Explainable Artificial Intelligence (XAI) has been identified as a viable method for determining the importance of features when making predictions using Machine Learning (ML) models. In this study, we created models that take an individual's health information (e.g. their drug history and comorbidities) as inputs, and predict the probability that the individual will have an Acute Coronary Syndrome (ACS) adverse outcome. Using XAI, we quantified the contribution that specific drugs had on these ACS predictions, thus creating an XAI-based technique for pharmacovigilance monitoring, using ACS as an example of the adverse outcome to detect. Individuals aged over 65 who were supplied Musculo-skeletal system (anatomical therapeutic chemical (ATC) class M) or Cardiovascular system (ATC class C) drugs between 1993 and 2009 were identified, and their drug histories, comorbidities, and other key features were extracted from linked Western Australian datasets. Multiple ML models were trained to predict if these individuals would have an ACS related adverse outcome (i.e., death or hospitalisation with a discharge diagnosis of ACS), and a variety of ML and XAI techniques were used to calculate which features -- specifically which drugs -- led to these predictions. The drug dispensing features for rofecoxib and celecoxib were found to have a greater than zero contribution to ACS related adverse outcome predictions (on average), and it was found that ACS related adverse outcomes can be predicted with 72% accuracy. Furthermore, the XAI libraries LIME and SHAP were found to successfully identify both important and unimportant features, with SHAP slightly outperforming LIME. ML models trained on linked administrative health datasets in tandem with XAI algorithms can successfully quantify feature importance, and with further development, could potentially be used as pharmacovigilance monitoring techniques.
