NCT07047768: AIRISE

Artificial Intelligence for Respiratory Infections SEverity Prediction

Active, enrolled Last updated 2 July 2025

What this trial tests

trial testing No intervention : data-based study in Infectious Respiratory Diseases in 52,000 participants. Participants enrolled and being followed up; not accepting new ones.

Timeline

7 January 2025

Primary endpoint
31 October 2025

30 April 2027

Quick facts

Lead sponsor	Hospices Civils de Lyon
Status	Active, enrolled
Study type	OBSERVATIONAL
Enrollment	52,000
Start date	7 January 2025
Primary completion	31 October 2025
Estimated completion	30 April 2027
Sites	1 location across France

Drugs / interventions tested

• No intervention : data-based study

Conditions studied

• Infectious Respiratory Diseases — all drugs for Infectious Respiratory Diseases →
• Hospitalized Patients — all drugs for Hospitalized Patients →
• Adult Patients — all drugs for Adult Patients →

Sponsor

Hospices Civils de Lyon — full company profile →

Who can join

18 and older, any sex, with Infectious Respiratory Diseases or Hospitalized Patients. Patients with the condition only — healthy volunteers not accepted.

Sponsor's own description

Health Data Warehouses (HDWs) are a major resource for the development of artificial intelligence (AI) applied to predictive and personalized medicine. We propose a project leveraging the HDW of the Hospices Civils de Lyon (HCL) to study acute lower respiratory tract infections (ALRTIs), a major public health issue due to their impact on morbidity, mortality, and healthcare costs. The COVID-19 pandemic has further highlighted their burden and complexity.

ALRTIs can be caused by viral agents (e.g., influenza, RSV, SARS-CoV-2) or bacterial pathogens (e.g., pneumococcus, mycoplasma, legionella), and may be acquired in the community or during hospitalization. Given their frequency and potential severity, early identification of patients at risk of clinical deterioration is crucial, especially those likely to require intensive care.

The recent deployment of the HCL HDW now allows for the structured extraction, linkage, and storage of administrative, clinical, biological, and pharmaceutical data. This system supports the reconstruction of each patient's care trajectory and clinical history, offering new opportunities for advanced modeling.

In recent years, several predictive tools have been developed to estimate the severity or prognosis of respiratory infections, including PSI/FINE, qSOFA, CURB-65, the EPIC sepsis model, and early warning systems (EWS). The COVID-19 crisis spurred the creation of new scores and models to predict clinical outcomes or mortality, as well as online tools and apps for clinicians. However, many of these tools rely on limited datasets (often single-center or small cohorts), static variables (e.g., comorbidities), and do not consider the temporal dynamics of patient data.

Some research teams have explored the use of multicenter data and machine learning (e.g., MLHO-Machine Learning to predict Health Outcomes), notably to model COVID-19 outcomes. Nonetheless, most models lack integration of longitudinal clinical and biological data, and few are generalizable to all respiratory infections. Additionally, existing tools rarely account for real-time contextual variables such as current levels of population immunity or vaccine availability.

Our project aims to develop a dynamic AI-based detection algorithm to predict the risk of ICU admission in patients with ALRTIs. The model will be trained on retrospective HDW data from the HCL, including the evolution of vital signs, laboratory values, treatments, and demographic factors. By capturing temporal trends and clinical trajectories, our algorithm will go beyond static scoring systems and offer real-time risk stratification.

Ultimately, this algorithm could be embedded in hospital information systems as a clinical decision support tool. By generating alerts for early signs of deterioration, it would enable more timely interventions, resource optimization, and improved patient outcomes.

This approach differs from existing models in two fundamental ways. First, it covers a broad patient population with viral and bacterial pneumonia of both community and hospital origin. Second, it explicitly incorporates the longitudinal dimension of health data, allowing the model to learn from dynamic changes in patient condition. This temporal perspective is key to improving prediction accuracy and enabling early detection of deterioration.

Publications & conference data

No peer-reviewed publications indexed yet for this trial.