Background:Lung cancer is an inflammatory disease with a high incidence of morbidity and mortality worldwide.In China,lung cancer mortality and morbidity topped the list,putting a huge burden on national health care.If the prognosis of a certain treatment can be accurately predicted in lung cancer patients,so that potential benefit groups would be screened out,and patients without obvious benefits would be eliminated.It will be very important for improv:ing the effect of some corresponding treatment clinically.Many predictors are currently proposed in a large number of studies,in which inflammatory composites occupy a very impo:rtant part,which is significantly associated with the prognosis of a variety of malignant tumors,especially digestive tract tumors.There are also many inflammatory composite markers that have been shown to be independent prognostic factors in lung cancer,including neutrophil to lymphocyte ratio(NLR),platelet to lymphocyte ratio(PLR),lymphocyte to monocyte ratio(LMR)etc.Here comes a problem of choosing the optimal indicator.How to find the optimal prognostic indicators for a specific patient population to achieve a most accurate predicting ability for the survival of lung cancer patients is clinicallly important.Surgical treatment is one of the main treatments for lung cancer patients to improve their survival time.However,different pathological types and stages of lung cancer have various sensitivities to surgical treatment.Accurately predicting the surgical outcomes of lung cancer with various pathological types and stages can help to indicate decision-making of surgical programs.Therefore,it is necessary to develop a reasonable selection strategy to select the optimal prognostic indicators for a specific lung cancer surgery patient population.Purposes:We develop an index selection strategy,the intersection of mathematical set operations,and examine its effectiveness and rationality of screening the optimal indicators in the lung cancer patients undergoing surgery.It is expected to help clinicians choose the best predictors.So far,we have not seen similar reports in lung cancer research.Methods:We systematically collected and followed up patients who underwent pulmonary tumor resection in the Department of Thoracic Surgery,Shandong Provincial Hospital from August 2013 to December 2015,and obtained pathological diagnosis,postoperative survival time,lymphocyte count(LY),neutrophil count(PMN),monocyte count(MNC),platelet count(PLT)and other detailed data.Based on these data,inflammatory composite indicators are calculated,including neutrophil to lymphocyte ratio(NLR),platelet to lymphocyte ratio(PLR),lymphocyte to monocyte ratio(LMR)etc.We divided the patients into symptom-positive group,symptom-negative group,adenocarcinoma group,squamous cell carcinoma group and other cancer group according to the complaint and pathological type respectively.Kaplan-Meier survival curve was used to analyze the relationship between overall survival(OS)and composite indicators.The Log-Rank(Mantel-Cox)test and the Breslow(Generalized Wilcoxon)test verify the long-term prognosis and short-term prognosis of the patients,respectively.The P value obtained by Log-Rank(Mantel-Cox)test is compared with P=0.05.The index with P<0.05 is defined as recommended predictive index,while the index with P?0.05 is defined as non-recommended predictive index.Corresponding recommended/non-recommended predictive indexes for each group are gathered,and recommended indexes are picked out as a set to prepare for the intersection operation.We divided the sample population into 12 sub-categories according to the complaint(positive/negative),pathological type(adenocarcinoma/squamous cell carcinoma/other types),and preoperative and postoperative indexes(2 X 3 X 2=12),corresponding to the 12 specific patient populations we analyzed(such as "symptom-positive adenocarcinoma patients with preoperative indicators only").Each set is analyzed according to the elements of 12 sub-categoriies,and the intersection is obtained under the rule of the mathematical set operations,and the recommended predictive indexes under each sub-category are obtained.Finally,the prediction efficiency of each predictive index in sub-category population is calculated and sorted by the area under the curve(AUC)of ROC.We executed the operation for two reasons.First,in order to find the optimal index;second,to further verify whether the obtained intersection indexes have greater predictive power,so as to verify the rationality of the selection strategy.Results:We followed up 535 postoperative lung cancer patients.The recommended predictive indexes of P(Log-Rank)<0.05 were summarized in each group,and the intersection was obtained to get 9 preoperative recommended predictive indexes for patients with symptomatic positive adenocarcinoma:LMR%,LY%,NLR,PMN%,PMN,MNC,PLR%,LMR,NLR%;10 preoperative predictive indicators recommended for patients with symptomatic positive squamous cell carcinoma:LMR%,NLR,PMN%,PLR%,PLR,LMR,LMR-PLR%,LMR-PLR,LY,NLR%.One preoperative predictor can be recommended for patients with symptomatic negative adenocarcinoma:LY%;one postoperative predictive index recommended for patients with symptomatic positive squamous cell carcinoma:LMR.Patients in adenocarcinoma group with positive symptoms and negative symptoms have no recommended postoperative predictive indicators;patients in other cancer group with symptomatic negative squamous cell carcinoma were not eligible for statistical calculation because of the small number of inclusions.Thus no satisfactory indicators were obtained in the rest groups.The results of ranking the predictive performance of the recommended indicators show that the predicted indicators of the intersections are ranked in the forefront of the list.Among the sub-classified populations who meets the statistical calculation requirements,the preoperative predictive indicator recommended for symptomatic positive adenocarcinoma patients is:NLR%,P(Log-Rank)=0.000<0.05.The preoperative predictor recommended for patients with symptomatic positive squamous cell carcinoma is:PLR%,P(Log-Rank)= 0.000<0.05.The preoperative predictor recommended for symptom-negative adenocarcinoma patients is:LY%,P(Log-Rank)=0.022.And the postoperative predictor recommended for symptomatic positive squamous cell carcinoma patients is:LMR,P(Log-Rank)=0.040.Conclusion:Among the patients undergoing lung cancer surgery,the mathematics set intersection algorithm is an effective strategy for rational selection of inflammatory composite indicators for predicting the prognosis,and can effectively improve the prediction accuracy of clinical survival of lung cancer patients after surgery.In this study,we found that the recommended preoperative index are more than postoperative indicators,and the range of choice is wider,the predictive power is higher.The predictive indicators selected by the algorithm have stronger predictive power and have good clinical applicability.However,the algorithm still needs to be improved,and further research should be warranted. |