Establishment method of malignant probability prediction model of isolated pulmonary nodules and method for comparing prediction rates of XJTUFAH, Mayo, VA and PKUPH models

Abstract

The invention relates to a method for establishing an isolated pulmonary nodule malignancy probability prediction model, which comprises the following steps of: acquiring basic information of a patient and tissues of suspicious pulmonary nodules, dividing the patient into a modeling group and a verification group in combination with a CT (Computed Tomography) imaging report, setting clinical variables, applying Logistic regression to significant factors of the modeling group to obtain independent prediction factors of a solid SPN (Specific Patient Nodule), and establishing a model for predicting the malignancy probability of the solid SPN. Establishing an SPN malignant probability prediction model, namely an XJTUFAH model; the invention further provides a method for comparing the malignant probability prediction rate of the isolated pulmonary nodules by using the XJTUFAH model with the Mayo model, the VA model and the PKUPH model, case data of a verification group are respectively input into the established XJTUFAH model, the Mayo model, the VA model and the PKUPH model, an ROC curve is drawn, the area AUC under the curve, the sensitivity and the specificity of a quantitative system are estimated, and then SPSS20.0 software (IBM, Armonk, Erk) is used for calculating the malignant probability prediction rate of the isolated pulmonary nodules. NewYork) is subjected to statistical analysis, and prediction rates corresponding to the four sets of models are calculated. The method is high in comparison data accuracy of the prediction rate, and is simple and easy to implement.

Description

Technical Field

[0001] This invention relates to the field of medical technology, and in particular to a method for establishing a malignancy probability prediction model for solitary pulmonary nodules. Background Technology

[0002] Lung cancer is one of the leading causes of cancer death worldwide, with 80% of cases diagnosed as advanced-stage lung cancer due to late-onset symptoms. Studies show that the 5-year survival rate for lung cancer is only 7-13%, while regular screening can detect early-stage and localized lung cancer. Among these, pulmonary nodules are early-stage lung cancers with curable potential and can be classified as solid or subsolid lung cancers. Currently, the academic community generally recognizes a round radiographic nodule (SPN) with a diameter of 3 cm or less and surrounded by lung tissue as a solitary pulmonary nodule.

[0003] According to the National Comprehensive Cancer Network guidelines, high-risk patients with stable disease should continue to undergo CT scans until they no longer require final treatments such as surgery, chemotherapy, and radiotherapy. Lung rads are currently used for lung cancer screening, with low-dose computed tomography (LDCT) detecting nodules. Although low-dose spiral CT can obtain high-quality chest images at 10% to 30% of the conventional radiation dose, Berrington DE Gonzalez et al. found a synergistic effect between radiation and smoking. While the Mayo and VA models screen all eligible SPNs based on chest X-ray results, small SPNs and other types of nodules may be missed. Furthermore, these models are designed for Western populations, limiting their applicability to Asian populations, and the PKUPH model also suffers from limited applicability and low predictive value. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art. This invention provides a method for establishing a malignancy probability prediction model for solitary pulmonary nodules. This method is highly accurate, simple and easy to implement, and can be well applied to Asian populations, especially in Northwest China. It can provide effective intermediate information for further diagnosis and treatment.

[0005] The technical solution of this invention: a method for establishing a malignancy probability prediction model for solitary pulmonary nodules. The establishment method is as follows: Basic patient information and tissue samples of suspected pulmonary nodules were collected, and combined with the patients' CT imaging reports. Patients were divided into a modeling group and a validation group. Clinical variables were set, and logistic regression was applied to the significant factors of the modeling group to obtain independent predictors of solid SPN. An SPN malignancy probability prediction model was established: the XJTUFAH model. The XJTUFAH model is: P = ex / (1 + ex), where x = 2.582 + (-1.620 lobes) + (-0.968 lobes) + (-1.252 x pleural-tag) + (1.080 x serum total protein) + (-1.192 x CEA) + (-0.724 x NSE), P is the probability of malignancy of solid SPN, and e is the natural logarithm.

[0006] Preferably, the clinical variables include patient gender, age, cough, smoking history, family history of cancer and history of collecting extrathoracic tumors, and radiation information from CT scans of the chest, including the location of the SPN, maximum lateral value, calcification, pleural label, whether the nodule margin is lobulated, spiculated, and whether the boundary is smooth or fluffy.

[0007] Preferably, univariate analysis is used to determine the factors associated with the malignancy probability of SPN. Parameters that show differences in univariate analysis are then entered into multivariate logistic regression analysis to determine the independent predictors of SPN.

[0008] Preferably, the independent predictors of SPN include lobulation, spikes, pleural markers, CEA, NSE, and serum total protein.

[0009] Another objective of this invention is to overcome the shortcomings of the prior art by providing a method for comparing the prediction rate of malignancy of solitary pulmonary nodules with the XJTUFAH model, Mayo model, VA model, and PKUPH model. This method has high accuracy in the comparison of prediction rates and is simple and easy to implement.

[0010] The technical solution of this invention: A method for comparing the prediction rate of malignancy probability of solitary pulmonary nodules using the XJTUFAH model with the Mayo model, VA model, and PKUPH model. The comparison method is as follows: The case data of the validation group were entered into the established XJTUFAH model, Mayo model, VA model and PKUPH model respectively, and ROC curves were plotted to estimate the area under the curve (AUC), sensitivity and specificity of the system. Then, SPSS 20.0 software (IBM, Armonk, New York) was used to perform statistical analysis to calculate the prediction rate of the four models respectively.

[0011] Preferably, R 3.1.2 is used to plot a nomogram of all statistically significant variables in the binary logistic regression analysis, where a p-value <0.5 is considered statistically significant and the confidence interval (CI) is 95%.

[0012] Preferably, the Hosmer-Lemeshow goodness-of-fit is used to evaluate the fit of the four models.

[0013] The present invention has the following beneficial effects: The predictive model developed in this study demonstrates high accuracy and a significant good fit in estimating the occurrence of malignant sphincters (SPNs), and validation set data was used to assess the model's effectiveness. Compared to the Mayo, VA, and PKUPH models, the XJTUFAH model exhibits a higher AUC and greater predictive efficiency. Clinicians are advised to sum the scores of each variable to obtain a total score, thereby determining the probability of malignancy. The predictive model established in this study is more effective than other predictive models in diagnosing SPNs. Attached Figure Description

[0014] Figure 1 Sensitivity-specificity plot for predicting benign or malignant solitary pulmonary nodules; Figure 2 A bar chart showing the number distribution of solitary pulmonary nodules in different diameter ranges; Figure 3 A comparison chart of ROC curves for the XJTUFAH model, Mayo model, VA model, and PKUPH model; Figure 4 Normograph for predicting the probability of malignancy in solid, solitary pulmonary nodules. Detailed Implementation Example

[0015] This study included 318 patients aged 18-85 years who were pathologically diagnosed with spastic nodules (SPN) by chest CT at the First Affiliated Hospital of Xi'an Jiaotong University from June 2013 to October 2019. Among them, 193 were male and 125 were female. Inclusion criteria: (a) SPN diameter less than 30 mm; (b) pathological diagnosis of benign or malignant nodules. Benign nodules were defined as disappearing, shrinking, or remaining stable after one year of radiological follow-up; complete clinical data. Exclusion criteria were as follows: (a) history of thoracic cancer within 5 years; (b) nodules with atelectasis, hilar enlargement, pleural effusion, or lymphadenopathy; (c) nodules with no definitive pathological diagnosis or no regular radiological follow-up after initial diagnosis of benign nodules; (d) incomplete clinical data.

[0016] Tissue samples from suspicious pulmonary nodules were obtained via surgical or non-surgical biopsy, including VATS, wedge resection and lobectomy, CT-guided transthoracic biopsy, and bronchoscopic non-surgical biopsy. Two independent pathologists were invited to evaluate the SPN tissue specimens. Immunohistochemical staining was performed on undetermined SPN pathological types.

[0017] In this study, clinical variables included patient sex, age, cough, smoking history, family history of cancer and history of collecting extrathoracic tumors, and radiation information from chest CT scans, including the location of the SPN (upper, middle, or lower lobe), which was determined by the SPN (maximum transverse size), cavity, calcification, pleural label, nodule margin (lobulated), spiculation, and border (smooth or fluffy). Clinical data and pathological findings were assessed by two senior radiologists unaware of the study. Laboratory parameters such as TP, serum lactate dehydrogenase (LDH), CEA, NSE, Cyfra21-1, magnesium ions, and d-dimer levels were also recorded.

[0018] Data were randomly divided into a modeling group (n=210) and a validation group (n=108). Logistic regression was applied to the significant factors in the modeling group to obtain independent predictors of solid-state SPN. First, univariate analysis was used to identify factors related to the likelihood of SPN malignancy. Parameters showing differences in univariate analysis were then subjected to multivariate logistic regression analysis to determine the independent predictors of SPN, establishing an SPN malignancy probability prediction model: the XJTUFAH model. The XJTUFAH model is: P = ex / (1 + ex), where x = 2.582 + (-1.620 lobes) + (-0.968 lobes) + (-1.252 x pleural-tag) + (1.080 x serum total protein) + (-1.192 x CEA) + (-0.724 x NSE), P is the probability of malignancy of solid SPN, and e is the natural logarithm.

[0019] Then, the case data of the validation group were entered into the established XJTUFAH model, Mayo model, VA model, and PKUPH model, respectively, and ROC curves were plotted to estimate the area under the curve (AUC), sensitivity, and specificity of the system. SPSS 20.0 software (IBM, Armonk, New York) was then used to perform statistical analysis to calculate the prediction rates of the four models. R3.1.2 was used to plot nomograms of all statistically significant variables in the binary logistic regression analysis, where a p-value <0.5 was considered statistically significant, and the confidence interval (CI) was 95%. Hosmer-Lemeshow goodness-of-fit was used to evaluate the goodness of fit of the four models.

[0020] Statistical analysis was performed using SPSS 20.0 software. Differences in continuous data were compared with the results of analysis of variance, and chi-square test was used for categorical variables.

[0021] 1.1 Univariate Analysis 318 patients with SPN were enrolled in this study, including 155 with malignant nodules and 163 with benign nodules. Figure 1 The average diameter of the nodules was 23.48 mm, and the nodule diameter was greater than 20 mm.

[0022] 35%–77% Furthermore, CYFRA 21-1 is also a sensitive tumor marker for non-small cell lung cancer (NSCLC). Lobe segmentation, needle puncture, pleural tag, carcinoembryonic antigen (CEA), neuron-specific enolase (NSE), and total serum protein (TP) were independent predictors of malignant lung nodules (P<0.05). Lobe segmentation (100 points) scored highest in the nomogram score, with a Hosmer-Lemeshow goodness-of-fit statistic of 0.805 (P>0.05). The area under the curve (AUC) for the logistic regression modeling group and the validation group were 0.859 (95% CI: 0.805–0.903) and 0.823 (95% CI: 0.738–0.890), respectively.

[0023] After removing confounding factors, the presence of lobulation, needle-like structures, pleural markers, irregular margins, CEA, and NSE was identified as pretest probability equations. Furthermore, in the PKUPH model, age, diameter, borders, calcification, spiculation, and family history of SPNs were independent predictors of malignancy. However, there are significant differences in economic and medical levels between Beijing and Northwest China. The incidence of tuberculosis in Northwest my country is showing an increasing trend year by year. In contrast, annual physical examinations are more popular in Beijing, but their applicability to small nodules is limited.

[0024] Pneumonia often presents with indistinct borders, making it difficult to distinguish between benign and malignant nodules. Wahidi et al.29 suggested that nodules with smooth margins have a 20%–30% chance of malignant transformation. In our study, margin smoothness was statistically significant as a univariate variable, but it was not an independent predictor after excluding confounding factors. This was because our study included a high proportion of pneumonia patients, accounting for 50.92% (83 / 163) of benign nodules. NSE is the preferred tumor marker for small cell lung cancer, particularly important for the diagnosis, treatment monitoring, and prognosis of SCLC, consistent with our findings. After excluding confounding factors, NSE was an important factor in our predictive model. Bombardieri et al.30 clearly demonstrated higher CYFRA21-1 levels in patients with benign lung diseases such as pneumonia and pulmonary fibrosis, suggesting that CYFRA21-1 is not included in the predictive model.

[0025] Each model has a different malignancy risk threshold. The XJTUFAH model established in this invention can preliminarily determine benign or malignant based on the calculated P-value. The higher the P-value, the greater the likelihood of malignancy. When the predicted probability is greater than 70%, surgical resection or thoracoscopic resection is recommended, with a probability of 85.07% for malignant lung nodules. When the predicted probability is less than 20%, with a probability of 89.29% for benign tumors, benign tumors can be considered. After 3 months of follow-up, if the lesion stabilizes, the follow-up frequency can be reduced. If the lesion increases, a 20%-70% probability of malignancy is considered, and bronchoscopy, percutaneous biopsy, or follow-up is recommended. Follow-up time and intervention measures can be referenced from Lung Rads / Fleischer. The AUC of the ROC curve of the validation samples shows that our model has a higher ability to identify malignant tumors than the above classification models.

Claims

1. A method for establishing a malignancy probability prediction model for solitary pulmonary nodules, characterized in that, The establishment method is as follows: Basic patient information and tissue samples of suspected pulmonary nodules were collected, and combined with the patients' CT imaging reports. Patients were divided into a modeling group and a validation group. Clinical variables were set, and logistic regression was applied to the significant factors of the modeling group to obtain independent predictors of solid SPN. An SPN malignancy probability prediction model was established: the XJTUFAH model. The XJTUFAH model is: P = ex / (1+ex), where x = 2.582 + (-1.620 lobes) + (-0.968 lobes) + (-1.252 x pleural tag) + (1.080 x serum total protein) + (-1.192 x CEA) + (-0.724 x NSE), and P is the probability of malignancy in solid SPN. e is the natural logarithm .

2. The method for establishing a malignancy probability prediction model for solitary pulmonary nodules according to claim 1, characterized in that, The clinical variables included patient sex, age, cough, smoking history, family history of cancer and history of collecting extrathoracic tumors, and radiation information from chest CT scans, including the location of the SPN, maximum lateral value, calcification, pleural label, lobulation of nodule margins, spiculation, and whether the borders were smooth or fluffy.

3. The method for establishing a malignancy probability prediction model for solitary pulmonary nodules according to claim 1, characterized in that, Univariate analysis was used to identify factors associated with the malignancy probability of SPN. Parameters that showed differences in the univariate analysis were then subjected to multivariate logistic regression analysis to identify independent predictors of SPN.

4. The method for establishing a malignancy probability prediction model for solitary pulmonary nodules according to claim 3, characterized in that, The independent predictors of SPN include lobulation, spikes, pleural markers, CEA, NSE, and serum total protein.

5. A method for comparing the prediction rate of malignancy probability of solitary pulmonary nodules using the XJTUFAH model with the Mayo model, VA model, and PKUPH model, wherein the XJTUFAH model is the XJTUFAH model proposed in any one of claims 1-4 above. The comparison method is as follows: The case data of the validation group were entered into the established XJTUFAH model, Mayo model, VA model and PKUPH model respectively, and ROC curves were plotted to estimate the area under the curve (AUC), sensitivity and specificity of the system. Then, SPSS 20.0 software (IBM, Armonk, New York) was used to perform statistical analysis to calculate the prediction rate of the four models respectively.

6. The method for comparing the prediction rate of malignancy probability of solitary pulmonary nodules with the XJTUFAH model, Mayo model, VA model, and PKUPH model according to claim 5, characterized in that, R3.1.2 software was used to plot nomograms of all statistically significant variables in the binary logistic regression analysis, where a p-value <0.5 was considered statistically significant, and the confidence interval (CI) was 95%.

7. The method for comparing the prediction rate of malignancy probability of solitary pulmonary nodules with the XJTUFAH model, Mayo model, VA model, and PKUPH model according to claim 5, characterized in that, The Hosmer-Lemeshow goodness-of-fit was used to evaluate the fit of the four models.

Images