Development and validation of a machine learning-based nomogram for predicting prognosis in lung cancer patients with malignant pleural effusion

With advancements in medical technology and a deeper understanding of the pathogenesis of MPE, novel therapies and drugs are continually being developed. For example, research is exploring the use of tumor-derived microparticles to encapsulate chemotherapeutic drugs, allowing for targeted tumor cell destruction while simultaneously activating immune responses to enhance therapeutic efficacy^12,13,14. Additionally, newly identified immunotherapeutic targets, signaling pathways, immunomodulatory cytokines, and non-coding RNAs delivered via exosomes hold promise for advancing immunotherapy in MPE¹⁵. Despite the availability of various treatment options, no consensus has been reached on the most effective approach. Therefore, further investigation into the mechanisms of MPE formation, early prognostic assessment, and personalized treatment strategies is essential.

This study found that the treatment regimen (targeted therapy, other treatment, or no treatment), presence of pericardial effusion, and total volume of pleural effusion were significantly associated with patients’ one-year survival rate. Targeted therapy, by specifically targeting molecular markers of tumors, can more precisely inhibit tumor growth and spread^16,17,18, thereby reducing the incidence of malignant pleural effusion. Due to its high efficacy and relatively low side effects, targeted therapy offers superior long-term outcomes compared to other treatment options, such as chemotherapy and radiotherapy¹⁹. In the cases examined in this study, cancer cells typically spread to the pleura and pericardium via the lymphatic system or bloodstream, leading to the formation of pleural and pericardial effusions^20,21. The coexistence of these conditions, along with larger volumes of malignant pleural effusion, often indicates advanced cancer progression and greater therapeutic challenges. Pericardial effusion directly affects cardiac function, potentially causing heart failure, hypotension, and circulatory collapse^22,23,24. Meanwhile, pleural effusion exerts a compressive effect on the lungs, impairing respiratory function, oxygen exchange, and overall cardiopulmonary health^7,20. As the volume of effusion increases, these compressive effects intensify. Large amounts of pleural effusion can further compress the heart, restricting its function⁶. Similarly, pericardial effusion can compress the lungs, limiting pulmonary expansion, reducing ventilation capacity, and exacerbating hypoxemia. Together, these factors contribute to poor patient prognosis.

This study identified three critical variables—treatment regimen, pericardial effusion, and total pleural effusion volume—for prognostic prediction in lung cancer patients with MPE. The LR model showed the best performance, with AUCs of 0.885 (training), 0.954 (internal testing), and 0.920 (external testing cohort 1). The nomogram achieved AUCs of 0.962 (external testing cohort 2) and 0.949 (temporal external validation cohort), with excellent calibration and clinical utility (net benefit thresholds: 0.01–0.95 and 0.02–0.96, respectively). The nomogram stratifies patients into four risk groups—low-risk (≤ 19 points), moderate-risk (19–88 points), high-risk (88–162 points), and very high-risk (≥ 162 points)—providing an effective tool for predicting one-year mortality and assessing overall survival prognosis.

The treatment of lung cancer with MPE involves both local therapy (symptom relief) and systemic therapy (control of primary tumors and metastases)²⁵. The goal of local therapy is primarily to alleviate dyspnea, reduce effusion recurrence, and improve quality of life. The main local treatment modalities include therapeutic thoracentesis, pleurodesis, hyperthermic intrathoracic chemotherapy, intrapleural drug perfusion, and indwelling pleural catheter. These approaches, which focus solely on managing MPE without targeting the primary tumor, were categorized as “untreated” in our study. In contrast, systemic therapy aims to inhibit tumor progression and prolong patient survival, encompassing systemic chemotherapy, targeted therapy, immune checkpoint inhibitors, and anti-angiogenic therapy. Among these, our study found that targeted therapy provided the greatest benefit to patients compared to all other treatment modalities. Based on the nomogram developed in our study and the risk stratification of patients (low-risk, moderate-risk, high-risk, and very high-risk groups), specific treatment recommendations can be made. For patients in the low-risk or moderate-risk groups, regular systemic therapy is strongly recommended, as these groups achieved median survival times of 27 months and 22 months, respectively. For patients in the high-risk group, long-term systemic therapy may be considered based on their financial situation, although the potential benefit is relatively limited, with a median survival time of only 8 months. For patients in the very high-risk group, the focus should be on reducing MPE and improving quality of life, given their short median survival time of just 2 months.

Despite the strong predictive performance of the nomogram, this study has several limitations. First, the analysis was limited to patients’ laboratory and basic clinical data, lacking an evaluation of lung cancer imaging features. Incorporating tumor imaging characteristics might further enhance the nomogram’s predictive accuracy. Second, although external validation was performed, the external dataset was relatively small. Therefore, additional validation with larger and more diverse samples from various regions is required to enhance the generalizability and reliability of the model.