Malignant pleural effusion (MPE) is the most common cause of unilateral pleural exudate.1 MPE accounts for over 150,000 admissions annually, representing a significant economic burden.2 The growing number of cancer cases expected in the coming years, in addition to the enhanced efficacy of systemic anti-cancer therapy, will lead to an increase in the prevalence of MPE.3 The presence of MPE suggests metastatic or advanced disease, with a mean survival ranging from 3 to 12 months.4

At the time of MPE diagnosis, an individual prognosis providing an accurate estimate of survival may help tailor the therapeutic approach. In cases of poor prognosis, patients should be spared unnecessary inconvenience at the end of their lives. In this context, palliative care emerges as the most appropriate approach. In patients with a better life expectancy, more aggressive strategies can be used. In recent years, new pleural procedures have become available for the management of MPE, leading to improved stratification of patients, progressive individualization of treatment, and diversification of outcomes.5

The MPE population is characterized by considerable heterogeneity, with substantial variations in life expectancy influenced by multiple factors. These factors include the type of underlying tumor, as malignant pleural mesothelioma generally presents a more favorable prognosis than metastatic MPE (MMPE), and the overall life expectancy in a specific cohort may be contingent upon the number of mesothelioma cases included.6 Additionally, the functional status of the patient, the prevalence of oncogenic mutations within a particular population or ethnicity, and the presence of various comorbidities also play significant roles.6,7

Currently, two prognostic scores for predicting survival in MPE have been validated (LENT and PROMISE).6,7 However, efforts are being made to develop more accurate predictive survival models that can overcome the limitations of existing scores and be tailored to the demographic characteristics of each region. This study aimed to develop a predictive model based on risk stratification or a continuous predictive scale that estimates life expectancy in unselected patients diagnosed with MMPE, with the potential to facilitate treatment personalization in the future.

In most settings, the GASENT survival predictive model based on a continuous scale yielded significantly higher AUC values than the GASENT scores for risk groups, LENT, and ECOG-PS at different time points (1, 3, and 6 months), and a higher Harrell's C index. Using this model, the probability of survival for a particular patient can be estimated without assigning the patient to a specific risk group. Therefore, this model will help tailor therapeutic approaches based on the prognosis of individual cases.

Consistent with previous studies, the survival range of MMPE patients was very broad, suggesting the influence of different variables on patient survival.6 Hence, because survival depends on multiple factors, the same treatment should not be administered to all the patients. In addition, the use of reliable survival prediction scores is crucial, as the clinical judgment of clinicians may fail to establish an accurate prognosis.15,16

On multivariate analysis, sex, serum NLR, ECOG-PS, cancer type, and albumin level were the only variables independently associated with survival (Table 6). The first four variables were included in the GASENT score because of robust evidence of their association with survival in neoplastic diseases.17–28,8,29–31 Serum albumin was found to be a relevant prognostic factor in the multivariate Cox model; however, the influence of this variable did not translate into a statistically significant effect in the predictive models (logistic regression) used to calculate the score. Consequently, the albumin level was excluded from the model. This inconsistency in the results is due to the methodological differences between the models. Whereas the Cox model includes both time-to-event and censoring, the binary analysis only considers the occurrence or absence of mortality, regardless of the time. Finally, although age did not maintain an independent prognostic impact on survival in the multivariate analysis (p=0.329), it was included in the model, as age is known to contribute to mortality in neoplastic diseases. Moreover, age is the most relevant risk factor for cancer in general and for many types of tumors.32

LENT and PROMISE scores have been validated for predicting survival in MPE.6,7 Several predictive models have been developed for specific tumor types (lung and breast).33,34 In light of the limitations of LENT and PROMISE (the former potentially lacks accuracy owing to its sole reliance on risk group classification, whereas the latter involves building a model for each tumor that needs to be updated as new mutations appear or new treatments become available), the GASENT model represents an alternative approach that provides enhanced results, overcomes the limitations of the two existing models, and adapts to the demographic characteristics of each region.

In recent years, several alternative prognostic scores applicable to all tumor types have been developed to estimate the survival of patients with MPE, and their results have been compared with those of the LENT score. Notably, the SELECT score, which considers factors such as Sex, Eastern Cooperative Oncology Group performance status, Leukocyte count, EGFR mutation status, Chemotherapy, and Type of primary tumor, has been utilized to identify patients with a high likelihood of 90-day survival.35 Additionally, the CONCH prognostic score, which includes CEA, monocyte count, NT-pro-BNP, and chloride values (the latter in PF), is recommended for guiding intervention selection and management of MPE.33 Furthermore, a meta-analysis of five randomized controlled trials involving 553 patients with MPE concluded that dyspnea, assessed using a visual analog scale before and after MPE treatment procedures, serves as a reliable predictor of survival in these patients.29 In summary, despite the advancements achieved with existing predictive models for MPE survival, there remains a continuous pursuit of alternative models to enhance their accuracy, address their limitations, and tailor them to the demographic characteristics of specific regions.

Our study differs notably from the LENT study in several aspects. First, cases of malignant pleural mesothelioma (accounting for over 20% of cases in the LENT study) were excluded, as the incidence of this disease varies across different regions of the world.36 Additionally, survival is higher in mesothelioma than in MMPE, which may result in an increased median survival in each risk group. Second, in the GASENT study, pleural cytology or biopsy results were positive for malignancy in all patients. In contrast, in the LENT study, 28% of cases in the derivation cohort (221 patients) had a negative test result (204 had an effusion of unknown etiology with confirmed malignancy in an organ other than the pleura, and 17 had radiological evidence of malignancy that was not confirmed histologically). This is relevant because the differences in median survival would result from including patients with non-neoplastic PE in the LENT study. Finally, the LENT and PROMISE models only consider three or four mean survival times, which is a limitation for individualized predictions and clinical decision-making.

In our study, the probability of survival at one month in high-risk patients was 55% (Fig. 4A). Does this mean that all patients in this group have the same probability of survival at that time point? The answer is no, as these prognostic scores provide the median probability of survival for the group, but not for each case in that group. The Excel spreadsheet (eFig. 1) details the probability of survival at 1 month for three patients with high-risk MMPE with GASENT scores of 77, 118, and 147 points. This continuous-scale prediction model provides an enhanced prediction of survival in future cases. Higher AUCs were derived from the GASENT model than from the LENT model in the derivation cohort. In the validation cohort, the differences were significant only at 3 months. This was probably due to the high AUC of the LENT model for the validation cohort at one month (0.795 vs. 0.777 for the derivation cohort). Nevertheless, the two scoring systems are solid and consistent with their purpose.37

The GASENT model was solely compared with the LENT model. A comparison with the PROMISE model was not performed because it is more complex, and one of the variables, tissue inhibitor of metalloproteinase 1 (TIMP1), is not available in all hospitals, which prevents its widespread use. The GASENT model is a continuous-scale prediction score. However, the variable “tumor type” was also categorized into groups; otherwise, the prediction model would have severe limitations in low-incidence tumors owing to the broad variability in predictive estimations. The variable “oncogenic mutation” was not included in the score because we aimed to develop a model that was valid for all tumor types. Oncogenic mutations were only available for some cases of lung cancer and breast cancer. If the missing cases had been excluded, the model would have provided less-precise predictions. Finally, biological variability in each neoplasm was not considered in any of the studies (GASENT and LENT), which may have reduced the accuracy of the model.38

A similar study was recently conducted to externally validate the ability of the LENT and PROMISE scores to provide a prognosis for MPE. The study suggests using statistical techniques that identify non-linear relationships between potential biomarkers and disease prognosis, which spares the need for risk categorization and the resulting loss of information.39 Thus, the performance of future predictive models can be improved.

This study had some limitations. The GASENT score was not designed for suspected MPE or paramalignant effusions. In addition, as this was a single-center study, 100% of the study population was Caucasian. This limits the generalizability of the results to populations of other geographic regions or ethnicities. Mutations in lung and breast cancers were only tested from 2016. Although a significant difference was observed in survival, it was not considered for the reasons mentioned above, which probably overestimates the risk of death, since receiving targeted treatment is an independent protective factor against recurrence of MPE.40 As tumor staging was challenging, disease spread was excluded from the analysis. Finally, the model requires external validation.

In conclusion, in patients with MMPE, the GASENT model, a validated continuous-scale predictive score, predicts survival at 1, 3, and 6 months more accurately than currently available models. The use of these models is not widespread in clinical practice, as therapeutic decisions are made based on patient preferences, clinician experience and skills, and equipment available in the hospital. A predictive model that provides perfectly measurable results is necessary to improve clinical practice. However, such a model has not yet been developed. Predictive survival models are expected to be used more frequently in the future, as patients increasingly request accurate estimates of their life expectancy, which clinicians should be able to provide.41 The continuous-scale predictive model developed in our study offers individualized survival predictions that are sufficiently accurate to provide individual information that allows for effective therapeutic decision-making.

A continuous-scale predictive model, such as the GASENT model, may provide sufficiently accurate individual predictions that allow for effective therapeutic decision-making based on individual information.

Juan Súarez-Antelo. Author and drafting. Conception and design. Reviewed intellectual content. Approval of the final manuscript.

Lucía Ferreiro. Co-author and drafting. Reviewed intellectual content. Approval of the final manuscript.

José M. Porcel. Co-author and drafting. Reviewed intellectual content. Approval of the final manuscript.

María E. Toubes. Co-author. Reviewed intellectual content. Approval of the final manuscript.

Óscar Lado Baleato. Co-author. Statistical analysis. Reviewed intellectual content. Approval of the final manuscript.

Nuria Rodríguez-Núñez. Co-author. Reviewed intellectual content. Approved final manuscript.

Jorge Ricoy. Co-author. Data Collect. Reviewed intellectual content. Approval of the final manuscript.

Adriana Lama. Co-author. Reviewed intellectual content. Approved final manuscript.

Antonio Golpe. Co-author. Reviewed intellectual content. Approved final manuscript.

José Manuel Álvarez-Dobaño. Co-author. Reviewed intellectual content. Approved final manuscript.

Francisco Gude. Co-author. Conception and design. Statistical analysis. Reviewed intellectual content. Approval of the final manuscript.

Luis Valdés. Author and drafting. Conception and design. Reviewed intellectual content. Approval of the final manuscript.

The manuscript has not been produced partially or totally with the help of any artificial intelligence software or tool.

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

The authors declare no conflicts of interest.

We wish to confirm that there are no known conflicts of interest associated with this publication and that there has been no significant financial support for this work that could have influenced its outcome. We confirm that the manuscript has been read and approved by all the authors. We further confirm that the order of the authors listed in the manuscript has been approved by all the authors. We confirm that we have given due consideration to the protection of intellectual property associated with this work, and that there are no impediments to its publication.

The Corresponding Author is the sole contact for the editorial process (including the Editorial Manager and direct communication with the office). She is responsible for communicating with the other authors about the progress, submission of revisions, and final approval of proofs. We confirm that we have provided a current and correct email address that is accessible to the Corresponding Author (lferfer7@gmail.com).

We confirm that this manuscript has not been published elsewhere in any other language and is not currently being considered for publication in any other journal.

Lucía Ferreiro on behalf of all the co-authors.