Medication Adherence Is the Variable Undermining Clinical Trial Outcomes

Bernard Rienz, CEO and Scientific Lead, Aardex Group.

What Non-Adherence Actually Does to a Clinical Trial

Medication non-adherence is not a binary event. It encompasses missed doses, delayed doses, extra doses, drug holidays and variable timing, often in combinations that differ between patients and shift across the duration of a study. The consequence of this variability is not simply a reduction in average drug exposure. It is an increase in the noise around that exposure, which directly reduces the statistical power of the trial to detect a genuine treatment effect. A study designed with 90% power to detect the expected effect at full adherence may have substantially less power to detect that same effect when 30% of patients are taking their medication inconsistently.

The effect size compression that accompanies non-adherence has a second-order consequence that compounds across the development programme. When effect sizes are smaller than they would be under adequate adherence, dose selection models are calibrated to a population that over-represents poor responders. The doses selected on that basis are higher than necessary for patients who would take the drug reliably. Higher doses increase the burden on safety, increase discontinuation rates, and in many cases become the primary reason regulatory submissions are rejected or delayed. The problem does not start at the regulatory review. It starts at the design table of a phase two study where adherence was treated as background noise rather than a measurable variable.

Why Traditional Monitoring Methods Are Not Sufficient

Pill counts, drug accountability logs, patient diaries and sparse pharmacokinetic sampling all share a common limitation: they provide incomplete or unreliable information about what the patient actually did, and they provide it too late to intervene effectively. A pill count at a site visit establishes that a certain number of tablets left the bottle. It establishes nothing about when they were taken, whether they were taken at all, or whether the count reflects reality rather than the patient’s expectation of what the investigator wants to see. Patient diaries are more granular but are subject to retrospective completion, missing data and the same social desirability bias. Sparse PK sampling captures a moment in the patient’s exposure history, not the pattern.

The practical implication is that sponsors often arrive at the end of a trial with efficacy data that underperforms expectations and no reliable mechanism for determining whether the drug failed or adherence failed. Post-hoc adherence adjustment in this situation is not a valid analytical rescue. Regulatory agencies recognise the approach for what it is, and the scrutiny it invites is at least as damaging as the underlying data problem. The only effective intervention is prospective: design the measurement of adherence into the trial, collect it continuously, and use it in real time.

Electronic Monitoring and the Case for Continuous Dosing Data

Smart packaging technology addresses the core measurement problem by recording automatically when a patient accesses the drug container, generating a continuous dosing history without adding burden to the patient. The data produced by these systems is qualitatively different from what pill counts or patient diaries provide: it is timestamped, complete across the full duration of the trial, and available for both adherent and non-adherent patients. This last point matters considerably. The patients whose data are most informative for understanding the relationship between adherence and outcome are precisely the patients whose adherence is worst. Traditional monitoring systematically underrepresents this group because non-adherent patients are the most likely to have missing or inaccurate diary entries.

Chronology plots of this data reveal the behavioural texture of non-adherence in ways that aggregate measures cannot. A patient who misses doses randomly across a year presents a very different risk profile to a patient who stops dosing completely for eight weeks and then restarts. A patient whose evening doses are consistently erratic while their morning doses are reliable is systematically under-dosed in ways that a mean adherence figure of 75% obscures entirely. These distinctions matter because they have different pharmacokinetic consequences, different implications for endpoint measurement, and different intervention requirements.

Rakhi Kilaru, Executive Director Statistical Science, PPD/Thermo Fisher Scientific.

Integrating Adherence Into RBQM and the Estimands Framework

ICH E6(R3), adopted in early 2025, places quality integrity and risk-based quality management at the centre of trial conduct requirements. Adherence data sits directly within this framework as a key risk indicator. The guidance is explicit that mitigation strategies for identified risks should operate on an ongoing basis throughout the trial rather than being assessed retrospectively. For adherence, this means monitoring tools that generate actionable signals in real time, proportionate intervention protocols that respect patient autonomy without being intrusive, and governance structures that ensure the right people are reviewing the right data at the right time.

Statistical visualisation tools make this practical at scale. Funnel plots of site-level adherence rates identify sites performing outside expected ranges given their patient volumes, flagging where investigator training or process review is needed without requiring manual review of individual patient profiles. Heat maps of daily dosing across the trial population surface patterns of systematic non-adherence that may be linked to specific patient subgroups, dosing times, or disease characteristics. These screening tools triage the patient-level chronology plots to the cases where detailed review adds genuine value, making adherence monitoring a workable component of the regular RBQM cadence rather than a separate and burdensome process.

The estimands framework introduced in ICH E9(R1) provides the analytical counterpart to this operational infrastructure. An estimand defines precisely what question a trial is trying to answer, including how intercurrent events such as missed doses, early discontinuation and rescue medication use are to be handled in the analysis. Where adherence data is collected continuously and reliably, the estimand can be constructed to reflect the actual exposure experienced by patients rather than the exposure assumed by the protocol. This enables sponsors to distinguish between the treatment policy estimand, which reflects real-world adherence, and the hypothetical estimand, which addresses what would have happened had patients taken the drug as prescribed. Both are scientifically and regulatorily meaningful. Neither is answerable without reliable adherence data.

Aardex Group is the developer of the MEMS electronic monitoring system and a long-standing specialist in medication adherence science for clinical research, with three decades of experience in capturing and analysing dosing history data across therapeutic areas. The organisation works with pharmaceutical sponsors, contract research organisations and academic research groups to integrate adherence monitoring into trial design and apply the resulting data within statistical and regulatory frameworks.