阅读说明：本技术 一种重症脓毒症患者一过性和持续性aki的预测模型及其构建方法 (Prediction model of transient and persistent AKI of severe sepsis patient and construction method thereof ) 是由 段绍斌 罗晓琴 晏萍 邓颖豪 刘茜 康怡昕 张宁雅 王梅 伍婷 吴皙 王鸿燊 于 2021-07-21 设计创作，主要内容包括：本发明提供了一种重症脓毒症患者一过性和持续性AKI的预测模型及其构建方法,所述预测模型包括连续性变量和分类变量；所述预测模型还包括预测概率表达式P的截断值为0.6316,用于界定一过性和持续性AKI的风险。所述构建方法包括步骤S1收集临床数据、步骤S2选择排名前20位的特征用于进一步特征筛选、步骤S3筛选14位变量、步骤S4使用logistic回归构建重症脓毒症患者一过性和持续性AKI的预测模型以及步骤S5构建的预测模型。本发明为临床医师提供了一个简单、实用的一过性和持续性AKI风险预测工具。(The invention provides a predictive model of transient and persistent AKI of a patient with severe sepsis and a construction method thereof, wherein the predictive model comprises a continuous variable and a classification variable; the prediction model further comprises a prediction probability expression)

1. A predictive model of transient and persistent AKI in patients with severe sepsis, comprising the following continuous variables:

age, Age in years; partial pressure of arterial blood oxygen PaO₂In mm of mercury; arterial blood partial pressure of carbon dioxide PaCO₂In mm of mercury; anion gap, in millimoles per liter; lactate in millimoles per moleLifting; international normalized ratio INR, unitless; partial thromboplastin time, Ptt, in seconds;

the predictive model includes the following categorical variables:

diabetes mellitus, Congestive heart failure, Chronic kidney disease, Mechanical ventilation required and renal replacement therapy RRT induction started, wherein the five variables are selected to be present or absent according to the actual condition of a patient, the corresponding value is 1 when the five variables are selected, and the corresponding value is 0 when the five variables are not selected;

the predictive model also includes the following categorical variables:

the AKI urine volume stage UO stage and AKI creatinine stage SCR stage, the two variables are selected from stage 0, stage1, stage2 or stage 3 according to the actual condition of the patient, when any stage is selected, the corresponding value of the stage is 1, and the corresponding values of the other stages are 0;

the prediction model comprises a prediction probability expression (1), wherein the prediction probability expression (1) is specifically as follows:

wherein, A is 0.006161 × Age-0.002077 × PaO₂+0.009274×PaCO₂+0.026074 × Anion gap +0.020870 × Lactate +0.068955 × 0INR +0.002655 × 1Ptt +0.259688 × 2Diabetes mellitus +0.320812 × 3 constructive heart failure +0.147493 × 4Chronic kit failure +0.270744 × 5Mechanical discovery +1.361824 × 6RRT indication +0 × UO Stage 0+0.494265 × UO Stage 1+1.279487 × UO Stage 2+2.169034 × UO Stage 3+0 × SCr Stage 0+0.856714 × SCr Stage 1+2.333892 × SCr Stage 2+2.585134 × SCr Stage 3-2.877650; p represents the probability score, with a cutoff value of 0.6316, when the patient probability score is below the cutoff value, a low risk patient is at risk for having a condition of transient AKI and recovering within 48 hours of AKI occurring; when the patient probability score is above the cutoff value, a high risk patient, the patient's condition progresses to persistent AKI 48 hours after AKI occurs.

2. A method for constructing a predictive model of the transient and persistent AKI of a severe sepsis patient according to claim 1, comprising the steps of:

step S1, collecting clinical data of sepsis patients who have AKI within 48 hours after entering an intensive care unit from an intensive care database;

step S2, obtaining a feature variable importance ranking for distinguishing transitional and persistent AKI after processing clinical data by using an extreme gradient boosting Xgboost algorithm, and selecting the feature with the top 20 bits of ranking for further feature screening;

step S3, the 20-bit features obtained in the step S2 are further screened by using the least absolute contraction and selection operator LASSO algorithm to obtain 14-bit variables, namely Age and arterial blood oxygen partial pressure PaO₂Arterial blood partial pressure of carbon dioxide PaCO₂Anion gap, Lactate, international normalized ratio INR, partial thromboplastin time Ptt, Diabetes mellitus Diabetes mellitis, Congestive heart failure concestic heart failure, Chronic renal disease Chronic kidney disease, Mechanical ventilation needed, initiation of renal replacement therapy RRT indication, AKI urine volume staging UO stage and AKI creatinine staging SCr stage;

step S4, selecting the 14-bit variable obtained in the step S3, and constructing a prediction model of the transient and persistent AKI of the patient with severe sepsis by using logistic regression;

and step S5, converting the prediction model into a prediction probability expression (1) by using R software according to the prediction model constructed in the step S4, and calculating the risk of the severe sepsis patient to progress to the transient and persistent AKI according to the prediction probability expression (1).

3. The method of constructing a predictive model of transient and persistent AKI in severe sepsis patients according to claim 2, wherein in step S1 the clinical data includes demographic characteristics, complications, vital signs, laboratory examinations, treatment, and AKI staging.

4. The method for constructing a predictive model of the transient and persistent AKI of a severe sepsis patient according to claim 2, further comprising selecting the hyper-parameters of the extreme gradient boost Xgboost algorithm by using 5-fold cross validation in step S2, wherein the final hyper-parameters are as follows: the learning rate learning _ rate is 0.01, the number of trees n _ estimators is 650, the maximum depth max _ depth of the trees is 3, the minimum weight min _ child _ weight of the child nodes is 3, the sampling rate subsample is 0.6, the characteristic sampling rate colsample _ byte of each tree is 0.7, and the minimum loss split gamma is 1.

5. The method for constructing a predictive model of the transient and persistent AKI of a patient with severe sepsis according to claim 4, wherein the first 20 th place in step S2 is characterized by the AKI urine volume stage UO stage1, AKI urine volume stage UO stage2, AKI urine volume stage UO stage 3, AKI creatinine stage SCR stage1, AKI creatinine stage SCR stage2, AKI creatinine stage SCR stage 3, Congestive heart failure coherent heart failure, Anion gap, and partial arterial blood pressure and partial arterial blood oxygen pressure PAO stage₂Initiating renal replacement therapy RRT induction, Diabetes mellitus, Lactate, Chronic renal disease Chronic Kidney disease, ICU type, International normalized ratio INR, arterial blood partial pressure of carbon dioxide, PaCO₂Arterial blood pH, Mechanical ventilation required, partial thromboplastin time Ptt and Age.

6. The method of constructing a predictive model of the transient and persistent AKI in severe sepsis patients according to any of claims 2-5, further comprising processing the least absolute contraction and selection operator LASSO algorithm using cross-validation to select a cross-validation determination parameter λ to obtain a 14-bit variable, wherein λ has a value of 0.009601, in step S3.

Technical Field

The invention relates to the technical field of kidney disease and intensive care, in particular to a model for predicting transient and persistent AKI of a patient with severe sepsis and a construction method thereof.

Background

Acute Kidney Injury (AKI) is one of the common and serious complications of severe sepsis patients. Studies have shown that approximately 60% of severe sepsis patients have incorporated AKI. However, the occurrence of AKI is associated with an increased risk of short and long term adverse outcomes in patients, such as death, cardiovascular complications, and chronic renal insufficiency, among others. Since both the duration of AKI and the recovery of AKI from renal function affect the prognosis of AKI patients, the 16 th acute disease quality initiative conference suggests that AKI is classified as transient AKI and persistent AKI based on whether it persists for more than 48 hours after AKI has occurred. Persistent AKI sepsis patients respond to a stronger inflammatory and procoagulant response, more severe loss of vascular integrity, and a higher mortality rate and risk of long-term poor prognosis than transient AKI sepsis patients. Early prediction of the onset of persistent AKI helps clinicians to stratify patients on risk, to select individualized treatment strategies such as making rational fluid management regimens to prevent the onset of harmful fluid overload, to assess the need for renal replacement therapy and to determine optimal initiation timing, and to dynamically monitor patients for volume status, renal function, and the occurrence of AKI-related complications. Early studies attempted to differentiate transient and persistent AKI patients early using laboratory indicators such as sodium excretion fraction, urea excretion fraction, and urine urea creatinine ratio, as well as imaging methods such as renal doppler ultrasound, but the differentiation was poor. Recent studies have evaluated the role of renal function and renal injury-associated biomarkers in early prediction of persistent AKI, with results showing that most markers are less potent and few are still validated in further large-scale clinical studies. There is currently no reliable method to early differentiate patients with transient AKI from persistent AKI sepsis.

In view of the above, there is an urgent need for a model for predicting transient and persistent AKI of patients with severe sepsis and a method for constructing the same to solve the problems in the prior art.

Disclosure of Invention

The invention aims to provide a model for predicting transient and persistent AKI of a patient with severe sepsis, and the specific technical scheme is as follows:

a predictive model of transient and persistent AKI in patients with severe sepsis, comprising the following continuous variables:

age, Age in years; partial arterial blood oxygen pressure PaO2 in mm hg; arterial blood partial pressure of carbon dioxide PaCO2 in millimeters of mercury; anion gap, in millimoles per liter; lactate in millimoles per liter; international normalized ratio INR, unitless; partial thromboplastin time, Ptt, in seconds;

the predictive model includes the following categorical variables:

diabetes mellitus, Congestive heart failure, Chronic kidney disease, Mechanical ventilation required and renal replacement therapy RRT induction started, wherein the five variables are selected to be present or absent according to the actual condition of a patient, the corresponding value is 1 when the five variables are selected, and the corresponding value is 0 when the five variables are not selected;

the predictive model also includes the following categorical variables:

the AKI urine volume stage UO stage and AKI creatinine stage SCR stage, the two variables are selected from stage 0, stage1, stage2 or stage 3 according to the actual condition of the patient, when any stage is selected, the corresponding value of the stage is 1, and the corresponding values of the other stages are 0;

the prediction model comprises a prediction probability expression (1), wherein the prediction probability expression (1) is specifically as follows:

wherein A is 0.006161 × Age-0.002077 × PaO2+0.009274 × PaCO2+0.026074 × Anion gap +0.020870 × Lactate +0.068955 × INR +0.002655 × Ptt +0.259688 × Diabetes mellitus wells +0.320812 × Congettive heart failure +0.147493 × Chronic kit failure +0.270744 × Mechanical transfer +1.361824 × RRT indication +0 × UO Stage 0+0.494265 × UO Stage 1+1.279487 × UO Stage 2+2.169034 × UO Stage 3+0 × SCR Stage 0+0.856714 × SCr Stage 1+2.333892 × SCr Stage 2+2.585134 × SCr Stage 3-2.877650; p represents the probability score, with a cutoff value of 0.6316, when the patient probability score is below the cutoff value, a low risk patient is at risk for having a condition of transient AKI and recovering within 48 hours of AKI occurring; when the patient probability score is above the cutoff value, a high risk patient, the patient's condition progresses to persistent AKI 48 hours after AKI occurs.

The second purpose of the invention is to provide a method for constructing a predictive model of transient and persistent AKI of a patient with severe sepsis, which comprises the following specific technical scheme:

a method for constructing a prediction model of transient and persistent AKI of a patient with severe sepsis comprises the following steps:

step S1, collecting clinical data of sepsis patients who have AKI within 48 hours after entering an intensive care unit from an intensive care database;

step S2, obtaining a feature variable importance ranking for distinguishing transitional and persistent AKI after processing clinical data by using an extreme gradient boosting Xgboost algorithm, and selecting the feature with the top 20 bits of ranking for further feature screening;

step S3, using a least absolute contraction and selection operator LASSO algorithm to further screen the 20-bit characteristics obtained in the step S2 to obtain 14-bit variables, namely Age, arterial blood oxygen partial pressure PaO2, arterial blood carbon dioxide partial pressure PaCO2, Anion gap, Lactate, international normalized ratio INR, partial thromboplastin time Ptt, Diabetes mellitus mellitis, Congestive heart failure constructive failure, Chronic renal disease Chronic kit disease, Mechanical ventilation required, renal replacement therapy starting RRT indication, AKI urine volume fraction UO stage and AKI creatinine fraction SCR stage;

step S4, selecting the 14-bit variable obtained in the step S3, and constructing a prediction model of the transient and persistent AKI of the patient with severe sepsis by using logistic regression;

and step S5, converting the prediction model into a prediction probability expression (1) by using R software according to the prediction model constructed in the step S4, and calculating the risk of the severe sepsis patient to progress to the transient and persistent AKI according to the prediction probability expression (1).

Preferably, in step S1, the clinical data includes demographic characteristics, complications, vital signs, laboratory examinations, treatments, and AKI staging.

Preferably, step S2 further includes selecting a hyper-parameter using a 5-fold cross validation epipolar gradient boost Xgboost algorithm, and the resulting hyper-parameter is as follows: the learning rate learning _ rate is 0.01, the number of trees n _ estimators is 650, the maximum depth max _ depth of the trees is 3, the minimum weight min _ child _ weight of the child nodes is 3, the sampling rate subsample is 0.6, the characteristic sampling rate colsample _ byte of each tree is 0.7, and the minimum loss split gamma is 1.

Preferably, the top 20 ranked in step S2 are characterized by AKI urine volume fraction UO stage1, AKI urine volume fraction UO stage2, AKI urine volume fraction UO stage 3, AKI creatinine fraction SCR stage1, AKI creatinine fraction SCR stage2, AKI creatinine fraction SCR stage 3, Congestive heart failure collective failure, Anion gap, arterial blood oxygen PaO2, onset renal replacement therapy RRT indication, Diabetes mellitus markers, Lactate, Chronic renal disease Chronic kit disease, ICU type, international normalized ratio INR, arterial blood carbon dioxide partial pressure PaCO2, arterial blood pH, Mechanical ventilation required Ptilation, partial clotting enzyme time and Age, respectively. And setting a dummy variable when processing the AKI urine volume stage and the AKI creatinine stage of the multi-classification variable, and selecting one of the classifications as a dummy variable reference. The invention selects the UO stage 0 of AKI urine volume stage as reference, and 1 is the three categories of the UO stage1 stage, the UO stage2 stage and the UO stage 3 stage of the AKI urine volume stage. Therefore, 1 variable corresponds to 3 characteristics in the stage of AKI urine volume. Similarly, the invention selects the SCr stage 0 of the AKI creatinine stage as a reference, and the three classifications of the SCr stage1, SCr stage2 and SCr stage 3 of the AKI creatinine stage are all 1 characteristic. Thus, 1 variable in the AKI creatinine stage corresponds to 3 features.

Preferably, the step S3 further includes processing the least absolute contraction and selection operator LASSO algorithm using cross-validation, and selecting a cross-validation determination parameter λ to obtain a 14-bit variable, where λ has a value of 0.009601.

The technical scheme of the invention has the following beneficial effects:

the clinical data indexes brought into the prediction model constructed by the method are indexes conventionally collected and recorded clinically, and the method has the advantages of easiness in obtaining, low price, convenience in use and the like. In the construction method of the prediction model, an advanced machine learning algorithm (such as an extreme gradient boosting Xgboost algorithm and a least absolute shrinkage and selection operator LASSO algorithm) and a logistic regression method are combined, the machine learning algorithm is used for identifying important variables related to the AKI persistence in an unbiased mode, the logistic regression is used for integrating the variables, specific influence coefficients of the variables on the AKI persistence are obtained, and therefore the prediction model has strong interpretability and prediction efficiency. The invention provides a simple and practical transient and persistent AKI risk prediction tool for clinicians, is beneficial to risk stratification of severe sepsis AKI patients, early identifies high-risk patients possibly developing persistent AKI, and further formulates individualized treatment strategies and management schemes, thereby improving clinical prognosis of patients and improving medical quality and medical safety.

Detailed Description

The technical solutions in the present invention will be described clearly and completely with reference to the following embodiments of the present invention, and it should be understood that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments that can be derived by one of ordinary skill in the art from the embodiments given herein are intended to be within the scope of the present invention.

Example 1:

5984 patients with sepsis AKI were screened according to current medical techniques and diagnostic criteria for sepsis AKI, with 3805 patients with transient AKI and 2179 patients with persistent AKI. The diagnostic criteria for sepsis AKI are as follows:

1) the diagnosis of sepsis is defined according to sepsis 3 rd edition, i.e. suspected infection (in fluid culture with antibiotics) combined with organ dysfunction (SOFA score ≧ 2);

2) diagnosis of AKI is based on 2012 tissue guidelines for improving global kidney disease prognosis, i.e. serum creatinine increases by more than or equal to 0.3mg/dL from baseline or rises to more than or equal to 1.5 fold baseline, or urine volume <0.5mL/kg/hr for more than 6 hours;

3) diagnosis of transient AKI, defined as complete reversal within 48 hours after AKI onset and maintained for at least 48 hours, and persistent AKI, according to the 16 th conference on acute disease quality initiative; persistent AKI is defined as AKI that persists for more than 48 hours, or reverses within 48 hours after AKI occurs, but recurs within the next 48 hours. Thus, the patient condition can be determined to be transient AKI, based on the patient condition not meeting the diagnostic criteria for AKI described in 2) within 48-96 hours after AKI diagnosis; determining the patient condition as persistent AKI based on the patient condition AKI being diagnosed within 48-96 hours of AKI meeting the criteria for AKI set forth in 2).

Randomly dividing 5984 sepsis AKI patients into a training set and a testing set, constructing a prediction model by using clinical data of the training set patients, and verifying the prediction model by using the clinical data of the testing set patients to verify the prediction efficiency of the prediction model. The number of patients in the training set was 4188 and the number of patients in the test set was 1796.

The prediction model and the construction method thereof in embodiment 1 are specifically as follows:

a predictive model of transient and persistent AKI in patients with severe sepsis, comprising the following continuous variables:

age, Age in years; partial pressure of arterial blood oxygen PaO₂In mm of mercury; arterial blood partial pressure of carbon dioxide PaCO₂In mm of mercury; anion gap, in millimoles per liter; lactate in millimoles per liter; international normalized ratio INR, unitless; partial thromboplastin time, Ptt, in seconds;

the predictive model includes the following categorical variables:

diabetes mellitus, Congestive heart failure, Chronic kidney disease, Mechanical ventilation required and renal replacement therapy RRT induction started, wherein the five variables are selected to be present or absent according to the actual condition of a patient, the corresponding value is 1 when the five variables are selected, and the corresponding value is 0 when the five variables are not selected;

the predictive model also includes the following categorical variables:

the AKI urine volume stage UO stage and AKI creatinine stage SCR stage, the two variables are selected from stage 0, stage1, stage2 or stage 3 according to the actual condition of the patient, when any stage is selected, the corresponding value of the stage is 1, and the corresponding values of the other stages are 0;

the prediction model comprises a prediction probability expression (1), wherein the prediction probability expression (1) is specifically as follows:

wherein, A is 0.006161 × Age-0.002077 × PaO₂+0.009274×PaCO₂+0.026074 × Anion gap +0.020870 × Lactate +0.068955 × 0INR +0.002655 × 1Ptt +0.259688 × 2Diabetes mellitus +0.320812 × 3 constructive heart failure +0.147493 × 4Chronic kit failure +0.270744 × 5Mechanical discovery +1.361824 × 6RRT indication +0 × UO Stage 0+0.494265 × UO Stage 1+1.279487 × UO Stage 2+2.169034 × UO Stage 3+0 × SCr Stage 0+0.856714 × SCr Stage 1+2.333892 × SCr Stage 2+2.585134 × SCr Stage 3-2.877650; p represents a probability score, the cutoff value of P is 0.6316, the cutoff value is selected by firstly drawing a working characteristic curve of a subject according to a series of different cutoff values in a training set, taking sensitivity as a vertical coordinate, taking 1-specificity (1-specificity represents 1 minus specificity) as a horizontal coordinate, determining the working characteristic curve according to a maximum jordan index in the training set, and when the probability score of a patient is lower than the cutoff value, the patient is a low-risk patient, the condition of the patient is transient AKI and recovers within 48 hours of the occurrence of AKI; when the patient probability score is above the cutoff value, a high risk patient, the patient's condition progresses to persistent AKI 48 hours after AKI occurs.

A method for constructing a prediction model of the transient and persistent AKI of the severe sepsis patient comprises the following steps:

step S1, collecting clinical data of sepsis patients who have AKI within 48 hours after entering an intensive care unit from an intensive care database;

step S2, obtaining a feature variable importance ranking for distinguishing transitional and persistent AKI after processing clinical data by using an extreme gradient boosting Xgboost algorithm, and selecting the feature with the top 20 bits of ranking for further feature screening;

step S3, the 20-bit features obtained in the step S2 are further screened by using the least absolute contraction and selection operator LASSO algorithm to obtain 14-bit variables, namely Age and arterial blood oxygen partial pressure PaO₂Arterial blood partial pressure of carbon dioxide PaCO₂Anion gap, Lactate, international normalized ratio INR, partial thromboplastin time Ptt, Diabetes mellitus, Congestive heart failure, Chronic renal disease, renal failure disease, Mechanical ventilation, renal activationSubstitution treatment RRT induction, AKI urine volume stage UO stage and AKI creatinine stage SCR stage;

step S4, selecting the 14-bit variable obtained in the step S3, and constructing a prediction model of the transient and persistent AKI of the patient with severe sepsis by using logistic regression;

and step S5, converting the prediction model into a prediction probability expression (1) by using R software according to the prediction model constructed in the step S4, and calculating the risk of the severe sepsis patient to progress to the transient and persistent AKI according to the prediction probability expression (1).

In step S1, the clinical data includes demographic characteristics (age, sex, race, ICU type and admission type), complications (hypertension, diabetes, congestive heart failure, peripheral vascular disease, chronic lung disease, liver disease, AIDS, metastatic tumors and chronic kidney disease), vital signs (body temperature, heart rate, respiratory rate and mean arterial pressure), laboratory tests (hemoglobin, white blood cell count, platelet count, bilirubin, albumin, arterial blood pH, arterial blood oxygen partial pressure, arterial blood carbon dioxide partial pressure, anion space, serum sodium, serum potassium, serum chloride, serum bicarbonate, lactate, international normalized ratio, and thrombin time), treatment (requiring mechanical ventilation, requiring vasopressor medication, initiating renal replacement therapy, using diuretics, and daily average infusion volume), and AKI staging (AKI staging according to creatinine and urine volume standards).

Step S2 further includes selecting a hyper-parameter of the extreme gradient boost Xgboost algorithm using 5-fold cross validation, and the resulting hyper-parameter is as follows: the learning rate learning _ rate is 0.01, the number of trees n _ estimators is 650, the maximum depth max _ depth of the trees is 3, the minimum weight min _ child _ weight of the child nodes is 3, the sampling rate subsample is 0.6, the characteristic sampling rate colsample _ byte of each tree is 0.7, and the minimum loss split gamma is 1.

The top 20 ranks in step S2 are characterized by AKI urine volume fraction UO stage1, AKI urine volume fraction UO stage2, AKI urine volume fraction UO stage 3, AKI creatinine fraction SCR stage1, AKI creatinine fraction SCR stage2, AKI creatinine fraction SCR stage 3, and congestiveCardiac failure, Anion gap, partial pressure of arterial blood oxygen PaO₂Initiating renal replacement therapy RRT induction, Diabetes mellitus, Lactate, Chronic renal disease Chronic Kidney disease, ICU type (coronary artery disease or Heart surgical Care Unit), International normalized ratio INR, arterial blood partial pressure of carbon dioxide PaCO₂Arterial blood pH, Mechanical ventilation required, partial thromboplastin time Ptt and Age. And setting a dummy variable when processing the AKI urine volume stage and the AKI creatinine stage of the multi-classification variable, and selecting one of the classifications as a dummy variable reference. The invention selects the UO stage 0 of AKI urine volume stage as reference, and 1 is the three categories of the UO stage1 stage, the UO stage2 stage and the UO stage 3 stage of the AKI urine volume stage. Therefore, 1 variable corresponds to 3 characteristics in the stage of AKI urine volume. Similarly, the invention selects the SCr stage 0 of the AKI creatinine stage as a reference, and the three classifications of the SCr stage1, SCr stage2 of the AKI creatinine stage and SCr stage 3 of the AKI creatinine stage are all 1 characteristic. Thus, 1 variable in the AKI creatinine stage corresponds to 3 features.

The step S3 further includes processing the least absolute contraction and selection operator LASSO algorithm using cross-validation, selecting a cross-validation determination parameter λ to obtain a 14-bit variable, where λ has a value of 0.009601.

And (3) predicting the patients in the test set by using a prediction model constructed by the clinical data of the patients in the training set to obtain the risk value of the disease progression of each patient to the continuous AKI. The area under the subject working characteristic curve of the predictive model in the test set was 0.76. A cutoff value 0.6316 was also used in the test set at which the sensitivity of the predictive model was 0.63 and specificity was 0.76.

Example 2:

a sepsis patient with AKI within 48 hours after ICU entry, Age 85 years (year); partial pressure of arterial blood oxygen PaO₂60 millimeters of mercury (mmHg); arterial blood partial pressure of carbon dioxide PaCO₂41 millimeters of mercury (mmHg); anion gap of 21 millimoles per liter (mmol/L); lactate was 1.3 mmoles per liter(mmol/L); the international normalized ratio INR is 1.1; partial thromboplastin time Ptt of 36.9 seconds (sec); diabetes mellitus free Diabetes mellitus; congestive heart failure (Congestive heart failure) is known as the heart failure; chronic kidney disease free chrono kidney disease; mechanical ventilation is not required; kidney replacement therapy RRT induction is not initiated; the UO stage of AKI urine volume stage is stage 3; the AKI creatinine stage SCR stage is stage 1.

And (3) calculating the value of the variable according to the prediction probability expression (1) to obtain a prediction probability of 0.88045 when the patient's disease condition progresses to continuous AKI, wherein the patient is a high risk (high risk) patient, and the patient's disease condition progresses to continuous AKI 48 hours after the AKI occurs.

Example 3:

a sepsis patient with AKI within 48 hours after ICU entry, Age 84 years (year); partial pressure of arterial blood oxygen PaO₂110 millimeters of mercury (mmHg); arterial blood partial pressure of carbon dioxide PaCO₂46 millimeters of mercury (mmHg); anion gap is 10 millimoles per liter (mmol/L); lactate was 2.2 millimoles per liter (mmol/L); the international normalized ratio INR is 1.4; partial thromboplastin time Ptt was 41.1 seconds (sec); diabetes mellitus free Diabetes mellitus; congestive heart failure-free Congestatic heart failure; chronic kidney disease free chrono kidney disease; mechanical ventilation is required; kidney replacement therapy RRT induction is not initiated; the UO stage of AKI urine volume stage is stage 3; the AKI creatinine stage SCR stage is stage 1.

And (3) calculating the value of the variable according to the prediction probability expression (1) to obtain a prediction probability of 0.83846 when the patient's disease condition progresses to continuous AKI, wherein the patient is a high risk (high risk) patient, and the patient's disease condition progresses to continuous AKI 48 hours after the AKI occurs.

Example 4:

a sepsis patient with AKI within 48 hours after ICU entry, Age 57 years (year); partial pressure of arterial blood oxygen PaO₂101 millimeters of mercury (mmHg); arterial blood partial pressure of carbon dioxide PaCO₂45 millimeters of mercury (mmHg); anion gap is 16 millimoles per liter (mmol/L); lactate was 2.2 millimoles per liter (mmol/L); international standardizationThe ratio INR is 1.5; partial thromboplastin time Ptt is 53.5 seconds (sec); diabetes mellitus is known; congestive heart failure-free Congestatic heart failure; chronic kidney disease free chrono kidney disease; mechanical ventilation is not required; kidney replacement therapy RRT induction is not initiated; the UO stage of AKI urine volume stage is stage 3; the AKI creatinine staging SCR stage is stage 3.

And (3) calculating the value of the variable according to the prediction probability expression (1) to obtain a prediction probability of 0.96781 when the patient's disease condition progresses to continuous AKI, wherein the patient is a high risk (high risk) patient, and the patient's disease condition progresses to continuous AKI 48 hours after the AKI occurs.

Example 5:

a sepsis patient with AKI within 48 hours after ICU entry, Age, Age 68 years (year); partial pressure of arterial blood oxygen PaO₂91 millimeters of mercury (mmHg); arterial blood partial pressure of carbon dioxide PaCO₂58 millimeters of mercury (mmHg); anion gap is 16 millimoles per liter (mmol/L); lactate is 1.9 millimoles per liter (mmol/L); the international normalized ratio INR is 1.3; partial thromboplastin time Ptt is 24.1 seconds (sec); diabetes mellitus free Diabetes mellitus; congestive heart failure-free Congestatic heart failure; chronic kidney disease free chrono kidney disease; mechanical ventilation is required; kidney replacement therapy RRT induction is not initiated; the UO stage of AKI urine volume stage is stage 2; the AKI creatinine stage SCR stage is stage 0.

And (3) calculating the values of the variables according to the prediction probability expression (1) to obtain a prediction probability of 0.51272 when the patient has a persistent AKI, wherein the patient is a low risk (high risk) patient, and the patient has a transient AKI and recovers within 48 hours of the AKI.

Example 6:

a sepsis patient with AKI within 48 hours after ICU entry, Age 71 years (year); partial pressure of arterial blood oxygen PaO₂103 millimeters of mercury (mmHg); arterial blood partial pressure of carbon dioxide PaCO₂43 millimeters of mercury (mmHg); anion gap is 12 millimoles per liter (mmol/L); lactate is 2.5 millimoles per liter (mmol/L); the international normalized ratio INR is 2.1; in partThromboplastin time Ptt 39.5 seconds (sec); diabetes mellitus free Diabetes mellitus; congestive heart failure-free Congestatic heart failure; chronic kidney disease free chrono kidney disease; mechanical ventilation is required; kidney replacement therapy RRT induction is not initiated; the UO stage of AKI urine volume stage is stage 1; the AKI creatinine stage SCR stage is stage 0.

And (3) calculating the values of the variables according to the prediction probability expression (1) to obtain a prediction probability of 0.29409 when the patient has a persistent AKI, wherein the patient is a low risk (high risk) patient, and the patient has a transient AKI and recovers within 48 hours of the AKI.

Example 7:

a sepsis patient with AKI within 48 hours after ICU entry, Age 57 years (year); partial pressure of arterial blood oxygen PaO₂76 millimeters of mercury (mmHg); arterial blood partial pressure of carbon dioxide PaCO₂54 millimeters of mercury (mmHg); anion gap is 14 millimoles per liter (mmol/L); lactate is 1.5 millimoles per liter (mmol/L); the international normalized ratio INR is 1.2; partial thromboplastin time Ptt was 33.3 seconds (sec); diabetes mellitus free Diabetes mellitus; congestive heart failure (Congestive heart failure) is known as the heart failure; chronic kidney disease free chrono kidney disease; mechanical ventilation is required; kidney replacement therapy RRT induction is not initiated; the UO stage of AKI urine volume stage is 0 stage; the AKI creatinine stage SCR stage is stage 1.

And (3) calculating the values of the variables according to the prediction probability expression (1) to obtain a prediction probability of 0.45816 when the patient has a persistent AKI, wherein the patient is a low risk (high risk) patient, and the patient has a transient AKI and recovers within 48 hours of the AKI.

Comparative example 1 (from the literature "Titeca-Beauport D, Daubin D, Van Vong L, Belliard G, Bruel C, Alaya S, Chaoui K, Andrieu M, Rouquette-Vinnti I, Godde F et al: U-line cell cycle operators diagnosis library from between the two channels and between AKI in the early section show: a proactive, multicenter study. Critical Care 2020,24 (1.):

the results of the Titeca-Beauport D et al study, in which 184 patients with septic shock who developed AKI were enrolled, investigated the predictive value of the tissue inhibitor of metalloproteinases-2 (TIMP-2). times.insulin-like growth factor binding protein 7(IGFBP7) for the persistence of AKI, showed that the area under the working characteristic curve of their subjects was 0.67, the sensitivity was 0.74, the specificity was 0.59, and the predictive potency was low.

Comparative example 2 (from the literature "Darmon M, Bourmaud A, Reynaud M, Rouleau S, Meziani F, Boivin A, Benyamine M, Vincent F, Lautrett A, Leroy C et al: Performance of Doppler-based reactive index and semi-reactive feedback in compressing catalytic AKI: results of Multi-reactive Multi-site study. Intensive Care Medicine 2018,44(11): 1904. 1913."):

the study of Darmon M et al, which included 233 critically ill AKI patients, demonstrated predictive value of doppler-based Resistance Index (RI) and semi-quantitative renal perfusion (SQP) for AKI persistence, with subjects having an under-operating characteristic score of 0.58 and 0.59, respectively, sensitivity of 0.50 and 0.39, specificity of 0.68 and 0.75, respectively, and low predictive efficacy.

Comparative example 3 (from the literature "Pons B, Lautrette A, Oziel J, Dellamonica J, Vermesch R, Ezinger E, Mariat C, Bernardin G, Zeni F, Cohen Y et al: Diagnostic access of early urinary indexes in differential training from laboratory access kit in critical intake in critical activities of pages: multicenter scientific. Critical Care 2013,17(2): R56."):

pons B and other researches are carried out on 107 severe AKI patients, the prediction values of urine indexes (sodium excretion fraction, urea excretion fraction, ratio of urea to plasma urea, and ratio of creatinine to plasma creatinine) on the AKI persistence are researched, and the prediction effects of the urine indexes are found to be poor, and the areas under the working characteristic curves of the subjects are all less than or equal to 0.65.

Compared with comparative examples 1-3, the clinical data indexes included in the prediction model constructed in example 1 are all indexes conventionally collected and recorded clinically, and have the advantages of easy acquisition, low price, convenient use and the like. In the construction method of the prediction model, in example 1, an advanced machine learning algorithm (such as the extreme gradient boosting Xgboost algorithm and the least absolute shrinkage and selection operator LASSO algorithm) and a logistic regression are combined, the machine learning algorithm is used to identify important variables associated with the AKI persistence unbiased, and logistic regression is used to integrate the variables to obtain specific influence coefficients of the variables on the AKI persistence, so that the prediction model has strong interpretability and prediction efficiency. Unlike the use of a single predictor in comparative examples 1-3, the predictive model in example 1 combines multiple clinical indicators with a subject working characteristic curve with an area of up to 0.76, sensitivity of up to 0.63 and specificity of up to 0.76. In addition, in the construction of the prediction model, 5984 patients are included in example 1, the sample size is large, total samples are randomly divided into a training set and a testing set according to the proportion of 7:3, and the model is further verified to have good prediction efficiency through the setting of the testing set, so that the stability and the practicability of the model are ensured. By combining the analysis of examples 1-7 and comparative examples 1-3, the invention provides a simple and practical transient and persistent AKI risk prediction tool for clinicians, is beneficial to risk stratification of severe sepsis AKI patients, early identifies high-risk patients who may progress to persistent AKI, and further formulates individualized treatment strategies and management schemes, thereby improving clinical prognosis of patients and improving medical quality and medical safety.

The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.