Establishment of discriminative models for predicting the infiltration degree of patients with lung adenocarcinoma based on clinical laboratory indicators

Abstract

Objective·To establish a multifactorial discriminative model for predicting the degree of infiltration in patients with non-small cell lung adenocarcinoma based on clinically accessible laboratory indicators, such as tumor markers, coagulation function indicators, routine blood count indicators, and biochemical indicators.Methods·A retrospective study was conducted on 202 patients with lung adenocarcinoma admitted to Shanghai Chest Hospital in 2022. Multifactorial Logistic regression analysis was applied to screen independent factors that influenced the predictive infiltration degree of lung adenocarcinoma and to establish a regression model. In addition, machine learning was used to construct a discriminative model, and the area under the receiver operating characteristic curve (AUC) was used to evaluate the discriminative ability of the model to discriminate the degree of infiltration in lung adenocarcinoma patients.Results·A total of 202 patients with lung adenocarcinoma were included in the study, and divided into pre-invasive lesion group (n=59) and invasive lesion group (n=143). Multifactorial Logistic regression analysis revealed that urea, percentage of basophilic granulocytes, and albumin were independent factors for predicting the degree of infiltration of lung adenocarcinoma (all P<0.05). The predictive model expression was P = eX/ (1 + eX), where X = (0.534×urea) + (1.527×percentage of basophilic granulocytes) - (1.916×albumin) + 6.373. Machine learning results showed that the model performed best when urea, fibrinogen, albumin, percentage of basophilic granulocytes, prealbumin and carcino embryonic antigen (CEA) were included. After comparing the performance of 8 machine learning algorithms (based on ridge regression, least absolute shrinkage and selection operator, neural network, random forest, k-nearest neighbors, support vector machine, decision tree, and adaptive boosting algorithms) using the DeLong test, the ridge regression algorithm with the highest AUC was selected. The AUC of the predictive model was calculated to be 0.744 (95% CI 0.656-0.832), with a sensitivity of 70.8% and a specificity of 70.2%.Conclusion·A comprehensive differentiation model constructed by urea, fibrinogen, albumin, percentage of basophilic granulocytes, prealbumin and CEA can effectively predict the infiltration degree of the enrolled lung adenocarcinoma patients, holding the potential to provide more precise guidance for the clinical grading and adjunctive treatment of lung adenocarcinoma.

Keywords

• lung adenocarcinoma
• tumor markers
• coagulation function indicators
• clinical biochemical indicators
• predictive discriminative model