Related Factors Of Benign And Malignant Superficial Lymph Nodes And Construction Of Diagnostic Model

Background and Objective Superficial lymph node enlargement is a clinical symptom that can occur in many diseases such as bacterial infection,virus invasion,and tumor.Different etiologies lead to completely different clinical prognosis.Therefore,the differential diagnosis and malignant grading of enlarged lymph nodes are very important.The diagnosis of enlarged superficial lymph nodes mainly depends on medical history,physical examination,hematological examination,B-ultrasound,MRI,PET-CT and pathological examination.Pathological diagnosis is the gold standard for differentiating benign and malignant lymph nodes,but lymph node biopsy or excision biopsy is an invasive procedure and waiting a long time for the result.The PET-CT could reflect the metabolic activity of enlarged lymph nodes through the standard uptake value（SUV）of isotope fluorine-18-labeled glucose（18F-FDG）,indirectly infer its benign and malignant,and can evaluate whether there is metastasis in other parts of the body,however it is too expensive for some patients to afford it.B-ultrasound has the advantages of economy,non-invasiveness,portability,etc.It is the first choice of primary screening for patients with superficial lymphadenopathy.B-ultrasonography can accurately and conveniently measure the long and short diameters,morphological characteristics,borders,changes of the cortex and medulla of the lymph nodes and the characteristics of the lymph node hilum,which provide an important reference for the diagnosis of benign and malignant lymph nodes.Routine blood test,the classification and counting of whole blood cells,which is the most common method to evaluate the inflammatory state,hematopoietic function and nutritional status of the body.The purpose of this study is to comprehensively analysis the related clinical characteristics between benign and malignant lymph nodes,to screen core characteristics through machine learning algorithms,and to establish a new diagnostic model based the integration of B-ultrasound and routine blood test characteristics,and to assist the diagnosis of benign and malignant superficial lymph nodes more conveniently and accurately.Methods Part Ⅰ:Systematic retrieval of the clinical data of all patients with superficial lymphadenopathy in our hospital from January 2018 to June 2019,who had a definite pathological diagnosis and received B-ultrasound and routine blood tests within 30 days before the pathological examination was submitted（a total of 866 patients were included,including 275 patients with benign lymph nodes and 591 patients with malignant lymph nodes）.We comprehensively analyzed the different characteristics（B-ultrasound and clinical features,as well as routine blood test,blood coagulation,blood biochemistry and other hematological indicators）between the two groups of patients with benign and malignant superficial lymph nodes,and explored the potential indicators which related to the identification of benign and malignant lymph nodes.Part Ⅱ:A total of 866 patients were randomly divided into training set（n=433）and testing set（n=433）,and the baseline data of the two data sets were balanced.Firstly,we analyzed the diagnostic performance of a single factor（clinical,B-ultrasound as well as the routine blood test characteristics）for benign and malignant superficial lymph nodes in all of the training set,testing set and the overall data set.In the training set,the core characteristics of clinical and B-ultrasound was selected by machine learning LASSO regression for identifying benign and malignant lymph nodes.The diagnostic model（P_B score）based on the selected core characteristics was established.Similarly,the diagnostic model（Pb score）based on routine blood test indexes was established.We further combined clinical,B-ultrasound and routine blood test indicators to establish a comprehensive diagnostic model(P_Bb score).We also constructed a P_Bbscore visual nomogram to facilitate the prediction of benign and malignant lymph nodes.In order to ensure the accuracy of the above three models,the diagnostic performance was validated in the testing set and the overall data set.Results Part Ⅰ:Difference analysis of clinical and B-ultrasound characteristics showed that in the malignant lymphoid group,elderly patients（83.6%vs.16.4%）,unilateral lymphadenopathy（70.2%vs.29.8%）,short diameter greater than 1.0 cm（74.8%vs.25.2%）,long diameter/short diameter less than 1.8（92.2%vs.7.8%）,irregular shape（86.6%vs.13.4%）,ill-defined（80.0%vs.20.0%）and ill-defined cortico-medulla（80.5%vs.19.5%）were significantly higher than those in the benign lymph node group（all P<0.05）.Difference analysis of hematological indexes showed that in malignant lymphoid group,white blood cell count[6.21（2.63）vs.5.64（2.50）],neutrophil/lymphocyte[2.62（1.86）vs.2.17（1.59）],neutrophil percentage（0.65±0.11 vs.0.60±0.13）,red blood cell distribution width SD（44.33±5.04 vs.43.29±4.78）,fibrinogen degradation products[1.87（2.18）vs.1.85（1.55）],prothrombin activity（97.18±15.36 vs.93.65±16.50）,carbon dioxide（24.91±2.73 vs.23.92±2.63）and creatinine（78.25±21.09 vs.71.66±17.91）were significantly higher than those in benign lymph node group（all P<0.05）,while lymphocyte count（0.26±0.10 vs.0.29±0.11）,monocyte count（0.07±0.03 vs.0.08±0.03）,activated partial prothrombin time（25.59±4.82 vs.27.25±6.18）,prothrombin time（18.13±1.26 vs.18.38±1.15）and serum Albumin（42.29±4.55 vs.43.69±3.59）was significantly lower than that in benign lymph node group（all P<0.05）.Part Ⅱ:Single factor analysis of the diagnostic performance of benign and malignant shown that,the long/short diameter of clinical and B-ultrasound features has the highest accuracy in diagnosing benign and malignant lymph nodes,with an accuracy of 69.5%;the percentage of neutrophils in routine blood test indicators can diagnose lymph nodes Benign and malignant have the highest accuracy,with an accuracy of 60.5%.In addition,the results of the diagnostic model（P_B score）based on clinical and B-ultrasound features showed that the diagnostic efficacy of the P_B score in the overall dataset was AUC=0.729,the sensitivity was 69.9%,the specificity was 62.9%,the positive predictive value was 80.2%,and the negative predictive value was 49.3%.0.741,70.9%,66.4%,82.4%and 50.6%in training set;0.707,68.8%,59.6%,77.9%and 48.0%in testing set,respectively.The results of the diagnostic model（P_b score）based on routine blood test indicators showed that the diagnostic efficacy of P_b score in the overall data set was AUC=0.636,the sensitivity was 44.7%,the specificity was 72.3%,the positive predictive value was 80.0%,and the negative predictive value was 39.0%;0.673,48.5%,79.9%,84.3%,and 41.0%in the training set;0.600,40.8%,72.3%,75.3%,and 37.1%in the testing set,respectively.The results of the comprehensive diagnostic model(P_Bb score)based on the combination of clinical,B-ultrasound and routine blood test showed that the diagnostic efficacy AUC=0.745 in the overall data set,the sensitivity of the P_Bb score was 75.0%,the specificity was 61.8%,and the positive predictive value was 80.8%.The negative predictive value was 53.5%;AUC=0.757,75.9%,66.2%,82.6%,and 56.4%in the training set;AUC=0.735,74.1%,57.4%,79.1%,and 50.3%in the testing set,respectively.Conclusion This study comprehensively analyzed the characteristics of clinical and B-ultrasound characteristics,as well as the value of routine blood test,blood coagulation,liver and kidney function and other hematological indicators in the differentiation of benign and malignant superficial lymph nodes,and further explored the single factor of the above indicators in the diagnosis efficacy of benign and malignant lymph nodes.This study innovatively established a diagnostic model（P_B score）based on clinical and B-ultrasound features,a diagnostic model（P_b score）based on routine blood test indicators,and a comprehensive diagnostic model(P_Bb score)based on a combination of clinical,B-ultrasound and routine blood test.The P_B score has high sensitivity but low specificity;while P_b score has high specificity but low sensitivity;P_Bb score model can more accurately distinguish benign and malignant superficial lymph nodes,and its diagnostic accuracy is 74.5%,sensitivity is 75.0%and specific is 61.8%.The visual nomogram of P_Bb score with accessibility indicators,which is easy for clinical promotion and application,as well as for guiding the follow-up diagnosis and treatment.It might have more clinical application value in"first-level"and"second-level"hospitals with relatively poor medical resources.