A method for constructing a MASH multimodal prediction model based on MRI-PDFF

Abstract

The present invention discloses a method for constructing a multimodal prediction model of MASH based on MRI-PDFF, comprising the following steps: S1, collecting MASH patient data: including basic information, comorbidities, and metabolic variables; S2, obtaining predictive variables: performing univariate analysis and multivariate regression analysis on the data collected in step S1 in sequence to obtain predictive variables; S3, data modeling: using a machine learning algorithm, constructing a MASH multimodal prediction model formula composed of the predictive variables obtained in step S2, the formula being: Logit score = -14.257 + 0.167MRI-PDFF + 0.066age + 0.070WC + 0.031AST; S4, model evaluation: drawing a receiver operating characteristic curve ROC curve, and calculating the area under the ROC curve to evaluate the effectiveness of the prediction model. The prediction model of the present invention is based on MRI-PDFF technology, displayed in the form of a nomogram, and the diagnosis probability of MASH is calculated by multiple indicators combined with ROC, which is easy to operate and provides a reference basis for non-invasive diagnosis of MASH; the prediction model has high discrimination, good calibration, and stable test results.

Description

Technical Field

[0001] The present invention belongs to the field of risk prediction, and in particular relates to a method for constructing a MASH multimodal prediction model based on MRI-PDFF. Background Art

[0002] Obesity promotes the occurrence and development of metabolic dysfunction-associated fatty liver disease (MASLD). The prevalence of metabolic dysfunction-associated steatohepatitis (MASH) in obese people is much higher than that in non-obese people.

[0003] Currently, the gold standard for the diagnosis of MASH is still liver biopsy pathology, however, its invasiveness, high cost and sampling errors make it unfavorable for diagnosis and long-term follow-up observation of the disease in clinical practice. Therefore, there is an urgent need to find a non-invasive diagnostic method for MASH in obese patients in clinical practice. Currently, non-invasive tests (NITs) are available for the diagnosis of steatohepatitis, however, there are no blood markers that can accurately distinguish between MASH and simple steatosis (SS). In addition, some blood-based or imaging-based scores (e.g., NIS2+, FAST [FibroScan-AST] score, MAST [based on MRI and serological indicators]) are used to identify "high-risk" MASH (MASH with stage 2 or higher fibrosis confirmed by biopsy), but they cannot cover the entire MASH population. Screening for the entire MASH population can avoid missing any patient at risk of progression to a higher stage of fibrosis and adverse clinical outcomes.

[0004] MRI-Based Proton Density Fat Fraction (MRI-PDFF) is the ratio of hydrogen proton density in adipose tissue to all liver signal sources (the sum of hydrogen proton density in adipose tissue and water). It is currently recognized as the best technology for non-invasive quantitative assessment of fatty liver, and can objectively assess the fat content of the entire liver and its dynamic changes. MRI-PDFF is more accurate than VCTE-based CAP in diagnosing fatty changes, does not require patients to be exposed to ionizing radiation (such as CT), and the sensitivity and specificity of the examination do not decrease with the increase of body mass index (BMI). Summary of the invention

[0005] Based on this, the present invention provides a method for constructing a MASH multimodal prediction model based on MRI-PDFF, and the specific scheme is as follows: A method for constructing a MASH multimodal prediction model based on MRI-PDFF, comprising the following steps: S1. Collect MASH patient data: including basic information, comorbidities, and metabolic variables; S2. Obtaining predictor variables: performing univariate analysis and multivariate regression analysis on the data collected in step S1 to obtain predictor variables; S3, data modeling: using machine learning algorithms, construct a MASH multimodal prediction model formula composed of the prediction variables obtained in step S2, the formula is: Logit score = - 14.257+0.167MRI-PDFF+0.066age+0.070WC+0.031AST; S4. Model evaluation: Plot the receiver operating characteristic (ROC) curve and calculate the area under the ROC curve to evaluate the effectiveness of the prediction model.

[0006] Preferably, the basic information in S1 includes: age, male or not, smoker or not, systolic blood pressure SBP, diastolic blood pressure DBP; height, weight, body mass index BMI, waist circumference, hip circumference HC, waist-to-hip ratio WHR, magnetic resonance proton density fat fraction MRI-PDFF.

[0007] Preferably, the comorbidities in S1 include: whether there is hypertension, whether there is hyperlipidemia, and glucose metabolism. NGT represents normal glucose metabolism, Pre-DM represents prediabetes, and DM represents diabetes.

[0008] Preferably, the metabolic variables in S1 include: glycated hemoglobin HbA1c, fasting blood glucose FBG, fasting insulin FINS, insulin resistance level HOMA-IR, triglycerides TG, total cholesterol TC, high-density lipoprotein cholesterol HDL-C, low-density lipoprotein cholesterol LDL-C, uric acid UA, alanine transaminase ALT, serum aspartate aminotransferase AST, FIB-4 index, total bilirubin TBIL, direct bilirubin DBIL, total bile acid TBA, alpha-fetoprotein AFP, platelet PLT, C-reactive protein CRP, interleukin-6 IL-6, interleukin-4 IL-4, interferon γ IFN-γ, interleukin-17 IL-17, interferon α IFN-α, and lactic acid Lac.

[0009] Preferably, the predictor variables in S2 include: age, waist circumference, MRI-PDFF, and AST.

[0010] Beneficial effects: 1. The prediction model of the present invention is based on MRI-PDFF technology and is displayed in the form of a nomogram. Multiple indicators are combined with ROC to calculate the diagnostic probability of MASH. It is easy to operate and provides a reference for non-invasive diagnosis of MASH. 2. The prediction model of the present invention has high discrimination, good calibration and stable detection results. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 A nomogram model for the development of MASH in obese patients.

[0012] Figure 2 ROC curve for evaluating the discrimination of the prediction model.

[0013] Figure 3 Hosmer-Lemeshow curve for evaluating the calibration of the prediction model.

[0014] Figure 4 DCA curve to evaluate the clinical validity of the prediction model. DETAILED DESCRIPTION

[0015] The present invention will be further explained below in conjunction with the accompanying drawings and embodiments.

[0016] The present invention can be better understood according to the following examples. However, it is easy for those skilled in the art to understand that the specific material ratios, process conditions and results described in the examples are only used to illustrate the present invention.

[0017] Example 1: Data Collection Data of obese patients who underwent liver biopsy in Nanjing Drum Tower Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College from May 2023 to September 2023 were collected, including basic information, comorbidities, and metabolic variables. Basic information: Age, male or not, smoker or not, systolic blood pressure (SBP), diastolic blood pressure (DBP); height and weight, and body mass index (BMI) was calculated, BMI = body weight / (height × height) (kg / m 2 Waist circumference (WC) and hip circumference (HC) were measured and waist-to-hip ratio (WHR) (WHR = waist circumference / hip circumference) and MRI-Based Proton Density Fat Fraction (MRI-PDFF) were calculated.

[0018] Comorbidities: Hypertension, hyperlipidemia, glucose metabolism, NGT stands for normal glucose metabolism, Pre-DM stands for prediabetes, DM stands for diabetes Metabolic variables: glucose metabolism indices [glycosylated hemoglobin (HbA1c), fasting blood glucose (FBG), fasting insulin (FINS), insulin resistance level was assessed by the insulin resistance index (HOMA-IR), HOMA-IR = fasting blood glucose (mmol / L) × fasting insulin (μU / L) / 22.5], lipid metabolism indices [triglycerides (TG), total cholesterol (TC), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C)], liver-related indices [alanine aminotransferase (ALT), serum aspartate aminotransferase (AST), calculation of FIB-4 index, FIB-4 = (age×AST) / (PLT×(ALT^(1 / 2))], total bilirubin (TBIL), direct bilirubin (DBIL), total bile acid (TBA), alpha-fetoprotein (AFP), platelets (PLT)], uric acid (UA), inflammatory markers [C-reactive protein (CRP), interleukin-6 (IL-6), interleukin-4 (IL-4), interferon γ (IFN-γ), interleukin-17 (IL-17), interferon α (IFN-α)], and lactic acid (Lac).

[0019] The inclusion criteria were as follows: (1) BMI ≥ 27.5 kg / m 2 ; (2) Liver steatosis ≥5% confirmed by biopsy; (3) Voluntarily participate in and complete abdominal MRI examination.

[0020] The exclusion criteria were as follows: (1) Pregnancy or suspected pregnancy; (2) The patient has acute infection, such as acute cholecystitis, appendicitis, pancreatitis, acute abdominal infection, acute gastroenteritis, acute upper respiratory tract infection, acute urinary tract infection (within 2 weeks); (3) A history of malignant tumor; (4) Presence of acute cardiovascular and cerebrovascular disease (within 6 months); (5) Patients who are too large to undergo MRI examination due to metal implants in the body or claustrophobia; (6) The clinical case data are incomplete.

[0021] According to the inclusion and exclusion criteria, a total of 109 subjects were included, and the interval between MRI examination and pathological biopsy was less than 7 days. The data collection of this invention was approved by the Ethics Committee of Nanjing Drum Tower Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College (2023-547), and all subjects signed informed consent before the examination.

[0022] According to whether the pathology of the research subject was MASH, the data were divided into MASH (44 cases) and non-MASH (65 cases). The statistical results are shown in Table 1. The P values ​​of WC, PDFF, HbA1C, FBG, FINS, HOMA-IR, AST, and ALT between the MASH group and the non-MASH group were < 0.05, indicating that there were significant differences between the data.

[0023] Table 1 Comparison of clinical characteristics between non-MASH group and MASH group Example 2: Obtaining predictor variables The data collected in Example 1 were statistically analyzed using R software. Univariate analysis was used to screen out variables with significant differences between MASH and non-MASH, and multivariate regression analysis was performed. The results are shown in Table 2. Patient age (OR=1.071, 95%CI: 1.002~1.145, P=0.043), waist circumference (OR=1.137, 95%CI: 1.048~1.234, P=0.002), MRI-PDFF (OR=1.198, 95%CI: 1.089~1.271, P<0.001), and AST (OR= 1.034, 95%CI: 1.004~1.065, P=0.025) were independent influencing factors for the diagnosis of MASH. OR is the odds ratio, and 95%CI is the 95% confidence interval. P -value is the error P value.

[0024] Table 2 Results of univariate analysis and multivariate regression analysis Example 3: Data Modeling According to the results of multivariate regression analysis, age, WC, MRI-PDFF, and AST were finally included in the nomogram model as predictor variables, such as Figure 1As shown in the figure, after incorporating the four indicators of age, WC, MRI-PDFF, and AST, the R package rms was used to establish a nomogram method score to establish a prediction model for non-invasive diagnosis of MASH risk in obese patients. Each line segment of an independent variable has a corresponding scale, indicating the numerical range of the independent variable, and the length of the line segment reflects the contribution of the independent variable to the dependent variable. The longer the line segment, the greater the contribution; the shorter the line segment, the smaller the contribution; the Points at the top of the nomogram represent the corresponding scores when each independent variable takes different values, and the sum of all independent variable scores is the total score (Total Points); the total score is used to infer the probability of the dependent variable, that is, to predict the risk of the event.

[0025] The Logit score was calculated by the following equation and replaced the original variable for subsequent analysis: Logit score = -14.257+0.167MRI-PDFF+0.066age+0.070WC+0.031AST. The scores of each predictor variable are added together to obtain the probability corresponding to the total score, which is the predicted probability of MASH diagnosis by this model.

[0026] Example 4: Prediction Model Evaluation 1. Discrimination: Use the R software "PRROC" function to draw the ROC curve. The Y axis of the curve is sensitivity and the X axis is specificity. The ROC curve can be used to evaluate the predictive performance of the statistical model. The AUC range is 0.80-0.89, indicating that the model has a high degree of discrimination. The results of this prediction model are as follows Figure 2 As shown in the figure, the AUC was 0.8764 (95%CI: 0.765-0.903), indicating that the nomogram prediction model had good discrimination.

[0027] 2. Calibration: The Hosmer-Lemeshow goodness-of-fit test was performed using the “risk Regression” function of R software. The P value was 0.593, which was greater than 0.05, indicating that the model had a good fit. The calibration curve was drawn using the Bootstrap method. The results are shown in the figure. Figure 3 As shown, the calibration curve of the model is close to the standard curve. The above results show that the predicted probability of the prediction model of the present invention is well consistent with the actual probability of occurrence, and the model has a good calibration degree.

[0028] 3. Clinical effectiveness: The DCA curve was drawn using the R software “rmda” function. The results are as follows: Figure 4As shown in the figure, the calibration curve was used to evaluate the clinical effectiveness of the model. It showed that the actual curve and ideal curve of the nomogram prediction model were in good agreement, indicating that the prediction model was consistent with clinical practice. The horizontal axis of the DCA curve represented the probability threshold, and the vertical axis represented the net benefit. The results showed that the model had a high clinical value.

[0029] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for constructing a MASH multimodal prediction model based on MRI-PDFF, characterized in that: The following steps are involved: S1. Collect MASH patient data: including basic information, comorbidities, and metabolic variables; S2. Obtaining predictor variables: performing univariate analysis and multivariate regression analysis on the data collected in step S1 to obtain predictor variables; S3, data modeling: using machine learning algorithms, construct a MASH multimodal prediction model formula composed of the prediction variables obtained in step S2, the formula is: Logit score = - 14.257+0.167MRI-PDFF+0.066age+0.070WC+0.031AST; S4. Model evaluation: Plot the receiver operating characteristic (ROC) curve and calculate the area under the ROC curve to evaluate the effectiveness of the prediction model.

2. The method for constructing a MASH multimodal prediction model based on MRI-PDFF according to claim 1, characterized in that: The basic information in S1 includes: age, male or not, smoker or not, systolic blood pressure SBP, diastolic blood pressure DBP; height, weight, body mass index BMI, waist circumference, hip circumference HC, waist-to-hip ratio WHR, magnetic resonance proton density fat fraction MRI-PDFF.

3. The method for constructing a MASH multimodal prediction model based on MRI-PDFF according to claim 1, characterized in that: The comorbidities in S1 include: whether there is hypertension, whether there is hyperlipidemia, and glucose metabolism. NGT represents normal glucose metabolism, Pre-DM represents prediabetes, and DM represents diabetes.

4. The method for constructing a MASH multimodal prediction model based on MRI-PDFF according to claim 1, characterized in that: The metabolic variables in S1 include: glycated hemoglobin HbA1c, fasting blood glucose FBG, fasting insulin FINS, insulin resistance level HOMA-IR, triglycerides TG, total cholesterol TC, high-density lipoprotein cholesterol HDL-C, low-density lipoprotein cholesterol LDL-C, uric acid UA, alanine transaminase ALT, serum aspartate aminotransferase AST, FIB-4 index, total bilirubin TBIL, direct bilirubin DBIL, total bile acid TBA, alpha-fetoprotein AFP, platelet PLT, C-reactive protein CRP, interleukin-6 IL-6, interleukin-4 IL-4, interferon γ IFN-γ, interleukin-17 IL-17, interferon α IFN-α, and lactic acid Lac.

5. The method for constructing a MASH multimodal prediction model based on MRI-PDFF according to claim 1, characterized in that: The predictor variables in S2 include: age, waist circumference, MRI-PDFF, and AST.

Images