How predictive safety can help patients

Adverse events due to drug consumption can damage the physical and mental wellbeing of patients and are a huge economic burden for the healthcare system (Sultana et al., 2013). Properly monitoring the effects of medical drugs, both during clinical stages and after a product launches, is a critical component of a company’s effective drug regulation system. That’s why we’re enthused about predictive safety, a technology-driven and proactive strategy that allows pharmaceutical companies to forecast which groups of patients could experience adverse events and prevent them before they happen.

Signal management has long been a key function for pharmacovigilance departments, which assess risks to patients. Traditionally, signal management is based on data that details adverse events and involves learning more about a drug’s side effects, based on adverse events during the clinical trials and the small number of patients reporting adverse events after the drug launch. This process isn’t ideal because it exposes patients to unknown side effects before the adverse event is properly analysed and documented, and it’s the reason the industry is exploring proactive solutions—like predictive safety.

Predictive safety in a patient-centric world

As pharmaceutical companies increase their focus on patient centricity, they are viewing safety through that lens. Until recent years, their safety processes were executed for the primary purpose of complying with regulations. Today, however, pharmacovigilance is becoming proactive and companies are aggressively investigating potential issues with their therapies, even if it means having to take costly and time-consuming action. They are using predictive safety to anticipate what signals might occur based on known science, especially for new products being introduced in the market.

For the pharmaceutical industry, this pivot is crucial. The focus on predictive safety follows reports of tragedies such as the 1961 Thalidomide incident, and more recently the 2005 Bextra incident which caused the high fatality Stevens-Johnson syndrome (SJS). These incidents hurt patients and damaged the industry’s reputation. Public trust in pharma companies has always been lower than other industries, and this is primarily due to concerns around the safety of pharma products—regardless of whether these concerns are based on real or perceived issues.

Predictive safety can have broader consequences in the healthcare ecosystem and can reduce the economic, social, and health burdens brought by adverse drug reactions. Given the lack of trust between physicians and patients in the US (Connelly & Campbell, 1987; Gupta et al., 2020), and with adverse drug reactions being estimated as the fifth-most common cause of hospital death in the EU (Montané & Castells, 2021), successfully predicting and preventing adverse events can improve the relationship between patients, physicians, and the pharmaceutical industry.

Using predictive safety in the near term

Even though there are questions about the feasibility of predictive safety, recent evidence shows technology and data sciences have matured enough to provide useful information to predict signals. For example, in 2021 a pharma company was able to use machine learning algorithms to predict the probability of a patient developing non-alcoholic fatty liver disease (NAFLD). This meant that, rather than using invasive procedures, the company could use common lab values as an input to predict which patients have NAFLD. By leveraging data sciences and training a convolution neural network, the company was able to identify the key biomarkers that were significant predictors of patients developing NAFLD. Their algorithm was then validated to provide explanations by using the XGBoost algorithm (Chen & Guestrin, 2016). While this company didn’t use the algorithm to predict adverse events, they did use it to predict a condition of a patient. It stands to reason that identical approaches can be used to predict adverse events.

Even more recently, a pharma company was able to use predictive safety to forecast which groups of patients may potentially develop adverse events in a clinical trial. Their analysis predicted signals through three major components: patient clustering, treatment response prediction, and patient characteristics and patient journey through the trial. This company achieved patient clustering though hierarchical clustering methods on genetics data, such as k-means, and then built a logistic regression-based prediction model using historical data based on patient group treatment responses and patient characteristics. This shows that, on a smaller scale, safety prediction is already helping clinical trial participants. Predictive safety is possible thanks to data collected during preclinical studies and clinical studies, as well as real-world data obtained from the market for products of a similar nature.

Standardising the process

In a heavily regulated and conservative industry like pharma, how can companies increase the prominence and use of predictive safety? First, they will need to adapt a risk-based approach to determine where predictive safety will be most beneficial in their product portfolio. For long-existent products or products treating mild illnesses, predictive signalling may not be beneficial.

Next, companies should initiate processes to determine the thresholds of confidence at which they would take action. This is essential because data sciences and artificial intelligence approaches will predict potential signals with a certain level of confidence. Needless to say, if there is low confidence in a signal prediction, it may not warrant action, and if there is enough confidence to take action, actionable steps must be defined. This means pharma companies must outline a process to mitigate risk for events that have been predicted—with high confidence—to cause serious events.

All of this can only happen at companies that are technologically mature. In addition to robust technology, these companies need the appropriate human resources and talent to manage and operate the technology. For instance, an increasingly complex landscape of data sources leveraged by pharmacovigilance today have to be mined, while machines have to be monitored to ensure they deliver the right predictions.

Finally, alignment with regulators is important. Companies must show regulators the benefits of predictive safety. Their engagement and encouragement will be vital, as companies use guidelines and incentives to develop processes.

Now is the time for predictive safety

Pharmacovigilance and safety signal management have a profound impact on public health and patient wellbeing. Predictive safety is an improvement that has long been desired by patients, providers, and regulatory officials, and there is already evidence showing its potential and benefits. As technology matures, we believe it’s only a matter of time before predictive safety enters the mainstream. It’s critical for leaders in drug safety to lead the way and shape this important landscape.

About the authors

Kumail Fazal is an associate principal at ZS Associates, enabling his clients to be data- and insights-driven. He’s managed teams across the globe and worked in multiple domains, including commercial, meetings & events, and enterprise analytics.

Siva Thiagarajan is a leader of the R&D practice at ZS Associates. His expertise includes life sciences, patient safety, FDA regulations, clinical trials, and patient engagement.

Vartika Pandya is a consultant at ZS Associates, focused on pharma R&D, drug safety, and regulatory affairs.
 

Lucy Liu is a management consultant specializing in life sciences. She’s currently a strategy consultant at ZS Associates and works with top pharmaceutical companies around the world.