CN-122024992-A - Dynamic health management method, system and storage medium based on diabetic nephropathy

CN122024992ACN 122024992 ACN122024992 ACN 122024992ACN-122024992-A

Abstract

The invention relates to the technical field of health management, in particular to a dynamic health management method, a system and a storage medium based on diabetic nephropathy, which comprises the steps of establishing a multidimensional tag set system comprising patient, self-management, risk and scheme tag sets; the method comprises the steps of generating a current personalized label combination of a patient, matching an initial health management scheme, acquiring action data and physiological health index data of a patient execution scheme, inputting the data into a pre-trained dynamic self-adaptive update model, constructing the model based on Bi-LSTM and an attention mechanism, and directly outputting an updated personalized health management scheme through multi-objective loss function optimization. The invention realizes the end-to-end self-updating of the health management scheme and improves the individuation degree of the scheme.

Inventors

SHI JIAN
WEI ZHIHAO
WANG KESHAN
WANG YAQI
WANG HUAN
ZHANG XIAOPING

Assignees

华中科技大学同济医学院附属协和医院

Dates

Publication Date: 20260512
Application Date: 20260319
Priority Date: 20250320

Claims (10)

1. A dynamic health management method based on diabetic nephropathy, which is characterized by comprising the following steps: In a dynamic health management system, dividing influence factors according to a personalized health management scheme of a type 2 diabetic nephropathy patient into a plurality of tag sets, and matching each tag set with personalized tags reflecting personalized characterization of the type 2 diabetic nephropathy patient on the influence factors, wherein the influence factors comprise health information, self-management information, risk grade information and expert consensus information of the type 2 diabetic nephropathy patient, the health information, the self-management information and the expert consensus information are all obtained by directly acquiring the system, and the risk grade information is obtained by indirectly measuring and calculating a pre-established disease risk prediction model of the type 2 diabetic nephropathy; the dynamic health management system acquires real-time personalized tags of the type 2 diabetic nephropathy patients on the tag sets, and matches a real-time personalized health management scheme of the type 2 diabetic nephropathy patients based on the real-time personalized tags of the type 2 diabetic nephropathy patients on the tag sets; The dynamic health management system monitors the action data of the real-time personalized health management scheme executed by the type 2 diabetic nephropathy patient and monitors health indexes of the type 2 diabetic nephropathy patient after the real-time personalized health management scheme is executed; The dynamic health management system dynamically adjusts the real-time personalized tag of the type 2 diabetic nephropathy patient based on the fed-back health index and the action data to obtain a new personalized tag, and matches a new personalized health management scheme of the type 2 diabetic nephropathy patient by utilizing the adjusted new personalized tag; the dynamic health management system establishes health indexes and action data of a real-time personalized health management scheme by using a neural network, and a mapping relation between the real-time personalized health management scheme and a new personalized health management scheme to obtain a dynamic self-adaptive update model of the personalized health management scheme; After the dynamic self-adaptive update model is built, the dynamic health management system inputs the health index and the action data obtained by monitoring and the real-time personalized health management scheme into the dynamic self-adaptive update model to obtain a new personalized health management scheme for the type 2 diabetic nephropathy patient.
2. The method of claim 1, wherein the dynamic adaptive update model employs an encoder-decoder architecture, wherein the encoder is a Bi-directional long-short-term memory network Bi-LSTM for processing the sequence of motion data and the sequence of health indicator data comprising a plurality of points in time of history, and the decoder is a fully connected network with residual connection for generating the new personalized health management scheme based on hidden states output by the encoder; A time attention mechanism is introduced between the encoder and the decoder, and the attention mechanism calculates attention weights according to the similarity between the hidden state and the current decoding state of each time step of the encoder, so that the model focuses on historical data points with the greatest influence on the scheme adjustment, and noise data is ignored.
3. The method of claim 2, wherein the dynamically adaptive update model is trained with a multi-objective loss function comprising: health improvement loss, using mean square error MSE to measure updated health index With the expected health goal The gap between ; Scheme smoothing loss, using cosine similarity Weighting updated schemes Differences from the historical scheme F encourage continuous changes in content ; Patient preference loss, establishing a preference matrix based on implicit feedback of patient history to a scheme, obtaining a patient preference vector P through matrix decomposition, and using an implicit attribute vector of a new scheme The resulting Euclidean distance from the preference vector is taken as a penalty ; Risk control loss, inputting a new personalized health management scheme into a risk prediction model, calculating expected change of patient risk level after implementing the new scheme, and encouraging the scheme to adjust to the direction of reducing risk , For risk scores corresponding to the new personalized health management program, For the current risk score, max is the maximum operator; the multiple loss terms are balanced by weighted summation so that the model generated solution achieves the best compromise between health improvement, patient acceptance, solution continuity and risk control Wherein For each lost weight coefficient.
4. The method of claim 3, wherein the constructing of the patient preference penalty comprises collecting a plurality of health management protocols each patient has historically performed and their corresponding execution durations and punch-out completion rates, constructing a patient-protocol implicit feedback matrix; Matrix decomposition is carried out by adopting a weighted alternating least square WALS algorithm, so as to obtain an implicit preference vector of a patient and an implicit attribute vector of a scheme; The patient preference penalty is a negative cosine similarity between the new personalized health management scheme implicit attribute vector and the patient preference vector.
5. The method according to claim 1, wherein the method for constructing the risk prediction model for developing type 2 diabetic nephropathy comprises: Screening a plurality of characteristic factors which have obvious influence on the onset of type 2 diabetic nephropathy from historical cases by utilizing a multi-factor Logistic regression model; Carrying out principal component analysis on each characteristic factor to obtain the contribution degree of each characteristic factor; Based on the contribution, introducing a game theory, and determining the optimal feature weight of each feature factor; Multiplying each characteristic factor by the corresponding optimal characteristic weight, and training to obtain the risk prediction model by taking the characteristic factors as input characteristics, wherein the risk prediction model is used for outputting the risk score of a patient; the risk prediction model is a double-flow time sequence static fusion network based on attention guidance, inputs a static feature vector weighted by an optimal feature weight set and feature sequences of a plurality of follow-up time points of a patient, outputs a disease risk score of diabetic nephropathy, and is used for dynamically evaluating risk trend of the patient, and a static feature encoder and a time sequence feature encoder are introduced into the double-flow time sequence static fusion network based on attention guidance, so that baseline static data and follow-up dynamic data can be processed simultaneously.
6. The method of claim 5, wherein the optimal feature weights re-perform the contribution analysis with newly added case data every other preset period to update the weights of the feature factors to enable the risk prediction model to adapt to changes in patient population.
7. The method of claim 5, wherein the method of determining the risk level in the risk tag set comprises: Comparing the morbidity risk score output by the risk prediction model with a preset dynamic threshold curve, and dividing a high risk, medium risk or low risk interval according to the relative position of the score curve and the threshold curve; The dynamic threshold curve is adaptively adjusted according to the change of the quantile of the historical crowd risk score over time.
8. The method of claim 1, wherein matching the real-time personalized wellness management solution for type 2 diabetic nephropathy patients based on the real-time personalized tags on the respective tag sets for type 2 diabetic nephropathy patients is achieved using deep metric learning, comprising: Mapping each scheme in the scheme library to an embedding space which is the same as the tag combination through a twin network in advance, embedding cosine similarity with each scheme through calculating the tag combination, and selecting a scheme with highest similarity as a matching result; The twin network is trained through triple loss, wherein the triple consists of an anchor point label combination, a positive example scheme and a negative example scheme, so that the effective scheme is embedded closer to the label combination, and the ineffective scheme is embedded far away.
9. A dynamic health management system based on diabetic nephropathy, applied to the method of any one of claims 1-8, comprising: The label system construction unit is used for establishing and storing the multi-dimensional label set system; the initial scheme matching unit is used for generating a current multidimensional personalized label combination of the patient and matching an initial health management scheme; The data monitoring unit is used for monitoring and acquiring action data and health indexes of a patient in real time; The scheme self-adaptive updating unit is internally provided with the dynamic self-adaptive updating model and is used for receiving feedback data and directly outputting an updated health management scheme; the preference modeling unit is used for learning and storing personalized preference information of the patient and participating in loss calculation; and the risk dynamic evaluation unit is used for dynamically dividing the risk grade according to the risk score curve output by the risk prediction model.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the method according to any of claims 1-8.

Description

Dynamic health management method, system and storage medium based on diabetic nephropathy Technical Field The invention relates to the technical field of health management, in particular to a dynamic health management method, a system and a storage medium based on diabetic nephropathy. Background Type 2 diabetic nephropathy (Diabetic KIDNEY DISEASE, DKD) is one of the most common complications of type 2 diabetic nephropathy patients, and is also the leading cause of End stage renal disease (End-STAGE RENAL DISEASE, ESRD), accounting for 59% of the new ESRD. As the incidence of diabetes rises year by year, the high incidence of DKD presents a tremendous pressure and challenge to the medical system and socioeconomic. At present, the health management scheme of DKD patients is subjectively set by doctors according to clinical experience, and the technical problems that ① standards are different, the effect depends on subjective experience, the health management scheme is lack of unified standards due to experience differences of different doctors, the health benefits of patients are difficult to guarantee, ② scheme is lagged, timeliness is insufficient, scheme adjustment is usually carried out after discomfort or invalid management is generated on the patients, dynamic change of the states of the patients cannot be responded in real time, ③ updating link is long, efficiency is low, label analysis and scheme matching are required to be repeated in a traditional monitoring-analyzing-re-matching scheme updating mode, processing link is long, management efficiency is affected, and ④ lacks personalized consideration, the existing scheme is difficult to consider personalized preference of the patients, patient compliance is poor, and long-term management effect is affected. Therefore, the current health management scheme of DKD patients has long updated links, lacks individuation and has insufficient long-term management performance. Disclosure of Invention The invention aims to provide a dynamic health management method, a system and a storage medium based on diabetic nephropathy, which are used for solving the technical problems of long update link, lack of individuation and insufficient long-term management performance of a health management scheme in the prior art. In order to solve the technical problems, the invention specifically provides the following technical scheme: A dynamic health management method based on diabetic nephropathy, comprising the steps of: In a dynamic health management system, dividing influence factors according to a personalized health management scheme of a type 2 diabetic nephropathy patient into a plurality of tag sets, and matching each tag set with personalized tags reflecting personalized characterization of the type 2 diabetic nephropathy patient on the influence factors, wherein the influence factors comprise health information, self-management information, risk grade information and expert consensus information of the type 2 diabetic nephropathy patient, the health information, the self-management information and the expert consensus information are all obtained by directly acquiring the system, and the risk grade information is obtained by indirectly measuring and calculating a pre-established disease risk prediction model of the type 2 diabetic nephropathy; the dynamic health management system acquires real-time personalized tags of the type 2 diabetic nephropathy patients on the tag sets, and matches a real-time personalized health management scheme of the type 2 diabetic nephropathy patients based on the real-time personalized tags of the type 2 diabetic nephropathy patients on the tag sets; The dynamic health management system monitors the action data of the real-time personalized health management scheme executed by the type 2 diabetic nephropathy patient and monitors health indexes of the type 2 diabetic nephropathy patient after the real-time personalized health management scheme is executed; The dynamic health management system dynamically adjusts the real-time personalized tag of the type 2 diabetic nephropathy patient based on the fed-back health index and the action data to obtain a new personalized tag, and matches a new personalized health management scheme of the type 2 diabetic nephropathy patient by utilizing the adjusted new personalized tag; the dynamic health management system establishes health indexes and action data of a real-time personalized health management scheme by using a neural network, and a mapping relation between the real-time personalized health management scheme and a new personalized health management scheme to obtain a dynamic self-adaptive update model of the personalized health management scheme; After the dynamic self-adaptive update model is built, the dynamic health management system inputs the health index and the action data obtained by monitoring and the real-time personalized health management scheme into the dynamic self-adaptive