AI (Artificial Intelligence), ARDS (Acute Respiratory Distress Syndrome)
Conditions
Keywords
ARDS, chest CT, AI models
Brief summary
By using multi-center chest CT data, an intelligent assessment model for the severity of ARDS was constructed. Based on CT quantitative features and clinical characteristics, a prediction model for short-term critical events (such as mechanical ventilation decisions, prone position strategies, death, ECMO use, etc.) was established. The disease was staged and quantified, and a diagnosis and risk stratification model for ARDS was developed to assist in guiding the diagnosis and treatment strategies for ARDS.
Interventions
DIAGNOSTIC_TESTCT scan
CT scan
Sponsors
Shanghai Zhongshan Hospital
Study design
Observational model
CASE_ONLY
Time perspective
RETROSPECTIVE
Eligibility
Sex/Gender
ALL
Age
18 Years to 100 Years
Healthy volunteers
No
Inclusion criteria
* Meets the diagnostic criteria for ARDS * Be admitted to the intensive care unit * There are chest CT images
Exclusion criteria
* Age less than 18 years old * Missing medical records * No chest CT images
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Accuracy of ARDS severity classification | Baseline, defined as within 24 hours of index chest CT acquisition during ICU admission. | Accuracy of the artificial intelligence-based model in classifying ARDS severity (mild, moderate, or severe), using the reference clinical classification defined by the 2023 global ARDS criteria as the ground truth. |
| Treatment plan matching rate between model-recommended and actual clinical management. | Baseline, defined as within 24 hours of index chest CT acquisition during ICU admission. | Concordance rate between model-recommended treatment strategies and actual clinical management decisions across five predefined intervention modalities: mechanical ventilation, high-flow nasal oxygen therapy, non-invasive ventilation, prone positioning, and neuromuscular blockade. |
| Accuracy of 28-day in-hospital mortality prediction. | Up to 28 days from ICU admission, or until hospital discharge, whichever occurs first. | Accuracy of the model in predicting all-cause in-hospital mortality within 28 days, based on integrated chest CT imaging features and clinical variables. |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Comparative performance improvement over baseline AI models. | Baseline for severity classification and treatment plan matching; up to 28 days from ICU admission for mortality prediction | Absolute performance improvement of the proposed model compared with three commonly used baseline artificial intelligence models across ARDS severity classification, treatment plan matching, and 28-day mortality prediction. |
| Association between treatment concordance and 28-day in-hospital mortality. | Up to 28 days from ICU admission, or until hospital discharge, whichever occurs first. | Association between concordance of model-recommended interventions and actual clinical treatments and 28-day in-hospital mortality, evaluated using multivariable logistic regression adjusted for key imaging-derived structural metrics. |
| Calibration performance of 28-day mortality prediction. | Up to 28 days from ICU admission, or until hospital discharge, whichever occurs first. | Calibration of the mortality prediction model assessed using calibration curves and calibration statistics to evaluate agreement between predicted and observed 28-day in-hospital mortality. |
| Model interpretability based on imaging and clinical feature contributions. | Baseline for feature extraction; up to 28 days from ICU admission for outcome association analysis. | Quantification of the relative contributions of imaging-derived features and clinical variables to mortality prediction using Shapley Additive Explanations (SHAP). Feature importance will be analyzed overall and stratified by ARDS severity. |
Countries
China
Outcome results
None listed