Abstract 387P
Background
Malignant pleural effusion (MPE) is a common condition associated with lung cancer. Prognosis is poor, especially given the challenges associated with accurately predicting survival. Thus, we constructed machine-learning models based on metabolomics to differentially diagnose MPE and predict 1-year mortality in lung cancer patients with MPE.
Methods
A total of 140 (85 with MPE and 55 with benign pleural effusion (BPE)) and 69 (40 and 29 with MPE and BPE, respectively) participants were collected in the discovery and independent multicentre validation cohorts, respectively. Following untargeted metabolomic analysis, logistic analysis and machine learning were utilised to identify metabolites associated with diagnosis and 1-year mortality in the MPE group. We established diagnostic algorithms in the discovery set and applied them to evaluate the results in the validation set. The functions of the candidate metabolites were confirmed in lung adenocarcinoma cell lines.
Results
Differential metabolites were observed in BPE and MPE and non-surviving and surviving groups. Seventy-one metabolites significantly differed between the MPE and BPE groups. There were 26 significantly altered metabolites in the non-surviving and surviving groups, particularly in arginine biosynthesis and phenylalanine metabolism. We developed algorithms for identifying MPE and BPE and predicting 1-year mortality in lung cancer patients with MPE. Palmitic acid significantly inhibited the proliferation of lung adenocarcinoma cell lines, indicating its anti-lung cancer function as a protective metabolite.
Conclusions
Based on novel machine learning-assisted metabolomics, we established models of higher accuracy than existing methods for predicting 1-year mortality in lung cancer patients with MPE.
Clinical trial identification
NCT05906823
Funding
National Natural Science Foundation of China
Disclosure
All authors have declared no conflicts of interest.