Monsana’s GenAI assistant boosts patient identification at AZ Delta

AZ Delta plays a key role in the Belgian clinical trial landscape, with a strong track record in both academic and commercial studies. However, they face a challenge common to all trial sites: efficiently identifying eligible patients from among the thousands who visit the hospital each year.

The cardiology department was supported by Monsana's AI assistant to pre-screen medical records in order to identify suitable trial candidates within their center. During this pilot, we showed that Monsana identified every eligible patient with no misses, while reducing healthcare staff workload by 85%.

Our estimates indicate that nearly two‐thirds of eligible patients are overlooked, while clinical trial coordinators can spend up to a quarter of their time searching for potential participants. Monsana's platform screens all patients in the background, saving time without compromising accuracy.

In this article, we explain how we made this possible.

From theory to practice: methodology of our pilot

To validate Monsana’s technology and showcase its impact, we conducted testing using real-world data at the cardiology department of AZ Delta. We screened 239 cardiology patients for 2 clinical trials ongoing at AZ Delta using Monsana.AI. To ensure a comprehensive evaluation, we also manually reviewed each patients' eligibility. This approach enabled us to compare the effectiveness of AI screening against traditional manual methods. *

Input data

The dataset consisted of unprocessed medical records in Dutch from 239 randomly selected cardiology patients, generated by physicians during real-world clinical workflows in an unstructured format. 2 to 3 medical records were included for each patient. On average, a consultation report contained around 775 words. All data was pseudonymized before analysis by Monsana.AI.

The clinical trial data included inclusion and exclusion criteria from two ongoing trials, each with roughly 32 eligibility criteria.

Results

Patient-trial level evaluations: identifying every patient with zero misses, in a fraction of the time

Based on the combined outcomes of all patient-criterion evaluations, the AI model classifies each patient-trial pair into one of four categories, based on the overall matching score: highly likely to be eligible, likely to be eligible, uncertain, or unlikely to be eligible.

In our evaluation of approximately 480 screenings, the model was able to automatically and correctly exclude 420 patient-trial pairs. The remaining 60 pairs were flagged as highly likely, likely, or uncertain, indicating to the evaluator that these patients may still be worth reviewing for trial eligibility. Of these, 40 were confirmed as ture trial candidates. Furthermore, a review of the 420 patient-trial pairs classified as unlikely confirmed that no potential trial candidates were overlooked. *

	Eligible	Ineligible
Highly likely, likely, uncertain	40	20
Unlikely	0	420

These results demonstrate that Monsana’s AI can identify all eligible trial candidates, while maintaining a low false positive rate, with 2 out of 3 flagged patients confirmed as truly eligible. This level of performance was achieved while reducing manual review workload by 85%, significantly easing the burden on healthcare professionals.

Patient-criterion level evaluations: deep dive into accuracy

The model assesses each trial criterion individually based on the patient’s medical record. For every patient-criterion pair, it returns a result: eligible, uncertain, or ineligible. A result is marked uncertain if the file lacks relevant information or if further testing is required. In tests on 1,645 patient-criterion pairs, the model achieved an accuracy of 96.9%. **

To support efficient human review, the model includes an explanation for each decision and highlights the part of the file that supports its reasoning.

Here are some examples that illustrate the model’s human-like decision-making:

Criterion	Explanation	Snippet
Patient is diagnosed with hypertension	Results: inclusion criteria match --> eligible Reason: the patient has a systolic blood pressure above 157, which indicates hypertension.	BD systolisch (mmHg): 157
The patient is eligible to receive anticoagulation therapy.	Results: inclusion criteria match --> eligible Reason: The patient is currently on anticoagulation therapy (Eliquis), indicating eligibility for this criterion.	Eliquis tablet filmomhuld 5mg; oraal; 2 x per dag 1 stuk
Known presence of atrial myxoma or left ventricular thrombus	Results: exclusion criteria mismatch --> eligible Reason: The note states that the left atrial appendage is free of thrombus, indicating no presence of left ventricular thrombus or atrial myxoma.	Laa vrij van thrombus
History of coronary revascularization	Results: inclusion criteria match --> eligible Reason: The patient had an urgent coronary angiography showing critical proximal LAD stenosis, treated with PCI and a drug-eluting stent — confirming prior coronary revascularization.	Urgente coronarografie met vaststellen van kritische Lad stenose proximaal waarvoor Pci met Des.
Patient has contraindications for anticoagulant therapy/	Results: exclusion criteria match --> ineligible Reason: The patient has a history of recurrent gastrointestinal bleeding, which may indicate a contraindication for anticoagulant therapy.	recidiverende lage GI bloeding onder Doac
Occurrence of any of the following conditions within the past 3 months: hemorrhagic stroke, subarachnoid hemorrhage, subdural hematoma, spinal cord hemorrhage.	Results: exclusion criteria mismatch --> ineligible Reason: The patient has a history of subarachnoid hemorrhage in 2016, but there is no indication of a recent occurrence within the past 3 months.	2016 : syncope met subarachnoidale bloeding, konservatief behandeld.

Reviewing smarter, not harder

To streamline the review process, all results are presented directly within the Monsana dashboard, enabling fast and efficient evaluation. The dashboard offers a clear, user-friendly overview of all patients, making it easy to identify and prioritize the most relevant patient-trial matches. Detailed information for each patient supports quick, informed decision-making.

As a result, the time required to review each case is estimated to be up to three times shorter than traditional methods. 

Conclusion & Future work

These preliminary results highlight the powerful potential of GenAI to transform patient identification in real-world clinical settings.

Our pilot demonstrated that Monsana can reduce manual review efforts by up to 85% while maintaining full accuracy in identifying eligible trial candidates. Impressively, two out of every three recommendations provided to doctors are confirmed as true potential participants.

We are now finalising three additional pilots across Belgian hospitals to validate the adaptability of our solution across varied workflows, data systems, and languages.

Interested in staying updated on our progress? Reach out to the Monsana team or follow us on LinkedIn to learn more.

*Note: Results are based on a single evaluator’s manual review, which may limit objectivity and completeness. Interpret as indicative, not definitive.

**Accuracy was tested on a separate real-world dataset. For details, contact valerie.vandeweerd@monsana.ai