Multi-parameter pre-eclampsia risk assessment prediction model and application thereof

Abstract

The invention provides a construction method of a multi-parameter pre-eclampsia risk assessment prediction model, a prediction model, a prediction product, a prediction system and related applications thereof. Specifically, six metabolic markers including 4-hydroxyphenyllactic acid, homovanillic acid, ornithine, sarcosine, adenosine monophosphate and 3-methylbutyryl carnitine and two preeclampsia-related protein markers IGFBP1 and PLGF are measured, the blood pressure of both arms of an individual is measured, the average arterial pressure is calculated, clinical information of a maternal body is combined, and the blood pressure of both arms of the individual is calculated. A metabolism and protein marker multi-index combined prediction model for the onset risk of preeclampsia of pregnancy 12-17 + 6 weeks is established. The effect of the random forest model for screening early-onset and late-onset eclampsia is better than that of using a single protein or metabolite marker or other metabolism and protein marker combined models, so that the work of predicting the preeclampsia occurrence risk in 12-17 + 6 weeks of pregnancy is realized, and clinical doctors are guided to carry out advanced intervention.

Description

Technical Field

[0001] The present invention relates to the field of preeclampsia analysis, and particularly relates to a method for constructing a multi-parameter preeclampsia risk assessment prediction model, a prediction product, a prediction system and related applications thereof. Background Art

[0002] Preeclampsia (PE) ranks second among the causes of maternal death due to the hypertensive diseases associated with this disease, accounting for about 14% of the total deaths. The definition of PE is new-onset hypertension accompanied by proteinuria or damage to other organs (such as the kidney, liver or brain) after 20 weeks of pregnancy. Its typical symptoms are hypertension, proteinuria and edema. The current prevalence of PE in China is about 2.3%. Although it is lower than the global overall prevalence of 2% - 4%, the proportion of severe preeclampsia in China is much higher than that in other countries. Preeclampsia can be divided into two categories according to the onset time: early-onset preeclampsia when it is less than 34 weeks and late-onset preeclampsia when it is greater than or equal to 34 weeks. PE may lead to serious complications, including internal bleeding, seizures, stroke, premature birth and death, and also cause a series of complications in pregnant women and newborns after childbirth. Currently, 60% of the cases of PE death are due to missed diagnosis or late diagnosis.

[0003] As early as 2003, research scholars found that preeclampsia is related to the elevation of soluble FMS-like tyrosine kinase 1 (sFlt-1). For the prediction of preeclampsia, the currently widely used method is to use the ratio of sFlt-1 and placental growth factor (PLGF). When the ratio is 38 or lower, the negative predictive value of predicting no preeclampsia within one week is 99.3%, the sensitivity is 80%, and the specificity is 78.3%; when the ratio is higher than 38, the positive predictive value of predicting the occurrence of PE within 4 weeks is 36.7%, the sensitivity is 66.2%, and the specificity is 83.1%. However, the sFlt-1 / PlGF ratio is mainly for the prediction of preeclampsia risk in the second and third trimesters of pregnancy, and its predictive value in the first trimester is limited. Akolekar et al. used the mean arterial pressure in the first trimester, uterine artery pulsatility index, PLGF and PAPP-A to predict early-onset preeclampsia. The detection rate of early-onset preeclampsia in pregnant women in Hong Kong, China by this mixed model reached 72% at a 10% false positive rate, and the detection rate of late-onset preeclampsia was 55%, which still cannot meet the clinical requirements for the accuracy of disease screening. Therefore, further research and development of earlier and more accurate clinical prediction methods for PE is of great significance for the early prevention and effective management of maternal PE. Summary of the Invention

[0004] Based on the limitations of the prior art, the object of the present invention is to develop a more accurate preeclampsia risk assessment and prediction model by finding metabolic markers and protein markers that can better correlate with preeclampsia, through in-depth research on these markers, and combining the mean arterial pressure of the pregnant woman and other clinical information, so as to judge the risks of early-onset and late-onset preeclampsia during pregnancy, and to guide clinicians to carry out drug interventions to reduce the risk of preeclampsia.

[0005] In a first aspect, the present invention provides a method for constructing a multi-parameter preeclampsia risk assessment and prediction model, the method comprising obtaining clinical information, mean arterial pressure (MAP), and data of metabolic markers and / or protein markers of pregnant women known to have preeclampsia and healthy pregnant women, constructing a model through machine learning with the relevant data to obtain a preeclampsia risk assessment and prediction model, and determining a threshold.

[0006] Among them, the metabolic markers include at least one of 4-hydroxy-phenyllactic acid, homovanillic acid, ornithine, sarcosine, adenosine monophosphate, and 3-methylbutyrylcarnitine.

[0007] The protein markers include IGFBP1 and / or PLGF.

[0008] The clinical information of the pregnant woman includes age, height, weight at 12-17 +6 weeks of gestation, parity, number of deliveries, history of adverse pregnancy, history of adverse past events (including previous gestational diabetes, preeclampsia history, chronic hypertension history, systemic lupus erythematosus, and antiphospholipid syndrome), and clinical data such as whether there is in vitro fertilization.

[0009] In a specific embodiment of the present invention, the present invention measures six preeclampsia-related metabolic markers in the plasma of pregnant women - including 4-hydroxy-phenyllactic acid, homovanillic acid, ornithine, sarcosine, adenosine monophosphate, and 3-methylbutyrylcarnitine - and two preeclampsia-related protein markers IGFBP1 and PLGF, calculates the ratio of metabolic markers and protein markers, combines the mean arterial pressure of the mother and clinical information, constructs a prediction model for early-onset and late-onset preeclampsia, and conducts disease prediction. The present invention uses the random forest algorithm to construct prediction models for early-onset and late-onset preeclampsia respectively to predict the probability of a pregnant woman developing preeclampsia in the early pregnancy, determines the risk threshold according to the Receiver operating characteristic curve (ROC), and judges the risks of early-onset and late-onset preeclampsia during pregnancy.

[0010] In a second aspect, the present invention provides the use of a combination of metabolic markers and protein markers related to preeclampsia in the preparation of a detection product related to preeclampsia, wherein the metabolites are selected from at least three of 4-hydroxyphenyllactic acid, homovanillic acid, ornithine, sarcosine, adenosine monophosphate, and 3-methylbutyrylcarnitine, and the protein markers are IGFBP1 and PLGF.

[0011] In a third aspect, the present invention provides a marker detection kit related to preeclampsia, and the markers include 4-hydroxyphenyllactic acid, homovanillic acid, ornithine, sarcosine, adenosine monophosphate, 3-methylbutyrylcarnitine, IGFBP1, and / or PLGF.

[0012] In a fourth aspect, based on the prediction model obtained in the first aspect above, the present invention provides a multi-parameter preeclampsia risk assessment prediction method (as Figure 10 shown), and this method includes:

[0013] 1) Obtain the clinical information, mean arterial pressure, data of metabolic markers, and protein markers of the pregnant woman to be tested;

[0014] 2) Input the obtained data of the clinical information, mean arterial pressure, metabolic markers, and protein markers of the pregnant woman into the prediction model for processing. When the risk value is higher than the set threshold, it is determined that the pregnant woman has a high risk of preeclampsia.

[0015] In a fifth aspect, based on the prediction method in the fourth aspect above, the present invention provides a prediction system for multi-parameter preeclampsia risk assessment, and this system includes:

[0016] 1) A device for obtaining the clinical information, mean arterial pressure, data of metabolic markers, and protein markers of the pregnant woman to be tested;

[0017] 2) A device for performing prediction model processing on the obtained data of the clinical information, mean arterial pressure, metabolic markers, and protein markers of the pregnant woman in 1);

[0018] 3) A device for outputting the prediction result.

[0019] In a sixth aspect, the present invention provides a prediction product for multi-parameter preeclampsia risk assessment, and this product includes: a memory and a processor. The memory is used to store a program; the processor is used to implement the preeclampsia risk assessment prediction method mentioned in the fourth aspect above by executing the program stored in the above memory.

[0020] Meanwhile, the present invention also provides a computer-readable storage medium, and a program is stored on this medium, and this program can be executed by a processor to implement the preeclampsia risk assessment prediction method mentioned in the fourth aspect above.

[0021] In addition, the present invention also provides a computer-readable storage medium, on which a prediction model obtained by the construction method of the first aspect is stored.

[0022] The beneficial effects of the present invention are as follows:

[0023] Based on maternal clinical information, mean arterial pressure (MAP), and the combined markers of Homovanillic acid*Sarcosine*Hydroxyphenyllactic acid*Ornithine / IGFBP1 / PLGF ratio, the present invention jointly constructs a prediction model for early-onset preeclampsia. Meanwhile, based on maternal clinical information, MAP, and Ornithine*Isovelarylcarnitine*Adenosine monophosphate / PLGF ratio, a prediction model for late-onset preeclampsia is constructed. The above two models perform better than single protein markers such as PLGF and IGFBP1, single MAP, single metabolite combinations, and other metabolic and protein combination models. The present invention does not rely on the uterine artery pulsatility index, realizes the early pregnancy prediction of preeclampsia, and the detection of relevant metabolic and protein markers and the acquisition of maternal clinical information data are simple, which is convenient for clinical popularization and application.

[0024] The present invention uses a relatively large sample size (see the examples) to evaluate the prediction effects of the above early-onset preeclampsia and late-onset preeclampsia models. The results show that the sensitivity and specificity of the early-onset preeclampsia prediction model are 77% and 90.14% respectively, and the sensitivity and specificity of the late-onset preeclampsia prediction model are 59.66% and 90.14% respectively. The present invention has relatively high accuracy in predicting early-onset and late-onset preeclampsia, has important value for the prevention and early diagnosis of the disease, and is of great significance for improving maternal and child outcomes and reducing maternal and infant mortality rates. Description of the Drawings

[0025] Figure 1 It is a difference graph of the multiples of the median (MoM) values of four different metabolites, namely Hydroxyphenyllactic acid (A graph), Sarcosine (B graph), Homovanillic acid (C graph), and Ornithine (D graph), and the product of the MoM values of four metabolic markers (E graph) in early-onset preeclampsia and healthy control samples in the training set (train), validation set (validation), and test set (test). Among them, EPE: early-onset preeclampsia, control: healthy control;

[0026] Figure 2For the differences in the multiples of median (MoM) values of three different metabolic markers, namely ornithine (A figure), adenosine monophosphate (B figure), isovelarylcarnitine (C figure), and the product of the MoM values of the three metabolic markers, in late-onset preeclampsia and healthy control samples in the training set, validation set, and test set. Also shown is the difference plot (D figure). Here, LPE represents late-onset preeclampsia, and control represents healthy control;

[0027] Figure 3 For the changes in the concentrations of two protein markers at different sampling gestational ages. Among them, figure A shows that the expected median concentration of IGFBP1 protein increases significantly with the increase in sampling gestational age, and figure B shows that the expected median concentration of PLGF protein increases significantly with the increase in sampling gestational age;

[0028] Figure 4 For the comparison of the differences in IGFBP1 MoM values between preeclampsia and healthy controls in the training set, validation set, and test set. Here, EPE represents early-onset preeclampsia, LPE represents late-onset preeclampsia, and control represents healthy control;

[0029] Figure 5 For the comparison of the differences in PLGF MoM values between preeclampsia and healthy controls in the training set, validation set, and test set. Here, EPE represents early-onset preeclampsia, LPE represents late-onset preeclampsia, and control represents healthy control;

[0030] Figure 6 For the comparison of the differences in the ratios of metabolic markers and protein markers between preeclampsia and healthy controls in the training set, validation set, and test set. Figure A shows the comparison of the differences in the ratio of Homovanillic acid*Sarcosine*Hydroxyphenyllactic acid*Ornithine / IGFBP1 / PLGF between early-onset preeclampsia and healthy control samples (each data is the MoM value); Figure B shows the comparison of the differences in the ratio of Ornithine*Isovelarylcarnitine*Adenosinemonophosphate / IGFBP1 / PLGF between late-onset preeclampsia and healthy control samples (each data is the MoM value);

[0031] Figure 7 For the prediction effects of the preeclampsia prediction model in the training set, validation set, and test set. Figure A shows the ROC curves of the early-onset preeclampsia prediction model in the training set, validation set, and test set; Figure B shows the ROC curves of the late-onset preeclampsia prediction model in the training set, validation set, and test set;

[0032] Figure 8Ranking of the importance of different prediction features in the early-onset preeclampsia prediction model. Note: MAP is the MAP MoM value, Biomarker_ratio is the ratio of Homovanillic acid * Sarcosine * Hydroxyphenyllactic acid * Ornithine / IGFBP1 MoM / PLGF MoM (each metabolite is the MoM value), BMI is the BMI MoM value, age is the age MoM value, IVF indicates whether there is in vitro fertilization, PMH indicates whether there is one or more of a history of gestational diabetes, preeclampsia, chronic hypertension, systemic lupus erythematosus, and antiphospholipid syndrome, abortion indicates whether there are more than 2 abortions or inductions, and Birth_history is the parity;

[0033] Figure 9 Ranking of the importance of different prediction features in the late-onset preeclampsia prediction model. Note: MAP is the MAP MoM value, Biomarker_ratio is the ratio of Ornithine * Isovelarylcarnitine * Adenosine monophosphate / IGFBP1MoM / PLGF MoM (each metabolite is the MoM value), BMI is the BMI MoM value, age is the age MoM value, IVF indicates whether there is in vitro fertilization, PMH indicates whether there is one or more of a history of gestational diabetes, preeclampsia, chronic hypertension, systemic lupus erythematosus, and antiphospholipid syndrome, abortion indicates whether there are more than 2 abortions or inductions, and Birth_history is the parity;

[0034] Figure 10 Schematic diagram of the multi-parameter preeclampsia risk assessment and prediction method provided by the present invention. Detailed implementation manners

[0035] Based on the R & D data in the exploration stage, the present invention analyzes the metabolomics of plasma samples of pregnant women at 12 - 17 +6 weeks of pregnancy, discovers and validates 6 metabolic markers that can be used to predict the risk of preeclampsia, as shown in Table 1. At the same time, through proteomics and enzyme-linked immunosorbent assay techniques, 2 protein markers that can be used to predict the risk of preeclampsia are discovered and validated, including IGFBP1 and PLGF. The physiological significance of the two indicators is as shown in Table 2 below.

[0036] Table 1 Metabolites and Their Clinical Significances

[0037]

[0038]

[0039] Table 2. Detection Indicators and Clinical Significance

[0040]

[0041] The above protein markers PLGF and IGFBP1 can be specifically detected by a variety of methods. In the embodiments of the present invention, the protein detection technology used is the ELISA enzyme-linked immunosorbent assay technology. Those skilled in the art should understand that replacing with other protein detection technologies, such as protein arrays, proteomics, expression proteomics, mass spectrometry (such as liquid chromatography-mass spectrometry (LC-MS), multiple reaction monitoring (MRM), selected reaction monitoring (SRM), scheduled MRM, scheduled SRM), 2DPAGE, 3D PAGE, electrophoresis, protein chips, proteomic microarrays, Edman degradation, direct or indirect ELISA, immunosorbent assay, immunological PCR, proximity extension assay, Luminex assay or homogeneous assay, time-resolved fluorescence (TRF), fluorescence oxygen channel immunoassay (FOCI) or luminescence oxygen channel immunoassay and other targeted protein detection technologies of liquid mass spectrometers, chemiluminescent immunoassay protein detection technologies, fluorescent immunoassay protein detection technologies may all have similar detection results. And the above metabolic markers were simultaneously quantitatively detected for 6 metabolite markers by liquid chromatography and tandem mass spectrometry technology in the embodiments of the present invention. It should be understood that other detection methods such as chemiluminescence method are also applicable to the present invention.

[0042] Furthermore, based on the obtained marker concentration data of the above pregnant women, the present invention combines the above markers with maternal factors for data analysis, develops a construction method for constructing a multi-parameter preeclampsia risk assessment prediction model in the first trimester of pregnancy, and obtains a relevant model and determines a threshold that can be used as an interpretation criterion for the detection results through this construction method. In the present invention, based on the differences between early-onset preeclampsia (EPE, early-onset eclampsia) and late-onset preeclampsia (LPE, late-onset eclampsia), according to the differences in data analysis, different markers are used in the construction of different models (mainly the metabolic markers are different in the two types) to obtain a more accurate prediction model.

[0043] In a specific embodiment of the present invention, pregnant women are divided into two case groups according to the presence or absence of early-onset and late-onset preeclampsia, including the early-onset preeclampsia group and the late-onset preeclampsia group. At the same time, as a control, a healthy control group is established. The control group is composed of pregnant women randomly selected during the same period, with full-term pregnancy without pregnancy complications, the fetus growing well at birth, and no obstetric, medical or surgical complications during pregnancy. The clinical information, mean arterial pressure, metabolic markers and protein marker-related information of the three groups of pregnant women are retrieved respectively. It should be particularly noted that in the present invention, compared with the general clinical information of pregnant women, the importance of mean arterial pressure is relatively special. Therefore, mean arterial pressure is classified separately in the present invention. Among them, (1) the clinical information of pregnant women includes pregnant women's age, height, weight at 12-17 +6 weeks of pregnancy, parity, number of deliveries, history of adverse pregnancy, history of adverse past (including previous gestational diabetes, preeclampsia, chronic hypertension, systemic lupus erythematosus and antiphospholipid syndrome), and clinical data such as whether there is in vitro fertilization. (2) Mean arterial pressure. The detection method is as follows: before the pregnant woman draws blood, after sitting and resting for 5 minutes, use an electronic sphygmomanometer to measure the blood pressure of both arms of the pregnant woman at least 2 times, and calculate the MAP value by measuring the systolic blood pressure and diastolic blood pressure. The calculation formula (Formula 1) is as follows:

[0044]

[0045] where diastolic blood pressure (DBP) is the diastolic blood pressure, and PP is the systolic blood pressure - diastolic blood pressure. (3) Concentrations of metabolic markers. In a specific embodiment of the present invention, six preeclampsia-related metabolic markers in the plasma of pregnant women are determined by liquid chromatography and tandem mass spectrometry technology, including the contents of 4-hydroxyphenyllactic acid, homovanillic acid, ornithine, sarcosine, adenosine monophosphate, and 3-methylbutyrylcarnitine. 4) Protein markers. In a specific embodiment of the present invention, the protein concentrations of IGFBP1 and PLGF in the plasma samples of the three groups of pregnant women are determined by enzyme-linked immunosorbent assay technology.

[0046] After performing corresponding processing on the above data as input and using whether each pregnant woman has preeclampsia as output, a prediction model is constructed through machine learning. In the present invention, according to the different correlations of metabolic markers in different types of preeclampsia, random forest models for early-onset and late-onset preeclampsia are respectively constructed. Specifically, corresponding processing is performed on the various data of the three groups of pregnant women obtained above. The independent variables used in the model specifically include:

[0047] 1) Clinical information: age, BMI, number of deliveries, whether there are more than 2 abortions or inductions, whether there is a history of adverse past, and whether there is in vitro fertilization. Among them, the above BMI is calculated from the height at 12-17 +6 weeks of pregnancy and the weight at 12-17 +6The weekly weight is further calculated, where BMI = weight ÷ height². Both age and BMI are the age and BMI MoM values calibrated to the median of the population age and BMI. The calculation formula (Formula 2) is as follows:

[0048]

[0049] Where Clin is the original value of the pregnant woman's age or BMI, and P(Median) Clin is the median of the population age or BMI obtained from healthy control pregnant women in this example, which are respectively assigned 29 and 20.69, and MoM Clin is the age or BMI MoM value calibrated to the median. The parity is the actual parity of the pregnant woman before this pregnancy and delivery. The in vitro fertilization variable value is 0 or 1, representing whether the pregnant woman's current pregnancy is a natural pregnancy or an in vitro fertilization respectively. The value of having had more than 2 abortions or induced labors is 0 or 1, representing whether the pregnant woman does not have or has more than 2 abortions or induced labors respectively. The previous medical history is 0 or 1, representing whether the pregnant woman has no or has one or more of the previous histories of gestational diabetes, preeclampsia, familial preeclampsia, chronic hypertension, systemic lupus erythematosus and antiphospholipid syndrome.

[0050] 2) MAP MoM value: For the MAP of each pregnant woman calculated above, it is further calibrated using the median of the population MAP, that is, through the statistics of the mean arterial pressure values of healthy pregnant women and the population median, the measured values of the mean arterial pressure of all pregnant women are divided by the population mean arterial pressure median to obtain the MAP MoM value of each pregnant woman. The calculation formula (Formula 3) is as follows:

[0051]

[0052] Where MAP is the original value of the mean arterial pressure calculated after measurement of the pregnant woman, and P(Median) MAP is the population median, which is the MAP median calculated using the MAP of healthy control pregnant women in the embodiment of the present invention, and can be specifically assigned 82.98, and MoM MAP is the MAP MoM value after calibration to the median.

[0053] 3) Ratio of metabolic marker and protein marker (Biomarker_ratio)

[0054] (1) Median multiple calibration of metabolic marker concentration: The MoM correction of the concentration of each metabolic marker of each pregnant woman is to divide the original value of the concentration of this metabolic marker by the median of the concentration of this metabolite in the same batch of samples loaded onto the machine, and calculate the median multiple MoM value. The calculation formula (Formula 4) is as follows:

[0055]

[0056] Among them, Concentration metabolite is the original value of the concentration of each metabolic marker for each pregnant woman, and BatchMedian metabolite is the median concentration of this metabolic marker for the samples run on the same batch. MoM metabolite is the calibrated multiple of median (MoM) value of the metabolic marker.

[0057] (2) Calibration of the multiple of median of protein marker concentration: The gestational age MoM correction of the IGFBP1 and PLGF protein concentrations of each pregnant woman is to divide the original value of the protein concentration by the median protein concentration of the corresponding gestational age, and calculate the MoM value of the gestational age-calibrated protein median multiple. The calculation formula (Formula 5) is as follows:

[0058]

[0059] Among them, Concentration protein is the original value of the IGFBP1 and PLGF protein concentrations of each pregnant woman. Median protein,GA is the median concentration of IGFBP1 and PLGF proteins for each corresponding gestational age. In the embodiments of the present application, the median protein concentration of each sampling gestational age is calculated by calculating the IGFBP1 and PLGF protein concentrations of healthy pregnant women (see Table 5 of the embodiments). MoM protein is the MoM value calibrated by gestational age.

[0060] (3) Calculation of the ratio of metabolic marker and protein marker

[0061] Substitute the relevant parameters in (1) and (2) above into the following formula to calculate the ratio of metabolic marker and protein marker:

[0062] In the early-onset preeclampsia model, the following formula is used to calculate the ratio of four metabolic markers (homovanillic acid, sarcosine, ornithine, and hydroxyphenyllactic acid) and two protein markers:

[0063] Biomarker_ratio_EPE = Homovanillic acid * Sarcosine * Hydroxyphenyllacticacid * Ornithine / IGFBP1 / PLGF

[0064] In the late-onset preeclampsia model, the ratios of three metabolic markers (3-methylbutyrylcarnitine, adenosine monophosphate, and ornithine) and the PLGF protein marker were calculated using the following formula:

[0065] Biomarker_ratio_LPE = Ornithine * Isovelarylcarnitine * Adenosine monophosphate / PLGF

[0066] Among them, Biomarker_ratio_EPE and Biomarker_ratio_LPE are the ratios of metabolic markers and protein markers in the early-onset and late-onset models respectively. The values of the metabolic markers and protein markers used in this formula are their MoM values.

[0067] 4) Construction of the random forest model

[0068] The above data was used to predict the samples to be classified using the random forest model and the predict function. The function used was predict(RandomForest_model, data = unknown_sample_data). Here, RandomForest_model is the random forest model, and unknown_sample_data is the maternal clinical information, MAP MoM value, and the ratios of metabolic markers and protein markers of the samples to be classified. The ROC curve was constructed using the risk probability of the above constructed prediction model - the random forest model. The optimal cut-off value and the area under the curve (AUC) were determined based on the ROC curve, and then the high-risk cut-off values for early-onset preeclampsia or late-onset preeclampsia were confirmed. At the same time, the sensitivity and specificity of the above models in predicting early-onset preeclampsia and late-onset preeclampsia could be calculated respectively. It should be understood that in addition to the above random forest model, other machine learning methods can also be applicable to the present invention.

[0069] When the model is constructed by the above method, during the prediction of preeclampsia risk assessment, the clinical information, MAP, metabolic markers, and protein marker data of the pregnant woman to be tested are input into the constructed prediction model according to the processing method in the above model construction method to predict whether preeclampsia occurs in this sample. The output result of the model is the risk values of early-onset and late-onset preeclampsia. According to the defined threshold, if the disease occurrence probability of the pregnant woman to be tested exceeds the set high-risk cut-off value of early-onset preeclampsia or late-onset preeclampsia, it is determined that the pregnant woman has a high risk of early-onset or late-onset preeclampsia. In an embodiment of the present invention, the threshold of early-onset preeclampsia is 0.31, and the threshold of late-onset preeclampsia is 0.53. When inputting the relevant data of the pregnant woman to be tested, if the risk value output by the established random forest model is higher than the set threshold, it can be determined that the pregnant woman has a high risk of early-onset or late-onset preeclampsia.

[0070] It should be understood that when the model is constructed by the above method, during the prediction of preeclampsia risk assessment, the prediction models of early-onset preeclampsia and late-onset preeclampsia can be used independently, that is, independently predict early-onset preeclampsia and late-onset preeclampsia for the same pregnant woman respectively. At this time, the data input of the relevant pregnant woman can be independent, that is, input the data of different metabolic marker products and protein markers for different types of prediction models respectively. Or the two models can be combined into one. After uniformly inputting the data of the metabolic marker products and protein markers of the pregnant woman, through data recognition and retrieval, the model respectively predicts early-onset preeclampsia and late-onset preeclampsia.

[0071] The present invention will be further described in detail below with reference to the specific embodiments in conjunction with the accompanying drawings.

[0072] Example:

[0073] 1. Sample collection

[0074] 1.1 Inclusion criteria

[0075] Inclusion criteria for preeclampsia: After 20 weeks of pregnancy, the pregnant woman shows systolic blood pressure ≥ 140 mmHg and / or diastolic blood pressure ≥ 90 mmHg, accompanied by any one of the following: urinary protein quantification ≥ 0.3 g / 24 h, or urinary protein / creatinine ratio ≥ 0.3, or random urinary protein ≥ (+) (examination method when protein quantification is not possible); no proteinuria but accompanied by any one of the following organ or system involvements: important organs such as the heart, lungs, liver, and kidneys, or abnormal changes in the hematological system, digestive system, and nervous system, and involvement of the placenta-fetus, etc. Preeclampsia can be divided into two categories according to the onset time, less than 34 weeks is early-onset preeclampsia, and greater than or equal to 34 weeks is late-onset preeclampsia.

[0076] Healthy control inclusion criteria: The samples were full-term pregnancies without pregnancy complications, the fetuses had good growth at birth, and there were no obstetric, medical, or surgical complications during pregnancy. Exclusion criteria: ① Concurrent other pregnancy complications; ② Severe heart, liver, and kidney insufficiency; ③ Patients with autoimmune diseases and malignant tumor diseases. Exclude abnormal pregnant women caused by chromosomal, congenital abnormalities, preterm birth, and multiple pregnancies.

[0077] 1.2 Participants:

[0078] This invention adopted a retrospective case-control study method. Through the clinical medical record system, the prenatal examination records and hospitalization history information of pregnant women during pregnancy were retrieved. According to the above preeclampsia diagnosis criteria and healthy control inclusion criteria, pregnant women were screened. A total of 187 cases of early-onset preeclampsia, 341 cases of late-onset preeclampsia, and 934 healthy control pregnant women were included, as shown in Table 3. Healthy control pregnant women were used to calculate the median concentrations of IGFBP1 and PLGF proteins, the median population MAP, the median age, and the median BMI at each gestational week. All pregnant women participated in non-invasive prenatal genetic testing, and the remaining plasma samples collected at 12 - 17 +6 weeks of pregnancy were stored in a -80°C refrigerator for subsequent biochemical analysis. The included samples were divided into a training set, a validation set, and a test set. The numbers of early-onset preeclampsia, late-onset preeclampsia, and healthy controls corresponding to each data set are shown in Table 3.

[0079] Table 3. Introduction to the sample sources

[0080] Dataset Training set Validation set Test set Early-onset preeclampsia 44 43 100 Late-onset preeclampsia 135 30 176 Healthy control 237 180 517

[0081] 1.3 Measurement and calculation of mean arterial pressure

[0082] After the pregnant women sat in a resting position for 5 minutes, an electronic sphygmomanometer (Omron J710) was used to measure the blood pressure at least twice on both arms of the pregnant women. The measured systolic and diastolic blood pressures were input into the mean arterial pressure calculation formula to calculate the MAP value.

[0083]

[0084] Among them, diastolic blood pressure (DBP) is the diastolic blood pressure, and PP is the systolic blood pressure - diastolic blood pressure. In the training set, validation set, and test set samples, the MAP values of early-onset and late-onset preeclampsia samples were significantly higher than those of the healthy control group (the P values were all less than 0.05, Table 4). Then, the mean arterial pressure of each pregnant woman was divided by the median population mean arterial pressure of 82.89 to generate the MAP MoM value, which was used for the construction and validation of the random forest model.

[0085] 1.4 Clinical information analysis

[0086] By comparing the clinical information among the training set, validation set, early-onset and late-onset preeclampsia, and healthy controls, the Wilcoxon rank sum test was used to statistically analyze the differences in continuous variables among the three groups of pregnant women, and the Fisher exact test was used to statistically analyze the differences in categorical variables. The results showed that the ages, BMIs, and mean arterial pressures of pregnant women with early-onset and late-onset preeclampsia were higher than those of the control group, and the differences were statistically significant (P values were all < 0.05, Table 4). Pregnant women with early-onset and late-onset preeclampsia had a higher proportion of more than 2 abortions or inductions, adverse previous histories, and in vitro fertilization (IVF) than the control group (P values were all < 0.05, Table 4). These results indicated that higher age, BMI, mean arterial pressure, primiparity, more than 2 abortions or inductions, adverse previous history, IVF, and twin pregnancy were all high-risk factors for the occurrence of preeclampsia during pregnancy, significantly increasing the risk of early-onset and late-onset preeclampsia, as shown in Table 4.

[0087] Table 4. Comparison of clinical characteristics of pregnant women with early-onset and late-onset preeclampsia and healthy controls in the training set and validation set

[0088]

[0089] *, **, and *** represent P values lower than 0.05, 0.01, and 0.001, respectively. Adverse previous histories include previous gestational diabetes, preeclampsia history, family preeclampsia history, chronic hypertension history, systemic lupus erythematosus, and antiphospholipid syndrome.

[0090] 2. Median multiple correction of metabolite marker concentrations

[0091] Retrieve 12 - 17 of pregnant women +6For maternal plasma samples, liquid chromatography-tandem mass spectrometry was used to measure six metabolite markers in the peripheral blood samples of each pregnant woman, including 4-hydroxyphenyllactic acid, homovanillic acid, ornithine, sarcosine, adenosine monophosphate, and 3-methylbutyrylcarnitine. The original concentration value of each metabolite in each pregnant woman was divided by the median concentration of the same metabolite in the samples analyzed in the same batch to obtain the MoM value. The product of the MoM values of four metabolites, 4-hydroxyphenyllactic acid, homovanillic acid, ornithine, and sarcosine, was calculated. The results showed that both the MoM values of the four individual metabolites and the product of the MoM values of the four metabolites were significantly upregulated in the samples of early-onset preeclampsia (P values were all < 0.05, Wilcoxon rank sum test); the product of three metabolites, ornithine, adenosine monophosphate, and 3-methylbutyrylcarnitine, was calculated. The results showed that both the individual metabolites and the product of the three metabolites were significantly upregulated in the samples of late-onset preeclampsia (P values were all < 0.05, Wilcoxon rank sum test), as shown in Figure 1 - 2 .

[0092] 3. Median multiple correction of protein marker concentration

[0093] The remaining plasma samples of pregnant women were retrieved, and the protein concentrations of IGFBP1 and PLGF in the peripheral blood samples of each pregnant woman were measured using enzyme-linked immunosorbent assay. The present invention analyzed the relationship between the protein concentrations of IGFBP1 and PLGF and the sampling gestational age in healthy control pregnant women. The expression levels of IGFBP1 and PLGF proteins were both positively correlated with the sampling gestational age of healthy pregnant women, and the correlation coefficients were 0.14 and 0.25, respectively, as shown in Figure 3 . The median concentrations of IGFBP1 and PLGF proteins at each gestational age are shown in Table 5. The MoM values of IGFBP1 and PLGF proteins for each sample were calculated, that is, the concentrations of IGFBP1 and PLGF proteins in each sample were divided by the median protein concentration at the corresponding gestational age. In the training set, validation set, and test set samples, the MoM values of IGFBP1 and PLGF proteins were significantly downregulated in the samples of early-onset and late-onset preeclampsia (P values were all lower than 0.05, Figure 4 - 5) Calculate the ratios of Homovanillic acid*Sarcosine*Hydroxyphenyllactic acid*Ornithine / IGFBP1 / PLGF (each data is in MoM value) in early-onset preeclampsia and healthy control samples, and the ratios of Ornithine*Isovelarylcarnitine*Adenosinemonophosphate / IGFBP1 / PLGF (each data is in MoM value) in late-onset preeclampsia and healthy control samples. In the training set, validation set, and test set samples, the above metabolic marker and protein marker ratios were significantly up-regulated in both early-onset and late-onset preeclampsia samples (P values were all lower than 0.05, Figure 6 )

[0094] Table 5. Median concentrations of IGFBP1 and PLGF proteins corresponding to different gestational weeks

[0095] Gestational age Median IGFBP1 protein concentration (pg / mL) Median PLGF protein concentration (pg / mL) 12W 104673.8 48.03 13W 110965.7 57.08 14W 85122.35 95.83 15W 111901.28 133.87 16W 146695.27 149.56 17W 156001.73 166.44

[0096] 4. Comparison of AUCs for screening preeclampsia with different marker combinations

[0097] For better comparison, in this example, models were constructed for each metabolic marker and protein marker alone and in different combinations, as shown in Table 6 and Figure 7 Compare the MoM values of two protein markers, IGFBP1 and PLGF, and the MoM values of six metabolic markers, Homovanillic acid, Sarcosine, Hydroxyphenyllactic acid, Ornithine, Isovelarylcarnitine, and Adenosinemonophosphate, and their different combination methods in the training set, validation set, and test set samples to screen for early-onset and late-onset preeclampsia AUCs.

[0098] The results showed that the ratio of Homovanillic acid*Sarcosine*Hydroxyphenyllactic acid*Ornithine / IGFBP1 / PLGF (each data is in MoM value) had the highest AUC for screening early-onset preeclampsia, and the AUC of this ratio for screening early-onset preeclampsia was higher than that of single metabolic or protein markers or other combinations of metabolic and protein markers. The ratio of Ornithine*Isovelarylcarnitine*Adenosine monophosphate / IGFBP1 / PLGF (each data is in MoM value) had the highest AUC for screening late-onset preeclampsia, and the AUC of this ratio for screening late-onset preeclampsia was higher than that of single metabolic or protein markers or other combinations of metabolic and protein markers. See Tables 6 and 7.

[0099] Table 6. Comparison of AUC for screening early-onset preeclampsia with different combinations of metabolic and protein markers

[0100] Early-onset preeclampsia (AUC) Training set Validation set Test set IGFBP1 0.6088 0.6594 0.6459 PLGF 0.7172 0.6963 0.6943 Homovanillic acid 0.5979 0.6786 0.5724 Sarcosine 0.7188 0.6587 0.5446 Hydroxyphenyllactic acid 0.6574 0.5705 0.6316 Ornithine 0.6147 0.6225 0.5886 Isovelarylcarnitine 0.6186 0.5804 0.6095 Adenosine monophosphate 0.6535 0.6196 0.572 Product of 4 metabolites 0.7322 0.6672 0.627 Ratio of 4 metabolites to 2 proteins 0.7979 0.7522 0.7801 Product of 4 metabolites / IGFBP1 0.741 0.6894 0.7055 Product of 4 metabolites / PLGF 0.7979 0.7262 0.7134

[0101] Note: The product of four metabolites is Homovanillic acid * Sarcosine * Hydroxyphenyllactic acid * Ornithine, and the ratio of 4 metabolites to 2 proteins is Homovanillic acid * Sarcosine * Hydroxyphenyllactic acid * Ornithine / IGFBP1 / PLGF ratio. All the above data are MoM values.

[0102] Table 7. Comparison of AUC for screening late-onset preeclampsia with different combinations of metabolic and protein markers

[0103]

[0104]

[0105] Note: The product of three metabolites is Ornithine * Isovelarylcarnitine * Adenosinemonophosphate, and the ratio of 3 metabolites to 2 proteins is Ornithine * Isovelarylcarnitine * Adenosinemonophosphate / IGFBP1 / PLGF ratio. All the above data are MoM values.

[0106] 5. Construction of random forest models for early-onset and late-onset preeclampsia

[0107] Using the data obtained in 1 - 3 above, random forest models for early - onset pre - eclampsia and late - onset pre - eclampsia were established respectively. Among them, the diagnosis of early - onset pre - eclampsia or late - onset pre - eclampsia and healthy controls were assigned 1 or 0 respectively as the dependent variable. The predictive features included the MoM value of maternal age, the MoM value of body mass index (BMI), parity, whether there were more than 2 abortions or inductions, whether there was in vitro fertilization, whether there was a history of previous gestational diabetes, a history of pre - eclampsia, a history of chronic hypertension, one or more of systemic lupus erythematosus and antiphospholipid syndrome, the MoM value of MAP, and the MoM values of biomarker predictive features including different combinations of metabolic and protein biomarker combinations. In the estimation process of the random forest, the value of m was specified, that is, m variables were randomly generated for the binary tree at the node. The selection of variables for the binary tree still satisfied the principle of minimum node impurity. In this model, m = 2. The Bootstrap resampling method was used to randomly draw k sample sets with replacement from the original dataset to form k decision trees, and the samples not drawn were used for the prediction of a single decision tree. In this model, when k = 300, the model was basically stable.

[0108] Using the same principle, 8 early - onset pre - eclampsia prediction models were constructed respectively (refer to Table 8):

[0109] Model E1: Clinical information + single combination of MAP MoM value

[0110] Model E2: Clinical information + MAP MoM value + IGFBP1 MoM value two - combination

[0111] Model E3: Clinical information + MAP MoM value + PLGF MoM value two - combination

[0112] Model E4: Clinical information + MAP MoM value + PLGF MoM value + IGFBP1 MoM value three - combination

[0113] Model E5: Clinical information + MAP MoM value + Homovanillic acid*Sarcosine*Hydroxyphenyllactic acid*Ornithine two - combination

[0114] Model E6: Clinical information + MAP MoM value + Homovanillic acid*Sarcosine*Hydroxyphenyllactic acid*Ornithine / PLGF MoM ratio two - combination

[0115] Model E7: Clinical information + MAP MoM value + Homovanillic acid*Sarcosine*Hydroxyphenyllactic acid*Ornithine / IGFBP1 MoM ratio two - combination

[0116] Model E8: Clinical information + MAP MoM value + Homovanillic acid*Sarcosine*Hydroxyphenyllactic acid*Ornithine / IGFBP1 MoM / PLGF MoM ratio dual

[0117] Similarly, using the same principle, 8 late-onset preeclampsia prediction models were constructed separately (refer to Table 8):

[0118] Model L1: Clinical information + MAP MoM value single

[0119] Model L2: Clinical information + MAP MoM value + IGFBP1 MoM value dual

[0120] Model L3: Clinical information + MAP MoM value + PLGF MoM value dual

[0121] Model L4: Clinical information + MAP MoM value + PLGF MoM value + IGFBP1 MoM value triple

[0122] Model L5: Clinical information + MAP MoM value + Ornithine*Isovelarylcarnitine*Adenosinemonophosphate dual

[0123] Model L6: Clinical information + MAP MoM value + Ornithine*Isovelarylcarnitine*Adenosinemonophosphate / PLGF MoM ratio dual

[0124] Model L7: Clinical information + MAP MoM value + Ornithine*Isovelarylcarnitine*Adenosinemonophosphate / IGFBP1 MoM ratio dual

[0125] Model L8: Clinical information + MAP MoM value + Ornithine*Isovelarylcarnitine*Adenosinemonophosphate / IGFBP1 MoM / PLGF MoM ratio dual

[0126] 6. Evaluation of the accuracy of screening early-onset and late-onset preeclampsia in the training set samples of the present invention

[0127] Predict the classification samples in the validation set and test set using the RandomForest_model composed of k decision trees. The function used is predict(RandomForest_model, data=validation), and the prediction principle is simple uniformity.

[0128] The results show that for the model E8 constructed in the present invention: the dual early-onset preeclampsia prediction model of clinical information + MAP MoM value + Homovanillic acid*Sarcosine*Hydroxyphenyllactic acid*Ornithine / IGFBP1 MoM / PLGF MoM ratio, the AUC for predicting early-onset preeclampsia in the training set can reach 0.9895, and the AUC in the validation set and test set can reach 0.9122 and 0.9063 respectively, as shown in Figure 7 A.

[0129] For the model L6 constructed in the present invention: the dual late-onset preeclampsia model of clinical information + MAP MoM + Ornithine*Isovelarylcarnitine*Adenosine monophosphate / PLGF ratio, the AUC for predicting late-onset preeclampsia in the training set reaches 0.9166, and the AUC in the validation set and test set can reach 0.8834 and 0.8529 respectively, as shown in Figure 7 B.

[0130] The Importance function calculates the importance of model variables to determine which variables contribute the most to the model. The results show that the MoM value of mean arterial pressure, the ratio of metabolic markers and protein markers, and the MoM value of BMI rank among the top three in the random forest model, indicating that the above characteristics contribute the most to the prediction of early-onset and late-onset preeclampsia, as shown in Figure 8 - 9 。

[0131] 7. Comparison of the effects of different metabolic and protein combination random forest models in screening preeclampsia

[0132] Compare the accuracy of protein models for predicting early-onset preeclampsia and late-onset preeclampsia with different metabolic and protein combinations in the validation set and the test set, and determine that the AUC, sensitivity, specificity, positive predictive value, and negative predictive value of the dual model of maternal clinical information + MAP MoM value + Homovanillic acid * Sarcosine * Hydroxyphenyllactic acid * Ornithine / IGFBP1 MoM / PLGF MoM ratio for screening early-onset preeclampsia are better than those of the single maternal clinical information + MAP MoM and other maternal clinical information + MAP MoM + metabolite or / and protein combination models; when the cut-off value is set at 0.31, the sensitivity and specificity of the early-onset preeclampsia model are 77% and 90.14% respectively, as shown in Table 8. In the late-onset preeclampsia model, the dual model of maternal clinical information + MAP MoM + Ornithine * Isovelarylcarnitine * Adenosine monophosphate / IGFBP1 / PLGF ratio performs better than the single maternal clinical information + MAP MoM and other maternal clinical information + MAP MoM + metabolite or / and protein combination models, but is not as good as the dual model of maternal clinical information + MAP MoM + Ornithine * Isovelarylcarnitine * Adenosine monophosphate / PLGF ratio. When the cut-off value is set at 0.53, the sensitivity and specificity of the dual late-onset preeclampsia model of maternal clinical information + MAP MoM + Ornithine * Isovelarylcarnitine * Adenosine monophosphate / PLGF ratio are 59.66% and 90.14% respectively, as shown in Table 8.

[0133] These results show that the model constructed by the present invention has a better performance in predicting early-onset preeclampsia at 12-17 +6 weeks of pregnancy. It should be understood that existing research data show that the complication rate of pregnant women with early-onset preeclampsia is much higher than that of late-onset preeclampsia. In particular, due to the frequent occurrence of liver and placental damage in pregnant women with early-onset preeclampsia, the incidence of adverse perinatal outcomes is also much higher than that of late-onset preeclampsia. Therefore, the significance and role of predicting early-onset preeclampsia are more important.

[0134] Table 8. Comparison of the effects of different protein combinations in the test set on predicting early-onset and late-onset preeclampsia

[0135]

[0136]

[0137] The above uses specific examples to illustrate the present invention, which is only for helping to understand the present invention and is not intended to limit the present invention. For those skilled in the art of the present invention, according to the idea of the present invention, several simple deductions, deformations or substitutions can also be made.

Claims

1. A method for constructing a multi-parameter preeclampsia risk assessment and prediction model, characterized in that, the construction method includes: obtaining clinical information, MAP, and data of metabolic markers and / or protein markers of pregnant women; constructing a prediction model for the data through a machine learning method to obtain a prediction model; the metabolic markers are selected from at least one of 4-hydroxyphenyllactic acid, homovanillic acid, ornithine, sarcosine, adenosine monophosphate, and 3-methylbutyrylcarnitine; the protein markers include IGFBP1 and / or PLGF; the pregnant women include pregnant women with early-onset preeclampsia or late-onset preeclampsia and healthy pregnant women.

2. The construction method according to claim 1, characterized in that, the machine learning method is a random forest model.

3. The construction method according to claim 1, characterized in that, after obtaining the prediction model, determine the threshold according to the receiver operating characteristic curve.

4. The construction method according to claim 1, characterized in that, the clinical information includes the age, height, weight, parity, number of deliveries, history of adverse pregnancy, history of adverse past events, and whether there is in vitro fertilization of pregnant women; the history of adverse past events includes past gestational diabetes, preeclampsia history, chronic hypertension history, systemic lupus erythematosus, and antiphospholipid syndrome; Optionally, perform median calibration on the MAP of the pregnant woman to obtain the MAP MoM value; Optionally, for the risk assessment and prediction of early-onset preeclampsia, the metabolic markers are 4-hydroxyphenyllactic acid, homovanillic acid, ornithine, and sarcosine; for the risk assessment and prediction of late-onset preeclampsia, the metabolic markers are 3-methylbutyrylcarnitine, adenosine monophosphate, and ornithine; Preferably, for the risk assessment and prediction of early-onset preeclampsia, use the product of the MoM values of the four metabolic markers / IGFBP1 MoM value / PLGF MoM value as the independent variable to construct a prediction model for early-onset preeclampsia; for the risk assessment and prediction of late-onset preeclampsia, use the product of the MoM values of the three metabolic markers / PLGF MoM value as the independent variable to construct a prediction model for late-onset preeclampsia.

5. Application of a combination of metabolic markers and protein markers related to preeclampsia in the preparation of a detection product related to preeclampsia, wherein the metabolites are selected from at least three of 4-hydroxyphenyllactic acid, homovanillic acid, ornithine, sarcosine, adenosine monophosphate, and 3-methylbutyrylcarnitine, and the protein markers are IGFBP1 and / or PLGF.

6. A marker detection kit related to preeclampsia, characterized in that, the markers include metabolic markers and protein markers, wherein the metabolic markers are selected from at least three of 4-hydroxyphenyllactic acid, homovanillic acid, ornithine, sarcosine, adenosine monophosphate, and 3-methylbutyrylcarnitine, and the protein markers are IGFBP1 and / or PLGF.

7. A multi-parameter preeclampsia risk assessment and prediction method, characterized in that includes: obtaining clinical information, MAP, and data of metabolic markers and / or protein markers of the pregnant woman to be tested; Input the data of the obtained clinical information, MAP, and metabolic markers and / or protein markers of the pregnant woman into a prediction model for processing. When the risk value is higher than the set threshold, it is determined that the pregnant woman has a high risk of preeclampsia; The metabolite is selected from at least one of 4-hydroxyphenyllactic acid, homovanillic acid, ornithine, sarcosine, adenosine monophosphate, and 3-methylbutyrylcarnitine; the protein markers include IGFBP1 and / or PLGF; The prediction model is obtained by the prediction model construction method according to any one of claims 1-4.

8. The multi-parameter preeclampsia risk assessment prediction method according to claim 7, characterized in that The clinical indicators are the age, height, weight at 12-17 +6 weeks of pregnancy, parity, number of deliveries, history of adverse pregnancy, history of adverse past events, and whether there is in vitro fertilization; the history of adverse past events includes previous gestational diabetes, preeclampsia, chronic hypertension, systemic lupus erythematosus, and antiphospholipid syndrome.

9. A prediction system for multi-parameter preeclampsia risk assessment, characterized in that comprising: a device for obtaining the data of the clinical information, MAP, and metabolic markers and / or protein markers of the pregnant woman to be tested; a device for performing prediction model processing on the data of the clinical information, MAP, and metabolic markers and / or protein markers of the pregnant woman; a device for outputting a prediction result; The metabolite is selected from at least one of 4-hydroxyphenyllactic acid, homovanillic acid, ornithine, sarcosine, adenosine monophosphate, and 3-methylbutyrylcarnitine; the protein markers include IGFBP1 and / or PLGF; The prediction model is obtained by the prediction model construction method according to any one of claims 1-4.

10. A prediction product for multi-parameter preeclampsia risk assessment, characterized in that comprising: a memory for storing a program; a processor for implementing the method according to claim 7 or 8 by executing the program stored in the memory.

11. A computer-readable storage medium, characterized in that a program is stored on the medium, and the program can be executed by a processor to implement the method according to claim 7 or 8.

12. A computer-readable storage medium, characterized in that a prediction model obtained by the construction method according to any one of claims 1-4 is stored on the medium.

Images