Medication adherence has long been a silent driver of outcomes and cost. In the US, nonadherence is associated with up to 125,000 deaths annually and an estimated $100–300 billion dollars in avoidable healthcare costs. Yet across chronic conditions, only about half of patients remain on therapy after one year. For sponsors, those numbers translate into trial noise, under‑estimated efficacy, and post‑launch performance gaps.[1]
Despite this impact, adherence is still poorly captured and reported. A recent review of pivotal trials found that nearly one‑third made no reference to medication adherence in either the main document or supplementary materials, and many that did relied on coarse measures such as “non‑compliance” counts or average intake. In other words, the variable that can make or break trial validity and commercial success is often measured once, late, and imprecisely.[2][3]
The industry is ready for a reframing: medication adherence is a data pipeline problem before it is a patient problem. If your adherence data is sparse, biased, or siloed, your ability to design interventions or demonstrate value is constrained from day one.
The problem: why adherence data matters
When adherence is treated primarily as a patient education or engagement issue, sponsors and providers tend to deploy generic interventions such as reminder calls, SMS nudges, simplified materials, etc., without robust evidence of where, when, and why behavior breaks down. That’s because traditional adherence data sources are inherently limited.
Across trials, up to 50% of participants do not adhere fully to the dosing regimen, and by day 100 in many Phase III studies, roughly 20% have stopped following the protocol and another 12% show sub‑optimal adherence. Yet these deviations are rarely integrated rigorously into statistical analysis plans, leading to underestimated efficacy, distorted pharmacoeconomic models, and in some cases trial failure.[3][6][5]
For digital innovators and pharma leaders, the implications are stark:
You are optimizing protocols and engagement strategies on incomplete, lagging indicators.
You cannot reliably distinguish a “failed drug” from a “failed behavior signal.”
You struggle to build data‑driven business cases for adherence interventions, because the measurement itself is noisy.
Until adherence is instrumented as a continuous, high‑resolution data stream, behavior‑level insights will remain shallow, and intervention design will stay reactive.
Several macro trends are pushing adherence from a soft issue to a strategic data asset:
Digital biomarkers and real‑world evidence. Digital biomarkers derived from apps, sensors, and platforms are reshaping clinical research and patient‑centric care, promising real‑time insights into how patients live with their condition. Adherence is a foundational behavioral biomarker, yet it often remains unmeasured or siloed from other digital signals.[7][8][9][10]
Regulatory expectations for quality data. Regulators increasingly expect trials to generate robust, interpretable data on treatment exposure. Reviews of RCTs across chronic diseases show that adherence is assessed in only about one‑third of trials, and even then the data is often missing from statistical analyses. As regulators push for better evidence on benefit‑risk and real‑world performance, adherence measurement will become non‑negotiable.[3]
Shift to decentralized and hybrid trials. As trials move closer to the patient’s home, the opportunity to capture behavior data expands—but so does the risk of losing control over dose‑level adherence. Traditional site‑based monitoring tools (pill counts, logs) don’t scale well into decentralized designs.
Value‑based, outcomes‑based contracting. Payers and health systems are increasingly tying reimbursement to demonstrable outcomes. Poor adherence can be responsible for between 33% and 69% of hospital admissions related to medications, and good adherence has been linked to a 21% reduction in long‑term mortality. Without high‑quality adherence data, it is difficult to prove that an intervention—or a therapy—is delivering outcomes as promised.[11][1]
At the same time, the data landscape remains fragmented:
Multiple sources—claims, EHRs, pharmacy history, digital tools—apply different coding schemas and standards, making data mapping complex.[4]
Many digital adherence solutions operate as standalone apps, generating useful data that rarely flows back into core clinical, safety, or commercial analytics.
Commercial incentives often favor building new point solutions rather than integrating adherence as a common behavioral data layer that powers trial design, market access, and lifecycle management.[7]
The result is a paradox: the industry is rich in potential adherence data but poor in adherence insight. Unlocking that insight requires treating adherence as a data engineering and AI problem, not just a patient education challenge.
Adherence as a behavioral data stream
At AICure, we approach medication adherence as a high‑frequency behavioral data stream that must be captured, interpreted, and operationalized with the same rigor as any primary endpoint. That informs both our technology stack and our philosophy.
Using computer vision and AI, AICure transforms medication ingestion into a verifiable digital event. Rather than relying on inferred signals (fills, refills, self‑report), our platform captures dose‑level data at the exact moment the patient takes—or fails to take—the medication. This yields:
Precise timestamps for each dose.
Confirmation of the correct patient and correct medication.
A longitudinal record of adherence patterns at an individual and cohort level.
From a data perspective, this is equivalent to upgrading from a periodic survey to a continuous sensor, dramatically increasing the resolution of adherence behavior.
Analytics that turn behavior into insight
Raw ingestion events are only the start. AICure’s analytics layer focuses on turning behavioral sequences into actionable signals:
Pattern detection (dose timing drift, clustering of missed doses, early discontinuation).
Risk scoring that flags emerging nonadherence before it manifests as endpoint noise.
Cohort‑level segmentation to identify sub‑populations that respond differently to interventions or have distinct adherence profiles.
This enables sponsors and digital innovators to ask more sophisticated questions: How does adherence vary by site, by device, by socio‑demographic cluster? Which behavioral patterns correlate most strongly with response or adverse events? Where should you focus intervention resources for maximal impact?
Integrations that break data silos
AICure is designed to integrate with core clinical systems, EDC, and real‑world data platforms. By aligning ingestion data with protocol schedules, exposure windows, and endpoint measurements, sponsors can:
Incorporate adherence into prespecified statistical analyses, rather than treating it as a late exploratory variable.
Interpret efficacy and safety outcomes in the context of actual treatment exposure.
Feed adherence insights into broader digital biomarker frameworks, connecting behavior to physiology and outcomes.
From a digital innovator’s standpoint, AICure becomes a data service: a reliable adherence signal that can be consumed by clinical analytics, patient engagement engines, and commercial performance dashboards.
Interventions informed by real behavior
Because AICure sees adherence behavior at the dose level, interventions can be personalized and timed precisely. Instead of blanket reminders, sponsors can design targeted workflows (e.g., dose‑time nudges, clinical outreach, digital coaching) driven by real‑time risk signals. Over time, this creates a closed loop:
Measure adherence with high fidelity.
Detect early patterns of risk.
Deploy tailored interventions.
Learn which interventions work for which behavioral phenotypes.
Crucially, the emphasis remains on data: interventions are not generic engagement tactics but experiments grounded in well‑characterized behavioral signals.
Treat adherence as infrastructure
For digital health innovators and pharma executives, reframing medication adherence as a data problem has several concrete implications:
Design adherence in, don’t bolt it on. Treat dose‑level adherence capture as core infrastructure in trial and program design. Build it into protocols, analytics plans, and downstream commercial strategies from the start.
Upgrade from inferred to observed data. Move beyond proxies like claims and pill counts toward direct, verifiable ingestion events. Higher‑resolution data unlocks better models, better interventions, and more defensible evidence of drug performance.
Integrate adherence across the lifecycle. Connect adherence signals from trials to post‑approval use, value‑based contracts, and digital companion strategies. A continuous behavioral data layer helps align R&D, medical, and commercial teams around the same reality.
Use AI to interpret behavior, not just to automate reminders. The power of AI in adherence is not merely automation; it is the ability to recognize patterns, predict risk, and inform targeted interventions that respect patient context.
AICure exists to solve medication adherence as a data problem. By capturing dose‑level behavior with AI, integrating it into clinical and commercial analytics, and enabling targeted, evidence‑driven interventions, we help sponsors replace adherence guesswork with adherence intelligence. When you use adherence as a reliable data stream, you’re no longer asking, “Why aren’t patients taking their meds?” You’re using precise behavioral insight to design better trials, deliver better outcomes, and prove the value of your therapies with confidence.
References:
https://www.dialoghealth.com/post/patient-adherence-statistics
https://www.tandfonline.com/doi/full/10.1080/14737167.2024.2363401
https://acrpnet.org/2025/08/19/unlocking-the-power-of-digital-biomarkers-in-clinical-trials
https://phil.us/advancing-medication-adherence-with-data-driven-interventions/
https://www.tevausa.com/news-and-media/article-pages/ai-driven-digital-biomarkers/
https://www.ejinme.com/article/S0953-6205(23)00367-9/fulltext
https://www.amjmedsci.com/article/S0002-9629(15)37996-9/fulltext
This blog post was developed with the assistance of artificial intelligence for research, outlining, and/or drafting. It was reviewed, edited, and approved by a human author to help ensure accuracy, clarity, and alignment with our organization’s views.