Model Prediction Of Radiation Pneumonia In Ⅲ Stage Non-small Cell Lung Cancer Treated With IMRT/VMAT

Background objective: Radiotherapy and chemotherapy are important ways to treat patients with non-small cell lung cancer who can receive radical radiotherapy but cannot be operated.But the benefits of radiotherapy are sometimes complicated by adverse reactions in normal tissues.For example,the emergence of radiation pneumonia may control the increase of dose and disturb the state of life.Although there are previous restrictions on V20,V30,mean lung dose(MLD)and other indicators obtained in the three-dimensional conformal radiotherapy mode,radiation pneumonia still occurs.In addition,with the development of radiotherapy technology,the prediction of radiation pneumonia in intensity modulated radiotherapy mode needs to be further studied.Therefore,this study discussed the clinical and dosimetric factors under intensity modulated radiotherapy(VMAT/IMRT)to build a predictive model of radiation pneumonia to provide support for clinical treatment.Materials and methods : Methods 161 patients were studied retrospectively.According to the symptom grade of radiation pneumonitis,they were divided into two groups,one group was less than grade 2,and the other group was more than grade 2.Univariate analysis was performed on the collected clinical and dosimetric factors by t/z/χ~2test.Lasso-logistic regression is used to solve the problem of collinearity among research factors,and then a regression model is established.C-index value,Calibration curve,ROC curve,decision curve analysis(DCA)to evaluate the model,internal data random sampling bootstrap method to verify the prediction ability of the model.Research results: For RP patients with grade 2 or more,univariate analysis showed that there were significant differences in clinical factors,such as age(t=-2.254,P=0.026),basic lung disease(χ~2=8.187,P=0.004)and chemotherapy(sequential/synchronous)(χ~2=4.529,P=0.033)(P<0.05).Among the dosimetric factors,univariate analysis showed that patients with diseased side V10(t=-2.127,P=0.037),diseased side MLD(Z=-2.344,P=0.019),opposite side MLD(Z=-2.437,P=0.015)、whole lung V5(t=-1.990,P=0.049),whole lung V10(t=-2.683,P=0.009),whole lung V20(Z=-3.720,P=<0.001),whole lung V30(Z=-3.017,P=0.003)were statistically significant(P < 0.05).A single factor of analytical significance in lasso-logistic regression equations,and after excluding the correlation between variables,the results included age(B=0.071,P=0.005,95%CI: 1.021,1.129),basic lung disease(B=1.068,P=0.007,95%CI: 1.336,6.342),chemotherapy(sequential /synchronous)(B=1.032,P=0.018,95%CI: 1.191,6.619)and whole lung V20(B=0.500,P=0.036,95%CI: 1.033,2.631)became the independent prognostic factors of grade 2 RP patients.Verified by ROC curve,the maximum AUC value of the compound model of 4 indexes(whole lung V20 + age +basic lung disease + chemotherapy)was 0.754(95%CI : 0.675,0.833,P <0.001).Sensitivity was 58.9% and specificity was 81%.The calibration curve shows a good agreement between the prediction model and the ideal model.When it is lower than 37% and higher than 69%,the model slightly overestimates the prediction probability,while between 37% and 69%,the model slightly underestimates the prediction probability.The decision curve shows that it is beneficial to use the RP risk model in the positive range of 4%-84.4%.The C index of the prediction model is 0.755(95%CI: 0.676,0.834).Finally,through the random sampling verification of internal data,it is confirmed that the C-index value is 0.729,suggesting that the model is still applicable after verification.Conclusion: Age,basic lung disease,chemotherapy(sequential/synchronous),diseased side V10,diseased side MLD,opposite side MLD,whole lung V5,whole lung V10,whole lung V20,and whole lung V30 are the risk factors for patients with ≥2grade RP.Age,basic lung disease,chemotherapy(sequential/synchronous)and whole lung V20 were independent prognostic factors in patients with ≥2grade RP.The compound model of four indexes(whole lung V20 + age + basic lung disease +chemotherapy)can be used as the best predictive model of RP≥2 grade.