CN-121983311-A - Early warning method, early warning device, early warning system and storage medium for sepsis

CN121983311ACN 121983311 ACN121983311 ACN 121983311ACN-121983311-A

Abstract

The invention discloses a sepsis early warning method, a device, a system and a storage medium, which comprise the steps of obtaining patient observation data of a target patient in continuous time steps, obtaining hidden state vectors and corresponding individualized baseline reference intervals of the target patient in current time steps, generating a counterfactual observation result in each counterfactual etiology scene, calculating etiology attribution vectors and infection specific residual errors of the target patient in the current time steps, calculating event window risk values of the target patient for occurrence of sepsis in a preset time window, and outputting risk prompt information of the target patient in the current time steps. By adopting the technical scheme of the invention, the problem that the existing early warning method is difficult to distinguish etiology in patients with multiple basic diseases is solved.

Inventors

CHEN ZHANHENG
WU HANCHEN
YANG SHUO
Li Hanbai
XIE XIN

Assignees

中国人民解放军海军军医大学

Dates

Publication Date: 20260505
Application Date: 20260126

Claims (10)

1. An early warning method for sepsis is characterized by comprising the following steps: s1, acquiring patient observation data of a target patient in continuous time steps, wherein the patient observation data comprises physiological observation data, treatment action data and static background data; s2, constructing and online updating an individualized heart-metabolism baseline digital twin model of the target patient based on the patient observation data to obtain a hidden state vector of the target patient in the current time step and a corresponding individualized baseline reference interval; S3, constructing a plurality of counterfactual etiology scenes based on the personalized heart-metabolism baseline digital twin model, and generating counterfactual observation results under each counterfactual etiology scene; s4, calculating an etiology attribution vector and an infection specific residual error of the target patient in the current time step according to the difference between the real observation data and each counter fact observation result; s5, calculating an event window risk value of the target patient for sepsis occurrence within a preset time window based on the infection specific residual error, the etiology attribution vector and the hidden state vector; and S6, based on the event window risk value, combining a preset value driving alarm decision rule, and outputting risk prompt information of the target patient in the current time step.
2. The method of claim 1, wherein the static background data includes a heart failure diagnosis marker and a type two diabetes diagnosis marker, and steps S2 to S6 are performed only when the heart failure diagnosis marker and the type two diabetes diagnosis marker satisfy a preset condition at the same time.
3. The method of claim 1, wherein the personalized cardiac-metabolic baseline digital twin model is a state space model, the hidden state vector includes at least a heart failure sub-state, a metabolic sub-state, and an infectious sub-state, and the personalized cardiac-metabolic baseline digital twin model performs predictive and observational updates on the hidden state vector based on the therapy action data.
4. The method of claim 1, wherein the adverse etiology scenes include at least an infection driving scene, a heart failure circulation driving scene, and a metabolic abnormality driving scene, wherein different adverse observations are generated by scene gating of state transition increments of sub-states under different adverse etiology scenes.
5. The method of claim 1, wherein the etiology attribution vector is calculated from a cost function of a difference residual between real observations and each of the counterfactual observations, and the etiology attribution vector characterizes a relative contribution of infectious inflammation drives, heart failure circulatory drives, and metabolic abnormality drives to current observation deviations by normalized mapping.
6. The method for early warning of sepsis according to claim 1, wherein the infection-specific residual error is obtained by combining a difference residual error between real observation data and a counter-fact observation result corresponding to a heart failure circulation driving scene and a difference residual error between real observation data and a counter-fact observation result corresponding to a metabolism abnormality driving scene, and is used for representing an observation deviation degree which cannot be interpreted by heart failure circulation factors and metabolism abnormality factors.
7. The method of claim 1, wherein the value driven alarm decision rule makes a comprehensive decision based on an event window risk value, a model uncertainty indicator, a preset benefit parameter, and a preset cost parameter, wherein the model uncertainty indicator is derived from a state uncertainty of the personalized heart-metabolic baseline digital twin model or an uncertainty of the event window risk value.
8. An early warning device for sepsis, comprising: the first processing module is used for acquiring patient observation data of a target patient in continuous time steps, wherein the patient observation data comprises physiological observation data, treatment action data and static background data; The second processing module is used for constructing and online updating an individualized heart-metabolism baseline digital twin model of the target patient based on the patient observation data to obtain a hidden state vector of the target patient under the current time step and a corresponding individualized baseline reference interval; the third processing module is used for constructing a plurality of counterfactual etiology scenes based on the personalized heart-metabolism baseline digital twin model and generating counterfactual observation results under each counterfactual etiology scene; The fourth processing module is used for calculating the cause attribution vector and the infection specific residual error of the target patient under the current time step according to the difference between the real observation data and each counter fact observation result; A fifth processing module, configured to calculate an event window risk value for the target patient to develop sepsis within a preset time window based on the infection specific residual, the etiology attribution vector, and the hidden state vector; And the sixth processing module is used for outputting risk prompt information of the target patient under the current time step based on the event window risk value and in combination with a preset value driving alarm decision rule.
9. An early warning system for sepsis, comprising a memory and a processor, the memory having stored thereon a computer program for execution by the processor, the computer program, when executed by the processor, performing the early warning method for sepsis of any one of claims 1-7.
10. A storage medium having stored thereon a computer program which, when run, performs the sepsis early warning method of any one of claims 1-7.

Description

Early warning method, early warning device, early warning system and storage medium for sepsis Technical Field The invention belongs to the field of clinical risk early warning technology, and particularly relates to a sepsis early warning method, a sepsis early warning device, a sepsis early warning system and a sepsis early warning storage medium. Background Sepsis is a systemic reaction state caused by infection, and the occurrence and development processes of sepsis have the characteristics of hidden onset and rapid progress. In order to realize early recognition of sepsis, various early warning methods based on physiological indexes, laboratory test data and scoring rules, such as a rule judging method based on a fixed threshold, a risk assessment method based on a scoring system and a prediction method based on a machine learning model, have been proposed in the prior art. Such methods typically output risk levels or cues by analyzing vital sign data, inflammation-related indicators, and organ function indicators, for assisting medical personnel in assessing patient status. However, there are significant limitations to the prior art described above in the heart failure combined type two diabetic patient population. The patients have circulatory dysfunction and abnormal metabolic regulation, and vital signs and laboratory indexes of the patients can obviously fluctuate under a non-infection state and are highly overlapped with early expression of sepsis. Most of the existing early warning methods are based on crowd unified threshold or integral feature modeling, lack of distinguishing capability for long-term states and common disease backgrounds of individuals of patients, and are difficult to effectively distinguish abnormal changes caused by infection from physiological changes caused by heart failure compensation or metabolic abnormality, so that non-infection related changes are easily misjudged as infection risks in the early warning process, and reliability of early warning results is affected. Disclosure of Invention In order to solve the problems in the prior art, the invention provides a sepsis early warning method, a sepsis early warning device, a sepsis early warning system and a sepsis early warning storage medium, and solves the problems that the existing early warning method is mostly based on crowd unified threshold or integral feature modeling and lacks the capability of distinguishing the long-term state and the common disease background of individuals of patients. In order to achieve the above object, the present invention provides the following solutions: an early warning method for sepsis, comprising the following steps: s1, acquiring patient observation data of a target patient in continuous time steps, wherein the patient observation data comprises physiological observation data, treatment action data and static background data; s2, constructing and online updating an individualized heart-metabolism baseline digital twin model of the target patient based on the patient observation data to obtain a hidden state vector of the target patient in the current time step and a corresponding individualized baseline reference interval; S3, constructing a plurality of counterfactual etiology scenes based on the personalized heart-metabolism baseline digital twin model, and generating counterfactual observation results under each counterfactual etiology scene; s4, calculating an etiology attribution vector and an infection specific residual error of the target patient in the current time step according to the difference between the real observation data and each counter fact observation result; s5, calculating an event window risk value of the target patient for sepsis occurrence within a preset time window based on the infection specific residual error, the etiology attribution vector and the hidden state vector; and S6, based on the event window risk value, combining a preset value driving alarm decision rule, and outputting risk prompt information of the target patient in the current time step. Preferably, the static background data includes a heart failure diagnosis flag and a type two diabetes diagnosis flag, and the steps S2 to S6 are performed only when the heart failure diagnosis flag and the type two diabetes diagnosis flag satisfy a preset condition at the same time. Preferably, the personalized heart-metabolism baseline digital twin model is a state space model, the hidden state vector includes at least a heart failure sub-state, a metabolism sub-state, and an infection sub-state, and the personalized heart-metabolism baseline digital twin model performs a prediction update and an observation update on the hidden state vector based on the treatment action data. Preferably, the anti-facts etiology scenes at least comprise an infection inflammation driving scene, a heart failure circulation driving scene and a metabolism abnormality driving scene, wherein under different anti-facts