Better Trials Begin With Better Data
Clinical research is generating more data than ever before. Digital health technologies can collect physiological and behavioral information remotely; decentralized approaches can extend research beyond traditional sites; and artificial intelligence can help study teams interpret signals at a scale that would have been difficult to imagine a decade ago.
But more data does not automatically mean better data.
For sponsors, the real challenge is ensuring that clinical trial data is complete, reliable, interpretable, and connected to what participants are actually experiencing. Regulators are emphasizing the same principle. ICH E6(R3), which modernizes Good Clinical Practice for an increasingly digital research environment, puts greater emphasis on quality by design, fit-for-purpose systems and risk-proportionate approaches to generating reliable trial results.
This is where AI clinical trials have an important opportunity. Used responsibly, AI can do more than accelerate analysis. It can help improve the quality of information at its source—from understanding whether participants are following a protocol to identifying emerging risks before they become larger data-quality problems.
Here are five ways I believe AI can help us get there.
The Problem: Clinical Trial Data Has a Human Context
Data quality is often discussed as a technical challenge. In practice, it is also a human one.
Consider medication adherence. A participant may report that they took an investigational therapy as directed. A site may record the information accurately. The database may function exactly as intended. But if the participant did not actually take the medication—or took it incorrectly—the resulting dataset can still give researchers an incomplete picture of treatment exposure.
The same challenge applies to assessments performed outside the clinic. Participants may miss measurements, use technology inconsistently or disengage gradually between site visits. Traditional data review may identify some of these problems only after they have accumulated.
Missing or unreliable information can introduce noise into a study and make it harder to distinguish the true treatment signal. It can also increase the operational burden on sites, which must investigate discrepancies, contact participants and resolve data queries while continuing to manage the many other demands of trial execution.
As trials become more decentralized and incorporate more digital health technologies, these questions become even more important. FDA guidance on remote data acquisition recognizes the potential of digital technologies to make participation more convenient while also emphasizing considerations such as usability, validation, data integrity, and participant compliance.
The objective, therefore, should not simply be to digitize existing processes.
We should ask a more ambitious question: How can technology help us understand the context behind clinical trial data while the study is actually happening?
That is where AI can have a meaningful impact.
Industry Perspective: Five Ways AI Can Strengthen Clinical Trial Data Quality
1. AI can provide more objective insight into medication adherence
Medication adherence is one of the clearest examples of the difference between recorded data and meaningful data.
Traditional approaches—including diaries, pill counts, and self-reporting—can leave uncertainty about whether and when medication was taken. AI-enabled computer vision can add another layer of objective observation by helping verify medication-taking behavior through a participant's smartphone.
That matters because treatment exposure is fundamental to interpreting outcomes. If researchers have greater confidence in adherence data, they have greater context for evaluating both efficacy and safety signals.
AI should not replace the relationship between participants and study teams. Instead, it can provide additional evidence that helps teams understand what is happening between scheduled interactions.
2. AI can identify risks earlier
Clinical trials produce enormous numbers of individual data points, but not every data point deserves equal attention.
Machine-learning models can help identify patterns associated with behaviors such as declining adherence or disengagement. Rather than waiting for a missed visit or repeated protocol deviation, study teams can potentially recognize changes earlier and determine whether an intervention is appropriate.
This represents an important shift from retrospective data cleaning toward prospective data quality management.
The distinction matters. Finding a problem after database lock may help explain what happened. Identifying the risk while a participant is still enrolled creates an opportunity to do something about it.
3. AI can reduce variability in difficult-to-measure assessments
Many clinically meaningful symptoms are inherently difficult to quantify.
Changes in speech, facial expression or movement, for example, may be relevant in neurological, psychiatric and other therapeutic areas. Traditional assessments remain essential, but they may occur only at specific visits and can include elements of subjective interpretation.
AI-enabled digital biomarkers create the possibility of repeatedly and objectively quantifying selected behavioral characteristics such as vocal acoustics, speech patterns, facial expressivity and movement.
The value is not simply that these measurements are digital. Their potential lies in creating standardized, longitudinal information that can complement established clinical assessments and help researchers understand how participants change over time.
4. AI can help teams focus on the data that matters most
One of the unintended consequences of digitization is information overload.
More sensors, apps, and remote assessments can produce more data, but they can also make it harder for clinical teams to identify which signals require attention. AI can help aggregate and analyze information so that teams can prioritize meaningful patterns rather than manually reviewing every data point with equal intensity.
That approach aligns with the broader evolution of Good Clinical Practice. ICH E6(R3) encourages proportionate, risk-based approaches and emphasizes that systems and processes supporting clinical trial information should be fit for purpose.
For AI, that creates an important standard: success should not be measured by how much data an algorithm processes. It should be measured by whether the technology helps produce information that is sufficiently reliable and useful for better decision-making.
5. AI can connect data quality with patient engagement
We sometimes discuss engagement and data quality as separate clinical trial objectives. They are closely connected.
A participant who understands what is expected, receives appropriate support and remains engaged with the study is more likely to complete required activities consistently. Conversely, declining engagement can appear in the dataset as missed doses, incomplete assessments, gaps in data or eventually early termination.
AI can help identify those patterns and enable study teams to focus support where it is most needed.
This is especially important because the goal should not be to automate the human element out of clinical research. The strongest applications of AI can help study teams determine where human attention will have the greatest impact.
AICure's Perspective: From Data Collection to Better Observation
At AiCure, we believe improving clinical trial data quality starts with improving our ability to understand the participant experience.
Our approach combines AI-enabled technology with clinical and operational expertise. Through a smartphone, our H.Code platform can support medication adherence, participant engagement, predictive analytics and the collection of behavioral measures. Computer vision can help provide objective information around dosing, while digital biomarker capabilities can quantify characteristics including movement, facial expression and vocal or speech-related variables.
But technology alone is not the answer.
A prediction that stays on a dashboard does not improve a trial. Data becomes useful when it enables an informed action.
That is why we view AI as part of a broader feedback loop: observe, understand and act.
First, technology can help create a more continuous picture of what is occurring between clinical visits. Second, analytics can identify patterns and potential risks within those observations. Finally, clinical teams can use those insights to determine whether a participant or site needs additional support.
Our own research across 28 Phase I-IV clinical trials involving more than 4,000 participants in 43 countries has examined the relationship between AI-enabled medication-adherence data and early termination risk. This work reinforces an important principle: behavioral information collected during a study may tell us not only what a participant has already done, but also where future trial risk may be developing.
The industry should nevertheless be disciplined about how it implements AI.
Every algorithm does not belong in every study. Technology should be selected because it addresses a clearly defined research or operational need. Systems need appropriate validation and governance. Data collection should be proportionate to the trial's objectives. And sponsors should be able to explain how an AI-enabled measure or prediction contributes to decision-making.
As regulatory frameworks increasingly accommodate digital technologies, decentralized approaches and new data sources, that discipline will become even more important.
The future of AI in clinical research is not about collecting everything simply because we can.
It is about collecting the right information, with sufficient quality and context, at the right time.
Key Takeaways: Data Quality Should Be Designed Into the Trial
AI has the potential to improve clinical trial data quality in five interconnected ways: by creating more objective adherence information, detecting emerging risks earlier, enabling consistent measurement of difficult-to-quantify behaviors, helping teams prioritize meaningful signals and supporting participant engagement.
But the technology is only part of the equation.
Sponsors considering AI for clinical trials should begin with the question they are trying to answer—not with the algorithm they want to deploy. They should define which data are critical to trial quality, evaluate whether the technology is fit for purpose and determine how AI-generated insights will translate into appropriate action.
Most importantly, we should remember what clinical trial data represents.
Behind every data point is a participant giving their time and effort to help answer an important scientific question. Improving data quality means making that contribution as meaningful as possible.
AI gives us an opportunity to see more of the participant journey between clinical visits and to respond to risk sooner. Used thoughtfully, it can help us build clinical trials that are not simply more digital, but more observable, more responsive and ultimately more reliable.
