A risk prediction model for endometrial carcinoma occurring from endometrial atypical hyperplasia and a construction method thereof

Abstract

The present application relates to the field of biomedical technology, specifically relates to a risk prediction model for endometrial carcinoma caused by endometrial atypical hyperplasia and a construction method thereof; the construction method is based on the clinical data of patients with endometrial atypical hyperplasia, uses LASSO regression analysis and light GBM multi-factor logistic regression analysis to screen the influencing factors of endometrial carcinoma caused by endometrial atypical hyperplasia, constructs two different prediction models respectively to predict the risk of EC, and verifies the models; the EC risk prediction model constructed by the method can accurately predict the risk of endometrial carcinoma, has high sensitivity and negative predictive value, makes clinical intervention more targeted, thereby more reasonably allocates clinical resources, and can reduce the psychological burden of a part of patients with endometrial atypical hyperplasia, has important significance for postoperative management of such patients, and is suitable for large-scale popularization and application.

Description

Technical Field

[0001] This invention relates to the field of biomedical technology, specifically to a risk prediction model for endometrial cancer arising from atypical endometrial hyperplasia and its construction method. Background Technology

[0002] Endometrial cancer (EC) is one of the most common malignant tumors in women, with an incidence rate of approximately 3%. Endometrial atypical hyperplasia (AEH) refers to an excessive proliferation of endometrial glands and abnormal changes in cell morphology (i.e., cellular atypia). AEH has a high risk of progressing to EC; statistics show that the proportion of AEH complicated by EC is as high as 43%. Currently, total hysterectomy is the preferred radical treatment for AEH. Since the survival rate of EC has decreased from 95% in stage I to 14% in stage IV, early diagnosis is particularly important. Accurate prediction of EC risk factors helps in the early diagnosis of EC and is of great significance for patient treatment decisions. Furthermore, because endometrial sampling cannot reliably detect the presence of EC in AEH patients, or because the disease is in a very early stage, accurate detection is not possible. Therefore, identifying risk factors that increase the risk of AEH progressing to cancer has important clinical value. Studies have shown that metabolic abnormalities and endocrine disorders are high-risk factors for EC, especially in perimenopausal women.

[0003] Studies have found that the incidence of endometrial cancer (EC) may be related to factors such as the extent of surgery, geographical location, and differences in patient physical condition, but an effective EC risk prediction model is still lacking. Chinese patent CN118937679A discloses a biomarker combination reagent and its application in preparing a kit for predicting endometrial cancer. This patent uses data from CA125, CA153, CA199, CEA, and SCC intrauterine cell suspension supernatant to construct an endometrial cancer risk prediction model through multivariate logistic regression analysis, developing an endometrial cancer screening and diagnostic kit. This kit demonstrates good clinical application potential in the field of endometrial cancer screening and diagnosis, establishing an endometrial cancer prediction model and solving a bottleneck problem in endometrial cancer screening. It is of great significance for the early diagnosis and screening of endometrial cancer and has broad clinical application prospects.

[0004] Chinese patent CN114023448A discloses a method for constructing a predictive diagnostic model for endometrial cancer, as well as the diagnostic model and diagnostic instrument. The method for constructing the predictive diagnostic model for endometrial cancer includes the following steps: 1) data acquisition; 2) data processing; 3) model building; 4) interpretation of model results; 5) verification of diagnostic predictive efficacy; and 6) determination of feature weight values. This patented method for constructing a predictive diagnostic model for endometrial cancer only requires commonly used clinical test results, without increasing the burden on patients. All required examinations are non-invasive, effectively reducing unnecessary invasive trauma to patients and optimizing the necessity of rationally performing invasive examinations, making it suitable for rapid clinical application.

[0005] It is evident that current research on the risk of endometrial cancer (EC) is conducted while preserving the uterus (e.g., with progesterone therapy), comparing EC patients with control groups (matched women without cancer). This approach fails to differentiate the risks associated with different treatment methods and rarely analyzes patients who have undergone total hysterectomy. Therefore, there is an urgent clinical need for a simple and effective method to assist physicians in accurately assessing the probability of endometrial cancer in patients with atypical endometrial hyperplasia based on their specific characteristics during perimenopause. Summary of the Invention

[0006] The purpose of this invention is to overcome the problem that existing EC risk prediction models cannot accurately predict the EC risk in patients with atypical endometrial hyperplasia, and to propose a risk prediction model for the development of endometrial cancer in patients with atypical endometrial hyperplasia and its construction method.

[0007] To achieve the above-mentioned objective, this invention provides a method for constructing a risk prediction model for endometrial cancer arising from endometrial atypical hyperplasia, comprising the following steps:

[0008] S1. Obtain clinically relevant data of the samples and randomly divide the samples into a study group and a validation group; the samples are AEH patients undergoing routine hysterectomy; the clinically relevant data include general information, metabolic syndrome status, laboratory test data, and immunohistochemical molecular indicators;

[0009] S2. The clinical data of the study group were statistically analyzed using independent samples t-test and χ2 test, and expressed as mean and standard deviation to obtain the statistical data of the sample.

[0010] S3. Using the occurrence of endometrial cancer in the study group as the dependent variable and the statistical data of the study group sample as the independent variable, univariate analysis and logistic regression analysis were performed to obtain the strongly correlated factors for the occurrence of endometrial cancer in patients with atypical endometrium; the strongly correlated factors were those with P < 0.05 in the analysis results.

[0011] We constructed a Light GBM model and used it to screen out the core influencing factors for the occurrence of EC from the clinically relevant data of the study group sample.

[0012] S4. Perform backward LASSO regression analysis and multivariate logistic regression analysis on the strongly correlated factor variables to obtain the feature variables with statistically significant differences in the analysis results, and construct the first prediction model based on the feature variables; the first prediction model is a nomogram prediction model.

[0013] The core influencing factors were reintroduced into the light GBM model to construct a second prediction model;

[0014] S5. Based on the clinical data of the validation group, the first prediction model and the second prediction model are validated, and the first prediction model and the second prediction model are optimized according to the validation results; the optimized first prediction model and the second prediction model are the risk prediction models for endometrial cancer caused by atypical endometrial hyperplasia of the present invention.

[0015] This invention discloses a method for constructing a risk prediction model for endometrial cancer arising from endometrial atypical hyperplasia. Based on clinical data of patients with endometrial atypical hyperplasia, LASSO regression analysis and Light GBM multivariate logistic regression analysis are used to screen influencing factors for endometrial cancer in patients with endometrial atypical hyperplasia. Two different prediction models are constructed to predict the risk of EC, and the models are validated accordingly. The EC risk prediction model constructed by this method can accurately predict the risk of endometrial cancer, with high sensitivity and negative predictive value, making clinical intervention more targeted, thus allocating clinical resources more rationally, and alleviating the psychological burden of some patients with endometrial atypical hyperplasia. It is of great significance for the postoperative management of such patients and is suitable for large-scale application.

[0016] In step S1, preferably, the number of samples is not less than 100.

[0017] Preferably, the sample size of the research group accounts for 60-80%, and the sample size of the validation group accounts for 20-40%.

[0018] Preferably, the general information includes: age, menarche, weight, body mass index (BMI), pulse pressure, time interval between diagnostic curettage and surgery, age distribution, regional distribution, menarche distribution, BMI distribution, whether there are menstrual abnormalities, clinical symptoms (lower abdominal pain, uterine bleeding), parity, parity, fertility status, preoperative endometrial thickness, duration of disease ≥1 year, whether menopause has occurred, and severity of lesions (classified as simple type (<50%) or complex type (≥50%) based on whether the abnormality in glandular proportion exceeds 50%); preferably, the metabolic syndrome includes: diabetes, hyperlipidemia, hypertension, and polycystic ovary syndrome (PCOS); the laboratory test data includes carbohydrate antigen 125 (CA125), blood glucose (GLU), creatinine, uric acid, aspartate aminotransferase (AST), and uric acid distribution; the immunohistochemical molecular indicators include P53 (no deletion, deletion), phosphatase and tensin homolog (PTEN) (no deletion, deletion) and paired cassette gene 2 antibody (PAX2) (no deletion, deletion).

[0019] Preferably, the screening criteria for the samples include: Inclusion criteria: ① The patient's physical examination ultrasound results showed endometrial thickening or intrauterine lesions, or abnormal vaginal bleeding, and hysteroscopic curettage or post-curettage pathological diagnosis confirmed AEH, EC conformed to the WHO classification of female reproductive system tumors, and pathological diagnosis was confirmed; ② All patients had complete clinical data; ③ The study was approved by the hospital ethics committee, and the patient's family gave informed consent and signed the surgical consent form; Exclusion criteria: ① Patients who had received exogenous hormone therapy before the operation; ② Patients with other endocrine diseases or autoimmune diseases; ③ Patients with other types of malignant tumors; ④ Patients who had fertility requirements or no surgical indications.

[0020] In step S3, preferably, the Light GBM model includes at least 100 decision trees; the more decision trees there are, the higher the accuracy of the constructed Light GBM model and the better the accuracy of the predicted model.

[0021] Preferably, the screening method for the core influencing factors includes: sorting the clinically relevant data in descending order according to the feature importance (Gain score), and selecting the clinically relevant data with a total feature importance of not less than 70% as the core influencing factors.

[0022] In step S4, preferably, the nomogram prediction model includes 10 scales: the first scale is a score scale, with a score range of 0-100; the second scale is menopause, with a score of 35 for menopause and 0 for no menopause; the third scale is diabetes, with a score of 75 for diabetes and 0 for no diabetes; the fourth scale is polycystic ovary syndrome, with a score of 55 for polycystic ovary syndrome and 0 for no polycystic ovary syndrome; the fifth scale is endometrial lesions. The severity of the disease is as follows: complex hyperplasia is scored as 100, otherwise as 0; the sixth scale is P53 deletion, which is scored as 45 if there is a deletion, otherwise as 0; the seventh scale is PAX2 deletion, which is scored as 60 if there is a deletion, otherwise as 0; the eighth scale is the total score, ranging from 0 to 400; the ninth scale is the linear predictive value, ranging from -6 to 6; and the tenth scale is the risk of endometrial cancer, ranging from 0.1 to 0.9.

[0023] In step S5, the preferred method for validating the first prediction model includes: validating the model using LASSO regression analysis of strongly correlated factor variables, validating the model using multivariate logistic regression analysis with clinically relevant data from the control group sample, and validating the nomogram prediction model by repeatedly sampling 100 times using the Bootstrap method.

[0024] A more preferred method for validating strongly correlated variables using LASSO regression analysis is the 10-fold cross-validation method.

[0025] Preferably, the method for validating the second test model includes: performing Hosmer-Lemeshow validation.

[0026] More preferably, the Hosmer-Lemeshow validation method includes: validating the EC risk prediction model through discrimination, calibration, and fit; more preferably, the discrimination is evaluated using an ROC curve; the calibration is validated using a calibration curve; and the fit is validated using a fit curve.

[0027] Preferably, the standard for verifying the EC risk prediction model is that the AUC value of the prediction model's ROC curve is greater than or equal to 0.8.

[0028] Preferably, the optimization method of the EC risk prediction model includes: increasing the total number of samples, removing extreme data, or increasing the proportion of the sample size in the study group (reconstructing the model with new sample data).

[0029] To achieve the above-mentioned objectives, the present invention further provides a risk prediction model for the development of endometrial cancer from atypical endometrial hyperplasia, wherein the risk prediction model is constructed using the above-described method.

[0030] To further achieve the above-mentioned objectives, the present invention also provides an application of a risk prediction model for endometrial cancer arising from atypical endometrial hyperplasia, used to predict the risk of endometrial cancer arising from atypical endometrial hyperplasia.

[0031] Preferably, the method for predicting the risk of endometrial cancer in patients with atypical endometrial hyperplasia using a risk prediction model includes: using a first prediction model to predict the probability of endometrial cancer in patients with atypical endometrial hyperplasia, and using a second prediction model to predict the influencing factors of EC after hysterectomy in patients with AEH.

[0032] More preferably, the method for predicting the probability of endometrial cancer in patients with atypical endometrial hyperplasia using the first prediction model includes: mapping the patient's characteristic variables to coordinate points in the nomogram, finding the score corresponding to the coordinate point of the patient's characteristic variable vertically upwards, adding the scores corresponding to the patient's characteristic variables to obtain the total score, and obtaining the probability of endometrial cancer in patients with atypical endometrial hyperplasia by mapping the total score vertically downwards.

[0033] A risk prediction device for patients with atypical endometrial hyperplasia to develop endometrial cancer, the device being capable of running the aforementioned risk prediction model; the device can be used to predict the probability of endometrial cancer in patients with atypical endometrial hyperplasia and the influencing factors of EC after hysterectomy in patients with AEH.

[0034] Beneficial effects of this invention:

[0035] 1. The present invention provides a method for constructing a risk prediction model for endometrial cancer in patients with atypical endometrial hyperplasia. By using LASSO regression analysis and light GBM multivariate logistic regression analysis to screen influencing factors on the occurrence of endometrial cancer in patients with atypical endometrial hyperplasia, two different prediction models were constructed, which is beneficial for accurately predicting the risk of endometrial cancer in patients with atypical endometrial hyperplasia.

[0036] 2. The risk prediction model for endometrial cancer caused by endometrial atypical hyperplasia of the present invention has the advantages of high accuracy, high sensitivity and negative predictive value, which helps to carry out targeted clinical intervention for patients with endometrial atypical hyperplasia, thereby rationally allocating clinical resources.

[0037] 3. The risk prediction model of this invention is used to predict the risk of endometrial cancer in patients with atypical endometrial hyperplasia. It can accurately predict the risk of endometrial cancer, reduce the psychological burden of some patients with atypical endometrial hyperplasia, and is of great significance for the postoperative management of such patients. Attached Figure Description

[0038] Figure 1The LASSO regression analysis plot and validation plot are shown in Embodiment 1 of the present invention (A is the LASSO regression coefficient trajectory plot; B is the dimensionality reduction plot of 10-fold cross-validation LASSO regression).

[0039] Figure 2 This is a multifactor logistic regression forest plot in Embodiment 1 of the present invention;

[0040] Figure 3 This is the first prediction model (nomogram prediction model) in Embodiment 1 of the present invention.

[0041] Figure 4 The standard curve (A) and ROC curve (B) of the first prediction model in Embodiment 1 of the present invention are shown.

[0042] Figure 5 This is a descending order of importance of the influencing factors in the Light GBM model in Embodiment 1 of the present invention;

[0043] Figure 6 This is a diagram of the second prediction model in Embodiment 1 of the present invention;

[0044] Figure 7 This is the ROC plot of the second prediction model in Embodiment 1 of the present invention;

[0045] Figure 8 This is a forest plot of univariate logistic regression in Embodiment 1 of the present invention. Detailed Implementation

[0046] The technical solutions in some embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments provided in this disclosure are within the scope of protection of this disclosure.

[0047] Example 1

[0048] A risk prediction model for endometrial cancer arising from endometrial atypical hyperplasia; its construction method includes the following steps:

[0049] S1. Obtain clinical data from a sample (206 patients with AEH who underwent hysterectomy at the Third Affiliated Hospital of Guangzhou Medical University from December 2016 to June 2024 were selected as the study subjects, of which 84 cases had EC and 122 cases did not have EC). The sample was randomly divided into a study group (70%) and a validation group (30%). The clinical data (as shown in Tables 1, 2 and 3) included general information, metabolic syndrome status, laboratory test data and immunohistochemical molecular indicators.

[0050] The inclusion criteria for the sample were as follows: ① Patients' physical examination ultrasound showed endometrial thickening or intrauterine lesions, or abnormal vaginal bleeding, and hysteroscopic curettage or post-curettage pathological diagnosis confirmed AEH; EC conformed to the WHO classification of female reproductive system tumors, and the pathological diagnosis was confirmed; ② All patients had complete clinical data; ③ The study was approved by the hospital's ethics committee, and the patients' families gave informed consent and signed surgical consent forms. Exclusion criteria: ① Patients who had received exogenous hormone therapy before surgery; ② Patients with other endocrine diseases or autoimmune diseases; ③ Patients with other types of malignant tumors; ④ Patients who desired fertility or had no indication for surgery.

[0051] The general information includes: age, menarche, weight, body mass index (BMI), pulse pressure, time interval between diagnostic curettage and surgery, age distribution, regional distribution, menarche distribution, BMI distribution, whether there are menstrual abnormalities, clinical symptoms (lower abdominal pain, uterine bleeding), parity, parity, fertility status, preoperative endometrial thickness, duration of disease ≥1 year, whether menopause has occurred, and severity of lesions (divided into simple type (<50%) and complex type (≥50%) based on whether the abnormality of glandular proportion exceeds 50%).

[0052] Metabolic syndrome includes: diabetes, hyperlipidemia, hypertension, and polycystic ovary syndrome (PCOS);

[0053] Laboratory tests include: carbohydrate antigen 125 (CA125), blood glucose (GLU), creatinine, uric acid, aspartate aminotransferase (AST), and uric acid distribution.

[0054] Immunohistochemical molecular markers: P53 (no deletion, deletion), phosphatase and tensin homolog (PTEN) (no deletion, deletion), pairing cassette gene 2 antibody (PAX2) (no deletion, deletion).

[0055] S2. The clinical data of the study group were statistically analyzed using independent samples t-test and χ2 test (SPSS 25.0 software). The results are expressed as mean ± standard deviation (±) and [n (%)]. The statistical data of the sample are shown in Tables 1, 2 and 3.

[0056] S3. Using the occurrence of endometrial cancer in the study group as the dependent variable and statistical data of the study group sample as the independent variable, univariate analysis (as shown in Tables 1, 2, and 3) and logistic regression analysis (as shown in Tables 5 and 6) were performed. Figure 8 As shown in the figure, the variables with strong correlations to the development of endometrial cancer in patients with atypical endometrium were obtained; the variables with strong correlations were those with P < 0.05 in the analysis results.

[0057] Table 1 Univariate analysis of general clinical data [n (%) / (±)]

[0058] index No EC occurred (n=122) EC occurred (n=84) <![CDATA[χ 2 / t]]> P Age (years) 47.86±8.46 50.90±9.75 2.381 0.018 Menarche (age) 13.02±2.45 13.12±2.26 0.297 0.767 Weight (kg) 58.46±9.42 61.27±7.29 2.300 0.022 <![CDATA[BMI(kg / m 2 )]]> 24.36±3.25 23.68±2.75 1.569 0.118 pulse pressure difference 45.12±8.02 45.48±7.91 0.318 0.751 Interval between diagnostic curettage and surgical procedure (months) 24.05±5.68 24.32±6.36 0.319 0.750 Age distribution [n (%)] / / 3.667 0.056 <50 years old 60（49.18） 30（35.71） / / ≥50 years old 62（50.82） 54（64.29） / / Regional distribution [n (%)] / / 0.003 0.958 town 62（50.82） 43（51.19） / / rural areas 60（49.18） 41（48.81） / / Distribution of menarche [n (%)] / / 1.823 0.177 ≥13 years old 73（59.84） 58（69.05） / / <13 years old 49（40.16） 26（30.95） / / BMI distribution [n (%)] / / 0.134 0.715 <![CDATA[<24 kg / m 2 ]]> 80（65.57） 53（63.10） / / <![CDATA[≥24 kg / m 2 ]]> 42（34.43） 31（36.90） / / Is there any menstrual irregularity [n (%)]? / / 0.080 0.778 No abnormalities 27（22.13） 20（23.81） / / abnormal 95（77.87） 64（76.19） / / Clinical symptoms [n (%)] / / 0.713 0.398 lower abdominal pain 18（14.75） 9（10.71） / / Uterine bleeding 104（85.25） 75（89.29） / / Parity [n (%)] / / 2.442 0.295 0 3（2.46） 5（5.95） / / 1 34（27.87） 18（21.43） / / 2 85（69.67） 61（72.62） / / Production number [n (%)] / / 2.008 0.366 0 9（7.38） 8（9.52） / / 1 63（51.64） 35（41.67） / / 2 50（40.98） 41（48.81） / / Nurses [n (%)] / / 0.695 0.405 no 113（92.62） 75（89.29） / / yes 9（7.38） 9（10.71） / / Preoperative endometrial thickness (mm) 10.12±2.31 12.06±2.29 5.944 0.000 Duration of illness [n (%)] / / 4.421 0.035 <1 year 87（71.31） 48（57.14） / / ≥1 year 35（28.69） 36（42.86） / / Menopausal status [n (%)] / / 9.562 0.002 premenopausal 98（80.33） 51（60.71） / / menopause 24（19.67） 33（39.29） / / Disease severity [n (%)] / / 44.800 0.000 Mild 96（78.69） 27（32.14） / / Moderate to severe 26（21.31） 57（67.86） / /

[0059] Table 2 Univariate analysis of metabolic syndrome and laboratory findings [n (%) / (±)]

[0060] index No EC occurred (n=122) EC occurred (n=84) <![CDATA[χ 2 / t]]> P Diabetes [n (%)] 16（13.11） 32（38.10） 17.370 0.000 High blood lipids [n (%)] 33（27.05） 41（48.81） 10.234 0.001 Hypertension [n (%)] 38（31.15） 38（45.24） 4.242 0.039 PCOS[n(%)] 6（4.92） 25（29.76） 24.018 0.000 CA125 (U / mL) 16.49±4.85 17.32±5.73 1.120 0.264 GLU (mmol / L) 5.20±0.74 5.26±1.18 0.448 0.654 Creatinine (mmol / L) 61.02±7.17 61.45±7.58 0.413 0.680 Uric acid (μmol / L) 315.92±40.86 324.60±41.73 1.485 0.139 AST (U / L) 15.69±3.28 16.12±4.81 0.763 0.446 Uric acid distribution [n (%)] / / 9.591 0.002 <319.46 μmol / L 92（75.41） 46（54.76） / / ≥319.46 μmol / L 30（24.59） 38（45.24） / /

[0061] Table 3 Univariate analysis of immunohistochemical indicators [n (%) / (±)]

[0062] index No EC occurred (n=122) EC occurred (n=84) <![CDATA[χ 2 / t]]> P P53 positive [n (%)] 21（17.21） 41（48.81） 23.606 0.000 PTEN positive [n (%)] 34（27.87） 48（57.14） 17.792 0.000 PAX2 positive [n (%)] 38（31.15） 52（61.90） 19.129 0.000

[0063] Table 4 Results of univariate logistic regression analysis

[0064] term estimate std.error statistic p.value conf.low conf.high (Intercept) 19.6860783 5.8675287 0.50786487 0.61154811 0.00019939 1943608.08 Distribution 1 1048905.24 0.00044494 31157.7639 0 1047990.93 1049820.35 Distribution 2 8.9454E-11 5.2036E-14 -4.446E+14 0 8.9454E-11 8.9454E-11 Preoperative lesion severity 1 49.1072377 1.32404167 2.94099991 0.00327155 3.66538158 657.918075 Preoperative endometrial thickness (mm) 1.33671751 0.11598392 2.5022174 0.01234181 1.06491254 1.67789713 PTEN1 13.3546427 1.01702972 2.54846446 0.01081983 1.81942856 98.0233501 age 1.00482919 0.09499466 0.05071405 0.95955338 0.83412789 1.21046389 BMI 0.80301713 0.18338414 -1.1962824 0.23158636 0.56056754 1.15032796 CA125 0.99888783 0.02121952 -0.0524416 0.95817684 0.95819663 1.04130705 Menopause 1 2.08584938 1.62119445 0.45347808 0.65020452 0.08696004 50.0318037 uric acid 0.98186299 0.0075267 -2.4318095 0.01502361 0.96748481 0.99645486 1. Diabetes / 2. Abnormal blood sugar 23.2756636 1.72677347 1.82271062 0.06834723 0.78898697 686.648244 Hyperlipidemia 1 7.65333057 1.34280894 1.51558488 0.12962435 0.55061677 106.377924 Production Method 2 0.06153623 1.40237948 -1.9881418 0.04679601 0.00393934 0.96125464 Production Method 3 0.12726528 1.7033007 -1.2102863 0.22616905 0.00451708 3.58560242

[0065] The analysis results showed that strongly correlated variables included age, weight, preoperative endometrial thickness, menopause, hypertension, diabetes, hyperlipidemia, PCOS, uric acid distribution, lesion severity, P53, PTEN, and PAX2.

[0066] The results showed that urban and rural areas, complex lesion severity before total hysterectomy, preoperative endometrial thickness, PTEN1, uric acid, and PTEN2 were influencing factors for EC in AEH patients. Among them, preoperative endometrial thickness, complex lesion severity before total hysterectomy, urban location, and diabetes / dyslipidemia were all positively correlated with EC in AEH patients.

[0067] A Light GBM model (including 100 decision trees) was constructed, and the Light GBM model was used to screen out the core influencing factors for the occurrence of EC from the clinically relevant data of the study group sample (such as...). Figure 5 As shown); specifically: the clinical data of the study group sample were sorted in descending order of Gain score, and the factors with a total importance of more than 70% were selected as core influencing factors (including: urban and rural areas, complex lesion degree before total hysterectomy, preoperative endometrial thickness, PTEN1, uric acid, and production method).

[0068] S4. Perform backward LASSO regression analysis on strongly correlated variables (e.g., ...). Figure 1 As shown in A) and multivariate logistic regression analysis (as shown in Table 5). Figure 2 As shown), statistically significant feature variables (including menopause, diabetes, PCOS, lesion severity, P53, and PAX2) were obtained from the analysis results. A first predictive model was constructed based on these feature variables; the first predictive model was a nomogram predictive model (e.g., Figure 3 (as shown)

[0069] Table 5 Results of Multivariate Logistic Regression Analysis

[0070] Influencing factors β SE <![CDATA[Waldχ 2 ]]> OR 95% CI P age 0.316 0.639 0.245 1.372 0.392~4.800 0.621 weight 0.417 0.604 0.478 1.518 0.465~4.959 0.490 Preoperative endometrial thickness 0.491 0.451 1.185 1.634 0.675~3.955 0.276 menopause 1.016 0.473 4.604 2.761 1.092~6.984 0.032 Has diabetes 2.113 0.498 18.015 8.273 3.118~21.951 0.000 High blood lipids and high blood pressure 0.2010.169 0.5190.762 0.1500.592 1.2230.853 0.442~3.3820.267~2.724 0.6980.789 PCOS is available 1.567 0.665 5.556 1.209 1.057~1.768 0.018 Uric acid distribution ≥319.46 μmol / L 0.333 0.478 0.485 1.395 0.547~3.560 0.486 Moderate to severe lesion 2.822 0.455 38.497 16.813 6.894~41.004 0.000 P53 positive 1.221 0.432 7.993 3.391 1.454~7.908 0.005 PTEN positive 0.534 0.426 1.569 1.705 0.740~3.930 0.210 PAX2 positive 1.706 0.435 15.389 5.506 2.348~12.912 0.000 constant -3.460 0.513 45.495 0.031 - 0.000

[0071] The nomogram prediction model consists of 10 scales. The first scale is the score scale, ranging from 0 to 100. The second scale is menopause, with a score of 35 for menopause and 0 for non-menopause. The third scale is diabetes, with a score of 75 for diabetes and 0 for non-diabetes. The fourth scale is polycystic ovary syndrome (PCOS), with a score of 55 for PCOS and 0 for non-PCOS. The fifth scale is the degree of endometrial lesions, with a score of 100 for complex hyperplasia and 0 for non-complex hyperplasia. The sixth scale is P53 deletion, with a score of 45 for deletion and 0 for non-complex P53 deletion. The seventh scale is PAX2 deletion, with a score of 60 for deletion and 0 for non-complex PAX2 deletion. The eighth scale is the total score, ranging from 0 to 400. The ninth scale is the linear predictive value, ranging from -6 to 6. The tenth scale is the risk of endometrial cancer, ranging from 0.1 to 0.9.

[0072] The method for predicting the probability of endometrial cancer in patients with atypical endometrial hyperplasia using the first prediction model includes: mapping the patient's characteristic variables to coordinate points in the nomogram, finding the score corresponding to the coordinate point of the patient's characteristic variable vertically upwards, adding the scores corresponding to the patient's characteristic variables to obtain the total score, and obtaining the probability of endometrial cancer in patients with atypical endometrial hyperplasia by mapping the total score vertically downwards.

[0073] By reintegrating the core influencing factors into the light GBM model, a second predictive model is constructed (e.g., Figure 6 (As shown).

[0074] S5. Based on the clinically relevant data of the validation group, the first predictive model was validated: 10-fold cross-validation LASSO regression analysis was used to screen feature variables. When the optimal λ value was -3.274, the nomogram predictive model was validated by repeated sampling 100 times using the Bootstrap method (e.g., ...). Figure 1 As shown in B), the model fit and calibration were verified using the Hosmer-Lemeshow test and calibration curve (as shown in B). Figure 4 As shown in A), the model's discriminative power was verified using ROC curves (e.g., ...). Figure 4 (as shown in B)

[0075] The validation results show that the actual predicted curve and the corrected predicted curve fit well (χ2=4.128, P=0.816), indicating that the nomogram prediction model has good fit and calibration. The AUC of the ROC curve results is 0.898 (95% CI: 0.746~0.945), indicating that the nomogram prediction model has good discrimination.

[0076] The second predictive model was validated based on clinically relevant data from the validation group: through ROC curves (such as...). Figure 7 The predictive performance is shown in the figure (as shown) and the area under the curve (AUC), while the shap value (waterfall plot) is used to quantify the contribution of important factors.

[0077] The results show that auc=0.938, which is good performance, indicating that the model has good explanatory power for the prediction results.

[0078] Therefore, the aforementioned first and second prediction models are the risk prediction models for the development of endometrial cancer from atypical endometrial hyperplasia according to the present invention.

[0079] In the description of this specification, specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples.

[0080] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for constructing a risk prediction model for endometrial cancer arising from atypical endometrial hyperplasia, characterized in that, Includes the following steps: S1. Obtain clinically relevant data of the samples and randomly divide the samples into a study group and a validation group; the samples are AEH patients undergoing routine hysterectomy; the clinically relevant data include general information, metabolic syndrome status, laboratory test data, and immunohistochemical molecular indicators; S2. The clinical data of the study group were statistically analyzed using independent samples t-test and χ2 test, and expressed as mean and standard deviation to obtain the statistical data of the sample. S3. Using the occurrence of endometrial cancer in the study group as the dependent variable and the statistical data of the study group sample as the independent variable, univariate analysis and logistic regression analysis were performed to obtain the strongly correlated factors for the occurrence of endometrial cancer in patients with atypical endometrium; the strongly correlated factors were those with P < 0.05 in the analysis results. We constructed a Light GBM model and used it to screen out the core influencing factors for the occurrence of EC from the clinically relevant data of the study group sample. S4. Perform backward LASSO regression analysis and multivariate logistic regression analysis on the strongly correlated factor variables to obtain the feature variables with statistically significant differences in the analysis results, and construct the first prediction model based on the feature variables; the first prediction model is a nomogram prediction model. The core influencing factors were reintroduced into the light GBM model to construct a second prediction model; S5. Based on the clinical data of the validation group, the first prediction model and the second prediction model are validated, and the first prediction model and the second prediction model are optimized according to the validation results; the optimized first prediction model and the second prediction model are the risk prediction models for endometrial cancer caused by atypical endometrial hyperplasia of the present invention.

2. The construction method according to claim 1, characterized in that, In S1, the general information includes: age, menarche, weight, body mass index, pulse pressure, time interval between diagnostic curettage and surgery, age distribution, regional distribution, menarche distribution, BMI distribution, whether there are menstrual abnormalities, clinical symptoms, parity, number of deliveries, fertility status, preoperative endometrial thickness, duration of illness ≥1 year, whether menopause has occurred, and degree of lesion; the metabolic syndrome status includes: diabetes mellitus, hyperlipidemia, hypertension, and polycystic ovary syndrome; the laboratory test data includes carbohydrate antigen 125, blood glucose, creatinine, uric acid, aspartate aminotransferase, and uric acid distribution; the immunohistochemical molecular indicators include P53, phosphatase, tensin homolog, and paired cassette gene 2 antibody.

3. The construction method according to claim 1, characterized in that, In S3, the screening method for the core influencing factors includes: sorting the clinically relevant data in descending order of feature importance, and selecting the clinically relevant data with a total feature importance of not less than 70% as the core influencing factors.

4. The construction method according to claim 1, characterized in that, In S4, the nomogram prediction model includes 10 scales: the first scale is the score scale, with a score range of 0-100; the second scale is menopause, with a score of 35 for menopause and 0 for no menopause; the third scale is diabetes, with a score of 75 for diabetes and 0 for no diabetes; the fourth scale is polycystic ovary syndrome, with a score of 55 for polycystic ovary syndrome and 0 for no polycystic ovary syndrome; the fifth scale is the degree of endometrial lesions, with a score of 100 for complex hyperplasia and 0 for no complex hyperplasia; the sixth scale is P53 deletion, with a score of 45 for deletion and 0 for no deletion; the seventh scale is PAX2 deletion, with a score of 60 for deletion and 0 for no deletion; the eighth scale is the total score, with a score range of 0-400; the ninth scale is the linear predictive value, with a score range of -6 to 6; and the tenth scale is the risk of endometrial cancer, with a score range of 0.1-0.

9.

5. The construction method according to claim 1, characterized in that, In S5, the standard for verifying the EC risk prediction model is that the AUC value of the ROC curve of the first prediction model and the second prediction model is greater than or equal to 0.

8.

6. The construction method according to claim 1, characterized in that, In S5, the optimization methods for the EC risk prediction model include: increasing the total number of samples, removing extreme data, or increasing the proportion of the study group sample size.

7. The construction method according to claim 1, characterized in that, In S5, the methods for validating the first prediction model include: validating the model using LASSO regression analysis on strongly correlated factor variables, validating the model using multivariate logistic regression analysis with clinically relevant data from the control group sample, and validating the nomogram prediction model by repeatedly sampling 100 times using the Bootstrap method; the methods for validating the second prediction model include: performing Hosmer-Lemeshow validation.

8. A risk prediction model for endometrial cancer arising from atypical endometrial hyperplasia, characterized in that, The risk prediction model is constructed using the construction method described in any one of claims 1-7.

9. The application of the risk prediction model for endometrial cancer arising from endometrial atypical hyperplasia as described in claim 8 in predicting the risk of endometrial cancer arising from endometrial atypical hyperplasia.

10. The application according to claim 9, characterized in that, Methods for predicting the risk of endometrial cancer in patients with atypical endometrial hyperplasia using risk prediction models include: using a first prediction model to predict the probability of endometrial cancer in patients with atypical endometrial hyperplasia, and using a second prediction model to predict influencing factors of EC after hysterectomy in patients with AEH.

Images