CAKUT children uremia prediction method and system based on machine learning

By combining machine learning models with Cox regression and XGBoost, and based on the clinical and genetic characteristics of children with CAKUT, the shortcomings in predicting uremia in children with CAKUT were addressed, achieving accurate prediction of ESRD and interpretability of the model.

CN118919065AInactive Publication Date: 2024-11-08CHILDRENS HOSPITAL OF FUDAN UNIV
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202410678914.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-05-29
Publication Date
2024-11-08
Estimated Expiration
Not applicable · inactive patent

Smart Images

  • Figure CN118919065A_ABST
    Figure CN118919065A_ABST
Patent Text Reader

Abstract

The invention discloses a CAKUT children uremia prediction method based on machine learning, and the method comprises the following steps: obtaining an input disease course and clinical features of a current patient, the clinical features comprise clinical variables and gene variables, the clinical variables comprise first diagnosis CKD staging, CAKUT subtype, extrarenal malformation, short stature, premature delivery, prenatal phenotype, gender, first diagnosis CAKUT age and family history data, and the gene variables comprise single gene mutation, copy number variation and chromosome abnormality; the obtained disease course and the clinical features of the current patient are predicted and calculated through a pre-trained prediction model, the prediction model adopts a Cox regression model, and the clinical features input into the prediction model are determined based on an XGBoost model; and outputting the probability that the current patient progresses to ESRD at the disease course moment based on the prediction calculation result. The prediction model takes the ESRD as a unique outcome index, is very objective, and is higher in interpretability compared with a composite outcome.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of chip design, and in particular relates to a CAKUT children's uremia prediction method and system based on machine learning. Background Art

[0002] Nowadays, in clinical practice, for children with congenital malformations of the kidney and urinary tract (CAKUT), we can only roughly infer whether the children will have deterioration of renal function or even progress to uremia in the later stages of the disease based on existing known risk factors for the disease, such as specific malformation types, premature birth, combined malformations of other organs outside the kidney, and specific gene mutations. However, judging the prognosis of the disease based solely on recognized risk factors is biased, because one of the characteristics of the disease is the large individual heterogeneity. Even for the same type of malformation, the progression of the disease in the later stages may not be consistent. The treatment and follow-up plans formulated by clinicians based on this prognostic judgment may not be suitable for every child. Individualized prognostic risk assessment for children with CAKUT is not yet perfect.

[0003] Currently, machine learning technology is widely used in the field of pediatric kidney disease, including: using machine learning models to analyze urodynamic data, automatically identifying waveform characteristics of bladder detrusor overactivity, and improving diagnostic efficiency and accuracy; analyzing renal ultrasound images to predict the severity of hydronephrosis and determine whether radionuclide renal examination or surgical treatment is needed; analyzing uroflowmetry data or renal ultrasound images to diagnose posterior urethral valves, predict the progression of chronic kidney disease, and the risk of requiring renal replacement therapy or clean intermittent catheterization; analyzing and integrating clinical factors and uroflowmetry data to predict the ultrasound results or surgical effects of ureterovesical junction obstruction; using clinical, laboratory or imaging data to predict the presence and extent of ureteral reflux, the risk of recurrent urinary tract infection and the effectiveness of antibiotic prevention, as well as predicting renal function recovery after pediatric kidney transplantation and the occurrence of enuresis in children.

[0004] It can be found that the application of machine learning models in the field of pediatric renal disease is very extensive, covering all aspects of diagnosis, treatment and prognosis. However, there is no research on the prognosis prediction model of uremia in children with congenital kidney and urinary tract malformations. Summary of the invention

[0005] To solve the above problems, the purpose of the present invention is to provide a method and system for predicting CAKUT pediatric uremia based on machine learning. The prediction model takes the occurrence of end-stage renal disease (ESRD), i.e. uremia, as the only outcome indicator, which is very objective and has stronger interpretability than the composite outcome.

[0006] A technical solution provided by the present invention is: a method for predicting CAKUT uremia in children based on machine learning, comprising the following steps:

[0007] Obtain the input course of disease and clinical characteristics of the current patient, the clinical characteristics including clinical variables and genetic variables, the clinical variables including first-diagnosed CKD stage, CAKUT subtype, extrarenal malformation, short stature, premature birth, prenatal phenotype, gender, age at first diagnosis of CAKUT, family history data, the genetic variables including single gene mutations, copy number variations, and chromosomal abnormalities;

[0008] The obtained course of disease and the clinical characteristics of the current patient are predicted and calculated by a pre-trained prediction model, wherein the prediction model adopts a Cox regression model, wherein the Cox regression model is expressed as: Where x represents the clinical characteristics, which will not change with time during the follow-up period, t represents the survival time, h(t,x) is the risk function, which represents the instantaneous risk rate or mortality rate of individuals with covariate x at time t, and h 0 (t) is the baseline risk function, which represents the instantaneous risk rate or mortality rate of an individual at time t when all x are 0, β 1 , β 2 ...β k is the overall regression coefficient; the clinical characteristics input into the prediction model are determined based on the XGBoost model;

[0009] The probability of the current patient progressing to ESRD at the time of the disease course is output based on the prediction calculation result.

[0010] Preferably, obtaining the input disease course and the clinical characteristics of the current patient further includes: obtaining medical history data, collecting clinical data of CAKUT children, including at least the age, gender, family history, prematurity, prenatal phenotype, initial CKD stage, CAKUT subclass, extra-renal malformations, short stature, hyperuricemia, and positive information on whether it is ESRD evaluated according to serum creatinine level and / or estimated glomerular filtration rate and / or undergoing renal replacement therapy. The estimated glomerular filtration rate is calculated using the updated Schwartz equation; wherein the extra-renal malformations include: abnormal cognitive and behavioral development, motor retardation, congenital heart disease, skeletal malformations, reproductive system malformations, central nervous system malformations, and pulmonary abnormalities; the CAKUT is divided into the following 7 major categories: PUV with renal dysplasia, solitary kidney, bilateral renal dysplasia, unilateral renal dysplasia, multicystic dysplastic kidney, horseshoe kidney, and the rest. The rest encompasses other malformations with low incidence but also affecting renal function that do not belong to the above types, including ectopic kidney and other urinary tract malformations, specifically divided into: PUJO, UVJO, VUR, duplicated system, hydronephrosis, and ectopic kidney; obtaining medical history data, collecting detailed information on genetic diagnosis, including single-gene mutations, copy number variations, and chromosomal abnormalities diagnosed by WGS or WES or CNV.

[0011] Preferably, before performing predictive calculations on the obtained disease course and the clinical characteristics of the current patient through a pre-trained Cox regression model, it further includes: defining the Cox regression model using CoxPHFitter of the lifelines library, specifying period as the survival time and ESRD(1 / 0) as the survival status; when testing the predictive ability after "yr" years, for patients with "period < yr", if they are ill during "period", the outcome event at "yr" has occurred, so "esrd = 1"; if they are still not ill at "period" years, the illness situation at "yr" years cannot be known, so these data need to be deleted; for patients with "period > yr", the outcome events have not occurred, so "esrd = 0"; for patients with "period = yr", "esrd" is the value itself, so as to achieve the purpose of dynamic testing using static sample data.

[0012] Preferably, before performing predictive calculations on the obtained disease course and the clinical characteristics of the current patient through a pre-trained Cox regression model, the Cox regression model needs to be trained, specifically including: fitting the Cox regression model through the fit() method based on a large amount of the clinical data, the detailed information on genetic diagnosis, and the positive information on whether it is ESRD; determining all baseline hazard functions and overall regression coefficients in the Cox regression model to obtain the trained Cox regression model.

[0013] Preferably, after training the Cox regression model, the method further includes: obtaining pre-configured test set data, wherein the test set data and the training data originate from different renal treatment centers; for the trained Cox regression model, using the predict_survival_function() method on the test set, inputting the clinical characteristics of the test set data as covariates, thereby obtaining a risk function; specifying t as the period of the test set data, then the value of the risk function is the survival probability p at the period, and 1-p is the ESRD positive probability.

[0014] Preferably, outputting the probability of the current patient progressing to ESRD at the time of the disease course based on the prediction calculation result further includes: predicting and calculating the acquired disease course and the current patient's clinical characteristics through a pre-trained Cox regression model to obtain the survival probability p at the time period, then the probability of ESRD positivity at the time period is 1-p.

[0015] Preferably, before training the Cox regression model, the method further includes: subdividing CAKUT subtypes into categories, wherein the classification criteria are based on imaging criteria, and when multiple CAKUT subtype clinical phenotypes are combined simultaneously, they are classified preferentially according to the severity of the impact on renal function; determining the clinical features input into the prediction model through the XGBoost model: obtaining all collected clinical feature information, and the XGBoost model calculates the average weight of the clinical feature in all trees according to the information gain or Gini coefficient obtained when each clinical feature is selected as a division node, as the importance score of the clinical feature; sorting the corresponding clinical features according to the importance score, and taking the top 10 variables, namely, first-visit CKD stage, CAKUT subtype, extrarenal malformation, short stature, premature birth, prenatal phenotype, gender, age at first diagnosis of CAKUT, family history data, genetic variables, and course of disease variables, as input variables of the model, wherein the course of disease variables are added after manual screening because they are related to time and are closely related to disease progression.

[0016] Based on the same concept, the present invention also provides a CAKUT children's uremia prediction system based on machine learning, comprising:

[0017] A data acquisition module is used to obtain the input course of disease and clinical characteristics of the current patient, wherein the clinical characteristics include clinical variables and genetic variables, wherein the clinical variables include first-visit CKD stage, CAKUT subtype, extrarenal malformation, short stature, premature birth, prenatal phenotype, gender, age at first diagnosis of CAKUT, and family history data, and the genetic variables include single gene mutations, copy number variations, and chromosomal abnormalities;

[0018] The prediction calculation module is used to predict the acquired course of disease and the clinical characteristics of the current patient through a pre-trained prediction model, wherein the prediction model adopts a Cox regression model, wherein the Cox regression model is expressed as: Where x represents the clinical characteristics, which will not change with time during the follow-up period, t represents the survival time, h(t,x) is the risk function, which represents the instantaneous risk rate or mortality rate of individuals with covariate x at time t, and h 0 (t) is the baseline risk function, which represents the instantaneous risk rate or mortality rate of an individual at time t when all x are 0, β 1 , β 2 ...β k is the overall regression coefficient; the clinical characteristics input into the prediction model are determined based on the XGBoost model; and a result output module is used to output the probability of the current patient progressing to ESRD at the moment of the disease course based on the prediction calculation result.

[0019] Based on the same concept, the present invention also provides an electronic device, characterized in that it includes: a memory, the memory is used to store a processing program; and a processor, the processor implements any one of the above-mentioned CAKUT children's uremia prediction methods based on machine learning when executing the processing program.

[0020] Based on the same concept, the present invention also provides a readable storage medium, characterized in that a processing program is stored on the readable storage medium, and when the processing program is executed by a processor, any one of the above-mentioned CAKUT children's uremia prediction methods based on machine learning is implemented.

[0021] Due to the adoption of the above technical solution, the present invention has the following advantages and positive effects compared with the prior art:

[0022] The technical solution of the present invention predicts the probability of the current patient progressing to ESRD during the course of the disease based on the course of disease information input by the doctor and the clinical characteristic information of the current patient based on a pre-trained prediction model. The clinical characteristic information includes clinical variables and gene variables. The clinical variables include the first-visit CKD stage, CAKUT subtype, extrarenal malformation, short stature, premature birth, prenatal phenotype, gender, age of first diagnosis of CAKUT, and family history data. The gene variables include single gene mutations, copy number variations, and chromosomal abnormalities. All these clinical characteristic variables are obtained from a large amount of case data according to the XGBoost model combined with manual screening. The XGBoost model is based on the information gain or Gi obtained when each feature is selected as a partition node. ni coefficient, calculate the average weight of the feature in all trees as the feature importance score, and then sort according to the importance score. The top 10 variables: first-visit CKD stage, CAKUT subtype, extrarenal malformation, short stature, premature birth, prenatal phenotype, gender, age at first diagnosis of CAKUT, family history data, and gene variables are used as input clinical characteristic variables. The manually screened course of disease variables are added to the input clinical characteristic variables because they are related to time and are closely related to disease progression. At the same time, they can also be combined with the COX regression model to realize the probability of end-stage renal disease (ESRD) or uremia at the time of the disease as the only outcome indicator, which is very objective and has stronger interpretability than the composite outcome. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The specific embodiments of the present invention are further described in detail below with reference to the accompanying drawings, wherein:

[0024] Figure 1 This is a flow chart of the CAKUT children's uremia prediction method based on machine learning of the present invention;

[0025] Figure 2 It is a schematic diagram of dynamic prediction based on COX regression model of the present invention. DETAILED DESCRIPTION

[0026] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The advantages and features of the present invention will become more apparent from the following description and claims. It should be noted that the drawings are all in a very simplified form and use non-precise ratios, which are only used to conveniently and clearly assist in explaining the purpose of the embodiments of the present invention.

[0027] It should be noted that all directional indications in the embodiments of the present invention (such as up, down, left, right, front, back, etc.) are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0028] First embodiment

[0029] like Figure 1 As shown, this embodiment provides a CAKUT children uremia prediction method based on machine learning, comprising the following steps:

[0030] S100: Obtain the input course of disease and clinical characteristics of the current patient, wherein the clinical characteristics include clinical variables and genetic variables, wherein the clinical variables include first-diagnosed CKD stage, CAKUT subtype, extrarenal malformation, short stature, premature birth, prenatal phenotype, gender, age at first diagnosis of CAKUT, and family history data, and the genetic variables include single gene mutation, copy number variation, and chromosome abnormality;

[0031] S200: Predicting and calculating the acquired course of disease and the clinical characteristics of the current patient using a pre-trained prediction model, wherein the prediction model adopts a Cox regression model, wherein the Cox regression model is expressed as: Where x represents the clinical characteristics, which will not change with time during the follow-up period, t represents the survival time, h(t,x) is the risk function, which represents the instantaneous risk rate or mortality rate of individuals with covariate x at time t, and h 0 (t) is the baseline risk function, which represents the instantaneous risk rate or mortality rate of an individual at time t when all x are 0, β 1 , β 2 ...β k is the overall regression coefficient; the clinical characteristics input into the prediction model are determined based on the XGBoost model;

[0032] S300: Outputting the probability of the current patient progressing to ESRD at the disease course moment based on the prediction calculation result.

[0033] The technical solution of this embodiment predicts the probability of the current patient progressing to ESRD during the course of the disease based on the pre-trained prediction model based on the course of disease information input by the doctor and the clinical characteristic information of the current patient. The clinical characteristic information includes clinical variables and gene variables. The clinical variables include the first-visit CKD stage, CAKUT subtype, extrarenal malformation, short stature, premature birth, prenatal phenotype, gender, age at first diagnosis of CAKUT, and family history data. The gene variables include single gene mutations, copy number variations, and chromosomal abnormalities. All these variables are obtained from a large amount of case data according to the XGBoost model combined with manual screening. The XGBoost model calculates the average weight of each feature in all trees based on the information gain or Gini coefficient obtained when each feature is selected as a partition node, and uses it as the feature importance score. The model then sorts the features according to the importance score, and uses the top 10 variables: first-visit CKD stage, CAKUT subtype, extrarenal malformation, short stature, premature birth, prenatal phenotype, gender, age at first CAKUT diagnosis, family history data, and gene variables as input clinical characteristic variables. The manually screened course of disease variables are also added to the input clinical characteristic variables because they are time-related and closely related to disease progression. They can also be combined with the COX regression model to achieve the probability of end-stage renal disease (ESRD), i.e., uremia, occurring at any time during the course of the disease as the only outcome indicator, which is very objective and has stronger interpretability than the composite outcome.

[0034] Preferably, obtaining the input course of disease and clinical characteristics of the current patient further comprises: obtaining medical history information, collecting clinical data of CAKUT children, including at least the age of first diagnosis of CAKUT, gender, family history, premature birth, prenatal phenotype, first diagnosis CKD stage, CAKUT subtype, extrarenal malformation, short stature, hyperuricemia, and whether the patient is ESRD positive according to serum creatinine level and / or estimated glomerular filtration rate and / or renal replacement therapy, and the estimated glomerular filtration rate is calculated using the updated Schwartz equation; wherein the extrarenal malformation includes: abnormal cognitive behavioral development, motor delay, congenital heart disease, skeletal malformation, birth Reproductive system malformations, central nervous system malformations, and lung abnormalities; the CAKUT is divided into the following 7 categories: PUV with renal dysplasia, solitary kidney, bilateral renal dysplasia, unilateral renal dysplasia, polycystic renal dysplasia, horseshoe kidney, and others. The others include other malformations that have a low incidence but can also affect renal function and are not included in the above types, including ectopic kidney and other urinary tract malformations, which are specifically divided into six categories: PUJO, UVJO, VUR, duplication system, hydronephrosis, and ectopic kidney; obtain medical history data and collect detailed information on genetic diagnosis, including single gene mutations, copy number variations, and chromosomal abnormalities diagnosed by WGS or WES or CNV.

[0035] The technical solution of this embodiment takes into account the complexity of kidney diseases, strictly classifies the variables of the model, which is conducive to improving the interpretability and scientific nature of the model, thereby enhancing the readability and prediction accuracy of the model.

[0036] Preferably, before predicting and calculating the disease course and the clinical characteristics of the current patient through a pre-trained Cox regression model, it further includes: defining the Cox regression model using CoxPHFitter of the lifelines library, specifying period as the survival time, and ESRD(1 / 0) as the survival status; when testing the prediction ability after "yr" years, for patients with "period < yr", if they get sick during "period", the outcome event at "yr" has occurred, so "esrd = 1"; if they are still not sick at "period" years, the disease situation at "yr" years cannot be known, so these data need to be deleted; for patients with "period > yr", the outcome events have not occurred, so "esrd = 0"; for patients with "period = yr", "esrd" is its own value, so as to achieve the purpose of dynamic testing using static sample data.

[0037] See Figure 2 , which is a schematic diagram of dynamic prediction based on the Cox regression model. In this embodiment, the Cox regression model is defined using CoxPHFitter of the lifelines library, and then, on this basis, period is specified as the survival time, and ESRD(1 / 0) is specified as the survival status. Considering the application scenario of this embodiment, for patients with "period < yr", if they get sick during "period", the outcome event at "yr" has occurred, so "esrd = 1"; if they are still not sick at "period" years, the disease situation at "yr" years cannot be known, so these data need to be deleted; for patients with "period > yr", the outcome events have not occurred, so "esrd = 0"; for patients with "period = yr", "esrd" is its own value, thus realizing the transformation of the model and achieving the purpose of dynamic testing using static sample data.

[0038] Preferably, before predicting and calculating the disease course and the clinical characteristics of the current patient through a pre-trained Cox regression model, the Cox regression model needs to be trained. Specifically, it includes: fitting the Cox regression model through the fit() method based on a large amount of the clinical data, the detailed information of gene diagnosis, and whether it is ESRD positive information; determining all baseline risk functions and overall regression coefficients in the Cox regression model to obtain the trained Cox regression model.

[0039] In the technical solution of this embodiment, a solution is disclosed for the prediction model to predict and calculate the current patient based on the input disease course and the clinical characteristics of the current patient. The COX regression model is trained based on a large amount of data provided to obtain all baseline risk functions and overall regression coefficients, thereby determining the Cox regression model.

[0040] Preferably, after training the Cox regression model, the method further includes: obtaining pre-configured test set data, wherein the test set data and the training data originate from different renal treatment centers; for the trained Cox regression model, using the predict_survival_function() method on the test set, inputting the clinical characteristics of the test set data as covariates, thereby obtaining a risk function; specifying t as the period of the test set data, then the value of the risk function is the survival probability p at the period, and 1-p is the ESRD positive probability.

[0041] The technical solution of this embodiment predicts the input clinical feature data based on the trained Cox regression model. In order to improve credibility and verifiability, the data comes from 15 pediatric kidney disease research centers in 13 provinces and cities across the country, including Henan, Wuhan, Chongqing, Guangzhou, Hunan, Shenzhen, Xiamen, Zhejiang, and Nanjing.

[0042] In the technical solution of this embodiment, the data is pre-divided into a training set, a test set, and an external independent test set. In practical applications, the data of the National Children's Medical Center, which has the largest number of CAKUT children in China, is used as the source data for the training set and the test set, and 15 foreign kidney disease centers are used as the external independent test set. The internal and external verifications are sufficient, and the results are highly credible. The model is trained on the training set, the parameters are optimized on the internal test set, and the external independent test set is used to obtain the final results of the model.

[0043] Preferably, outputting the probability of the current patient progressing to ESRD at the time of the disease course based on the prediction calculation result further includes: predicting and calculating the acquired disease course and the current patient's clinical characteristics through a pre-trained Cox regression model to obtain the survival probability p at the time period, then the probability of ESRD positivity at the time period is 1-p.

[0044] More specifically, in contrast to the survival probability p, 1-p is the probability of ESRD positivity at the time period, that is, the probability that the current patient will progress to ESRD at the time period of the entered course of disease.

[0045] Preferably, before training the Cox regression model, the following steps are also included:

[0046] CAKUT subtypes are subdivided into categories based on imaging criteria. When multiple CAKUT subtypes are combined, they are prioritized according to the severity of the impact on renal function.

[0047] The clinical features input into the prediction model are determined by the XGBoost model:

[0048] All collected clinical feature information is obtained. The XGBoost model calculates the average weight of each clinical feature in all trees based on the information gain or Gini coefficient obtained when each clinical feature is selected as a partition node, which is used as the importance score of the clinical feature.

[0049] The corresponding clinical characteristics were ranked according to their importance scores, and the top 10 variables, including first-visit CKD stage, CAKUT subtype, extrarenal malformation, short stature, premature birth, prenatal phenotype, gender, age at first CAKUT diagnosis, family history data, genetic variables, and disease course variables, were used as input variables for the model. The disease course variable was added after manual screening because it is related to time and is closely associated with disease progression.

[0050] The technical solution of this embodiment further takes into account the complexity of kidney disease, and always prioritizes classifying the patient's CAKUT subtype into the CAKUT subtype classification that has a more serious impact on kidney function, so that the data standard of the variable can be more unified, which is conducive to improving the readability and accuracy of the model. Through the machine learning model XGBoost and combined with manual screening, the most influential clinical feature variables are screened out, which is beneficial to the accuracy of the final prediction results.

[0051] Second embodiment

[0052] Based on the same concept, the present invention also provides a CAKUT children's uremia prediction system based on machine learning, comprising:

[0053] A data acquisition module is used to obtain the input course of disease and clinical characteristics of the current patient, wherein the clinical characteristics include clinical variables and genetic variables, wherein the clinical variables include first-visit CKD stage, CAKUT subtype, extrarenal malformation, short stature, premature birth, prenatal phenotype, gender, age at first diagnosis of CAKUT, and family history data, and the genetic variables include single gene mutations, copy number variations, and chromosomal abnormalities;

[0054] The prediction calculation module is used to predict the acquired course of disease and the clinical characteristics of the current patient through a pre-trained prediction model, wherein the prediction model adopts a Cox regression model, wherein the Cox regression model is expressed as:

[0055] Where x represents the clinical characteristics, which will not change with time during the follow-up period, t represents the survival time, h(t,x) is the risk function, which represents the instantaneous risk rate or mortality rate of individuals with covariate x at time t, and h 0 (t) is the baseline risk function, which represents the instantaneous risk rate or mortality rate of an individual at time t when all x are 0, β 1 , β 2 ...β k is the overall regression coefficient; the clinical characteristics input into the prediction model are determined based on the XGBoost model;

[0056] A result output module is used to output the probability of the current patient progressing to ESRD at the disease course moment based on the prediction calculation result.

[0057] The technical solution of this embodiment predicts the probability of the current patient progressing to ESRD during the course of the disease based on the pre-trained prediction model based on the course of disease information input by the doctor and the clinical characteristic information of the current patient. The clinical characteristic information includes clinical variables and gene variables. The clinical variables include the first-visit CKD stage, CAKUT subtype, extrarenal malformation, short stature, premature birth, prenatal phenotype, gender, age at first diagnosis of CAKUT, and family history data. The gene variables include single gene mutations, copy number variations, and chromosomal abnormalities. All these variables are obtained from a large amount of case data according to the XGBoost model combined with manual screening. The XGBoost model calculates the average weight of each feature in all trees based on the information gain or Gini coefficient obtained when each feature is selected as a partition node, and uses it as the feature importance score. The model then sorts the features according to the importance score, and uses the top 10 variables: first-visit CKD stage, CAKUT subtype, extrarenal malformation, short stature, premature birth, prenatal phenotype, gender, age at first CAKUT diagnosis, family history data, and gene variables as input clinical characteristic variables. The manually screened course of disease variables are also added to the input clinical characteristic variables because they are time-related and closely related to disease progression. They can also be combined with the COX regression model to achieve the probability of end-stage renal disease (ESRD), i.e., uremia, occurring at any time during the course of the disease as the only outcome indicator, which is very objective and has stronger interpretability than the composite outcome.

[0058] Based on the same concept, the present invention also provides an electronic device, comprising: a memory, the memory being used to store a processing program; and a processor, the processor implementing any one of the above-mentioned CAKUT children's uremia prediction methods based on machine learning when executing the processing program.

[0059] Based on the same concept, the present invention also provides a readable storage medium, on which a processing program is stored. When the processing program is executed by a processor, any one of the above-mentioned CAKUT children's uremia prediction methods based on machine learning is implemented.

[0060] If the CAKUT child uremia prediction method based on machine learning is implemented in the form of program instructions and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this embodiment is essentially or part of the contribution to the prior art or all or part of the technical solution can be embodied in the form of software, and the computer software is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present disclosure. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., and other media that can store program codes.

[0061] Those skilled in the art can clearly understand that, for the sake of convenience and brevity in description, the identification content specifically executed by the above-described system and device can refer to the corresponding process in the aforementioned method embodiment.

[0062] The embodiments of the present invention are described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above embodiments. Even if various changes are made to the present invention, if these changes fall within the scope of the claims of the present invention and their equivalents, they still fall within the protection scope of the present invention.

Claims

1. A method for predicting CAKUT uremia in children based on machine learning, characterized in that: It includes the following steps: Obtain the input disease course and the clinical characteristics of the current patient. The clinical characteristics include clinical variables and gene variables. The clinical variables include the initial diagnosis of CKD stage, CAKUT subtype, extra-renal malformations, short stature, prematurity, prenatal phenotype, gender, age at first diagnosis of CAKUT, family history data. The gene variables include single gene mutations, copy number variations, and chromosomal abnormalities; The obtained course of disease and the clinical characteristics of the current patient are predicted and calculated by a pre-trained prediction model, wherein the prediction model adopts a Cox regression model, wherein the Cox regression model is expressed as: Where x represents the clinical characteristics, which will not change with time during the follow-up period, t represents the survival time, h(t,x) is the risk function, which represents the instantaneous risk rate or mortality rate of the individual with covariate x at time t, h0(t) is the baseline risk function, which represents the instantaneous risk rate or mortality rate of the individual at time t when all x are 0, β1, β2...β k is the overall regression coefficient; the clinical characteristics input into the prediction model are determined based on the XGBoost model; Output the probability that the current patient progresses to ESRD at the moment of the disease course based on the prediction calculation result.

2. The CAKUT children's uremia prediction method based on machine learning according to claim 1 is characterized in that: The obtaining of the input disease course and the clinical characteristics of the current patient further includes: Obtain medical history data, collect the clinical data of CAKUT children, including at least the age, gender, family history, prematurity, prenatal phenotype, initial diagnosis of CKD stage, CAKUT subclass, extra-renal malformations, short stature, hyperuricemia at the first diagnosis of CAKUT, and positive information on whether it is ESRD evaluated according to serum creatinine level and / or estimated glomerular filtration rate and / or undergoing renal replacement therapy. The estimated glomerular filtration rate is calculated using the updated Schwartz equation; where the extra-renal malformations include: abnormal cognitive and behavioral development, motor retardation, congenital heart disease, skeletal malformations, reproductive system malformations, central nervous system malformations, pulmonary abnormalities; CAKUT is divided into the following 7 major categories: PUV with renal dysplasia, solitary kidney, bilateral renal dysplasia, unilateral renal dysplasia, multicystic dysplastic kidney, horseshoe kidney, others. The others include other malformations with low incidence but also affecting renal function and not belonging to the above types, including ectopic kidney and other urinary tract malformations, specifically divided into: PUJO, UVJO, VUR, duplication system, hydronephrosis, ectopic kidney six major categories; Obtain medical history data, collect the detailed information of gene diagnosis, including single gene mutations, copy number variations, and chromosomal abnormalities diagnosed by WGS or WES or CNV.

3. The CAKUT children's uremia prediction method based on machine learning according to claim 1 is characterized in that: Before predicting and calculating the obtained disease course and the clinical characteristics of the current patient through a pre-trained Cox regression model, it also includes: Define the Cox regression model using CoxPHFitter of the lifelines library, specifying period as the survival time and ESRD(1 / 0) as the survival status; When testing the predictive ability after "yr" years, for patients with "period < yr", if they are ill during the "period", the outcome event at "yr" has occurred, so "esrd = 1"; If they are not ill at the "period" year, the illness situation at "yr" cannot be known, so these data need to be deleted; For patients with "period > yr", the outcome events have not occurred, so "esrd = 0"; for patients with "period = yr", "esrd" is its own value, so as to achieve the purpose of dynamic testing using static sample data.

4. The CAKUT children's uremia prediction method based on machine learning according to claim 3 is characterized in that: Before predicting and calculating the obtained disease course and the clinical characteristics of the current patient through a pre-trained Cox regression model, the Cox regression model needs to be trained, specifically including: Fitting a Cox regression model using the fit() method based on a large amount of clinical data, detailed information on the genetic diagnosis, and whether the patient is ESRD positive; All baseline risk functions and overall regression coefficients in the Cox regression model are determined to obtain a trained Cox regression model.

5. The CAKUT children's uremia prediction method based on machine learning according to claim 4 is characterized in that: After training the Cox regression model, the following steps are further performed: Obtaining pre-configured test set data, wherein the test set data and training data originate from different renal treatment centers; For the trained Cox regression model, use the predict_survival_function() method on the test set, input the clinical characteristics of the test set data as covariates, and thus obtain the risk function; Specify t as the period of the test set data, then the value of the risk function is the survival probability p at the period time, and 1-p is the ESRD positive probability.

6. The CAKUT children's uremia prediction method based on machine learning according to claim 1 is characterized in that: Outputting the probability of the current patient progressing to ESRD at the disease course time based on the prediction calculation result further includes: The pre-trained Cox regression model is used to predict and calculate the acquired disease course and the current clinical characteristics of the patient to obtain the survival probability p at the period time. The ESRD-positive probability at the period time is 1-p.

7. The CAKUT children's uremia prediction method based on machine learning according to claim 2 is characterized in that: Also include before training the Cox regression model: CAKUT subtypes are subdivided into categories based on imaging criteria. When multiple CAKUT subtypes are combined, they are prioritized according to the severity of the impact on renal function. Determine the clinical features input into the prediction model by XGBoost model: All collected clinical feature information is obtained. The XGBoost model calculates the average weight of each clinical feature in all trees based on the information gain or Gini coefficient obtained when each clinical feature is selected as a partition node, which is used as the importance score of the clinical feature. The corresponding clinical characteristics were ranked according to their importance scores, and the top 10 variables, including first-visit CKD stage, CAKUT subtype, extrarenal malformation, short stature, premature birth, prenatal phenotype, gender, age at first CAKUT diagnosis, family history data, genetic variables, and disease course variables, were used as input variables for the model. The disease course variable was added after manual screening because it is related to time and is closely associated with disease progression.

8. A CAKUT children's uremia prediction system based on machine learning, characterized in that: include: A data acquisition module is used to obtain the input course of disease and clinical characteristics of the current patient, wherein the clinical characteristics include clinical variables and genetic variables, wherein the clinical variables include first-visit CKD stage, CAKUT subtype, extrarenal malformation, short stature, premature birth, prenatal phenotype, gender, age at first diagnosis of CAKUT, and family history data, and the genetic variables include single gene mutations, copy number variations, and chromosomal abnormalities; The prediction calculation module is used to predict the acquired course of disease and the clinical characteristics of the current patient through a pre-trained prediction model, wherein the prediction model adopts a Cox regression model, wherein the Cox regression model is expressed as: Where x represents the clinical characteristics, which will not change with time during the follow-up period, t represents the survival time, h(t,x) is the risk function, which represents the instantaneous risk rate or mortality rate of the individual with covariate x at time t, h0(t) is the baseline risk function, which represents the instantaneous risk rate or mortality rate of the individual at time t when all x are 0, β1, β2...β k is the overall regression coefficient; the clinical characteristics input into the prediction model are determined based on the XGBoost model; A result output module is used to output the probability of the current patient progressing to ESRD at the disease course moment based on the prediction calculation result.

9. An electronic device, characterized in that: include: A memory, the memory being used to store a processing program; A processor, wherein when executing the processing program, the processor implements the CAKUT children's uremia prediction method based on machine learning as described in any one of claims 1 to 7.

10. A readable storage medium, characterized in that: The readable storage medium stores a processing program, and when the processing program is executed by the processor, the CAKUT child uremia prediction method based on machine learning as described in any one of claims 1 to 7 is implemented.

Citation Information

Patent Citations

  • ACS risk prediction method and device and storage medium

    CN115512845A

  • Developmental dysplasia hip joint harness treatment effect prediction method and device

    CN117352125A

  • Systems and methods for predicting renal dysfunction

    CN118019494A