Search

CN-121983346-A - Real-time early warning system for risk of nosocomial infection based on multi-source heterogeneous data

CN121983346ACN 121983346 ACN121983346 ACN 121983346ACN-121983346-A

Abstract

The application relates to an intra-hospital infection risk real-time early warning system based on multi-source heterogeneous data, which relates to the technical field of infection risk early warning, and comprises a data acquisition module, a data processing module, a risk calculation module, an early warning trigger module and a tracing positioning module, wherein the data acquisition module is used for acquiring the multi-source heterogeneous data in real time, the data processing module is used for generating standardized processing data from the multi-source heterogeneous data, the risk calculation module is used for calculating an individual infection risk value and a group infection probability value according to the standardized processing data, the early warning trigger module is used for triggering a grading early warning signal when the individual infection risk value and the group infection probability value reach a preset early warning threshold, and the tracing positioning module is used for generating a propagation path and a visual tracing map according to the standardized processing data. The application has the effect of early warning the risk of nosocomial infection.

Inventors

  • DAI HUANGYI
  • QI YUE
  • SU LI

Assignees

  • 南通大学

Dates

Publication Date
20260505
Application Date
20260331

Claims (10)

  1. 1. The hospital infection risk real-time early warning system based on the multi-source heterogeneous data is characterized by comprising a data acquisition module, a data processing module, a risk calculation module, an early warning trigger module and a tracing positioning module, wherein the data acquisition module is used for acquiring the multi-source heterogeneous data in real time, the multi-source heterogeneous data comprises patient clinical data, medical equipment data, medical personnel data and environment data, the data processing module is used for generating standardized processing data, the standardized processing data is input into a preset risk calculation model in the risk calculation module to calculate individual infection risk values and group infection probability values, the preset risk calculation model comprises an input layer, a characteristic engineering layer, a double-branch decision tree prediction layer and an output layer, the early warning trigger module is used for triggering a grading early warning signal when the individual infection risk values and group infection probability values reach preset early warning thresholds, and the tracing positioning module is used for generating a propagation path and a visual tracing source map according to the standardized processing data.
  2. 2. The real-time early-warning system for risk of nosocomial infections based on multi-source heterogeneous data according to claim 1, wherein the patient clinical data comprises patient body temperature data, blood routine detection data and antibacterial drug use data, the medical device data comprises medical device use duration data, disinfection record data and contact patient information data, the medical personnel data comprises sanitary frequency data, protective wear data and diagnosis and treatment track data, and the environment data comprises air colony data and object surface disinfection record data.
  3. 3. The real-time early warning system for risk of nosocomial infection based on multi-source heterogeneous data according to claim 2, wherein the data processing module performs normalization processing on the multi-source heterogeneous data to generate normalized processed data, and the normalization processing comprises the following steps: S11, respectively carrying out format conversion on clinical data of a patient, data of medical equipment, data of medical staff and environmental data based on a preset data format and a coding rule; s12, eliminating abnormal values of the patient clinical data, the medical equipment data, the medical personnel data and the environment data after format conversion to generate standardized processing data.
  4. 4. The real-time early warning system for risk of nosocomial infection based on multi-source heterogeneous data according to claim 2, wherein: the input layer respectively performs dimension disassembly on the clinical data, the medical equipment data, the medical personnel data and the environmental data of the patient to generate a clinical feature matrix, a medical equipment feature matrix, a medical personnel feature matrix and an environmental feature matrix of the patient, and combines the clinical feature matrix, the medical equipment feature matrix, the medical personnel feature matrix and the environmental feature matrix of the patient to generate a standardized feature matrix; The feature engineering layer sequentially normalizes the standardized feature matrix, eliminates redundant features and performs 5-section equal-frequency bin processing to generate a standardized feature set; the method comprises the steps that an individual infection prediction branch based on a CART decision tree algorithm and a group infection prediction branch based on the CART decision tree algorithm are arranged in the double-branch decision tree prediction layer, a standardized feature set is input to the individual infection prediction branch to generate an individual infection initial risk value, and the standardized feature set is input to the group infection prediction branch to generate a group infection initial probability value; The output layer corrects the initial risk value of the individual infection and the initial probability value of the group infection to obtain an initial risk value of the individual infection and a probability value of the group infection, wherein the initial risk value of the individual infection is the corrected initial risk value of the individual infection, the risk value of the individual infection is positioned in a preset individual infection risk interval, the probability value of the group infection is the corrected initial probability of the group infection, and the probability value of the group infection is positioned in a preset group infection probability interval.
  5. 5. The real-time early-warning system for risk of nosocomial infection based on multi-source heterogeneous data according to claim 4, wherein said generating an initial risk value for infection of an individual comprises the steps of: S21, carrying out feature matching and extraction on individual dimensions of a standardized feature set, determining a feature combination uniquely bound with a single patient to be predicted, generating an individual exclusive feature subset comprising clinical features of the patient, diagnosis and treatment medical equipment features corresponding to the patient, behavior features of medical staff responsible for diagnosis and treatment of the patient and environmental features of a diagnosis and treatment area where the patient is located, and generating an individual prediction exclusive feature set based on a preset initial weight coefficient; S22, constructing an individual infection risk prediction binary tree model based on a CART decision tree algorithm by taking the corresponding feature set of the hospital infection cases and the non-infection cases of the hospital history diagnosis as a positive and negative training sample set and taking whether the hospital infection of the cases as a classification label; performing recursive binary division on the individual prediction exclusive feature set by taking the coefficient of the foundation as a node division core criterion until a preset node sample size threshold, a coefficient of the foundation drop threshold or a stopping condition of a maximum depth threshold of a decision tree is met, and completing model branch construction and cost complexity pruning optimization to obtain an individual infection risk prediction model with stable convergence; S23, inputting an individual prediction exclusive feature set into an individual infection risk prediction model with stable convergence, judging the step-by-step feature of each node of a decision tree to be matched with a path, and outputting probability distribution of the patient corresponding to different hospital infection risk grades; And S24, carrying out weighted summation calculation on probability distribution of each risk level based on a preset risk level and score mapping rule, and generating an individual infection initial risk value of the patient.
  6. 6. The real-time early warning system for risk of nosocomial infection based on multi-source heterogeneous data of claim 4, wherein said generating an initial probability value for a group infection comprises the steps of: S31, performing feature aggregation and statistics of group dimensions on a standardized feature set, performing time sequence feature summarization according to a preset sliding time window by taking a target high-risk department ward as a statistics unit, extracting a group basic feature subset comprising clinical features of a department patient group, total usage and disinfection features of department common medical equipment, general personnel behavior features of department medical personnel and general environment monitoring features of the department, and simultaneously superposing a patient diagnosis and treatment track cross feature, medical equipment common cross feature and medical personnel contact cross feature to generate a group prediction specific feature set; S32, taking the corresponding feature set of the hospital infection aggregation event and the non-aggregation period which occur in the hospital history as a positive training sample set, taking whether the hospital infection aggregation outbreak occurs in the target area as a classification label, and constructing a group infection probability prediction binary tree model based on a CART decision tree algorithm; Performing recursive binary division on the special feature set of the group prediction by using the minimum coefficient of the radix as the node division optimal criterion, and simultaneously introducing a time sequence fluctuation checking mechanism to remove sporadic interference features until a preset node sample size threshold, a minimum coefficient of the radix or a maximum depth threshold of a decision tree is met, and completing model branch construction and cost complexity pruning optimization to obtain a group infection probability prediction model with stable convergence; S33, inputting the special feature set of the group prediction into a group infection probability prediction model with stable convergence, judging the feature of each node of the decision tree step by step to match with the path, and outputting the probability value of the occurrence of the intra-hospital infection aggregation event of the target area in a preset future early warning time window, namely the initial probability value of the group infection.
  7. 7. The real-time early warning system for risk of nosocomial infection based on multi-source heterogeneous data according to claim 1, wherein the generating the propagation path comprises the steps of: S41, defining a continuous tracing time window by taking a time node corresponding to early warning trigger as a reference, and extracting the total standardized processing data in the tracing time window to generate a tracing analysis special data set; S42, based on a preset entity association method, carrying out time sequence alignment and association matching on the total data in the special data set for the traceability analysis, identifying independent entity units with direct association relations in the special data set for the traceability analysis, and building an entity association network; s43, taking a target entity corresponding to early warning trigger as a core node, executing full path traversal search in an entity association network, extracting full-quantity association links with direct or indirect association relation with the core node, and synchronously marking the time sequence relationship and association node information corresponding to each full-quantity association link; S44, carrying out risk weight assignment and validity verification on all the associated links one by one based on a preset propagation risk judging method, removing invalid associated links which do not accord with propagation logic, and integrating the valid associated links according to time sequence order to generate a propagation path.
  8. 8. The real-time early warning system for risk of nosocomial infection based on multi-source heterogeneous data of claim 7, wherein said predetermined entity association method comprises the steps of: S421, timing alignment is carried out on the total data in the special data set of the traceability analysis by taking the timing scale of the traceability time window as a unified reference, and data units to be matched corresponding to the same timing interval are divided; S422, setting a preset matching threshold, and judging that a direct association relationship exists between two independent entity units when the matching intersection of the data units to be matched corresponding to the two independent entity units in the same time sequence interval meets the preset matching threshold; s423, using all independent entity units as network nodes and using the confirmed direct association relationship as node connection edges to build an entity association network.
  9. 9. The real-time early warning system for risk of nosocomial infection based on multi-source heterogeneous data according to claim 7, wherein the method for determining the risk of spread comprises the following steps: S441, setting a grading weight assignment standard by taking a time sequence relation corresponding to the association links and a link level of the association nodes and the core nodes as core dimensions, and carrying out risk weight assignment on the whole association links one by one; S442, setting a preset risk weight threshold and a one-way time sequence propagation check rule, and carrying out validity check on the assigned total associated links; And S443, judging the associated links with the risk weight lower than a preset risk weight threshold and not conforming to the unidirectional time sequence propagation check rule as invalid associated links to be removed, and judging the rest links as valid associated links.
  10. 10. The real-time early warning system for risk of nosocomial infection based on multi-source heterogeneous data of claim 7, wherein the generating of the visual traceability map comprises the steps of: s51, extracting total associated links, corresponding time sequence relations, associated node information and risk weight assignment corresponding to the propagation paths, and constructing a basic element set of the visual tracing map; S52, taking a target entity corresponding to early warning trigger as a core node, carrying out hierarchical division on associated node information and effective associated links in a basic element set according to a time sequence and risk weight assignment, distinguishing visual presentation dimensions of direct associated links and indirect associated links, and building a basic layout framework of a tracing map; s53, setting differentiated visual identifications for associated node information, effective associated links and ineffective associated links in different risk weight assignment intervals, and synchronously mapping associated time sequence relationships at the visual positions of the corresponding nodes and links to finish basic drawing of a tracing map; and S54, embedding the meta information of the entity association network, the tracing time window and the tracing analysis special data set into the visual tracing map as the matched association information of the tracing map to generate the visual tracing map.

Description

Real-time early warning system for risk of nosocomial infection based on multi-source heterogeneous data Technical Field The invention relates to the technical field of infection risk early warning, in particular to an in-hospital infection risk real-time early warning system based on multi-source heterogeneous data. Background Nosocomial infections, also known as hospital acquired infections, refer to infections acquired by patients in hospitals, the incidence of which is a core indicator of the level of quality safety and fine management of medical institutions, and is also a medical related risk for important prevention and control in the global public health field. At present, the risk early warning of the nosocomial infection mainly takes clinical diagnosis and treatment data of a patient as a core to construct a risk judging system, clinical diagnosis and treatment data such as body temperature data, blood routine detection data, antibacterial drug use data and the like of the patient are collected in a mode of docking an electronic medical record system of a hospital, a rule judging model based on a fixed threshold is constructed on the basis of rules in related standards of the nosocomial infection diagnosis, or the clinical data is subjected to fitting training through a machine learning model with a single structure, so that hysteresis judgment on whether the nosocomial infection of the patient occurs is realized. On the basis of clinical diagnosis and treatment data, auxiliary data such as discrete sampling data of environmental monitoring, hand hygiene statistical data of medical staff and the like can be introduced, multiple types of data are simply matched and compared through a cross check rule preset manually, and retrospective statistical analysis and risk investigation are carried out on related data after a confirmed hospital infection case appears in a hospital. The occurrence and the transmission of nosocomial infection have the characteristics of multi-cause coupling, multi-path crossing and strong time sequence relevance, but the existing nosocomial infection risk early warning can only output a single risk judgment result aiming at the infection risk of a single individual, or can only carry out post-aggregation event statistics aiming at the infection condition of a specific department area, can not synchronously complete the two-dimensional quantitative calculation of the individual infection risk and the group infection probability based on the same data system, and can not complete the association analysis and the visual presentation of the infection transmission path based on the full data related to early warning after the risk prompt. Disclosure of Invention In order to obtain an individual infection risk value and a group infection probability value conveniently and achieve the effect of early warning the risk of nosocomial infection, the application provides a nosocomial infection risk real-time early warning system based on multi-source heterogeneous data. The application provides a real-time early warning system for risk of nosocomial infection based on multi-source heterogeneous data, which adopts the following technical scheme: The utility model provides an intra-hospital infection risk real-time early warning system based on heterogeneous data of multisource, includes data acquisition module, data processing module, risk calculation module, early warning trigger module and traceability location module, data acquisition module is used for acquireing heterogeneous data of multisource in real time, data processing module is used for generating standardized processing data with heterogeneous data of multisource, risk calculation module is used for calculating individual infection risk value and crowd infection probability value according to standardized processing data, early warning trigger module is used for triggering hierarchical early warning signal when individual infection risk value and crowd infection probability value reach the pre-set early warning threshold value, traceability location module is used for generating propagation path and visual traceability map according to standardized processing data. Preferably, the multi-source heterogeneous data includes patient clinical data including patient body temperature data, blood routine detection data, and antibacterial agent usage data, medical device data including medical device usage duration data, sterilization record data, and contact patient information data, medical device data including sanitary frequency data, protective wear data, and diagnosis track data, and environmental data including air colony data and object surface sterilization record data. Preferably, the data processing module performs a normalization process on the multi-source heterogeneous data to generate normalized processed data, the normalization process including the steps of: S11, respectively carrying out format conversion on clinical