Prediction method for progression from acute kidney injury to chronic kidney disease

Abstract

The invention relates to a method for predicting the progress from acute kidney injury to chronic kidney disease, and belongs to the field of medical prediction. The method comprises the following steps: constructing a clinical model containing age, gender, baseline serum creatinine, urinary albumin creatinine ratio and discharged serum creatinine; fitting the nonlinear relationship between the acute kidney injury staging and the chronic kidney disease progress by adopting a restrictive cubic spline function; detecting the level of cytokeratin 20 in urine to construct a biomarker model; and combining Logistic regression to analyze the nonlinear correlation between the cytokeratin 20 in the urine and the progress of the chronic kidney disease, and evaluating the performance of the model. Conventional clinical indexes and cytokeratin 20 in urine are integrated, and extra detection is not needed; the nonlinear relation between the acute kidney injury staging and the chronic kidney disease progress is accurately described by using a cubic spline function, and the method reduces the clinical application cost, improves the prediction efficiency, and is suitable for early risk stratification and individualized intervention.

Description

Technical Field

[0001] The present invention relates to the field of medical prediction, and in particular to a method for predicting the progression of acute kidney injury to chronic kidney disease. Background Art

[0002] Acute kidney injury is a major global public health problem with a high annual incidence rate, which not only significantly increases patient mortality but also leads to a sharp increase in medical resource consumption. Traditionally, acute kidney injury is considered a reversible disease, but large-scale population cohort studies in recent years have revealed that surviving patients with acute kidney injury are at significant risk of progression to chronic kidney disease. Currently, chronic kidney disease is prevalent in the global population and causes a large number of deaths each year. However, the existing clinical assessment system still lacks an effective means to accurately predict the transformation of acute kidney injury to chronic kidney disease. Summary of the Invention

[0003] Based on this, it is necessary to provide a method to predict the progression of acute kidney injury to chronic kidney disease in order to address the problem of the lack of effective means to accurately predict the transformation of acute kidney injury to chronic kidney disease.

[0004] A method for predicting the progression of acute kidney injury to chronic kidney disease, comprising:

[0005] S1. Construct a predictive clinical model, wherein the clinical factors of the predictive clinical model include age, sex, baseline serum creatinine, the patient's urine albumin-to-creatinine ratio at the time of diagnosis of acute kidney injury, and the serum creatinine level at discharge;

[0006] S2, obtain restricted cubic splines adjusted for acute kidney injury stage;

[0007] S3. Obtain the urine cytokeratin 20 level at the time of diagnosis and construct a biomarker model based on the urine cytokeratin 20 level;

[0008] S4. Logistic regression was used to examine the nonlinear pattern of the association between urine cytokeratin 20 at diagnosis and progression from acute kidney injury to chronic kidney disease using restricted cubic splines adjusted for age, sex, baseline serum creatinine, urine albumin-creatinine ratio at the time of acute kidney injury diagnosis, serum creatinine level at discharge, and acute kidney injury stage.

[0009] S5. Use the area under the receiver operating characteristic curve to evaluate and compare the discriminative performance of models for predicting the risk of progression of acute kidney injury to chronic kidney disease.

[0010] The present application discloses a method for predicting the progression of acute kidney injury to chronic kidney disease. By integrating age, gender, baseline serum creatinine level, urine albumin-creatinine ratio at the time of diagnosis of acute kidney injury and serum creatinine level at discharge, the patient's baseline renal function status and degree of damage can be quickly assessed. At the same time, all included indicators are routine clinical test items (such as blood and urine tests), and no additional special tests are required, which reduces the threshold and cost of clinical application. The relationship between acute kidney injury staging and chronic kidney disease progression is not simply linear. The restricted cubic spline function can be used to more accurately describe the risk change trend between different stages. Elevated levels of cytokeratin 20 in urine indicate damage to renal tubular epithelial cells and impaired repair. This persistent damage is a key pathological process in the progression of renal fibrosis and chronic kidney disease. Combined with clinical models, patients with subclinical renal damage that are difficult to capture with conventional indicators can be identified, such as patients with normal baseline serum creatinine but elevated cytokeratin 20 in urine. Logistic regression, incorporating nonlinear relationships adjusted for age, sex, baseline serum creatinine, urine albumin-creatinine ratio, discharge serum creatinine level, and acute kidney injury stage, allows for a comprehensive assessment of patients' overall risk. The area under the receiver operating characteristic curve objectively assesses the model's ability to discriminate between patients with progressive chronic kidney disease and those without.

[0011] In one embodiment, the baseline serum creatinine is the most recent level measured within 1 month before the patient is admitted to the hospital.

[0012] In one embodiment, the specific steps of step S5 are as follows:

[0013] S51. Analyze the association between predictive clinical models and biomarker models and construct models that combine clinical factors and biomarkers.

[0014] S52. Use the area under the receiver operating characteristic curve to assess and compare the discriminative performance of predictive clinical models, biomarker-based models, and models that combine clinical factors and biomarkers.

[0015] S53. Use the Youden index method to determine the optimal cutoff value. By comprehensively considering the sensitivity and specificity of the model, find the cutoff value that maximizes the sum of the two.

[0016] S54. Evaluate the degree of model improvement. Using the model containing only clinical factors as a reference, calculate the net reclassification improvement index and comprehensive discrimination improvement index of the model containing a combination of clinical factors and biomarkers.

[0017] Combining clinical factors with biomarkers for modeling can integrate multi-dimensional information. The performance of different models is quantified by the area under the receiver operating characteristic curve, providing a scientific basis for clinical selection. The Youden Index method optimizes the practicality of the model in clinical practice by finding the cutoff value with the largest sum of sensitivity and specificity. Evaluate whether the integrated model can more accurately reclassify patients to the correct risk level. Quantify the overall improvement of the integrated model's predicted probability for all patients. By combining clinical factors and biomarkers to construct an optimized prediction model, and using the receiver operating characteristic curve, Youden index, net reclassification improvement index and comprehensive discrimination improvement index for scientific evaluation, the model's predictive ability and clinical application value are enhanced. This method not only improves the accuracy of disease prediction, but also optimizes personalized medical decisions, contributes to the development of precision medicine, and ultimately improves patient prognosis and increases the efficiency of medical resource utilization.

[0018] In one embodiment, the specific steps of step S51 are as follows:

[0019] S511. Collect clinical factors including age, sex, baseline serum creatinine, urine albumin-creatinine ratio of patients at the time of acute kidney injury diagnosis, and serum creatinine level at discharge, denoted as vector X = (x1, x2, x3, x4, x5, x6);

[0020] S512. Collect data on the level of cytokeratin 20 in urine at the time of diagnosis, recorded as biomarker B;

[0021] S513. Using multiple imputation methods, based on the existing data distribution, a statistical model is used to generate missing value estimates multiple times to obtain multiple complete data sets;

[0022] S514. A logistic regression prediction model is constructed based on clinical factors, and spline transformation features are introduced to capture nonlinear relationships. The model formula of the clinical model is as follows:

[0023]

[0024] Among them, Y = 1 means that the patient progresses to chronic kidney disease; x j is the jth clinical factor; β0 is the intercept; β j is the regression coefficient; is the variable x r The lth basis function after spline transformation; γ l are the spline coefficients, is the contribution of the linear combination of clinical factors, is the nonlinear contribution of clinical factors;

[0025] S515. Prediction probability based on clinical factors

[0026] S515. Construct a prediction model based on biomarker B. The nonlinear spline method is used to construct the model formula:

[0027]

[0028] Where B is the biomarker; Θ0 is the model intercept, Θ1 is the linear coefficient; Θ i+1 is the spline transformation truncation coefficient; ξi is the preset split node; n is the number of nodes; Θ1B is the biomarker linear contribution,

[0029] is the nonlinear contribution of the biomarker;

[0030] Biomarker risk prediction probability

[0031] S516, using clinical and biomarker models to obtain the predicted probability p C and p B , retain the intermediate features of spline transformation to construct interaction terms;

[0032] S517. Construct a joint meta-model to capture synergy effects. The meta-model formula is:

[0033] logitP(Y=1|p C ,pB)=α0+α1p C +α2p B +α3(p C ×p B )

[0034] Among them, α0 is the intercept, α1 and α2 measure the independent contribution parameters; α3 captures the interaction effect parameter; the synergistic effect is captured by the interaction term pc×pB;

[0035] The final joint prediction probability expression is:

[0036]

[0037] S518, use K-fold cross validation to generate the first layer of predictions, and construct a prediction containing p based on the results. C 、p B The parameters α0, α1, α2, and α3 are determined by maximum likelihood estimation using the secondary training set and the true label Y. Nested cross-validation is used to ensure that tuning and evaluation are independent, and α0, α1, α2, and α3 are input to obtain the final joint prediction probability.

[0038] By generating missing value estimates multiple times, the original distribution characteristics of the data are retained, avoiding sample bias caused by deleting missing values. Spline transformation allows the relationship between clinical factors and chronic kidney disease progression to present a nonlinear pattern. The nonlinear spline model can capture the dose-response relationship between urinary cytokeratin 20 levels and chronic kidney disease progression. The joint meta-model is constructed by the interaction term (p C ×p B ) reveals the superimposed effect of clinical risk and biomarker risk. The outer K-fold cross-validation is used for model evaluation, and the inner cross-validation is used for parameter tuning to ensure that the model is completely independent during training and testing to avoid overfitting. The parameters α0, α1, α2, and α3 are determined by maximizing the likelihood function to ensure the best fitting effect of the model on the training data, and the generalization ability is verified by cross-validation. Multiple imputation is used to fill in missing data, combined with nonlinear spline transformation to capture complex relationships, and interaction term modeling is used to consider synergistic effects, and finally the robustness of the model is ensured by cross-validation. This method improves the accuracy of predicting the progression of acute kidney injury to chronic kidney disease, optimizes personalized risk assessment, makes it of great value in clinical applications, and also provides a reliable mathematical modeling method.

[0039] In one embodiment, the specific steps of step S53 are as follows:

[0040] S531. Calculate the sensitivity and specificity of the model at different cutoff values;

[0041] S532. Calculate the Youden index using the following formula:

[0042] Among them, TP is the number of true positives, FN is the number of false negatives, TN is the number of true negatives, and FP is the number of false positives; is sensitivity; is specificity;

[0043] S533. Calculate the corresponding sensitivity and 1-specificity at different cutoff values, and plot the receiver operating characteristic curve with the false positive rate as the horizontal axis and the sensitivity as the vertical axis;

[0044] S534. On the receiver operating characteristic curve, find the point that maximizes the Youden index. The cutoff value corresponding to this point is the optimal cutoff value.

[0045] By calculating the sensitivity and specificity at different cutoff values, we can fully understand the performance of the model at different critical points. Emergency scenarios may give priority to high sensitivity to reduce missed diagnoses, while chronic disease screening may pay more attention to high specificity to reduce the cost of misdiagnosis. The Youden Index (J) quantifies the comprehensive ability of the model through a formula, with a value range of -1 to 1. The larger the J value, the better the overall effect of the model in reducing missed diagnoses and misdiagnoses. Plotting a curve with the false positive rate (1-specificity) as the horizontal axis and the sensitivity as the vertical axis can intuitively show the performance of the model at all possible cutoff values. The closer the curve is to the upper left corner (sensitivity 100% and false positive rate 0%), the better the model performance. The point with the largest Youden Index corresponds to the cutoff value with the largest sum of sensitivity and specificity, at which time the combined loss of missed diagnosis and misdiagnosis is the smallest.

[0046] In one embodiment, the specific steps of obtaining the urine cytokeratin 20 level at the time of diagnosis are as follows:

[0047] S31. Urine sample from the patient within 24 hours after the diagnosis of acute kidney injury;

[0048] S32, centrifuging the urine sample for 8 to 15 minutes;

[0049] S33, collecting the supernatant after separation in the urine;

[0050] S34. Measure urine albumin and creatinine using an automatic analyzer, and normalize the concentrations of urine albumin and creatinine to urine creatinine, where the urine creatinine is the level of cytokeratin 20 in urine.

[0051] Collecting urine samples within 24 hours of the diagnosis of acute kidney injury maximizes the retention of early biomarker signals for renal tubular injury. Centrifugation (8-15 minutes) effectively removes cellular debris, red blood cells, and macromolecular impurities from the urine, preventing these components from interfering with the cytokeratin 20 assay. Normalizing urine albumin and creatinine concentrations to urine creatinine eliminates the effects of urine dilution or concentration on the test results. Rapidly obtaining cytokeratin 20 levels can help physicians identify high-risk patients early in the course of acute kidney injury.

[0052] In one embodiment, the specific steps of step S4 are as follows:

[0053] S41. Perform data cleaning, remove missing values ​​and outliers, and perform interpolation. At the same time, use the density peak clustering method to identify outliers in the data, and use the Mahalanobis distance for secondary screening to avoid extreme values ​​interfering with the regression analysis;

[0054] S42. Perform Box-Cox or Yeo-Johnson transformation on the urinary cytokeratin 20 level to reduce the effect of skewness;

[0055] S43. Use decision tree or LightGBM method to perform nonlinear binning of urinary cytokeratin 20 levels to make it more consistent with the pattern of kidney disease progression;

[0056] S44. Calculate the mutual information between the urine cytokeratin 20 level and the patient's urine albumin-to-creatinine ratio at the time of diagnosis of acute kidney injury to form an interaction term. Calculate the ratio of baseline serum creatinine to serum creatinine level at discharge to reflect the degree of recovery. Calculate the interaction effect term between age and urine cytokeratin 20 level to account for the modulatory effect of age on biomarkers.

[0057] S45. Stratified logistic regression was used, with variables introduced stepwise. The first layer used age, sex, baseline serum creatinine, the patient's urine albumin-creatinine ratio at the time of AKI diagnosis, and the serum creatinine level at discharge as the base model. The second layer introduced restricted cubic splines adjusted for AKI stage. The third layer introduced urine cytokeratin 20 levels, and its contribution was first analyzed using a linear term. The fourth layer added a nonlinear term for urine cytokeratin 20 levels to test whether the nonlinear effect was significant. The fifth layer added interaction terms, and the likelihood ratio test was used to screen for significant variables.

[0058] S46, Acute Kidney Injury Stage-Adjusted Restricted Cubic Splines Modeling the Nonlinear Relationship Between Urinary Cytokeratin 20 Levels and the Progression of Acute Kidney Injury to Chronic Kidney Disease;

[0059] S47. Perform variable selection and model optimization. Use LASSO regression for variable screening to avoid overfitting while retaining the most predictive features. Use Bayesian model averaging and weighted averaging between different models to improve stability and construct logistic regression to enhance the generalization ability of the model.

[0060] Density peak clustering automatically identifies outliers by calculating the local density and distance of data points, avoiding the limitations of traditional distance- or distribution-based methods. Mahalanobis distance, combined with intervariate correlations, can effectively identify multivariate outliers and reduce the interference of extreme values ​​on regression analysis. Missing values ​​can be filled using multiple imputation or model prediction to preserve data information and avoid sample bias caused by data deletion. Box-Cox or Yeo-Johnson transformations can effectively correct the skewed distribution of CK20 levels, making it closer to a normal distribution and satisfying the linear relationship assumption of logistic regression. Decision trees or LightGBM automatically find the optimal binning points for CK20 by recursively partitioning the data. Mutual information analysis can quantify the nonlinear association between CK20 and the urine albumin-to-creatinine ratio. The baseline to discharge serum creatinine ratio (reflecting the degree of renal function recovery) can be used as an independent variable to supplement the assessment of the reversibility of acute kidney injury. The interaction term between age and CK20 can reveal differences in the predictive value of biomarkers across different age groups; for example, the cutoff value of CK20 may be lower in elderly patients. The introduction of hierarchical variables allows for a systematic assessment of the independent impact of various factors on CKD progression. Restricted cubic splines allow the relationship between CK20 and CKD progression to exhibit different slopes in different intervals. LASSO regression uses a penalty coefficient to compress the coefficients of irrelevant variables to zero, retaining core predictors and improving model parsimony.

[0061] In one embodiment, the step S4 further includes the following steps:

[0062] S6. The association between urinary cytokeratin 20 and severe acute tubular injury during acute kidney injury was examined using logistic regression with tertiles and logarithmic transformation of urinary cytokeratin 20.

[0063] Tertile division and logarithmic transformation reduced inter-individual measurement bias and the impact of extreme values, improving data stability. The use of a logistic regression model effectively controlled for confounding factors, enhancing the scientific nature and credibility of the analysis results.

[0064] In one embodiment, the specific steps of step S6 are as follows:

[0065] S61. Collect urine cytokeratin 20 levels from patients at the time of diagnosis of acute kidney injury. Calculate the 25th, 50th, and 75th percentiles based on the distribution of urine cytokeratin 20 levels in the samples and categorize the data into three levels: low, medium, and high.

[0066] S62. Logarithmic transformation was used to make the data conform to normal distribution, and the transformed continuous variables were recorded so that they could be analyzed as continuous independent variables in the regression model;

[0067] S63. The low, medium, and high scores were included in the model as categorical variables. Logistic regression was used to analyze the strength of association between each group and severe acute renal tubular injury. The odds ratio and its 95% confidence interval were calculated for each group.

[0068] S64. Examine the regression coefficients and P values ​​of the grouped or log-transformed variables in the urine cytokeratin 20 model to determine whether there is a statistically significant association between the variable and severe acute renal tubular injury.

[0069] S65. Based on the regression results, interpret the association between urinary cytokeratin 20 levels and severe acute renal tubular injury.

[0070] Dividing urine cytokeratin 20 levels into three equal parts, from smallest to largest, allows for intuitive visualization of risk differences within these intervals. Logarithmic transformation improves data normality, fulfills the linearity assumption of logistic regression, and avoids parameter estimation errors caused by data distribution bias. Incorporating covariates such as age, sex, baseline serum creatinine, and urine albumin-creatinine ratio allows for the isolation of the independent effect of urine cytokeratin 20 levels on SATI.

[0071] In one embodiment, the step S5 further includes the following steps:

[0072] Construct a virtual patient cohort and simulate the effects of different intervention plans on the progression of acute kidney injury to chronic kidney disease through Monte Carlo simulation or proxy model. Use simulation to verify the adaptability and intervention effect of the prediction model in different scenarios, and provide theoretical basis and optimization suggestions for personalized treatment plans.

[0073] By constructing a virtual patient cohort, it is possible to train high-precision prediction models based on large sample data, thereby improving the reliability of disease progression predictions. Monte Carlo simulation can fully account for randomness and individual differences, making the prediction model closer to real clinical conditions and improving the accuracy of risk assessment. Proxy models can simulate disease progression under different intervention strategies, helping clinicians identify the optimal treatment plan. By simulating the treatment responses of different individuals, "one-size-fits-all" treatment plans can be avoided, providing a theoretical basis for personalized precision medicine. Early identification of high-risk patients and targeted intervention can help reduce the rate of acute kidney injury progressing to chronic kidney disease. By simulating the effects of interventions under different medical resource conditions, the allocation of medical resources can be optimized and the economic benefits of intervention strategies can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 Flowchart of a method for predicting progression of acute kidney injury to chronic kidney disease;

[0075] Figure 2Flowchart showing the specific steps for obtaining the urine cytokeratin 20 level at diagnosis;

[0076] Figure 3 Flowchart showing the specific steps for evaluating and comparing the discriminative performance of models for predicting the progression of acute kidney injury to chronic kidney disease using the area under the receiver operating characteristic curve;

[0077] Figure 4 Flowchart showing the steps involved in examining the association between urinary cytokeratin 20 and severe acute tubular injury in acute kidney injury using logistic regression with tertiles and logarithmic transformation of urinary cytokeratin 20. DETAILED DESCRIPTION

[0078] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features therein can be combined with each other.

[0079] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below.

[0080] The following describes the method for predicting the progression of acute kidney injury to chronic kidney disease according to some embodiments of the present invention with reference to the accompanying drawings.

[0081] Example 1

[0082] like Figure 1 As shown, a method for predicting the progression of acute kidney injury to chronic kidney disease includes the following steps:

[0083] S1. Construct a predictive clinical model, wherein the clinical factors of the predictive clinical model include age, sex, baseline serum creatinine, the patient's urine albumin-to-creatinine ratio at the time of diagnosis of acute kidney injury, and the serum creatinine level at discharge;

[0084] S2, obtain restricted cubic splines adjusted for acute kidney injury stage;

[0085] S3. Obtain the urine cytokeratin 20 level at the time of diagnosis and construct a biomarker model based on the urine cytokeratin 20 level;

[0086] S4. Logistic regression was used to examine the nonlinear pattern of the association between urine cytokeratin 20 at diagnosis and progression from acute kidney injury to chronic kidney disease using restricted cubic splines adjusted for age, sex, baseline serum creatinine, urine albumin-creatinine ratio at the time of acute kidney injury diagnosis, serum creatinine level at discharge, and acute kidney injury stage.

[0087] S5. Use the area under the receiver operating characteristic curve to evaluate and compare the discriminative performance of models for predicting the risk of progression of acute kidney injury to chronic kidney disease.

[0088] The present application discloses a method for predicting the progression of acute kidney injury to chronic kidney disease. By integrating age, gender, baseline serum creatinine level, urine albumin-creatinine ratio at the time of diagnosis of acute kidney injury and serum creatinine level at discharge, the patient's baseline renal function status and degree of damage can be quickly assessed. At the same time, all included indicators are routine clinical test items (such as blood and urine tests), and no additional special tests are required, which reduces the threshold and cost of clinical application. The relationship between acute kidney injury staging and chronic kidney disease progression is not simply linear. The restricted cubic spline function can be used to more accurately describe the risk change trend between different stages. Elevated levels of cytokeratin 20 in urine indicate damage to renal tubular epithelial cells and impaired repair. This persistent damage is a key pathological process in the progression of renal fibrosis and chronic kidney disease. Combined with clinical models, patients with subclinical renal damage that are difficult to capture with conventional indicators can be identified, such as patients with normal baseline serum creatinine but elevated cytokeratin 20 in urine. Logistic regression, incorporating nonlinear relationships adjusted for age, sex, baseline serum creatinine, urine albumin-creatinine ratio, discharge serum creatinine level, and acute kidney injury stage, allows for a comprehensive assessment of patients' overall risk. The area under the receiver operating characteristic curve objectively assesses the model's ability to discriminate between patients with progressive chronic kidney disease and those without.

[0089] In addition to the features of the above embodiment, this embodiment is further limited in that: the baseline serum creatinine is the most recent measurement level within 1 month before the patient is admitted to the hospital.

[0090] Example 2

[0091] like Figure 3 As shown, in addition to the features of the above embodiment, this embodiment further defines: S5, the specific steps of using the area under the receiver operating characteristic curve to evaluate and compare the discriminant performance of the risk model for predicting the progression of acute kidney injury to chronic kidney disease are as follows:

[0092] S51. Analyze the association between predictive clinical models and biomarker models and construct models that combine clinical factors and biomarkers.

[0093] S52. Use the area under the receiver operating characteristic curve to assess and compare the discriminative performance of predictive clinical models, biomarker-based models, and models that combine clinical factors and biomarkers.

[0094] S53. Use the Youden index method to determine the optimal cutoff value. By comprehensively considering the sensitivity and specificity of the model, find the cutoff value that maximizes the sum of the two.

[0095] S54. Evaluate the degree of model improvement. Using the model containing only clinical factors as a reference, calculate the net reclassification improvement index and comprehensive discrimination improvement index of the model containing a combination of clinical factors and biomarkers.

[0096] Combining clinical factors with biomarkers for modeling can integrate multi-dimensional information. The performance of different models is quantified by the area under the receiver operating characteristic curve, providing a scientific basis for clinical selection. The Youden Index method optimizes the practicality of the model in clinical practice by finding the cutoff value with the largest sum of sensitivity and specificity. Evaluate whether the integrated model can more accurately reclassify patients to the correct risk level. Quantify the overall improvement of the integrated model's predicted probability for all patients. By combining clinical factors and biomarkers to construct an optimized prediction model, and using the receiver operating characteristic curve, Youden index, net reclassification improvement index and comprehensive discrimination improvement index for scientific evaluation, the model's predictive ability and clinical application value are enhanced. This method not only improves the accuracy of disease prediction, but also optimizes personalized medical decisions, contributes to the development of precision medicine, and ultimately improves patient prognosis and increases the efficiency of medical resource utilization.

[0097] Example 3

[0098] In addition to the features of the above embodiment, this embodiment further defines: S51, analyzing the association between the predictive clinical model and the biomarker model, and constructing a model combining clinical factors and biomarkers. The specific steps are as follows:

[0099] S511. Collect clinical factors including age, sex, baseline serum creatinine, urine albumin-creatinine ratio of patients at the time of acute kidney injury diagnosis, and serum creatinine level at discharge, denoted as vector X = (x1, x2, x3, x4, x5, x6);

[0100] S512. Collect data on the level of cytokeratin 20 in urine at the time of diagnosis, recorded as biomarker B;

[0101] S513. Using multiple imputation methods, based on the existing data distribution, a statistical model is used to generate missing value estimates multiple times to obtain multiple complete data sets;

[0102] S514. A logistic regression prediction model is constructed based on clinical factors, and spline transformation features are introduced to capture nonlinear relationships. The model formula of the clinical model is as follows:

[0103]

[0104] Among them, Y = 1 means that the patient progresses to chronic kidney disease; x j is the jth clinical factor; β0 is the intercept; β j is the regression coefficient; is the variable x r The lth basis function after spline transformation; γ l are the spline coefficients, is the contribution of the linear combination of clinical factors, is the nonlinear contribution of clinical factors;

[0105] S515. Prediction probability based on clinical factors

[0106] S515. Construct a prediction model based on biomarker B. The nonlinear spline method is used to construct the model formula:

[0107]

[0108] Where B is the biomarker; Θ0 is the model intercept, Θ1 is the linear coefficient; Θ i+1 is the spline transformation truncation coefficient; ξi is the preset split node; n is the number of nodes; Θ1B is the biomarker linear contribution,

[0109] is the nonlinear contribution of the biomarker;

[0110] Biomarker risk prediction probability

[0111] S516, using clinical and biomarker models to obtain the predicted probability p C and p B , retain the intermediate features of spline transformation to construct interaction terms;

[0112] S517. Construct a joint meta-model to capture synergy effects. The meta-model formula is:

[0113] logitP(Y=1|p C , p B )=α0+α1p C +α2p B +α3(p C ×p B) where α0 is the intercept, α1 and α2 measure the independent contribution parameters; α3 captures the interaction effect parameter; the synergistic effect is captured by the interaction term pc×pB;

[0114] The final joint prediction probability expression is:

[0115]

[0116] S518, use K-fold cross validation to generate the first layer of predictions, and construct a prediction containing p based on the results. C 、p B The parameters α0, α1, α2, and α3 are determined by maximum likelihood estimation using the secondary training set and the true label Y. Nested cross-validation is used to ensure that tuning and evaluation are independent, and α0, α1, α2, and α3 are input to obtain the final joint prediction probability.

[0117] By generating missing value estimates multiple times, the original distribution characteristics of the data are retained, avoiding sample bias caused by deleting missing values. Spline transformation allows the relationship between clinical factors and chronic kidney disease progression to present a nonlinear pattern. The nonlinear spline model can capture the dose-response relationship between urinary cytokeratin 20 levels and chronic kidney disease progression. The joint meta-model is constructed by the interaction term (p C ×p B ) reveals the superimposed effect of clinical risk and biomarker risk. The outer K-fold cross-validation is used for model evaluation, and the inner cross-validation is used for parameter tuning to ensure that the model is completely independent during training and testing to avoid overfitting. The parameters α0, α1, α2, and α3 are determined by maximizing the likelihood function to ensure the best fitting effect of the model on the training data, and the generalization ability is verified by cross-validation. Multiple imputation is used to fill in missing data, combined with nonlinear spline transformation to capture complex relationships, and interaction term modeling is used to consider synergistic effects, and finally the robustness of the model is ensured by cross-validation. This method improves the accuracy of predicting the progression of acute kidney injury to chronic kidney disease, optimizes personalized risk assessment, makes it of great value in clinical applications, and also provides a reliable mathematical modeling method.

[0118] Example 4

[0119] In addition to the features of the above embodiment, this embodiment further defines: S53, using the Youden index method to determine the optimal cutoff value, by comprehensively considering the sensitivity and specificity of the model, finding the cutoff value that maximizes the sum of the two. The specific steps are as follows:

[0120] S531. Calculate the sensitivity and specificity of the model at different cutoff values;

[0121] S532. Calculate the Youden index using the following formula:

[0122] Among them, TP is the number of true positives, FN is the number of false negatives, TN is the number of true negatives, and FP is the number of false positives; is sensitivity; is specificity;

[0123] S533. Calculate the corresponding sensitivity and 1-specificity at different cutoff values, and plot the receiver operating characteristic curve with the false positive rate as the horizontal axis and the sensitivity as the vertical axis;

[0124] S534. On the receiver operating characteristic curve, find the point that maximizes the Youden index. The cutoff value corresponding to this point is the optimal cutoff value.

[0125] By calculating the sensitivity and specificity at different cutoff values, we can fully understand the performance of the model at different critical points. Emergency scenarios may give priority to high sensitivity to reduce missed diagnoses, while chronic disease screening may pay more attention to high specificity to reduce the cost of misdiagnosis. The Youden Index (J) quantifies the comprehensive ability of the model through a formula, with a value range of -1 to 1. The larger the J value, the better the overall effect of the model in reducing missed diagnoses and misdiagnoses. Plotting a curve with the false positive rate (1-specificity) as the horizontal axis and the sensitivity as the vertical axis can intuitively show the performance of the model at all possible cutoff values. The closer the curve is to the upper left corner (sensitivity 100% and false positive rate 0%), the better the model performance. The point with the largest Youden Index corresponds to the cutoff value with the largest sum of sensitivity and specificity, at which time the combined loss of missed diagnosis and misdiagnosis is the smallest.

[0126] Example 5

[0127] like Figure 2 As shown, in addition to the features of the above embodiment, this embodiment further defines the following specific steps for obtaining the urine cytokeratin 20 level during diagnosis:

[0128] S31. Urine sample from the patient within 24 hours after the diagnosis of acute kidney injury;

[0129] S32, centrifuging the urine sample for 8 to 15 minutes;

[0130] S33, collecting the supernatant after separation in the urine;

[0131] S34. Measure urine albumin and creatinine using an automatic analyzer, and normalize the concentrations of urine albumin and creatinine to urine creatinine, where the urine creatinine is the level of cytokeratin 20 in urine.

[0132] Collecting urine samples within 24 hours of the diagnosis of acute kidney injury maximizes the retention of early biomarker signals for renal tubular injury. Centrifugation (8-15 minutes) effectively removes cellular debris, red blood cells, and macromolecular impurities from the urine, preventing these components from interfering with the cytokeratin 20 assay. Normalizing urine albumin and creatinine concentrations to urine creatinine eliminates the effects of urine dilution or concentration on the test results. Rapidly obtaining cytokeratin 20 levels can help physicians identify high-risk patients early in the course of acute kidney injury.

[0133] Example 6

[0134] In addition to the features of the above embodiment, this embodiment further defines the following steps: S4, combining age, sex, baseline serum creatinine, the patient's urine albumin-creatinine ratio at the time of acute kidney injury diagnosis, serum creatinine level at discharge, and acute kidney injury stage adjusted restricted cubic splines to examine the nonlinear pattern of the association between urine cytokeratin 20 at diagnosis and the progression of acute kidney injury to chronic kidney disease through logistic regression:

[0135] S41. Perform data cleaning, remove missing values ​​and outliers, and perform interpolation. At the same time, use the density peak clustering method to identify outliers in the data, and use the Mahalanobis distance for secondary screening to avoid extreme values ​​interfering with the regression analysis;

[0136] S42. Perform Box-Cox or Yeo-Johnson transformation on the urinary cytokeratin 20 level to reduce the effect of skewness;

[0137] S43. Use decision tree or LightGBM method to perform nonlinear binning of urinary cytokeratin 20 levels to make it more consistent with the pattern of kidney disease progression;

[0138] S44. Calculate the mutual information between the urine cytokeratin 20 level and the patient's urine albumin-to-creatinine ratio at the time of diagnosis of acute kidney injury to form an interaction term. Calculate the ratio of baseline serum creatinine to serum creatinine level at discharge to reflect the degree of recovery. Calculate the interaction effect term between age and urine cytokeratin 20 level to account for the modulatory effect of age on biomarkers.

[0139] S45. Stratified logistic regression was used, with variables introduced stepwise. The first layer used age, sex, baseline serum creatinine, the patient's urine albumin-creatinine ratio at the time of AKI diagnosis, and the serum creatinine level at discharge as the base model. The second layer introduced restricted cubic splines adjusted for AKI stage. The third layer introduced urine cytokeratin 20 levels, and its contribution was first analyzed using a linear term. The fourth layer added a nonlinear term for urine cytokeratin 20 levels to test whether the nonlinear effect was significant. The fifth layer added interaction terms, and the likelihood ratio test was used to screen for significant variables.

[0140] S46, Acute Kidney Injury Stage-Adjusted Restricted Cubic Splines Modeling the Nonlinear Relationship Between Urinary Cytokeratin 20 Levels and the Progression of Acute Kidney Injury to Chronic Kidney Disease;

[0141] S47. Perform variable selection and model optimization. Use LASSO regression for variable screening to avoid overfitting while retaining the most predictive features. Use Bayesian model averaging and weighted averaging between different models to improve stability and construct logistic regression to enhance the generalization ability of the model.

[0142] Density peak clustering automatically identifies outliers by calculating the local density and distance of data points, avoiding the limitations of traditional distance- or distribution-based methods. Mahalanobis distance, combined with intervariate correlations, can effectively identify multivariate outliers and reduce the interference of extreme values ​​on regression analysis. Missing values ​​can be filled using multiple imputation or model prediction to preserve data information and avoid sample bias caused by data deletion. Box-Cox or Yeo-Johnson transformations can effectively correct the skewed distribution of CK20 levels, making it closer to a normal distribution and satisfying the linear relationship assumption of logistic regression. Decision trees or LightGBM automatically find the optimal binning points for CK20 by recursively partitioning the data. Mutual information analysis can quantify the nonlinear association between CK20 and the urine albumin-to-creatinine ratio. The baseline to discharge serum creatinine ratio (reflecting the degree of renal function recovery) can be used as an independent variable to supplement the assessment of the reversibility of acute kidney injury. The interaction term between age and CK20 can reveal differences in the predictive value of biomarkers across different age groups; for example, the cutoff value of CK20 may be lower in elderly patients. The introduction of hierarchical variables allows for a systematic assessment of the independent impact of various factors on CKD progression. Restricted cubic splines allow the relationship between CK20 and CKD progression to exhibit different slopes in different intervals. LASSO regression uses a penalty coefficient to compress the coefficients of irrelevant variables to zero, retaining core predictors and improving model parsimony.

[0143] Example 7

[0144] In addition to the features of the above embodiment, this embodiment further defines: S4, combining age, sex, baseline serum creatinine, the patient's urine albumin-creatinine ratio at the time of acute kidney injury diagnosis, serum creatinine level at discharge, and acute kidney injury stage after adjusting restricted cubic splines, and examining the nonlinear pattern of the association between urine cytokeratin 20 at diagnosis and the progression of acute kidney injury to chronic kidney disease through logistic regression, further comprising the following steps:

[0145] S6. The association between urinary cytokeratin 20 and severe acute tubular injury during acute kidney injury was examined using logistic regression with tertiles and logarithmic transformation of urinary cytokeratin 20.

[0146] Tertile division and logarithmic transformation reduced inter-individual measurement bias and the impact of extreme values, improving data stability. The use of a logistic regression model effectively controlled for confounding factors, enhancing the scientific nature and credibility of the analysis results.

[0147] Example 8

[0148] like Figure 4 As shown, in addition to the features of the above embodiment, this embodiment further defines the following specific steps for examining the association between urine cytokeratin 20 and severe acute renal tubular injury during acute kidney injury by using logistic regression and tertiles and logarithmic transformation of urine cytokeratin 20:

[0149] S61. Collect urine cytokeratin 20 levels from patients at the time of diagnosis of acute kidney injury. Calculate the 25th, 50th, and 75th percentiles based on the distribution of urine cytokeratin 20 levels in the samples and categorize the data into three levels: low, medium, and high.

[0150] S62. Logarithmic transformation was used to make the data conform to normal distribution, and the transformed continuous variables were recorded so that they could be analyzed as continuous independent variables in the regression model;

[0151] S63. The low, medium, and high scores were included in the model as categorical variables. Logistic regression was used to analyze the strength of association between each group and severe acute renal tubular injury. The odds ratio and its 95% confidence interval were calculated for each group.

[0152] S64. Examine the regression coefficients and P values ​​of the grouped or log-transformed variables in the urine cytokeratin 20 model to determine whether there is a statistically significant association between the variable and severe acute renal tubular injury.

[0153] S65. Based on the regression results, interpret the association between urinary cytokeratin 20 levels and severe acute renal tubular injury.

[0154] Dividing urine cytokeratin 20 levels into three equal parts, from smallest to largest, allows for intuitive visualization of risk differences within these intervals. Logarithmic transformation improves data normality, fulfills the linearity assumption of logistic regression, and avoids parameter estimation errors caused by data distribution bias. Incorporating covariates such as age, sex, baseline serum creatinine, and urine albumin-creatinine ratio allows for the isolation of the independent effect of urine cytokeratin 20 levels on SATI.

[0155] Example 9

[0156] In addition to the features of the above embodiment, this embodiment further defines that: S5, after the step of using the area under the receiver operating characteristic curve to evaluate and compare the discriminative performance of the risk model for predicting the progression of acute kidney injury to chronic kidney disease, the method further includes the following steps:

[0157] Construct a virtual patient cohort and simulate the effects of different intervention plans on the progression of acute kidney injury to chronic kidney disease through Monte Carlo simulation or proxy model. Use simulation to verify the adaptability and intervention effect of the prediction model in different scenarios, and provide theoretical basis and optimization suggestions for personalized treatment plans.

[0158] By constructing a virtual patient cohort, it is possible to train high-precision prediction models based on large sample data, thereby improving the reliability of disease progression predictions. Monte Carlo simulation can fully account for randomness and individual differences, making the prediction model closer to real clinical conditions and improving the accuracy of risk assessment. Proxy models can simulate disease progression under different intervention strategies, helping clinicians identify the optimal treatment plan. By simulating the treatment responses of different individuals, "one-size-fits-all" treatment plans can be avoided, providing a theoretical basis for personalized precision medicine. Early identification of high-risk patients and targeted intervention can help reduce the rate of acute kidney injury progressing to chronic kidney disease. By simulating the effects of interventions under different medical resource conditions, the allocation of medical resources can be optimized and the economic benefits of intervention strategies can be improved.

[0159] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0160] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the patent for this invention shall be determined by the appended claims.

Claims

1. A method for predicting the progression of acute kidney injury to chronic kidney disease, the method comprising: S1. Construct a predictive clinical model, wherein the clinical factors of the predictive clinical model include age, sex, baseline serum creatinine, the patient's urine albumin-to-creatinine ratio at the time of diagnosis of acute kidney injury, and the serum creatinine level at discharge; S2, obtain restricted cubic splines adjusted for acute kidney injury stage; S3. Obtain the urine cytokeratin 20 level at the time of diagnosis and construct a biomarker model based on the urine cytokeratin 20 level; S4. Logistic regression was used to examine the nonlinear pattern of the association between urine cytokeratin 20 at diagnosis and progression from acute kidney injury to chronic kidney disease using restricted cubic splines adjusted for age, sex, baseline serum creatinine, urine albumin-creatinine ratio at the time of acute kidney injury diagnosis, serum creatinine level at discharge, and acute kidney injury stage. S5. Use the area under the receiver operating characteristic curve to evaluate and compare the discriminative performance of models for predicting the risk of progression of acute kidney injury to chronic kidney disease.

2. The method for predicting progression of acute kidney injury to chronic kidney disease according to claim 1, wherein: The baseline serum creatinine level was the most recent measurement within 1 month before the patient's admission.

3. The method for predicting progression of acute kidney injury to chronic kidney disease according to claim 1, wherein: The specific steps of step S5 are as follows: S51. Analyze the association between predictive clinical models and biomarker models and construct models that combine clinical factors and biomarkers. S52. Use the area under the receiver operating characteristic curve to assess and compare the discriminative performance of predictive clinical models, biomarker-based models, and models that combine clinical factors and biomarkers. S53. Use the Youden index method to determine the optimal cutoff value. By comprehensively considering the sensitivity and specificity of the model, find the cutoff value that maximizes the sum of the two. S54. Evaluate the degree of model improvement. Using the model containing only clinical factors as a reference, calculate the net reclassification improvement index and comprehensive discrimination improvement index of the model containing a combination of clinical factors and biomarkers.

4. The method for predicting progression of acute kidney injury to chronic kidney disease according to claim 3, wherein: The specific steps of step S51 are as follows: S511. Collect clinical factors including age, sex, baseline serum creatinine, urine albumin-creatinine ratio of patients at the time of acute kidney injury diagnosis, and serum creatinine level at discharge, denoted as vector X = (x1, x2, x3, x4, x5, x6); S512. Collect data on the level of cytokeratin 20 in urine at the time of diagnosis, recorded as biomarker B; S513. Using multiple imputation methods, based on the existing data distribution, a statistical model is used to generate missing value estimates multiple times to obtain multiple complete data sets; S514. A logistic regression prediction model is constructed based on clinical factors, and spline transformation features are introduced to capture nonlinear relationships. The model formula of the clinical model is as follows: Among them, Y = 1 means that the patient progresses to chronic kidney disease; x j is the jth clinical factor; β0 is the intercept; β j is the regression coefficient; is the variable x r The lth basis function after spline transformation; γ l are the spline coefficients, is the contribution of the linear combination of clinical factors, is the nonlinear contribution of clinical factors; S515. Prediction probability based on clinical factors S515. Construct a prediction model based on biomarker B. The nonlinear spline method is used to construct the model formula: Where B is the biomarker; Θ0 is the model intercept, Θ1 is the linear coefficient; Θ i+1 is the spline transformation truncation coefficient; ξi is the preset split node; n is the number of nodes; Θ1B is the biomarker linear contribution, is the nonlinear contribution of the biomarker; Biomarker risk prediction probability S516, using clinical and biomarker models to obtain the predicted probability p C and p B , retain the intermediate features of spline transformation to construct interaction terms; S517. Construct a joint meta-model to capture synergy effects. The meta-model formula is: logitP(Y=1|p C ,p B )=α0+α1p C +α2p B +α3(p C ×p B ) Among them, α0 is the intercept, α1 and α2 measure the independent contribution parameters; α3 captures the interaction effect parameter; the synergistic effect is captured by the interaction term pc×pB; The final joint prediction probability expression is: S518, use K-fold cross validation to generate the first layer of predictions, and construct a prediction containing p based on the results. C 、p B The parameters α0, α1, α2, and α3 are determined by maximum likelihood estimation using the secondary training set and the true label Y. Nested cross-validation is used to ensure that tuning and evaluation are independent, and α0, α1, α2, and α3 are input to obtain the final joint prediction probability.

5. The method for predicting the progression of acute kidney injury to chronic kidney disease according to claim 3, wherein: The specific steps of step S53 are as follows: S531. Calculate the sensitivity and specificity of the model at different cutoff values; S532. Calculate the Youden index using the following formula: Among them, TP is the number of true positives, FN is the number of false negatives, TN is the number of true negatives, and FP is the number of false positives; is sensitivity; is specificity; S533. Calculate the corresponding sensitivity and 1-specificity at different cutoff values, and plot the receiver operating characteristic curve with the false positive rate as the horizontal axis and the sensitivity as the vertical axis; S534. On the receiver operating characteristic curve, find the point that maximizes the Youden index. The cutoff value corresponding to this point is the optimal cutoff value.

6. The method for predicting progression of acute kidney injury to chronic kidney disease according to claim 1, wherein: The specific steps for obtaining the urine cytokeratin 20 level at the time of diagnosis are as follows: S31. Urine sample from the patient within 24 hours after the diagnosis of acute kidney injury; S32, centrifuging the urine sample for 8 to 15 minutes; S33, collecting the supernatant after separation in the urine; S34. Measure urine albumin and creatinine using an automatic analyzer, and normalize the concentrations of urine albumin and creatinine to urine creatinine, where the urine creatinine is the level of cytokeratin 20 in urine.

7. The method for predicting progression of acute kidney injury to chronic kidney disease according to claim 1, wherein: The specific steps of the S4 step are as follows: S41. Perform data cleaning, remove missing values ​​and outliers, and perform interpolation. At the same time, use the density peak clustering method to identify outliers in the data, and use the Mahalanobis distance for secondary screening to avoid extreme values ​​interfering with the regression analysis; S42. Perform Box-Cox or Yeo-Johnson transformation on the urinary cytokeratin 20 level to reduce the effect of skewness; S43. Use decision tree or LightGBM method to perform nonlinear binning of urinary cytokeratin 20 levels to make it more consistent with the pattern of kidney disease progression; S44. Calculate the mutual information between the urine cytokeratin 20 level and the patient's urine albumin-to-creatinine ratio at the time of diagnosis of acute kidney injury to form an interaction term. Calculate the ratio of baseline serum creatinine to serum creatinine level at discharge to reflect the degree of recovery. Calculate the interaction effect term between age and urine cytokeratin 20 level to account for the modulatory effect of age on biomarkers. S45. Stratified logistic regression was used, with variables introduced stepwise. The first layer used age, sex, baseline serum creatinine, the patient's urine albumin-creatinine ratio at the time of AKI diagnosis, and the serum creatinine level at discharge as the base model. The second layer introduced restricted cubic splines adjusted for AKI stage. The third layer introduced urine cytokeratin 20 levels, and its contribution was first analyzed using a linear term. The fourth layer added a nonlinear term for urine cytokeratin 20 levels to test whether the nonlinear effect was significant. The fifth layer added interaction terms, and the likelihood ratio test was used to screen for significant variables. S46, Acute Kidney Injury Stage-Adjusted Restricted Cubic Splines Modeling the Nonlinear Relationship Between Urinary Cytokeratin 20 Levels and the Progression of Acute Kidney Injury to Chronic Kidney Disease; S47. Perform variable selection and model optimization. Use LASSO regression for variable screening to avoid overfitting while retaining the most predictive features. Use Bayesian model averaging and weighted averaging between different models to improve stability and construct logistic regression to enhance the generalization ability of the model.

8. The method for predicting progression of acute kidney injury to chronic kidney disease according to claim 1, wherein: The S4 step further includes the following steps: S6. The association between urinary cytokeratin 20 and severe acute tubular injury during acute kidney injury was examined using logistic regression with tertiles and logarithmic transformation of urinary cytokeratin 20.

9. The method for predicting progression of acute kidney injury to chronic kidney disease according to claim 8, characterized in that: The specific steps of step S6 are as follows: S61. Collect urine cytokeratin 20 levels from patients at the time of diagnosis of acute kidney injury. Calculate the 25th, 50th, and 75th percentiles based on the distribution of urine cytokeratin 20 levels in the samples and categorize the data into three levels: low, medium, and high. S62. Logarithmic transformation was used to make the data conform to normal distribution, and the transformed continuous variables were recorded so that they could be analyzed as continuous independent variables in the regression model; S63. The low, medium, and high scores were included in the model as categorical variables. Logistic regression was used to analyze the strength of association between each group and severe acute renal tubular injury. The odds ratio and its 95% confidence interval were calculated for each group. S64. Examine the regression coefficients and P values ​​of the grouped or log-transformed variables in the urine cytokeratin 20 model to determine whether there is a statistically significant association between the variable and severe acute renal tubular injury. S65. Based on the regression results, interpret the association between urinary cytokeratin 20 levels and severe acute renal tubular injury.

10. The method for predicting progression of acute kidney injury to chronic kidney disease according to claim 1, wherein: The step S5 further includes the following steps: Construct a virtual patient cohort and simulate the effects of different intervention plans on the progression of acute kidney injury to chronic kidney disease through Monte Carlo simulation or proxy model. Use simulation to verify the adaptability and intervention effect of the prediction model in different scenarios, and provide theoretical basis and optimization suggestions for personalized treatment plans.

Images