How AI is Transforming Pharmacovigilance and Drug Safety.

Pharmacovigilance – also known as PV, PhV, or drug safety – is a vital component of the drug development process, for protecting the health and safety of healthcare consumers and keeping drugmakers informed of any adverse drug reactions (ADRs) their products may cause in specific individuals.

PV involves “identifying, tracking, evaluating and preventing negative outcomes” from drug therapies. It has seen “huge growth” over the past few years: According to European Pharmaceutical Manufacturer Magazine, the PV market will surpass US$8 billion by 2024. That’s because of the sheer number of drugs in development or on the market today, along with the requirement that every manufacturer must prove a drug’s efficacy and safety to the U.S Food and Drug Administration (FDA) and similar agencies around the world.

The PV process begins early on in a drug’s development in the form of phased clinical trials designed to generate data on a drug’s efficacy and safety. Following approval by regulatory bodies, PV continues throughout the drug’s lifecycle and is generally composed of two main pillars:

• Single case processing of Individual Case Safety Reports (ICSRs) involves manually collecting, assessing, and reporting ADRs. Case processing usually takes up anywhere from 40 to 85 percent of a company’s PV budget and can divert resources that could otherwise work on more critical tasks. With PV data volumes increasing at a rate of 15 to 15 percent per year, the inflated cost of case processing has become an enormous issue.
• Postmarketing surveillance (PMS) or signal detection involves ongoing monitoring for ADRs. PMS depends on clinical data from sources such as medical assessments of ADRs, medical literature, health databases, and clinical trial data, along with “real-world evidence” such as electronic health records (EHRs), medical device data, customer surveys, warranty claims, and data from social media and other online sources.

Because PMS involves long-term monitoring, it plays a vital role in discovering rare benefits or problems in a drug that would otherwise go undetected for years. PMS has other advantages over pre-marketing clinical trials: (1) it’s indefinite (whereas clinical trials usually last just a few weeks or months) and (2) involves a much larger population group, including smaller subgroups not represented in a limited clinical trial.

There are three distinct types of PMS: Spontaneous reporting of cases to the FDA or in health literature; Postmarketing studies such as observational studies or randomized clinical trials; and active surveillance.

Traditional PV’s rising costs (and the business case for AI)

The cost of PV in both expenditure and resources is continually rising, thanks to increasing numbers of incoming ADR reports captured by ongoing PMS monitoring. Several factors, such as an aging population, increased awareness among the public, and the number of pharma products on the market increases the amount of ADRs. Combine this with increased regulatory requirements over the past few years, and it’s easy to see how the costs of PV have ballooned for pharmaceutical companies.

A recent Deloitte survey of mid- to large-cap global biopharma companies shows that PV practices are increasingly looking to improve efficiencies while cutting costs. According to the study, cutting case processing costs was the primary goal of 90 percent of respondents. The survey also showed that many companies are investing in AI and automation to tackle various PMS-related tasks. After all, the sheer volume of healthcare and other available data for analysis is practically impossible for humans to track and analyze manually and on time.

Applications of AI in PV

These are all reasons why the focus of many PV and PMS strategies, traditionally relatively reactive, have morphed into proactive (and even predictive) risk management using AI tools. Unlike human analysts, these tools can quickly identify, collect, and analyze massive amounts of data. Because free-form text data is prevalent in healthcare, machine learning (ML) and trained natural language processing (NLP) algorithms can detect, extract, and classify ADR data from this kind of unstructured data.

Perhaps the most important (and immediate) potential application of AI in PV – especially the PMS element of PV – is the automation of manual, repetitive, routine tasks with case processing. AI will decrease the cost of processing each case and free up valuable resources to work on more complex and value-added tasks.

Big data analytics using AI can also help discover drug-event associations for certain groups of people, improving the detection of potential events while improving risk-benefit assessments. Some estimate that AI  has reduced screening costs by 80 percent and general annotation costs by 50 percent while eliminating setup costs.

NLP algorithms can quickly analyze big datasets from social media sources, news articles, medical literature, medical records, and other text data. This approach uses AI and trained analysts to monitor for signals indicating unexpected benefits or adverse reactions. These signals provide valuable real-world intelligence that simply can’t be found by mining data from controlled clinical situations. Similarly, natural language generation (NLG) technology can be applied to generate reports – or, at least, its framework, with human experts freed up to provide further analysis and polish.

Other use cases of AI in PV include auto-coding of terms used by consumers and non-medical personnel to official medical ontologies. The automatic converting of ADRs recorded by audio or video into text, is another use case.

As AI becomes more entrenched in healthcare, and PV specifically, companies will be able to analyze PMS data faster than ever before. They’ll have the power to quickly discover nuanced insights from extensive (and more accurate) batches of data, improving patient safety while reducing risk among drug manufacturers. Real-time signal detection will lead to more timely risk minimization, better and more personalized treatments, and improved patient outcomes.