Prognostic model construction of lung infection in patients with chemoradiotherapy for esophageal cancer based on SMOTE algorithm
10.3760/cma.j.cn371439-20230621-00045
- VernacularTitle:基于SMOTE算法的食管癌放化疗患者肺部感染的预后模型构建
- Author:
Jing LIU
1
;
Qin LIU
;
Mei HUANG
Author Information
1. 四川省南充市中心医院中西医结合科，南充 637000
- Keywords:
Esophageal neoplasms;
Pneumonia;
Risk factors
- From:
Journal of International Oncology
2024;51(5):267-273
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To explore the independent risk factors of lung infection in patients with esophageal cancer treated with chemoradiotherapy and to establish an individualized early warning model based on synthetic minority oversampling technique (SMOTE) algorithm.Methods:A total of 197 patients with esophageal cancer who received concurrent chemoradiotherapy in Nanchong Central Hospital of Sichuan Province from January 2016 to March 2022 were selected as the study objects. Patients were categorized into the infected group ( n=23) and the uninfected group ( n=174) according to whether they developed lung infection during treatment. The clinical data of patients in both groups were collected, and independent risk factors for lung infection were screened using univariate and binary logistic regression analysis, and a logistic regression model ( P1) was established, while an early warning model ( P2) was constructed based on the improved dataset with the SMOTE algorithm, and the predictive efficiency of the model was compared by receiver operator characteristic (ROC) curve. Results:The incidence of lung infection in 197 patients was 11.7% (23/197), Univariate analysis showed that there were statistically significant differences in the age ( t=3.53, P=0.001), the proportion of patients with a smoking index of ≥200 cigarette-years ( χ2=7.64, P=0.006), the proportion of patients with concomitant radiological lung injury ( χ2=5.41, P=0.020), the proportion of patients with comorbid diabetes mellitus ( χ2=6.71, P=0.010), the proportion of patients with chronic obstructive lung disease ( χ2=3.92, P=0.048) and forced expiratory volume in one second/forced vital capacity (FEV 1/FVC) ( t=3.93, P<0.001) of patients between the infected group and the uninfected group. Logistic regression multivariate analysis found that increasing patient age ( OR=1.09, 95% CI: 1.02-1.16, P=0.008), decreased FEV 1/FVC ( OR=0.92, 95% CI: 0.87-0.98, P=0.005), combined diabetes mellitus ( OR=3.29, 95% CI: 1.22-8.91, P=0.019), smoking index ≥200 cigarette-years ( OR=4.02, 95% CI: 1.42-11.41, P=0.009) and combined radiation lung injury ( OR=4.75, 95% CI: 1.26-17.85, P=0.021) were independent risk factors for the occurrence of lung infection during simultaneous chemoradiotherapy in patients with esophageal cancer. Probabilistic prediction model logit ( P1) =-2.760+0.084×age-0.081×FEV 1/FVC+1.191×diabetes+1.392×smoking index+1.558×radiation lung injury. The early warning model logit ( P2) =-1.544-0.127×age-0.115×FEV 1/FVC+1.599×diabetes+1.434×smoking index+1.748×radiation lung injury. ROC curve analysis showed that the sensitivity of P1 and P2 models were 0.826 and 0.897, the specificity were 0.747 and 0.793, and the Youden index were 0.573 and 0.690, respectively. The area under curve of P2 model was 0.903 (95% CI: 0.872-0.934), which was significantly higher than 0.843 (95% CI: 0.763-0.923) of P1 model, with a statistically significant difference ( Z=13.23, P=0.002) . Conclusion:Increasing patient age, decreased FEV 1/FVC, smoking index ≥200 cigarette-years, combined diabetes mellitus and radiation lung injury are strongly associated with the occurrence of lung infection during simultaneous chemoradiotherapy in patients with esophageal cancer. The individualized early warning model established by SMOTE algorithm can significantly improve the predictive efficacy of patients' occurrence of lung infection.
