Nighttime hypertension risk prediction method, system, device, storage medium and product

Abstract

This invention relates to the field of medical artificial intelligence technology, and discloses a method, system, device, storage medium, and product for predicting the risk of nocturnal hypertension. It employs single-factor analysis combined with clinical relevance to screen several significant research factors. These significant research factors are then input into a nocturnal hypertension risk prediction model for prediction, which reduces fitting risk and noise, improves prediction accuracy, and achieves an optimal balance between predictive performance and clinical operability, facilitating rapid application in busy clinical environments. The nocturnal hypertension risk prediction model of this invention uses a table diffusion model, which can effectively capture the nonlinear relationships between clinical variables, and its predictive performance is significantly better than that of traditional logistic regression models. This invention also uses survival analysis for validation, which not only examines the model's generalization ability and shelf life over time but also effectively corrects for survivor bias caused by time camouflage.

Description

Technical Field

[0001] This invention relates to the field of medical artificial intelligence technology, and in particular to a method, system, device, storage medium, and product for predicting the risk of nighttime hypertension. Background Technology

[0002] Chronic kidney disease (CKD) has become a global public health challenge, with a very high prevalence of hypertension among its patients. Nocturnal hypertension (NH) is particularly common in CKD patients and is an independent risk factor that significantly increases the risk of kidney damage, cardiovascular events, and even death. Therefore, early identification and effective management of NH are crucial for improving the prognosis of CKD patients.

[0003] Currently, the gold standard for diagnosing nocturnal emission (NH) is 24-hour ambulatory blood pressure monitoring (ABPM). However, the application of ABPM in clinical practice is subject to many limitations, including limited device availability, high testing costs, potential disruption to patient sleep quality, inaccurate readings during daily activities, and patient discomfort. These limitations constitute a significant healthcare burden, especially for CKD patients who require frequent blood pressure monitoring to guide treatment adjustments. Therefore, there is an urgent clinical need for a practical tool based on readily available and routine clinical data to accurately predict NH risk, optimize antihypertensive medication adjustments, and reduce reliance on repeated ABPM.

[0004] Although various hypertension risk prediction models have been developed, most are not specifically designed and validated for CKD populations, or have limited predictive performance. Traditional statistical models, such as logistic regression, while computationally simple and easy to interpret, often fail to capture the complex, nonlinear interactions prevalent in clinical data. On the other hand, while modern deep learning architectures (such as convolutional neural networks (CNNs) and transformers) have achieved great success in image and text processing, they are not designed for handling the high-dimensional tabular data common in clinical practice, limiting their direct applicability to such tasks. Therefore, developing a customized NH prediction model that can effectively handle high-dimensional clinical data and provide superior predictive performance is a pressing technical challenge in the field. Summary of the Invention

[0005] The purpose of this invention is to provide a method, system, device, storage medium, and product for predicting the risk of nocturnal hypertension that has high prediction accuracy, strong clinical applicability, and is robust and reliable.

[0006] To achieve the above objectives, the present invention provides the following solution:

[0007] In a first aspect, the present invention provides a method for predicting the risk of nocturnal hypertension, comprising: Obtain clinical datasets of non-dialysis chronic kidney disease patients; Based on the clinical dataset, several significant research factors were identified using single-factor analysis combined with clinical relevance. A training dataset is generated from the clinical dataset based on the aforementioned significant research factors; A risk prediction model for nocturnal hypertension based on a table value diffusion model is constructed. The risk prediction model for nocturnal hypertension takes the significant research factors as input and outputs the predicted risk probability value of developing nocturnal hypertension. The training dataset is preprocessed, including forward noise addition, to obtain a preprocessed training dataset; the nighttime hypertension risk prediction model is trained based on the preprocessed training dataset to obtain a trained nighttime hypertension risk prediction model. The system receives significant research factors of the patient to be tested, inputs these factors into a trained nocturnal hypertension risk prediction model, and outputs the risk probability value of the patient to be tested developing nocturnal hypertension. Based on the risk probability value and a preset threshold, the system outputs the prediction result.

[0008] Secondly, the present invention provides a nighttime hypertension risk prediction system, comprising: The data acquisition module is used to acquire clinical datasets of non-dialysis chronic kidney disease patients. The predictor selection module is used to screen out several significant research factors based on the clinical dataset by using single-factor analysis and combining clinical relevance. A training dataset construction module is used to generate a training dataset from the clinical dataset based on the significant research factors. The model building module is used to construct a nighttime hypertension risk prediction model based on a table-valued diffusion model. The nighttime hypertension risk prediction model takes the significant research factors as input and outputs the predicted risk probability value of developing nighttime hypertension. The training dataset is preprocessed, including forward noise addition, to obtain a preprocessed training dataset. The nighttime hypertension risk prediction model is trained based on the preprocessed training dataset to obtain a trained nighttime hypertension risk prediction model. The prediction module is used to receive the significant research factors of the patient to be tested, input the significant research factors of the patient to be tested into the trained nocturnal hypertension risk prediction model, output the risk probability value of the patient to be tested developing nocturnal hypertension, and output the prediction result based on the risk probability value and a preset threshold.

[0009] Thirdly, the present invention provides an electronic device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the nighttime hypertension risk prediction method as described above.

[0010] Fourthly, the present invention provides a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the above-described method for predicting the risk of nocturnal hypertension.

[0011] Fifthly, the present invention provides a computer program product comprising a computer program that, when executed by a processor, implements the steps of the above-described method for predicting the risk of nocturnal hypertension.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention employs single-factor analysis combined with clinical relevance to screen several significant research factors. These significant factors are then input into a nocturnal hypertension risk prediction model, reducing the risk of fitting and noise, improving prediction accuracy, and achieving an optimal balance between predictive performance and clinical operability, facilitating rapid application in busy clinical environments. The nocturnal hypertension risk prediction model of this invention uses a table diffusion model, which effectively captures the nonlinear relationships between clinical variables, and its predictive performance is significantly superior to traditional logistic regression models. This invention also uses survival analysis for validation, which not only examines the model's generalization ability and shelf life over time but also effectively corrects for survivor bias caused by time camouflage. Attached Figure Description

[0013] Figure 1 This is a flowchart of the nighttime hypertension risk prediction method according to an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the results of screening significant research factors according to an embodiment of the present invention; Figure 3 The diagram shows the ROC curves, calibration curves, and decision curve (DCA) analysis of the nocturnal hypertension risk prediction model in the replication cohort and external validation cohort of this invention. Figure 4 This is a schematic diagram of a decision tree for risk classification based on model prediction probability in an embodiment of the present invention; Figure 5 This is a Kaplan-Meier survival curve for patients with different risk levels who experienced major adverse cardiovascular events (MACCE) and major adverse renal events (MAKE). Figure 6This is a block diagram illustrating the principle of the nighttime hypertension risk prediction system according to an embodiment of the present invention. Figure 7 This is a schematic diagram of the hardware operating environment involved in the nighttime hypertension risk prediction method in this embodiment of the invention. Detailed Implementation

[0014] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0015] Example 1 like Figure 1 As shown, a preferred embodiment of the present invention provides a method for predicting the risk of nocturnal hypertension, comprising: S1. Obtain clinical datasets of non-dialysis chronic kidney disease patients; S2. Based on the clinical dataset, use single-factor analysis and combine it with clinical relevance to screen out several significant research factors; S3. Generate a training dataset from the clinical dataset based on the significant research factors; S4. Construct a nighttime hypertension risk prediction model based on the table value diffusion model. The nighttime hypertension risk prediction model takes the significant research factors as input and outputs the predicted risk probability value of developing nighttime hypertension. S5. Preprocess the training dataset, including forward noise addition, to obtain a preprocessed training dataset; train the nighttime hypertension risk prediction model based on the preprocessed training dataset to obtain a trained nighttime hypertension risk prediction model. S6. Receive the significant research factors of the patient to be tested, input the significant research factors of the patient to be tested into the trained nocturnal hypertension risk prediction model, and output the risk probability value of the patient to be tested developing nocturnal hypertension; output the prediction result based on the risk probability value and the preset threshold.

[0016] This embodiment employs single-factor analysis combined with clinical relevance to screen several significant research factors. These significant factors are then input into the nocturnal hypertension risk prediction model for prediction. This reduces the risk of fitting and noise, improves prediction accuracy, and achieves an optimal balance between predictive performance and clinical operability, facilitating rapid application in busy clinical environments. The nocturnal hypertension risk prediction model of this invention uses a table-valued diffusion model, which effectively captures the nonlinear relationships between clinical variables, and its predictive performance is significantly superior to traditional logistic regression models.

[0017] Specifically, the clinical dataset in step S1 contains multiple research factors and corresponding nocturnal hypertension labels, which are binary labels of "present" or "absent" nocturnal hypertension determined based on ABPM results.

[0018] The study subjects in this embodiment are non-dialysis chronic kidney disease patients. Various research factors for the study of nocturnal hypertension were obtained based on the study subject set. These research factors include the subjects' demographic characteristics (such as age, sex, BMI, etc.), lifestyle characteristics (such as smoking, alcohol consumption, etc.), comorbidities (such as cardiovascular disease, diabetes, etc.), previous use of antihypertensive medications, and clinical characteristics (such as glomerular filtration rate, creatinine level, etc.).

[0019] In the definition of sample labels in the clinical dataset, the "label" (Ground Truth) used in the nocturnal hypertension risk prediction model of this embodiment is strictly defined as: presence or absence of nocturnal hypertension. Based on the gold standard ABPM data, if the patient's average systolic blood pressure during sleep is ≥120 mmHg or diastolic blood pressure is ≥70 mmHg, it is marked as "Positive / 1"; otherwise, it is marked as "Negative / 0".

[0020] After obtaining the clinical dataset of non-dialysis chronic kidney disease patients as described in step S1, multiple imputation is used to fill in missing values ​​for research factors with a missing rate higher than a preset threshold in the clinical dataset, resulting in a completed clinical dataset. If the missing rate of a significant research factor in the clinical dataset is high, multiple imputation is used to fill in the missing values. The preset threshold for the missing rate is 20%. In step S2, based on the completed clinical dataset, single-factor analysis is used in conjunction with clinical relevance to screen out several significant research factors.

[0021] Before using single-factor analysis to screen for significant research factors from all the aforementioned research factors, this embodiment also includes comparing the clinical characteristics and outcomes (outcomes refer to the long-term prognosis of the research subjects, including renal outcomes, cardiovascular outcomes, etc.) of nocturnal hypertension and non-nocturnal hypertension in the research subject set to achieve clinical relevance analysis. By comparing the nocturnal hypertension and non-nocturnal hypertension of the research subjects, it is possible to quickly determine whether there is an original distribution difference of the research factor between the two groups, achieving rapid pre-screening for single-factor analysis. If a research factor has no difference in distribution between the nocturnal hypertension and non-nocturnal hypertension groups (extremely large p-value), then it is almost impossible for it to be significant in single-factor logistic regression. Eliminating these research factors in advance can significantly reduce invalid calculations and reduce the risk of false positives from multiple comparisons. Furthermore, the results of clinical relevance analysis can be used as an aid. If single-factor analysis identifies a research factor as significant, but the mean of this research factor is almost the same in the nocturnal hypertension and non-nocturnal hypertension groups, then be wary of spurious significance and avoid the possibility of confounding or extreme value driving factors.

[0022] Single-factor analysis was used to screen out significant research factors from all research factors, specifically including: Single-factor analysis is performed on all research factors to assess the importance of each research factor. All research factors are then ranked in descending order of importance to obtain a research factor sequence. The top predetermined number of research factors in the research factor sequence are extracted to obtain a research factor set. The research factors in the intersection of the research factors are the significant research factors.

[0023] Specifically, in the process of screening significant research factors, single-factor analysis assesses the importance of the included candidate research factors based on the p-value, and finally screens significant research factors according to the importance of the research factors. For example, the top 25 most important research factors in single-factor analysis are listed to form a research factor set, and the research factors that rank higher in this research factor set are selected as significant research factors for the next step of constructing a nocturnal hypertension risk prediction model.

[0024] One-factor analysis is a statistical analysis method. In one-factor analysis, the linear relationship between each independent variable and the dependent variable is measured by the p-value; compared with multivariate analysis, one-factor analysis can more directly assess the impact of each independent variable on the dependent variable. Figure 2 The results show the results of using single-factor analysis to screen out significant research factors.

[0025] In step S2, the significant study factors include age, body mass index (BMI), office systolic blood pressure, office diastolic blood pressure, estimated glomerular filtration rate (eGFR), history of hypertension, and history of use of non-renin-angiotensin-aldosterone system (nRAAS) antihypertensive drugs.

[0026] In step S3, after screening out the significant research factors, the samples in the generated training dataset include significant research factors and binary labels of "present" or "absent" nocturnal hypertension.

[0027] Therefore, the input to the nocturnal hypertension risk prediction model constructed in step S4 is the seven selected significant research factors: age, BMI, office systolic blood pressure, office diastolic blood pressure, estimated glomerular filtration rate (eGFR), history of hypertension (binary classification), and history of nRAAS medication use (binary classification). The final output of the nocturnal hypertension risk prediction model is a probability value representing the likelihood that the patient has nocturnal hypertension. The prediction process of the nocturnal hypertension risk prediction model is as follows: the model receives the seven input variables, generates a distribution of nocturnal hypertension status matching the input conditions through a backdiffusion process in the latent space, and finally maps it to a specific risk probability through a classification head.

[0028] In step S5, since the tabular data contains both numerical and categorical data, the clinical dataset is preprocessed to address its heterogeneity. Preprocessing includes encoding and noise addition. Encoding includes standardizing numerical columns (e.g., blood pressure, age) and one-hot encoding of categorical columns (e.g., nRAAS usage history). Specifically, preprocessing also includes normalizing numerical variables and one-hot encoding categorical variables. Numerical variables include age, BMI, office systolic blood pressure, office diastolic blood pressure, and estimated glomerular filtration rate (eGFR), while categorical variables include history of hypertension and nRAAS medication usage history. Noise addition involves uniformly adding Gaussian noise to both the processed numerical columns and the encoded categorical columns during the forward process of the nighttime hypertension risk prediction model. Although categorical variables are inherently discrete, mapping them to a continuous space and applying Gaussian noise during the diffusion process of this model effectively simulates the potential distribution changes of the data. Specifically, in the forward noise addition process, Gaussian noise is iteratively added to the unified tabular data representation until it approximates a pure Gaussian noise distribution.

[0029] In some embodiments, step S5 divides the training dataset into a training set and an internal validation set according to a preset ratio, which can be 8:2. The nighttime hypertension risk prediction model is trained using significant research factors as variables. This model, as an advanced generative artificial intelligence technique based on a table-value diffusion model, learns the latent distribution of data through two processes: a "forward process" and a "backward process." Specifically: Forward process (noise addition): In this process, Gaussian noise is gradually added to the original tabular clinical data (the encoded latent representation) through a series of steps (time steps $t$) until the data eventually becomes pure random noise. This process is fixed and requires no learning.

[0030] Reverse Process (Denoising): This process is central to model learning. A deep neural network (e.g., a UNet architecture suitable for tabular data) is constructed to predict the noise added during the forward process, given a time step $t$ and conditional information (i.e., the patient's seven clinical variables). By iteratively and progressively subtracting the predicted noise from the pure noise, the model can "reverse" the noise-addition process, ultimately generating target data (i.e., the probabilities of the NH states) that matches the input conditions and conforms to the true data distribution.

[0031] The nocturnal hypertension risk prediction model described in this embodiment employs a CycleGAN architecture and incorporates privileged information for training. The model is optimized by minimizing a composite loss function, which includes a classification loss function, an adversarial loss function, and an alignment loss function. The training of this nocturnal hypertension risk prediction model using a CycleGAN architecture combined with privileged information significantly suppresses the generation of illusions by utilizing unique high-dimensional information during training, while maintaining the convenience of an unsupervised / unpaired architecture. Furthermore, privileged knowledge is encoded into the network weights, enabling the generation of outputs with high semantic fidelity using only low-dimensional inputs during testing.

[0032] The composite loss function for training the nocturnal hypertension risk prediction model is:

[0033] in, It is a composite loss function; For classification loss, the cross-entropy loss function is used to measure the difference between the model's output prediction probability of nighttime hypertension and the true label. To combat the adversarial loss, a binary cross-entropy objective based on the CycleGAN framework is used to ensure the consistency between the generated data distribution and the real data distribution. For Alignment Loss, Mean Squared Error (MSE) is used to ensure that the generated features are structurally consistent with the original input features.

[0034] During the model training phase, if the evaluation results (such as AUC, accuracy) are not ideal, the present invention adopts the following mechanism for adjustment: (1) Optimize the composite loss function: the model does not only optimize a single objective, but jointly minimizes the classification loss ( ), combating losses ( ) and alignment loss If the model's predictions are inaccurate (poor classification performance), increase... The weights. If the feature distribution generated by the model is not realistic (poor generalization ability), then... (Introducing a discriminator based on the CycleGAN architecture) involves adversarial training for adjustment. If the generated features deviate from the original input, [the process is modified / adjusted]. Forced feature alignment. (2) Adjust the architecture: Introduce the "hallucination branch" and privileged information strategies to use more auxiliary information to guide the model learning during training and to shield this information during testing, thereby improving robustness.

[0035] After obtaining the trained nocturnal hypertension risk prediction model, the method further includes validating the model's performance on at least one independent external validation cohort to assess its generalization ability. This invention uses at least one or more of the following performance evaluation metrics to evaluate the model: receiver operating characteristic (ROC) curve analysis, calibration curve analysis, and decision curve analysis (DCA).

[0036] The nocturnal hypertension risk prediction model was internally validated using significant research factors of the subjects in the internal validation set as variables to verify its performance. The nocturnal hypertension risk prediction model was externally validated using significant research factors of the subjects in the independent external validation cohort as variables to verify its generalization.

[0037] This embodiment incorporates the significant research factors selected in step S2 to construct the nighttime hypertension risk prediction model, and evaluates its predictive performance using the AUC value. The nighttime hypertension risk prediction model is used to explore potential nonlinear relationships and interactions among the significant research factors.

[0038] In this embodiment, based on the screening results of significant research factors in step S2 above, the present invention selected seven significant research factors for the table value diffusion model to construct a nighttime hypertension risk prediction model: outpatient systolic blood pressure, outpatient diastolic blood pressure, eGFR, BMI, history of nRAAS drug use, history of hypertension, and age.

[0039] The performance evaluation of the nighttime hypertension risk prediction model is specifically as follows: Based on the nighttime hypertension risk prediction model, the ROC curve analysis method was used to analyze the internal validation set and the external validation set respectively, and the corresponding training set ROC curve, internal validation set ROC curve and external validation set ROC curve were obtained. Calculate the area under the ROC curve of the internal validation set and the area under the ROC curve of the external validation set respectively, and obtain the area under the ROC curve of the internal validation set and the area under the ROC curve of the external validation set respectively. The predictive performance of the nighttime hypertension risk prediction model is evaluated based on the area under the ROC curve of the internal validation set and the area under the ROC curve of the external validation set.

[0040] Specifically, performance evaluation compares the predictive performance of the nighttime hypertension risk prediction model constructed using the table diffusion model with the area under the curve (AUC) value, and uses Hosmer data to evaluate the predictive performance. The Lemeshow test was used to evaluate the fit of the established nocturnal hypertension risk prediction model (i.e., whether the expected probability and the actual probability fit each other).

[0041] Based on the nighttime hypertension risk prediction model, the calibration curve analysis method is used to analyze the internal validation set and the external validation set respectively, and the corresponding internal validation set calibration curve and external validation set calibration curve are obtained. The consistency between the predicted values ​​and actual observed values ​​of the nocturnal hypertension risk prediction model is measured using the internal validation set calibration curve and the external validation set calibration curve. Based on the nighttime hypertension risk prediction model, the decision curve analysis method is used to analyze the internal validation set and the external validation set respectively, and the corresponding decision curves of the internal validation set and the external validation set are obtained. The overall net benefit of the nighttime hypertension risk prediction model is measured based on the decision curves of the internal validation set and the decision curves of the external validation set.

[0042] Specifically, the performance evaluation aims to measure the ability of the constructed nocturnal hypertension risk prediction model to predict nocturnal hypertension in patients. The method involves using patients diagnosed with nocturnal hypertension in the study subjects within the nocturnal hypertension risk prediction model constructed using a table value diffusion model. The consistency between the predicted values ​​of the table value diffusion model and the actual observed values ​​is observed. A calibration curve is used to measure the consistency between the predicted values ​​and the actual observed values. Decision curve analysis (DCA) is used to measure the overall net benefit of the constructed nocturnal hypertension risk prediction model.

[0043] The performance evaluation of the nocturnal hypertension risk prediction model in this invention essentially involves using ROC curve, calibration curve, and decision curve analysis methods to evaluate the goodness of fit, accuracy, and applicability of the model in the training set, internal validation set, and external validation set. The ROC curve, calibration curve, and decision curve (DCA) analysis of the nocturnal hypertension risk prediction model in the replication cohort and external validation cohort of this embodiment are shown in the figure below. Figure 3 As shown.

[0044] Further, step S6, which involves outputting the prediction result based on the risk probability value and the preset threshold, includes: classifying the risk level based on the risk probability value and the preset threshold, and outputting the risk level of the patient to be tested for nocturnal hypertension. In this embodiment, the preset threshold is a risk cutoff value determined using decision tree analysis. Based on the risk prediction result output by the nocturnal hypertension risk prediction model, decision tree analysis is used to determine the risk cutoff value, classifying the patient into high-risk, medium-risk, and low-risk risk levels. Figure 4 As shown in the diagram, this embodiment uses decision tree analysis to classify patients based on the risk probability value output by the nocturnal hypertension risk prediction model.

[0045] Therefore, after the nocturnal hypertension risk prediction model outputs the risk probability value, the prediction method further includes: using a classification tree to divide the model risk score into high-risk, intermediate-risk, and low-risk categories; establishing a KM curve between risk grouping and poor prognosis; and utilizing the real prognosis to assess the model's ability to differentiate patients, such as... Figure 5 As shown. Figure 5Throughout the study, at any given time point (0–60 months), the survival probability was consistently lowest in the high-risk group, highest in the low-risk group, and intermediate in the middle. Furthermore, the three curves showed no intersection or overlap, demonstrating that the scoring system stably and monotonously divides patients into three groups with drastically different prognoses, effectively widening the risk gradation. Moreover, from the short term (12 months) to the long term (60 months), the survival curve for the high-risk group exhibited a "continuously steep decline," and the decay rate of the "Number at risk" for each group perfectly matched the risk gradation (the high-risk group experienced the fastest decay). This indicates that the model's predictive efficacy is not limited to the early stages but remains effective throughout the 5-year follow-up period. Therefore, this embodiment, using the risk probability values ​​output by the nocturnal hypertension risk prediction model and applying decision tree analysis to determine the risk cutoff value for risk gradation, demonstrates excellent discriminative ability for major adverse cardiovascular events (MACCE) and major adverse renal events (MAKE), with clear stratified gradients and long-lasting effectiveness.

[0046] Furthermore, this embodiment verifies the clinical utility of the risk stratification using survival analysis methods, and adjusts the preset threshold based on the clinical utility. The clinical utility of the risk stratification is verified through survival analysis by demonstrating statistically significant differences in the incidence of adverse clinical outcomes (e.g., major adverse cardiovascular events or major adverse renal events) among patients of different risk levels. If the actual incidence of adverse events in the high-risk group is significantly higher than that in the low-risk group, the clinical effectiveness of the model's risk stratification strategy is confirmed.

[0047] In this invention, survival analysis does not directly participate in model parameter training or backpropagation; its role is to verify the clinical utility of the model's output. After the model outputs risk probabilities, it is divided into high, intermediate, and low-risk groups using a decision tree. Survival analysis is used to track the occurrence of cardiovascular (MACCE) and renal (MAKE) adverse events in these groups in the real world. The conclusions of the survival analysis (e.g., "the risk in the high-risk group is significantly higher than that in the low-risk group") are used to adjust the risk stratification threshold. For example, if the survival curve shows no difference in prognosis between the intermediate-risk and high-risk groups, it indicates that the model's risk stratification cut-off value needs to be recalibrated to ensure that the stratification results are discriminative of clinical outcomes.

[0048] Example 2 like Figure 6 As shown, a preferred embodiment of the present invention provides a nighttime hypertension risk prediction system, comprising: Data acquisition module 610 is used to acquire clinical datasets of non-dialysis chronic kidney disease patients; The predictor selection module 620 is used to screen out several significant research factors based on the clinical dataset by using single-factor analysis and combining clinical relevance. Training dataset construction module 630 is used to generate a training dataset from the clinical dataset based on the significant research factors; The model building module 670 is used to build a nighttime hypertension risk prediction model based on a table-value diffusion model. The nighttime hypertension risk prediction model takes the significant research factors as input and outputs the predicted risk probability value of nighttime hypertension. The training dataset is preprocessed, including forward noise addition, to obtain a preprocessed training dataset. The nighttime hypertension risk prediction model is trained based on the preprocessed training dataset to obtain a trained nighttime hypertension risk prediction model. The prediction module 650 is used to receive significant research factors of the patient to be tested, input the significant research factors of the patient to be tested into a trained nocturnal hypertension risk prediction model, output the risk probability value of the patient to be tested developing nocturnal hypertension, and output the prediction result based on the risk probability value and a preset threshold.

[0049] Example 3 like Figure 7 As shown, this embodiment of the invention also provides an electronic device, the device including: a memory 706, a processor 705, and a computer program stored in the memory 706 and executable on the processor 705. The computer program is configured to implement the steps of the above-described method for predicting the risk of nocturnal hypertension, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0050] For details, see Figure 7 The present invention also provides an electronic device, including a bus 701, a transceiver 702, an antenna 703, a bus interface 704, a processor 705, and a memory 706.

[0051] The transceiver 702 is used to acquire unstructured data, which includes at least one of data obtained based on user input information and data obtained based on configuration file scanning. The processor 705 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 705 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 705 performs the various methods and processes described above, such as a nocturnal hypertension risk prediction method.

[0052] exist Figure 7In this document, a bus architecture (represented by bus 701) is used. Bus 701 can include any number of interconnected buses and bridges, linking various circuits including one or more processors represented by processor 705 and memory represented by memory 706. Bus 701 can also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. Bus interface 704 provides an interface between bus 701 and transceiver 702. Transceiver 702 can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by processor 705 is transmitted over a wireless medium via antenna 703, which further receives data and transmits data to processor 705.

[0053] Processor 705 manages bus 701 and general processing, and also provides various functions, including timing, peripheral interface, voltage regulation, power management, and other control functions. Memory 706 can be used to store data used by processor 705 during operation.

[0054] Optionally, the processor 705 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a complex programmable logic device (CPLD).

[0055] Example 4 This invention also provides a storage medium, which is a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the steps of the above-described method for predicting the risk of nighttime hypertension and achieves the same technical effect. To avoid repetition, it will not be described again here. The computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0056] Example 5 This invention also provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the steps of the above-described method for predicting the risk of nocturnal hypertension, achieving the same technical effect. To avoid repetition, it will not be described again here.

[0057] In summary, this invention acquires clinical data from non-dialysis chronic kidney disease (CKD) patients, using ambulatory blood pressure monitoring (ABPM) results as a label for the presence or absence of nocturnal hypertension. Seven core significant research factors are identified through single-factor analysis and clinical expert knowledge screening. A nocturnal hypertension risk prediction model based on a table-value diffusion model is constructed and trained using a CycleGAN architecture. Numerical and categorical variables are encoded and uniformly subjected to Gaussian noise for forward diffusion. A reverse denoising process is used to establish the mapping relationship between clinical variables and nocturnal hypertension risk. During model training, a composite loss function including classification loss, adversarial loss, and alignment loss is used for optimization. The final model takes seven conventional variables as input and outputs the predicted probability of nocturnal hypertension. This invention employs an advanced diffusion model combined with a concise variable set containing only seven core predictors, demonstrating significantly higher predictive accuracy than traditional logistic regression models and other machine learning models in both external validation and repeated measures validation. This invention also combined survival analysis to verify the prognostic value of the model's risk stratification, confirming that the high-risk group was significantly associated with real-world adverse cardiovascular and renal events, providing a low-cost, high-precision tool for accurate clinical triage.

[0058] This invention employs a table-value diffusion model for predicting the risk of nocturnal hypertension, which effectively captures the nonlinear relationships between clinical variables. Its predictive performance (measured by the area under the ROC curve, AUC) significantly outperforms traditional logistic regression models in both internal validation and external testing cohorts. The model requires only seven clinically routine and readily available variables (age, BMI, office systolic / diastolic blood pressure, eGFR, history of hypertension, and history of nRAAS medication use), achieving an optimal balance between predictive performance and clinical operability, facilitating rapid application in busy clinical settings. Validation of this invention in external cohorts from different medical centers demonstrates the model's good generalization ability. This invention not only predicts the risk of nocturnal hypertension (NH) but also correlates risk stratification with real-world adverse patient outcomes (major adverse cardiovascular event MACCE and major adverse renal event MAKE). Results show that patients assessed as high-risk by the model have a significantly higher risk of adverse events, providing direct and robust prognostic information for clinical decision-making and helping to guide physicians in more proactive interventions and monitoring of high-risk patients.

[0059] This invention, through systematic development and rigorous multi-dimensional validation, provides a novel method and system for predicting nocturnal hypertension based on a 7-variable diffusion model. Utilizing advanced generative AI technology, this invention achieves superior predictive performance with minimal clinical input, and its significant clinical application value is confirmed through correlation with real-world prognostic data. This intelligent tool can effectively assist physicians in individualized blood pressure management, potentially improving clinical outcomes for CKD patients while conserving medical resources.

[0060] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of the present invention is not limited to performing functions in the order discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0061] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0062] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of the present invention.

Claims

1. A method for predicting the risk of nocturnal hypertension, characterized in that, include: Obtain clinical datasets of non-dialysis chronic kidney disease patients; Based on the clinical dataset, several significant research factors were identified using single-factor analysis combined with clinical relevance. A training dataset is generated from the clinical dataset based on the aforementioned significant research factors; A risk prediction model for nocturnal hypertension based on a table value diffusion model is constructed. The risk prediction model for nocturnal hypertension takes the significant research factors as input and outputs the predicted risk probability value of developing nocturnal hypertension. The training dataset is preprocessed, including forward noise addition, to obtain a preprocessed training dataset. The nighttime hypertension risk prediction model is trained based on the preprocessed training dataset to obtain a trained nighttime hypertension risk prediction model. The system receives significant research factors of the patient to be tested, inputs these factors into a trained nocturnal hypertension risk prediction model, and outputs the risk probability value of the patient to be tested developing nocturnal hypertension. Based on the risk probability value and a preset threshold, the system outputs the prediction result.

2. The method according to claim 1, characterized in that, The significant factors included age, body mass index, office systolic blood pressure, office diastolic blood pressure, estimated glomerular filtration rate, history of hypertension, and history of use of non-renin-angiotensin-aldosterone system antihypertensive drugs.

3. The method according to claim 1, characterized in that, After obtaining the clinical dataset of non-dialysis chronic kidney disease patients, multiple imputation is used to fill in the missing values ​​of research factors with a missing rate higher than a preset threshold in the clinical dataset to obtain the completed clinical dataset. Based on the completed clinical dataset, single factor analysis is used in combination with clinical relevance to screen out several significant research factors.

4. The method according to claim 1, characterized in that, The nighttime hypertension risk prediction model is trained using the CycleGAN architecture and combined with privileged information, and the model is optimized by minimizing a composite loss function, which includes a classification loss function, an adversarial loss function, and an alignment loss function.

5. The method according to claim 1, characterized in that, The step of outputting a prediction result based on the risk probability value and a preset threshold includes: Based on the risk probability value and the preset threshold, the risk level is classified, and the risk level of the patient to be tested for nocturnal hypertension is output.

6. The method according to claim 5, characterized in that, The clinical utility of the risk level classification is verified using survival analysis methods, and the preset threshold is adjusted based on the clinical utility.

7. A nighttime hypertension risk prediction system, characterized in that, include: The data acquisition module is used to acquire clinical datasets of non-dialysis chronic kidney disease patients. The predictor selection module is used to screen out several significant research factors based on the clinical dataset by using single-factor analysis and combining clinical relevance. A training dataset construction module is used to generate a training dataset from the clinical dataset based on the significant research factors. The model building module is used to build a nighttime hypertension risk prediction model based on the table value diffusion model. The nighttime hypertension risk prediction model takes the significant research factors as input and outputs the predicted risk probability value of developing nighttime hypertension. The training dataset is preprocessed, including forward noise addition, to obtain a preprocessed training dataset. The nighttime hypertension risk prediction model is trained based on the preprocessed training dataset to obtain a trained nighttime hypertension risk prediction model. The prediction module is used to receive the significant research factors of the patient to be tested, input the significant research factors of the patient to be tested into the trained nocturnal hypertension risk prediction model, and output the risk probability value of the patient to be tested developing nocturnal hypertension; based on the risk probability value and a preset threshold, the prediction result is output.

8. An electronic device, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the nocturnal hypertension risk prediction method as described in any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the nocturnal hypertension risk prediction method as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the nocturnal hypertension risk prediction method as described in any one of claims 1 to 6.

Images