Lung cancer typing risk screening method based on joint risk assessment

Abstract

The invention discloses a lung cancer typing risk screening method based on joint risk assessment, which relates to the technical field of immunoassay and mainly comprises the following steps: screening model variable basic factors of each lung cancer typing by combining single-factor regression analysis with multi-factor stepwise regression analysis; carrying out normalized distribution processing on the normalized marker detection data, analyzing the distribution difference of each marker between healthy people and each lung cancer type group, and screening model variable marker factors of each lung cancer type; according to the screened basic factors and marker factors, constructing a risk assessment model in which a marker likelihood probability and a sample basic risk probability are combined based on a Bayesian formula; and performing risk grade assessment of each lung cancer type under Cuff value optimization on sample data through the constructed risk assessment model. Specific prior model variables are screened for different types of lung cancer, so that the model can realize screening and evaluation of different types of lung cancer.

Description

Technical Field

[0001] The present invention relates to the technical field of immunoassay, and in particular to a lung cancer typing risk screening method based on joint risk assessment. Background Art

[0002] Primary lung cancer is one of the most common malignant tumors in my country, and its incidence and mortality rate both rank first among malignant tumors in my country. According to data from the China Cancer Registry, there were 787,000 new cases of lung cancer in my country in 2015, including 520,000 males and 267,000 females, accounting for 20.0% of all malignant tumors. The national incidence of lung cancer is about 57.3 / 100,000, including 73.9 / 100,000 males and 39.8 / 100,000 females, respectively. The incidence of lung cancer in urban areas is 59.7 / 100,000, and in rural areas it is 54.2 / 100,000. From the perspective of pathology and treatment, lung cancer can be roughly divided into two categories: non-small cell lung cancer (NSCLC) and small cell lung cancer (SCLC).

[0003] Non-small cell lung cancer accounts for about 80% to 85% of all lung cancer cases in my country, of which adenocarcinoma and squamous cell carcinoma are the two main pathological types. Depending on the stage, non-small cell lung cancer can be treated with a variety of treatment methods such as surgery, radiotherapy, chemotherapy, targeted therapy, and immunotherapy. If non-small cell lung cancer can be detected and treated in time at an early stage, it can achieve better treatment results. The five-year survival rate of non-small cell cancer is about 24%.

[0004] Small cell lung cancer is a more aggressive type of lung cancer, accounting for about 15% to 20% of all lung cancer cases in my country. Small cell lung cancer usually develops faster than non-small cell lung cancer, often metastasizes in the early stages of the disease or when diagnosed, and surgical treatment is limited. Small cell lung cancer is more sensitive to radiotherapy and chemotherapy, so even in the early stages of the disease, radiotherapy and chemotherapy are the main treatment methods, but small cell lung cancer is prone to secondary resistance to radiotherapy and chemotherapy, and the prognosis is poor. The five-year survival rate of small cell lung cancer is about 6%.

[0005] Early diagnosis of lung cancer The "Guidelines for Clinical Laboratory Applications of the National Institute of Clinical Biochemistry of the United States" recommends that CEA, SCC, NSE, CYFRA21-1, CA125, and ProGRP can be used as markers for lung cancer. Different combinations can improve the early diagnosis rate of lung cancer. Among them, CEA, NSE, and ProGRP can be used as a marker combination for small cell lung carcinoma (SCLC), while CEA, SCC, and CYFRA21-1 can be used as a marker combination for non-small cell lung carcinoma (NSCLC). The diagnostic sensitivity of these two combinations in SCLC and NSCLC is 80% and 82%, respectively. Therefore, in view of the above reports and application status, it is best to select the corresponding marker combination according to different lung cancer pathological types in clinical practice.

[0006] At present, a variety of lung cancer risk screening systems have been developed at home and abroad. The shortcomings of the existing risk screening systems are that there is generally no joint risk assessment between the patient's basic factors and tumor markers. When using basic factors alone for risk assessment, Logistic and COX models are generally used; when using tumor markers alone for risk assessment, ROMA index is more often used; a small number of systems use non-parametric machine learning models to combine basic factors with tumor markers for risk assessment. This type of non-parametric machine learning model has high requirements for the training sample size, and abnormal input variables have a greater impact on the model, making the model prone to overfitting and underfitting, and the generalization ability of the model is weak. In addition, the existing system mainly conducts comprehensive lung cancer risk assessment, and cannot simultaneously conduct three types of lung cancer classification assessments of lung adenocarcinoma, lung squamous cell carcinoma, and small cell carcinoma. Through the analysis of clinical sample data, it is found that the impact of individual basic factors on different types of lung cancer is different. Therefore, the risk assessment model constructed with comprehensive lung cancer as the screening target will weaken the role of some basic factors and serum tumor markers in risk screening, resulting in a lower detection rate of the screening system and an increase in the false positive rate. Summary of the invention

[0007] In order to achieve classification evaluation of different types of lung cancer in a single measurement and reduce the influence of individual basic factors on the evaluation results, the present invention proposes a lung cancer classification risk screening method based on joint risk assessment, comprising the steps of:

[0008] S1: Collect sample data and screen the basic factors of model variables for each lung cancer classification through univariate regression analysis combined with multivariate stepwise regression analysis;

[0009] S2: Based on the basic factors screened and the corresponding lung cancer classification, the multi-classification basic risk probability of lung cancer was calculated by Logistic regression;

[0010] S3: Based on the basic factors, the sample data corresponding to the lung cancer-free group is classified into two categories. Based on the analysis of the differences of each marker in different classification groups, the basic factors that significantly affect the detection results of the corresponding markers are screened out, and the corresponding marker detection data are normalized;

[0011] S4: The normalized marker detection data were processed for normal distribution, and the differences in the distribution of each marker between the healthy population and the population of each lung cancer classification were analyzed by t-test and Wilcoxon rank sum test to screen the model variable marker factors of each lung cancer classification;

[0012] S5: Based on the screened basic factors and marker factors, a risk assessment model is constructed by combining the marker likelihood probability with the basic risk probability based on the Bayesian formula;

[0013] S6: The constructed risk assessment model is used to evaluate the risk level of each lung cancer classification under the optimized Cutoff value of the sample data.

[0014] Furthermore, in the step S1, the univariate regression analysis is combined with the multivariate stepwise regression analysis to screen the basic factors, specifically: the basic factors of the model are preliminarily screened based on the difference quantitative indicators through the univariate regression analysis, and then the secondary screening is performed through the multivariate stepwise regression based on the preliminarily screened basic factors.

[0015] Furthermore, in the step S2, the multi-classification basic risk probability is calculated by the following formula:

[0016]

[0017] In the formula, k is the different types of lung cancer. When k is 1, it is adenocarcinoma; when k is 2, it is squamous cell carcinoma; when k is 3, it is small cell carcinoma. represents the probability that the sample source is the kth lung cancer classification, P N represents the probability that the sample source is free of lung cancer, is the exposure factor of the i-th type included in the model under the k-th lung cancer classification, is the weight coefficient of the i-th exposure factor under the k-th lung cancer classification, i is a constant ranging from 1 to n, and n is the total number of exposure factors.

[0018] Furthermore, in step S3, when the underlying factor is a continuous variable, the median is used for binary classification.

[0019] Furthermore, in the step S3, the normalization process is represented by the MOM value:

[0020]

[0021] Wherein, the Median() function is used to obtain the median of the marker test results after eliminating the influence of various basic factors while ensuring the uniformity of the region and the test instrument.

[0022] Furthermore, in the S4 step, based on the distribution differences obtained by the t-test and the Wilcoxon rank sum test, the marker detection data are first divided into negative and positive according to the marker reference value, and the negative and positive data are respectively classified into two categories according to their respective medians to achieve verification of the marker screening.

[0023] Furthermore, in the step S5, the marker likelihood probability is expressed as the following formula:

[0024]

[0025] In the formula, LR is the probability ratio of the marker detection value coming from the lung cancer-free population and the k-th lung cancer classification population, G N is the multidimensional Gaussian distribution function of the markers of the lung cancer-free population, is the multidimensional Gaussian distribution function of the k-th lung cancer typing group marker, the MOM value is the median multiple obtained after normalization, log10(MOM i ) is the logarithmic vector of MOM values ​​of multiple markers of the i-th sample, μ N is the mean of the logarithmic values ​​of the marker MOM values ​​in the lung cancer-free population, ∑ N is the marker covariance matrix calculated by standard deviation and correlation coefficient in the lung cancer-free population, is the mean of the logarithmic values ​​of the marker MOM values ​​in the k-th lung cancer classification group, is the marker covariance matrix calculated by standard deviation and correlation coefficient in the k-th lung cancer classification population.

[0026] Furthermore, the construction of the risk assessment model is expressed as the following formula:

[0027]

[0028] In the formula, is the kth lung cancer classification group, A N is a group without lung cancer, B is a group of marker detection values, P(A N ) is the basic risk probability of no lung cancer in the screened sample assessed by basic factors, The basic risk probability of the screened sample being classified as the kth type of lung cancer based on basic factors, is the posterior risk probability of the screened sample being assessed as the kth lung cancer type under the condition of the current marker detection value, P(B|A N ) is the probability of the current marker detection value occurring in a group without lung cancer, is the probability of the current marker detection value occurring in the k-th lung cancer classification group, is the probability ratio of the marker detection values ​​in the lung cancer-free population coming from samples of different groupings, It is the probability ratio that the marker detection values ​​in the lung cancer population come from samples of different groupings.

[0029] Furthermore, in step S6, the Cutoff value optimization objective is expressed as the following formula:

[0030]

[0031] Wherein, TPR is the positive detection rate of the model judgment result under a specific Cutoff value, Min_TPR is the minimum allowable value of the model, FPR is the false positive rate of the model judgment result under a specific Cutoff value, Max_FRP is the maximum allowable value of the model, α is the weight coefficient of the positive detection rate, and β is the weight coefficient of the false positive rate.

[0032] Furthermore, by traversing different Cutoff values, the ROC curve of each lung cancer classification in the risk assessment model is obtained, and the optimization interval range for the next optimization is obtained in combination with the limit requirements of TPR and FPR. When the optimization interval exists, the Cutoff value that satisfies TPR×α+(1-FPR)×β→Max is traversed in the interval.

[0033] Compared with the prior art, the present invention has at least the following beneficial effects:

[0034] (1) The lung cancer classification risk screening method based on joint risk assessment described in the present invention screens out the unique basic risk variables for three different types of lung cancer, namely adenocarcinoma, squamous cell carcinoma and small cell carcinoma, through univariate regression analysis and multivariate stepwise regression analysis. This not only improves the pertinence of the model, but also enables the model to achieve simultaneous screening and evaluation of different lung cancer classifications;

[0035] (2) The basic factors were normalized, effectively eliminating the impact of the basic factors on the marker test results, ensuring the comparability of marker test results between different individuals and improving the accuracy of screening;

[0036] (3) The Bayesian model was used to perform joint risk screening of basic lung cancer risk factors and serum tumor markers, which improved the detection rate of the risk screening model and could better combine individual background information with biomarker data to provide a more personalized risk assessment;

[0037] (4) A risk screening model for adenocarcinoma, squamous cell carcinoma, and small cell carcinoma was constructed, which strengthened the differences in the impact of risk factors on different types of lung cancer, improved the sensitivity of model screening, and facilitated the early detection of specific types of lung cancer, thereby improving the success rate of treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A step diagram of a lung cancer classification risk screening method based on joint risk assessment;

[0039] Figure 2 This is the variation pattern of the median logarithm of NSE concentration;

[0040] Figure 3 The frequency histogram of CEA in the lung cancer-free group and adenocarcinoma group. DETAILED DESCRIPTION

[0041] The following are specific embodiments of the present invention and the accompanying drawings to further describe the technical solution of the present invention, but the present invention is not limited to these embodiments.

[0042] In view of the shortcomings of the existing system, this method screens the significant differences between different lung cancer classification groups and healthy groups without lung cancer, and outputs the comprehensive risk assessment results of three classifications: adenocarcinoma, squamous cell carcinoma, and small cell carcinoma based on the Bayesian model combined with the analysis results of individual basic factors and serum tumor marker data. Figure 1 As shown, the present invention proposes a lung cancer classification risk screening method based on joint risk assessment, comprising the steps of:

[0043] S1: Collect sample data and screen the basic factors of model variables for each lung cancer classification through univariate regression analysis combined with multivariate stepwise regression analysis;

[0044] S2: Based on the basic factors screened and the corresponding lung cancer classification, the multi-classification basic risk probability of lung cancer was calculated by Logistic regression;

[0045] S3: Based on the basic factors, the sample data corresponding to the lung cancer-free group is classified into two categories. Based on the analysis of the differences of each marker in different classification groups, the basic factors that significantly affect the detection results of the corresponding markers are screened out, and the corresponding marker detection data are normalized;

[0046] S4: The normalized marker detection data were processed for normal distribution, and the differences in the distribution of each marker between the healthy population and the population of each lung cancer classification were analyzed by t-test and Wilcoxon rank sum test to screen the model variable marker factors of each lung cancer classification;

[0047] S5: Based on the screened basic factors and marker factors, a risk assessment model is constructed by combining the marker likelihood probability with the basic risk probability based on the Bayesian formula;

[0048] S6: The constructed risk assessment model is used to evaluate the risk level of each lung cancer classification under the optimized Cutoff value of the sample data.

[0049] In this embodiment, the data required for building the model are derived from hospital inpatient cases and samples from health examination centers, where basic information includes age, gender, height, weight, BMI, history of smoking, history of drinking, body temperature, pulse rate, respiratory rate, endocrine history, history of surgery or trauma, family history of tumors, history of exposure to harmful substances such as poisons and dust, etc. Serum tumor markers include soluble fragments of cytokeratin 19 (CYFRA211), neuron-specific enolase (NSE), squamous cell carcinoma antigen (SCC), carcinoembryonic antigen (CEA), carbohydrate antigen 125 (CA125), carbohydrate antigen 199 (CA199), and clinical diagnosis results include the absence of lung cancer, lung adenocarcinoma, lung squamous cell carcinoma, and small cell carcinoma. The basic information and serum tumor markers are used as potential independent variables of the risk model, and the clinical diagnosis results are used as the dependent variables of the risk model. Of course, according to the sample inclusion requirements, the data that does not meet the requirements must be cleaned up, among which:

[0050] Lung cancer screening inclusion criteria:

[0051] (1) Age ≥ 35 years old

[0052] (2) Small cell lung cancer or non-small cell lung cancer (differentiating between adenocarcinoma and squamous cell carcinoma) confirmed by imaging examination and pathological diagnosis

[0053] (3) The patients had not received anti-tumor treatment such as surgery, chemoradiotherapy, and drug therapy;

[0054] (4) No concurrent tumors in other organs and no history of tumors in other organs;

[0055] (5) Have relatively complete clinical data;

[0056] Inclusion criteria for normal control group without lung cancer:

[0057] (1) Aged ≥ 35 years;

[0058] (2) Be in good health;

[0059] (3) No malignant tumors of the lungs or other organs;

[0060] (4) No radiotherapy, chemotherapy, drug treatment or surgical treatment;

[0061] (5) Have relatively complete clinical data.

[0062] In order to ensure the applicability of the risk screening model, we allow the missing of some variables, and the missing variables will be replaced by the average level, and some variables will be formatted and converted as follows:

[0063] Conversion correction value = (current value - lower limit value) / (upper limit value - lower limit value)

[0064] Among them: the lower limit is the 5th percentile of the data distribution, and the upper limit is the 95th percentile.

[0065] The dependent variable of the model was marked as 0 for the group without lung cancer, 1 for the adenocarcinoma group, 2 for the squamous cell carcinoma group, and 3 for the small cell carcinoma group.

[0066] It can be seen that there is a lot of basic information in the initially collected data, and only some of the basic information is related to the lung cancer typing to be screened, so we first need to screen out the key basic information related to lung cancer typing risk screening. To this end, we formed three groups of training data with the lung cancer-free population and the lung adenocarcinoma, lung squamous cell carcinoma, and small cell carcinoma populations, and used univariate regression analysis + multifactor stepwise regression analysis to screen model variables. Specifically, we first preliminarily screened the basic factors of the model based on the difference quantitative indicators through univariate regression analysis, and then performed secondary screening through multifactor stepwise regression based on the preliminary screened basic factors. Then, based on the screened basic factors and the corresponding lung cancer typing, we used the unordered multi-classification Logistic regression model to calculate the multi-classification basic risk probability of lung cancer:

[0067]

[0068] In the formula, k is the different types of lung cancer. When k is 1, it is adenocarcinoma; when k is 2, it is squamous cell carcinoma; when k is 3, it is small cell carcinoma. represents the probability that the sample source is the kth lung cancer classification, P N represents the probability that the sample source is free of lung cancer, is the exposure factor of the i-th type included in the model under the k-th lung cancer classification, is the weight coefficient of the i-th exposure factor under the k-th lung cancer classification, i is a constant ranging from 1 to n, and n is the total number of exposure factors. Furthermore, exp(β0) represents the probability ratio of an individual to those who have cancer when all exposure factors are 0 or at the average level; exp(β i ) represents the odds ratio (OR value) of the ith exposure factor.

[0069] Among them, after screening the lung cancer-free group and adenocarcinoma group, the final model variables selected were gender, smoking history, and family tumor history. After screening the lung cancer-free group and squamous cell carcinoma group, the final model variables selected were age, gender, and smoking history. After screening the lung cancer-free group and small cell carcinoma group, the final model variables selected were age, BMI, smoking history, family tumor history, and breathing.

[0070] The model parameter optimization adopts the maximum likelihood estimation method, based on the collected clinical sample data set and R language software package to achieve parameter optimization. The optimization coefficients of the basic risk probabilities of three types of lung cancer, adenocarcinoma, squamous cell carcinoma, and small cell carcinoma, are shown in Table 1.

[0071] Table 1: Calculation formula for basic risk probability of lung cancer

[0072]

[0073]

[0074] At the same time, we take into account the possible impact of individual basic factors and medical history on marker detection results. The coupling relationship between variables in the model risk calculation must be eliminated, otherwise it will bias the model calculation results.

[0075] First, the basic factors were used as the basis for classification of the lung cancer-free group (if the basic factors were continuous variables, the median was used for binary classification), and the differences in markers in different classification groups were analyzed. The basic factors that significantly affected the marker test results were screened out, and the results are shown in Table 2.

[0076] Table 2: Basic factors that significantly affect marker detection values

[0077]

[0078] Then, we use normalization to eliminate the impact of population differences in different regions, different individual basic factors, and different testing equipment on the marker test results. The normalization method is to establish a method for obtaining the median concentration values ​​of different markers under different regional populations, different individual basic factors, and different testing equipment conditions, and use the ratio of the marker test value to the median as the normalized value, that is, the MOM value (Multiple of Median)

[0079]

[0080] In the formula, the Median() function is used to obtain the median of the marker test results without the influence of various basic factors while ensuring the uniformity of the region and the test instrument. The construction of the Median() function is explained by taking a group of people without lung cancer detected by a certain region and a certain type of instrument as an example. If the region or instrument type is changed, the corresponding median calculation model should be established in the same way. Specifically:

[0081] (1) First, through the changing relationship between the marker and age, the quantitative relationship between the median of the marker in different age groups and age is constructed. Figure 2 The figure shows the changing pattern of the median logarithm of NSE concentration in lung cancer-free groups of different age groups.

[0082] (2) Then, based on the changing pattern of the marker median and age, a suitable fitting model is selected. For example, the fitting model of NSE is a quadratic polynomial. An empirical calculation formula for the median test value of the marker in people of different ages is constructed to eliminate the influence of age factors on the test results.

[0083] (3) If there are other factors, then based on the data after age elimination, the influence of other factors on the marker detection results is gradually eliminated. Finally, the marker concentration detection value is normalized to the MOM value by formula (2).

[0084] After normalizing the tumor marker detection values, they were further logarithmized to make the data approximately follow a normal distribution. Here, we used t-test and Wilcoxon rank sum test to analyze whether there were significant differences in the distribution of markers among different populations. The analysis results are shown in Table 3 (the order of the data in the table is (mean (SD), median (first quartile, third quartile))):

[0085] Table 3: Difference analysis of MOM values ​​of different markers in normal population and adenocarcinoma population

[0086]

[0087]

[0088] In Table 3: P value is the result of t test, Px value is the result of Wilcoxon rank sum test, when P<0.05 or Px<0.05, the variable is judged to be significant. The statistical results should be considered comprehensively with clinical experience to finally determine whether the variable is included in the model.

[0089] According to the statistical results, the distribution differences of CYFRA211, SCC, and CEA in normal and adenocarcinoma populations are significant, among which CYF and SCC are protective factors, and CEA is a risk factor. However, CYF and SCC as protective factors are inconsistent with clinical experience. Therefore, based on the distribution differences obtained by the t-test and Wilcoxon rank sum test, we further converted the detection amounts of markers CYF, SCC, and CEA into categorical variables, which were divided into four categories: negative, weakly negative, weakly positive, and positive. The classification rules are: first divide into positive and negative according to the marker reference value, and then divide into two categories according to the median in the negative, and divide into two categories according to the median in the positive. The analysis results are shown in Table 4 (the data order in the table is (number of cases (percentage))).

[0090] Table 4: Difference analysis of CYF, SCC, and CEA classification variables in normal population and adenocarcinoma population

[0091]

[0092]

[0093] In Table 4: P value is the result of chi-square test, and Px value is the result of Fisher test. From the analysis results, SCC has a higher relative proportion in negative and positive adenocarcinoma populations. However, CYFRA211 is not significant in variable classification test. Therefore, we finally chose CYFRA211 variable as the variable to be considered in the adenocarcinoma model, and SCC and CEA as adenocarcinoma model variables.

[0094] Similarly, we selected CYFRA211, SCC, and CEA as squamous cell carcinoma model variables, and CA125 and CA199 as variables to be considered for the squamous cell carcinoma model; CYFRA211, NSE, SCC, CEA, and CA125 as small cell carcinoma model variables, and CA199 as variables to be considered for the small cell carcinoma model.

[0095] Based on the selected marker factors, when obtaining a set of marker concentrations, we also need to determine the source of the concentration data, that is, the probability ratio (likelihood ratio) of the concentration data coming from different groups. Therefore, we need a probability distribution function to quantify the probability of each concentration value of the marker appearing in different groups. Here, we take the logarithm of the MOM value based on the marker MOM value calculated previously, and statistically analyze the frequency histogram of the logarithmic value of each marker in different groups, for example Figure 3 Shown is a frequency histogram of CEA in the lung cancer-free group and the adenocarcinoma group.

[0096] The normal distribution function was used to fit the frequency histogram, and the characteristic parameters of the normal distribution curve of each marker in different groups and the correlation coefficients between markers were counted. And through the variable screening method, the marker factors with significant differences among lung cancer, adenocarcinoma, squamous cell carcinoma, and small cell carcinoma groups were further screened to form a multidimensional Gaussian probability model. Finally, a quantitative formula for calculating the likelihood ratio of the marker can be obtained.

[0097]

[0098] In the formula, LR is the probability ratio of the marker detection value coming from the lung cancer-free population and the k-th lung cancer classification population, G N is the multidimensional Gaussian distribution function of the markers of the lung cancer-free population, is the multidimensional Gaussian distribution function of the k-th lung cancer typing group marker, the MOM value is the median multiple obtained after normalization, log10(MOM i ) is the logarithmic vector of MOM values ​​of multiple markers of the i-th sample, μ N is the mean of the logarithmic values ​​of the marker MOM values ​​in the lung cancer-free population, ∑ N is the marker covariance matrix calculated by standard deviation and correlation coefficient in the lung cancer-free population, is the mean of the logarithmic values ​​of the marker MOM values ​​in the k-th lung cancer classification group, is the marker covariance matrix calculated by standard deviation and correlation coefficient in the k-th lung cancer classification population.

[0099] On the basis of the marker likelihood probability, we then combine the risk probability of lung cancer in the sample source, that is, the basic risk probability that the sample is a lung cancer sample, through the Bayesian formula to construct the required risk assessment model.

[0100]

[0101] In the formula, is the kth lung cancer classification group, A N is a group without lung cancer, B is a group of marker detection values, P(A N ) is the basic risk probability of no lung cancer in the screened sample assessed by basic factors, The basic risk probability of the screened sample being classified as the kth type of lung cancer based on basic factors, is the posterior risk probability of the screened sample being assessed as the kth lung cancer type under the condition of the current marker detection value, P(B|A N ) is the probability of the current marker detection value occurring in a group without lung cancer, is the probability of the current marker detection value occurring in the k-th lung cancer classification group, is the probability ratio of the marker detection values ​​in the lung cancer-free population coming from samples of different groupings, It is the probability ratio that the marker detection values ​​in the lung cancer population come from samples of different groupings.

[0102] After the risk assessment model is determined, the collected sample data is input into the risk screening model to obtain the risk screening output value. For the lung cancer risk assessment system, the initial output value of the risk model is the risk probability value, including adenocarcinoma risk value, squamous cell carcinoma risk value, and small cell carcinoma risk value. In this way, the model-predicted risk value and clinical diagnosis result of each sample can be obtained, as shown in Table 5.

[0103] Table 5: Cutoff value optimization input data

[0104] Sample No. Clinical diagnosis results Adenocarcinoma risk score Squamous cell carcinoma risk Small cell carcinoma risk 1 Adenocarcinoma 1 / 423 1 / 3354 1 / 4563 2 No lung cancer 1 / 8423 1 / 6356 1 / 12567 3 Small cell carcinoma 1 / 5084 1 / 1052 1 / 364 4 Squamous cell carcinoma 1 / 9886 1 / 258 1 / 2469

[0105] Based on these obtained Cutoff values, we also need to optimize and evaluate their risk levels. The target conditions for optimizing Cutoff values ​​at different risk levels are:

[0106]

[0107] In the formula, TPR is the positive detection rate of the model judgment result under a specific Cutoff value, Min_TPR is the minimum allowable value of the model, FPR is the false positive rate of the model judgment result under a specific Cutoff value, Max_FPR is the maximum allowable value of the model, α is the weight coefficient of the positive detection rate, and β is the weight coefficient of the false positive rate. Among them, Min_TPR and Max_FPR are expected values ​​set according to different risk levels of the model. Since these two are empirical values ​​related to clinical applications, there is uncertainty. Therefore, in the Cutoff value optimization, these two values ​​are externally settable. However, the requirements for α and β are different in different application scenarios and different risk levels. Therefore, in the Cutoff value optimization, these two values ​​are also externally settable.

[0108] Then, based on this optimization goal, we can obtain the ROC curves of the risk models for adenocarcinoma, squamous cell carcinoma, and small cell carcinoma by traversing different Cutoff values, and determine the interval range for the next optimization through the ROC curve and the limit requirements of TPR and FPR.

[0109] When the optimization interval is not empty, the Cutoff value is searched within the interval to make the formula TPR×α+(1-FPR)×β→Max hold. At this time, the optimal Cutoff value corresponding to formula (5) can be obtained.

[0110] In this way, the optimal Cutoff value for different risk levels of adenocarcinoma, squamous cell carcinoma, and small cell carcinoma risk models can be obtained by inputting different data sources and setting the corresponding Min_TPR, Max_FPR, α, and β values.

[0111] In summary, the lung cancer classification risk screening method based on joint risk assessment described in the present invention, through univariate regression analysis and multivariate stepwise regression analysis, screened out the unique basic risk variables for three different types of lung cancer, namely adenocarcinoma, squamous cell carcinoma, and small cell carcinoma. This not only improves the targetedness of the model, but also enables the model to realize simultaneous screening and evaluation of different lung cancer classifications.

[0112] The basic factors were normalized to effectively eliminate the impact of the basic factors on the marker detection results, ensure the comparability of marker detection results between different individuals, and improve the accuracy of screening.

[0113] The Bayesian model is used to conduct joint risk screening of basic risk factors for lung cancer and serum tumor markers, which improves the detection rate of the risk screening model and can better combine individual background information with biomarker data to provide a more personalized risk assessment.

[0114] A risk screening model for adenocarcinoma, squamous cell carcinoma, and small cell carcinoma has been constructed, which strengthens the differences in the impact of risk factors on different types of lung cancer, improves the sensitivity of model screening, and helps to detect specific types of lung cancer at an early stage, thereby increasing the success rate of treatment.

[0115] It should be noted that all directional indications in the embodiments of the present invention (such as up, down, left, right, front, back, etc.) are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0116] In addition, in the present invention, descriptions such as "first", "second", "one", etc. are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0117] In the present invention, unless otherwise clearly specified and limited, the terms "connection", "fixation", etc. should be understood in a broad sense. For example, "fixation" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements, unless otherwise clearly defined. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0118] In addition, the technical solutions between the various embodiments of the present invention can be combined with each other, but it must be based on the fact that ordinary technicians in the field can implement it. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

Claims

1. A lung cancer classification risk screening method based on joint risk assessment, characterized in that: Includes steps: S1: Collect sample data and screen the basic factors of model variables for each lung cancer classification through univariate regression analysis combined with multivariate stepwise regression analysis; S2: Based on the basic factors screened and the corresponding lung cancer classification, the multi-classification basic risk probability of lung cancer was calculated by Logistic regression; S3: Based on the basic factors, the sample data corresponding to the lung cancer-free group is classified into two categories. Based on the analysis of the differences of each marker in different classification groups, the basic factors that significantly affect the detection results of the corresponding markers are screened out, and the corresponding marker detection data are normalized; S4: The normalized marker detection data were processed for normal distribution, and the differences in the distribution of each marker between the healthy population and the population of each lung cancer classification were analyzed by t-test and Wilcoxon rank sum test to screen the model variable marker factors of each lung cancer classification; S5: Based on the screened basic factors and marker factors, a risk assessment model is constructed by combining the marker likelihood probability with the basic risk probability based on the Bayesian formula; S6: The constructed risk assessment model is used to evaluate the risk level of each lung cancer classification under the optimized Cutoff value of the sample data.

2. A lung cancer classification risk screening method based on joint risk assessment as claimed in claim 1, characterized in that: In the step S1, the basic factors are screened by combining univariate regression analysis with multivariate stepwise regression analysis. Specifically, the basic factors of the model are preliminarily screened based on the difference quantitative indicators through univariate regression analysis, and then secondary screening is performed through multivariate stepwise regression based on the preliminarily screened basic factors.

3. A lung cancer classification risk screening method based on joint risk assessment as claimed in claim 1, characterized in that: In the step S2, the multi-classification basic risk probability is calculated using the following formula: In the formula, k is the different types of lung cancer. When k is 1, it is adenocarcinoma; when k is 2, it is squamous cell carcinoma; when k is 3, it is small cell carcinoma. represents the probability that the sample source is the kth lung cancer classification, P N represents the probability that the sample source is free of lung cancer, is the exposure factor of the i-th type included in the model under the k-th lung cancer classification, is the weight coefficient of the i-th exposure factor under the k-th lung cancer classification, i is a constant ranging from 1 to n, and n is the total number of exposure factors.

4. The lung cancer typing risk screening method based on joint risk assessment according to claim 1, characterized in that: In the S3 step, when the underlying factor is a continuous variable, the median is used for binary classification.

5. The lung cancer classification risk screening method based on joint risk assessment according to claim 1, characterized in that: In the step S3, the normalization process is represented by the MOM value: Wherein, the Median() function is used to obtain the median of the marker test results after eliminating the influence of various basic factors while ensuring the uniformity of the region and the test instrument.

6. The lung cancer classification risk screening method based on joint risk assessment according to claim 1, characterized in that: In the step S4, based on the distribution differences obtained by the t-test and the Wilcoxon rank sum test, the marker detection data are first divided into negative and positive according to the marker reference value, and the negative and positive data are binary classified according to their respective medians to achieve verification of marker screening.

7. The lung cancer classification risk screening method based on joint risk assessment according to claim 3, characterized in that: In the step S5, the marker likelihood probability is expressed as the following formula: In the formula, LR is the probability ratio of the marker detection value coming from the lung cancer-free population and the k-th lung cancer classification population, G N is the multidimensional Gaussian distribution function of the markers of the lung cancer-free population, is the multidimensional Gaussian distribution function of the k-th lung cancer typing group marker, the MOM value is the median multiple obtained after normalization, log10(MOM i ) is the logarithmic vector of MOM values ​​of multiple markers of the i-th sample, μ N is the mean of the logarithmic values ​​of the marker MOM values ​​in the lung cancer-free population, ∑ N is the marker covariance matrix calculated by standard deviation and correlation coefficient in the lung cancer-free population, is the mean of the logarithmic values ​​of the marker MOM values ​​in the k-th lung cancer classification group, is the marker covariance matrix calculated by standard deviation and correlation coefficient in the k-th lung cancer classification population.

8. The lung cancer classification risk screening method based on joint risk assessment according to claim 7, characterized in that: The construction of the risk assessment model is expressed as the following formula: In the formula, is the kth lung cancer classification group, A N is a group without lung cancer, B is a group of marker detection values, P(A N ) is the basic risk probability of no lung cancer in the screened sample assessed by basic factors, The basic risk probability of the screened sample being classified as the kth type of lung cancer based on basic factors, is the posterior risk probability of the screened sample being assessed as the kth lung cancer type under the condition of the current marker detection value, P(B|A N ) is the probability of the current marker detection value occurring in a group without lung cancer, is the probability of the current marker detection value occurring in the k-th lung cancer classification group, is the probability ratio of the marker detection values ​​in the lung cancer-free population coming from samples of different groupings, It is the probability ratio that the marker detection values ​​in the lung cancer population come from samples of different groupings.

9. The lung cancer classification risk screening method based on joint risk assessment according to claim 1, characterized in that: In the step S6, the Cutoff value optimization objective is expressed as the following formula: Wherein, TPR is the positive detection rate of the model judgment result under a specific Cutoff value, Min_TPR is the minimum allowable value of the model, FPR is the false positive rate of the model judgment result under a specific Cutoff value, Max_FPR is the maximum allowable value of the model, α is the weight coefficient of the positive detection rate, and β is the weight coefficient of the false positive rate.

10. A lung cancer classification risk screening method based on joint risk assessment according to claim 9, characterized in that: By traversing different Cutoff values, the ROC curve of each lung cancer classification in the risk assessment model is obtained, and the optimization interval range for the next step of optimization is obtained in combination with the limit requirements of TPR and FPR. When the optimization interval exists, the Cutoff value that satisfies TPR×α+(1-FPR)×β→Max is traversed in the interval.

Images