AI-based Predictive and Interventional System for Early Detection of Non-compliance Risks With Oral Therapies in Lymphoma Patients. (LNH-AI-Tools)

AI-based Predictive and Interventional System for Early Detection of Non-compliance Risks With Oral Therapies in Lymphoma Patients, Integrating the Complete Care Pathway and an Interoperable Clinical Interface With Algorithms Paired With Explainability Tools.

This research forms part of a continuous quality improvement initiative. It aims to assess patient compliance of oral therapies by artificial intelligence. It could overcome the limitations of current practices and enhance the responsiveness and accuracy of clinical interventions.

Study Overview

• Status: Recruiting
• Conditions: Lymphoma, Non-Hodgkin; Care Coordination
• Intervention / Treatment: Other: Retrospective Group; Other: Prospective Group

Detailed Description

Non- Hodgkin Lymphomas require rigorous treatment protocols, including intensive intravenous chemotherapy or targeted oral therapies. Secondary immunosuppression necessitates oral anti-infective prophylaxis (such as valacyclovir or Bactrim forte) to prevent opportunistic complications. However, the literature reports figures of up to 50% of patients experiencing adherence difficulties on oral therapies, compromising treatment efficacy, increasing the risk of severe infections, prolonged hospitalizations, and consequently, additional costs for the healthcare system. This project proposes to develop an innovative artificial intelligence (AI) tool, based on real-world data, to detect early signs of non-adherence and enable targeted intervention by healthcare teams. Our approach combines analysis of clinical data (patient, disease, dispensing history, laboratory results, drug interactions) and machine learning algorithms (supervised machine learning and neural networks) to identify at-risk profiles. The tool will generate a real-time alert and offer the patient's referring physician and coordinating nurse tailored recommendations, such as an automated reminder, a dedicated nursing consultation, etc. An intuitive interface will allow clinicians and nurses to visualize compliance trends and act quickly. This project relies on a multidisciplinary team (hematologists, advanced practice nurses (APNs), data scientists, AI experts) and patient partners to validate the tool in real-world conditions.

• Study Type: Observational
• Enrollment (Estimated): 210

Contacts and Locations

• Study Contact: Name: Marie Detrait, MD, PhD; Phone Number: 0032 60 11 20 08; Email: marie.detrait@ghdc.be
• Study Contact Backup: Name: Aline Gillain, MedSciences; Phone Number: 0032 60 11 00 89; Email: aline.gillain@ghdc.be

Study Locations

• Belgium
  • Hainaut
    • Charleroi, Hainaut, Belgium, 6060
      • Recruiting
      • Grand Hôpital de Charleroi
      • Contact: Marie Detrait, MD, PhD; Phone Number: 0032 60 11 20 08; Email: marie.detrait@ghdc.be
      • Contact: Marie Detrait, MedSC; Phone Number: 0032 60 11 00 89; Email: aline.gillain@ghdc.be
      • Principal Investigator: Delphine Pranger, MD
      • Principal Investigator: Marie Detrait, MD, PhD
      • Sub-Investigator: Stéphanie De Prophetis, Nurse

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Adult; Older Adult
• Accepts Healthy Volunteers: No

• Sampling Method: Probability Sample

Study Population

Patients treated in the Haematology Department at Charleroi General Hospital for Non-Hodgkin lymphoma.

Description

Inclusion Criteria:

• All patients aged 18 and over who are treated in the Haematology Department at the Grand Hôpital de Charleroi from November 2025 onwards
• Treated for a lymphoma, Non Hodgkin
• Capable of giving informed consent

Exclusion Criteria:

• All other patients who did not meet the eligibility criteria

Study Plan

How is the study designed?

Number of groups / cohorts

2

Cohorts and Interventions

Group / Cohort	Intervention / Treatment
Retrospective cohort; A retrospective cohort from 2019 to 2024 comprising 350 lymphoma patients who were monitored on an empirical basis.	Other: Retrospective Group; For the retrospective group of 20 patients.
Prospective cohort; A prospective cohort study involving up to 210 consecutive patients, starting in November 2025, with the aim of developing a decision-support tool using machine learning.	Other: Prospective Group; Follow-up of the patients for the prospective group

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
ROC-AUC	Description: ROC-AUC : Receiver Operating Characteristic - Area Under the Curve is a performance metric for binary classification prediction algorithms. ROC Curve: Plots the True Positive Rate (sensitivity) against the False Positive Rate (1-specificity) at various classification thresholds. AUC: The area under this curve (ranging from 0 to 1). A higher AUC indicates better model performance-1.0 is perfect, 0.5 is random guessing. ROC-AUC evaluates how well the model distinguishes between classes, regardless of the classification threshold. Time Frame: When the data will be avalaible, at the end of 2027	2027

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
F1-score	F1-Score is a performance metric for classification algorithms, the harmonic mean of Precision (correct positive predictions / total positive predictions) and Recall (correct positive predictions / actual positives). Formula: F1 = 2 × (Precision × Recall) / (Precision + Recall) Range: 0 to 1, where 1 is perfect precision and recall, and 0 is the worst. F1-Score balances precision and recall, making it ideal when you need to avoid both false positives and false negatives.	When the data will be avalaible, at the end of 2027

Other Outcome Measures

Outcome Measure	Measure Description	Time Frame
Recall for the positive class	Recall for the positive Class is a metric for binary classification that answers: "What proportion of actual positives was correctly identified by the model?" Formula: Recall = True Positives / (True Positives + False Negatives) Range: 0 to 1, where 1 means all positives were correctly predicted, and 0 means none were. High recall means the model is good at capturing most positive cases, but it may also include more false positives. It's critical when missing a positive (false negative) is costly.	2027

Collaborators and Investigators

• Sponsor: Grand Hôpital de Charleroi
• Investigators: Principal Investigator: Marie Detrait, MD, PhD, Grand Hôpital de Charleroi

Study record dates

Study Major Dates

• Study Start (Actual): February 15, 2026
• Primary Completion (Estimated): December 31, 2027
• Study Completion (Estimated): February 15, 2029

Study Registration Dates

• First Submitted: April 8, 2026
• First Submitted That Met QC Criteria: April 20, 2026
• First Posted (Actual): April 22, 2026

Study Record Updates

• Last Update Posted (Actual): April 22, 2026
• Last Update Submitted That Met QC Criteria: April 20, 2026
• Last Verified: April 1, 2026

More Information

Terms related to this study

• Keywords: Artificial intelligence; Non-Hodgkin lymphoma; Care pathway
• Additional Relevant MeSH Terms: Neoplasms; Immune System Diseases; Neoplasms by Histologic Type; Lymphatic Diseases; Lymphoproliferative Disorders; Immunoproliferative Disorders; Lymphoma; Hemic and Lymphatic Diseases; Lymphoma, Non-Hodgkin
• Other Study ID Numbers: LNH-AI-Tools

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: NO
• IPD Plan Description: At present, this research is being carried out in-house; following analysis, this option could be considered if the model can be adapted for use elsewhere.

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No