Bayer’s Michael Levy examines how artificial intelligence (AI) can assist in both the routine and advanced analytical tasks involved in conducting pharmacovigilance and therefore better ensure the safety of pharmaceuticals and the health of patients.
Pharmacovigilance or Patient Safety aims to promote and protect the health and well-being of patients and other healthcare consumers. There are two important pillars supporting this mission: (1) the collection, assessment and reporting of adverse events- known as Single Case Processing; and (2) the continuous monitoring, interpretation and communication of product benefit-risk profiles to enable Signal Detection and Benefit-Risk Management.
Artificial intelligence holds great promise to address key challenges and to provide new opportunities relevant to both aspects of Pharmacovigilance.
Artificial intelligence systems can automate the manual and routine tasks associated with case processing, thus decrease overall costs for conducting pharmacovigilance
Key challenges for Single-Case Processing include the ever-growing amount of incoming adverse event reports due to the higher number of products in development and on the market, as well as increasing public and media awareness. Year-over-year growth rates are substantial in the high single-digit to low double-digit percentage.
Additionally, increasing regulatory requirements demanded by national health authorities further drives complexity. Together these two factors drive an increase in resource demands and costs which are typically addressed by outsourcing or offshoring of workforce. New sustainable and scalable approaches are needed to offset the increasing resource demands. With large amounts of high-quality historic adverse event data and highly standardized processes, Artificial Intelligence related approaches are suitable to address these challenges.
Machine learning algorithms can be trained to extract and classify information from incoming adverse event reports. The extraction capability includes data elements such as patient, product, event, or reporter. This extraction is done both on structured and unstructured information. The classification capability refers, for example, to the assessment of whether a report contains a fatal or life-threatening event, thus requiring expedited processing and reporting. The artificial intelligence-extracted and classified information is then proposed to a drug safety specialist for review and ultimately confirmation or correction.
This process can also be described as “augmented intelligence” of adverse event report processing. Corrected information then serves as input for subsequent machine learning rounds, improving the algorithms over time. It is believed that in the mid- to long-term, machine learning accuracy will achieve levels that will allow for the processing of selected adverse event reports without human touch. Taken together, these advancements have the potential to deliver a significant efficiency boost to the current pharmacovigilance operating model.
Artificial intelligence systems can also support activities that require medical knowledge and expertise, and advanced analytical skills
Whereas efficiency is the focus for processing incoming safety information, Artificial Intelligence in the area of Benefit-Risk Management will open opportunities to address classification and prediction problems. This will help drive effectiveness and the generation of new insights. Classification in this context refers to leveraging Artificial Intelligence for Signal Detection. Potential signals can be identified early and be confirmed or refuted with higher confidence. Using advanced analytics in the pharmacovigilance process will also aid better clustering of data and consequent discovery of associations, for instance, a selected patient group being more or less prone to developing an adverse event. In the long term, this also opens the opportunity to complement value-based reimbursement models with drug safety information.
Moreover, artificial intelligence-based systems have the potential to advance benefit-risk assessment through predictive capabilities. Once a risk is identified – for instance through incoming safety reports as described above – predictive algorithms could estimate the burden on a population or sub-population. In this example, the artificial intelligence-based system learns from historical data and integrates additional information to predict the effectiveness of risk-minimization measures.
Two other important features of Artificial Intelligence systems are the natural language generation (NLG) and natural language processing (NLP) capabilities. NLG could be utilized for medical writing and the generation of aggregate reports which are developed from individual case reports and the signal detection process. NLP is also useful because much information is only available in an unstructured and free-text form. For instance, medical content from additional data sources such as Electronic Health Records could be obtained by applying NLP to support Signal Verification.
Another value-adding case would be applying NLP to a broad set of data, such as free text in social media, news articles, literature, or medical records for the detection of unexpected benefits of a pharmaceutical product. This could lead to an expansion of indications for an already marketed product and provides an opportunity for Pharmacovigilance to improve patient care while contributing to the top-line revenues of a company.
Artificial intelligence systems truly have the potential to transform pharmacovigilance
Instead of focusing on the resource-intensive manual and repetitive tasks of processing adverse event reports, Pharmacovigilance can now focus on more analytic tasks with greater potential to improve the lives of patients through improved benefit-risk assessment and risk management programs. In the future, we expect to analyze patient safety data more quickly and detect trends within larger volumes of data.
Faster– ideally real-time – detection of relevant safety signals will contribute towards the optimum use of therapies and enhanced patient safety. Risk minimization measures can potentially be initiated faster and, thanks to increased accuracy, the scientific evidence generated should be more robust. As a result, the application of AI in PV has the potential to further improve our ability to promote and protect the health and well-being of patients and other healthcare consumers.