AARDEX Group · Company in Belgium Tech

Overview

AARDEX Group offers optimization services for drug development through effective monitoring of patient non-compliance in clinical trials. The company operates in the medical equipment manufacturing sector.

Key people

Bernard Vrijens

In the news

Interesting findings. One question is whether the study population may have been "too adherent" or "too engaged" to fully demonstrate the potential value of the intervention. Recruiting from highly motivated centers in Switzerland and enrolling patients willing to participate in multiple visits, ABPM monitoring and app-based follow-up inevitably selects for patients already engaged in their care. In addition, excluding drop outs from the primary analysis may underestimate the real-world adherence challenge. The true test will beLinkedIn · 17 Sept 2026
Should medication adherence be managed as a Critical-to-Quality factor in clinical trials? In this video, we look at when non-adherence can become a meaningful risk to trial interpretability and how to move from risk assessment to an appropriate monitoring strategy. Watch the video for a practical, study-focused perspective. #ClinicalTrials #MedicationAdherence #ClinicalResearch #ICH_E6R3 #QualityByDesign #AARDEXGroupLinkedIn · 15 Sept 2026
A timely reminder that the value of any prediction model is only as strong as the quality of the data and endpoints on which it is built. In medication adherence, reliable and precise measurement remains the foundation for generating actionable insights. If adherence itself is measured inaccurately, even the most sophisticated AI models risk producing biased or clinically misleading predictions. Before discussing algorithms, we should ensure we can accurately capture patient behavior in the real world. #MedicationAdherenceLinkedIn · 14 Sept 2026
QuotedGeorgi Gospodinov, Ph.D.@ggospodinov39 of 41 medication adherence prediction models reviewed in this npj Digital Medicine paper failed either development quality standards or evaluation bias checks under PROBAST + AI, and the failure mode is not algorithmic complexity, it is foundational design. 1. The outcome definition problem is upstream of everything. Twenty-five studies used pharmacy refill metrics (PDC or MPR), which cannot confirm ingestion and silently misclassify stockpiling as adherence. Twenty studies then dichotomized even those weak continuous signals, discarding variance that a downstream intervention system could actually act on. A prediction pipeline built on a poorly defined label propagates that error into every downstream risk score. 2. The validation gap is severe in deployment terms. Only 2 of 41 studies performed truly independent external validation, one temporal hold-out, one geographic. The remainder used internal split-sample or cross-validation schemes, and several described within-dataset hold-out as "external validation." A model that has never been tested on a different site, a different EHR schema, or a different payer-mix cannot be trusted to hold calibration under distribution shift. 3. Calibration reporting was treated as optional. Most studies reported AUC as the primary performance claim, with no reliability diagram, no Brier score, no clinical utility curve. Tengrium's architecture for longitudinal risk stratification requires calibration checks at inference time because a miscalibrated model that ranks patients correctly but assigns wrong absolute probabilities will misfired any threshold-based alert logic. 4. The "prior adherence paradox" noted in the paper is an integration surface problem. Early refill behavior is a dominant predictor in large administrative datasets but was surfaced in only 8 of 41 studies, largely because designs excluded early follow-up windows to enable baseline prediction. Any production architecture that wants to use this signal has to instrument the refill event stream from day one of enrollment, not as a retrospective pull. 5. Missing data handling was flagged as high concern in 58.5% of studies at the development domain. In EHR-derived cohorts, missingness is rarely random, patients with fewer encounters have systematically sparse records, so naive imputation or complete-case analysis encodes a selection bias that compounds at inference. The next study that wants to advance clinical readiness would need to pre-register its adherence outcome definition, justify the binary threshold or drop it, and treat external geographic validation as a minimum rather than a stretch goal. https://lnkd.in/eUxUPRra #DigitalHealth #ClinicalAI #PrecisionMedicine #HealthInformatics #MedicalAI+1
Have you ever remembered at lunchtime that you were supposed to take your medication that morning? Maybe you were rushing to work. Your routine changed. You simply forgot and remembered a few hours later. It happens. And this seemingly small difference between “medication prescribed” and “medication actually taken” can matter. According to the WHO, adherence to long-term therapies among patients with chronic diseases in developed countries averages around 50%. That doesn’t mean that half of all doses are forgotten, non-adherenceLinkedIn · 10 Sept 2026
Do we actually know whether adherence monitoring can help medication-taking become a sustainable habit? In a recent meta-analysis of two studies conducted in the US and Türkiye involving kidney transplant recipients, adherence to immunosuppressive medication was objectively measured using MEMS® Caps, with each bottle opening recorded electronically. And what do you think the researchers found? According to the study by Cynthia L. Russell Lippincott, Antoine Pironet, Yaprak Sarigol Ordin, Rebecca Bartlett Ellis, PhD, RN, ACNS-BC,LinkedIn · 8 Sept 2026
We track what matters. Why should medication-taking be different? In #ClinicalTrials, knowing whether and when a dose was taken can be critical to understanding treatment exposure and interpreting study outcomes. With objective #MedicationAdherence data, sponsors can replace assumptions with evidence and bring greater visibility to what happens between study visits. #ClinicalResearch #DigitalHealth #DataQuality #AARDEXGroupLinkedIn · 4 Sept 2026
Worth watching 🚀 The recording of the recent Statistical Considerations for Digitally-Derived Endpoints in Clinical Trials workshop, convened by the Duke-Margolis Institute for Health Policy under a cooperative agreement with the FDA, is now available. The discussion explores the growing role of digital health technologies in clinical trials, with a particular focus on data quality, statistical considerations, and the use of digitally-derived endpoints in regulatory decision-making. As clinical trials increasingly rely onLinkedIn · 31 Aug 2026
Nobody navigates with yesterday's location. Open Google Maps and you see exactly where you are, right now. Not where you were an hour ago. So why is medication adherence in clinical trials still measured with data that's already weeks old? Pill counts. Diaries. Patients trying to remember what they took last Tuesday. Up to half of trial participants don't take their medication as prescribed and that puts your data quality and your trial timeline at risk. At AARDEX Group, we think adherence monitoring deserves the same real timeLinkedIn · 27 Aug 2026

Alumni

Bernard VrijensGeneral Director→Invited Professor at University of Liège

Something wrong or missing? Send an update. Fixed within 24 hours.