Improved Drug Safety Needs Better Data Before AI
Contributed Commentary by Daniel O’Keeffe, Qinecsa
January 23, 2026 | Applying artificial intelligence (AI) to fragmented and incomplete safety data cannot fill the gap left when the right information has not been captured during initial patient adverse event reporting. And yet, software vendors position the technology as a quick fix to safety data complexity. Although AI definitely has a role to play, it would be far better to use it to refine the data capture process, driving richer safety insights and ultimately better safety outcomes. Regulators could help ensure this with a clearer mandate for professional data collectors.
The drive to ensure that pharmaceutical products and therapies are as safe as they can be—and deliver optimal outcomes—is a basic assumption in this industry. Yet too often the practical reality of ongoing drug safety monitoring is a good deal more complex. Capturing consistent and high-quality insights that are of meaningful value has arguably never been harder.
Unless the pharma industry gets a grip on this problem, it risks missing a huge opportunity. This includes the potential to harness AI to derive better intelligence. Although the technology is shown to be very good at combining data and creating order from complexity, AI can only work with what it has. Attempting to make it turn bad or incomplete data into something of greater value is misguided, then. That isn’t to say AI can’t help patient safety teams; rather that the technology needs to be harnessed from an earlier stage—to get better data in, when the opportunity is strongest.
Safety Tracking Shouldn’t be This Hard
Given how critical safety monitoring is to patient outcomes, the associated complexity and the challenges this causes are lamentable. Disparate and proliferating channels, and rising volumes of adverse event reports (due both to advanced new treatments and growing public awareness around how to report side-effects), have amplified the patient safety monitoring burden.
Very often, adverse event reports are patchy, incomplete, and difficult to follow up. Disjointed means of reporting, and inconsistency in what and how much is captured, render findings hard to combine in a way that is meaningful, e.g. as the basis for actionable intelligence, blended with data combed from scientific journals, online forums, and so forth. Anecdotally, large pharma organizations report that barely just 10% of attempts at information follow-up (to fill in gaps in the narrative) are successful once initial details of any side-effects have been reported. To miss this window is to forego those insights altogether.
If this problem could be overcome more systematically, richer insights and better decisions would be a surer bet. One option is that regulators should exert more influence around patient safety data capture, in the interests of establishing the richest possible understanding of each patient’s experience. The case for proactively improving original patient safety data capture is particularly strong where the facilitator is a paid third party—perhaps a specialty pharmacist or a service provider contracted to deliver a patient support program (a PSP vendor). As it stands, those partners’ main vehicles for receiving and registering adverse event notifications are typically an email address, paper form, or Word document, resulting in a painful process of manual data amalgamation and reconciliation for someone. When gaps in the narratives are found, it is generally too late to do anything about it—a gap that AI can’t fill retrospectively without risk of hallucination.
What Happened to Patient Centricity?
With so much claimed in life sciences about improving patient centricity, it is mystifying that only some pharma companies (paying for the data) and regulators (relied on to uphold quality and safety), formally insist on the capture of complete and high-quality data first time.
Better data would provide a much clearer picture of adverse events and what may be contributing to them (such as drug interactions and pre-existing conditions). Consider, for instance, the high volumes of incoming data ready to be recorded around weight-loss drugs traditionally associated with diabetes treatment—those targeting GLP-1 and/or GIP receptors to control appetite. Possible side-effects range from digestive issues to reduced muscle and bone mass. The opportunity to capture this information widely and draw trend information from it is rich and important, so assigned and especially paid professionals should be doing all they can to improve the consistency and value of this activity.
Until the authorities mandate that better data is captured at source wherever possible, pharma safety functions and their patients will be no better off. It should be a priority to empower respective actors with the right tools for the job, as well as a sense of accountability for the quality and onward value of the data being captured.
Where AI Could Play a Role
Given the complex, multi-channel landscape through which relevant patient safety data can flow now, it follows that strategies and approaches for improvements must be geared to standardization and consistency. AI can help here by prompting good and comprehensive data capture up front—for instance, guiding the user to provide additional information. AI could also be used to tailor and optimize the digital experience, e.g. for each inputter’s persona (such as patient, PSP nurse, staff member, and pharmacist), their likely medical knowledge, their native language, the device being used, and so on.
In the modern age, sorting through batches of data sent by email and then trying to chase down missing details is inadequate, given that more effective alternatives are available. It is inefficient, ineffective, and costly, and it serves no one. An optimized digital experience for the reporter, with pertinent questions or prompts to capture all of the preferred detail, has been shown in pharma company deployments to enable 70% overall improved efficiency, including reduced follow-up.
Personalized Treatments Make Robust Reporting Paramount
With the current pace of drug innovation, every safety data point is priceless and needs to be treated as such from the moment of capture through to signal detection and analysis. As personalized medicine continues to grow as a proportion of pharma pipelines, and as smart devices do more to track individual’s health, the data collected will inevitably become more patient-specific and critical. Developing better practices now will set the pharma industry in good stead for what’s to come.
There is no question of the criticality of breaking down safety data complexity now, to enable a more holistic, richer picture of a drug and its impact once in the market. And this must begin with capturing more via the earliest patient’s feedback.
Real transformation requires that the pharma industry demystifies the complexity, and deploys the right tools for the given situation, including AI, where it can ensure greater data richness from the outset. Companies will need to join up systems and overcome data silos, too, of course, so that more insights can flow into all the relevant downstream systems, where they can be analyzed and actioned, without the need for manual data re-entry. All of this will support more accurate triaging and onward decision-making, simultaneously boosting productivity and elevating patient outcomes.
Daniel O’Keeffe specializes in transforming pharmacovigilance through cutting-edge technology. Qinecsa is a global provider of innovative, digital pharmacovigilance solutions, including cloud-based analytics solutions and services for medical research and healthcare delivery. He can be reached out at daniel.okeeffe@qinecsa.com.