A hospital infection risk early warning system based on multi-source nursing data fusion

Abstract

The present application belongs to the technical field of hospital infection risk intelligent early warning, and discloses a hospital infection risk early warning system based on multi-source nursing data fusion, which comprises a multi-source nursing event acquisition module, a nursing contact dynamic graph construction module, a nursing transmission time sequence analysis module, a multi-path hospital infection risk analysis module, a nursing evidence credible fusion module and a hierarchical early warning and feedback updating module; a dynamic nursing contact graph among nursing staff, patients, instruments and environmental objects is constructed, the transmission state in a continuous nursing time window is analyzed, and multi-path risk results are dynamically fused according to the historical credibility of the data source and the current nursing evidence effectiveness; the transmission chain formed by cross-bed nursing and shared instruments can be identified, the early warning deviation caused by data loss, delayed supplementary recording and cross-source conflicts is reduced, and the accuracy and traceability of the hospital infection risk early warning are improved.

Description

Technical Field

[0001] This invention belongs to the field of intelligent early warning technology for hospital infection risks, and particularly relates to an early warning system for hospital infection risks based on the fusion of multi-source nursing data. Background Technology

[0002] During night shift handover, intensive care, and invasive care in the intensive care unit, nursing staff need to continuously perform tasks such as suctioning, catheter maintenance, urine bag handling, wound care, IV fluid replacement, and vital sign recording between different patient beds. They also come into contact with treatment carts, bed rails, shared ultrasound equipment, and other bedside instruments. Patients form a continuous contact relationship with nursing staff, instruments, and environmental objects. The risk of hospital-acquired infection is not only related to the patient's own condition but also affected by the sequence of nursing contact, the implementation of hand hygiene, the use of invasive instruments, and the circulation process of shared equipment. Therefore, it is necessary to integrate nursing records, personnel location, hand hygiene, instrument, and patient status data to continuously analyze the actual nursing transmission process and implement risk warnings.

[0003] Problems existing in the current technology: (1) Existing hospital infection risk warning systems usually use patient body temperature, infection-related monitoring indicators, nursing records and catheter indwelling time as independent risk inputs. They judge risks through fixed thresholds, statistical scores and single-patient time series models. They do not establish nursing staff, patients, instruments and environmental objects as a related structure that changes with nursing behavior, nor do they dynamically describe the transmission relationship between objects based on nursing contact duration, nursing operation exposure degree, hand hygiene status and contact time interval. When nursing staff perform continuous cross-bed operations in a short period of time or shared instruments are transferred between multiple patients, it is difficult to identify the nursing transmission chain that extends from the previous patient to the subsequent patient through nursing staff and shared instruments.

[0004] (2) Existing multi-source nursing data fusion methods usually directly splice different data or configure fixed fusion weights without distinguishing the long-term reliability of the data source and the validity of the evidence in the current nursing time window. When nursing records are added in a concentrated manner, location tags are temporarily lost, hand hygiene terminals are disconnected, and the results of multiple data source records conflict with each other, abnormal data may still participate in the final risk calculation according to fixed weights, which can easily lead to false or missed reports of hospital infection risk. At the same time, it is difficult to determine whether the risk results mainly come from the patient's own condition, invasive device exposure or nursing contact transmission process. Summary of the Invention

[0005] This invention aims to address the problem that existing hospital infection risk early warning systems typically use patient vital signs, nursing records, and invasive device data as independent inputs. These systems fail to construct a time-evolving propagation structure representing the actual nursing contact relationships formed by the continuous movement of nursing staff among different patients, devices, and environmental objects. This results in the difficulty of timely identifying potential transmission chains between patients during centralized nursing care, facilitated by nursing staff, shared devices, and environmental objects. Simultaneously, this invention addresses the problem that existing multi-source nursing data fusion methods, which typically employ fixed weights, direct splicing, and equal-weight fusion, fail to adjust evidence contribution based on the historical reliability of different nursing data sources, current data gaps, upload delays, and cross-source conflict states. This allows issues such as supplementary nursing records, device disconnections, and missing location information to easily be directly transmitted to the final risk outcome.

[0006] To address the aforementioned technical problems, this invention provides a hospital infection risk early warning system based on multi-source nursing data fusion, comprising a multi-source nursing event acquisition module, a nursing contact dynamic map construction module, a nursing transmission time sequence analysis module, a multi-path hospital infection risk analysis module, a nursing evidence credibility fusion module, and a graded early warning and feedback update module.

[0007] The multi-source nursing event acquisition module collects nursing record data, nursing staff location data, hand hygiene data, invasive instrument data, patient status data, and environmental contact data. It performs unified coding, timeline alignment, object identity binding, nursing task fragment generation, and data integrity status checks on nursing events from different sources to generate multi-source nursing event data.

[0008] Nursing record data includes nursing task type, task recipient, task start time, task end time, abnormal observation records, wound care records, and catheter maintenance records; nursing staff location data includes nursing staff number, time of entry into the bed area, time of departure from the bed area, and time spent on shared equipment; hand hygiene data includes hand hygiene execution time, drainage status, and nursing staff identity matching status; invasive device data includes central venous catheter indwelling time, urinary catheter indwelling time, duration of mechanical ventilation, device maintenance time, and number of unplanned operations; patient status data includes changes in body temperature, changes in drainage volume, wound status, and infection-related nursing abnormalities.

[0009] The nursing contact dynamic diagram construction module establishes nursing staff, patients, instruments, and environment objects as nursing staff nodes, patient nodes, instrument nodes, and environment nodes respectively according to the continuous nursing time window. Directed nursing contact edges are established based on the nursing staff-patient contact relationship, nursing staff-instrument operation relationship, instrument-patient use relationship, nursing staff-environment contact relationship, and environment-patient association relationship.

[0010] When multiple nursing contacts occur to the same group of objects within the same nursing time window, the corresponding nursing contact events are grouped into a nursing contact event set. The nursing contact transmission weight is determined based on the duration of the nursing contact, the degree of exposure to the nursing operation, the effectiveness of hand hygiene, and the time between the nursing contact and the current warning time. A nursing contact dynamic map is then regenerated after each nursing time window ends.

[0011] The nursing transmission time sequence analysis module reads the nursing contact dynamic graph corresponding to multiple consecutive nursing time windows, aggregates the neighborhood information of each node according to the nursing contact transmission edge weight, generates the patient graph state vector of each patient in different nursing time windows, and then performs time evolution memory update on the patient graph state vector according to the nursing time window order to generate patient transmission risk and main nursing transmission chain.

[0012] During the evolution of time, the continuous existence of the same nursing transmission chain enhances the corresponding historical transmission state, while the detection of effective hand hygiene, disinfection of shared instruments, and interruption of the nursing chain inhibits the corresponding historical transmission state, so that even when the patient does not currently show obvious abnormalities, the transmission risk can still be identified based on the previous nursing contact process.

[0013] The multi-pathway infection risk analysis module establishes nursing record analysis path, nursing contact analysis path, invasive device analysis path, and patient status analysis path. The nursing record analysis path generates nursing record risk probability based on abnormal observation records, wound care records, and catheter maintenance records. The nursing contact analysis path generates nursing contact risk probability based on the nursing transmission sequence and nursing contact transmission edge weight. The invasive device analysis path generates invasive device risk probability based on device indwelling time, maintenance interval, and unplanned operation status. The patient status analysis path generates patient status risk probability based on changes in body temperature, drainage, wound, and infection-related nursing status.

[0014] The nursing evidence credibility fusion module generates nursing evidence credibility based on the effective identification rate of each nursing data source in the retrospective of historical hospital infection events, the proportion of missing data in the current nursing time window, the delay between data generation and system reception, and the proportion of conflicts between different data sources.

[0015] The nursing evidence credibility fusion module further generates the current nursing evidence validity based on the proportion of valid nursing records in the current nursing time window and the cross-validation deviation between the corresponding analysis path and other nursing data, so that the long-term reliability of the nursing data source and the reliability of the current single evidence participate in risk fusion respectively.

[0016] When nursing records are updated in a concentrated manner, the credibility of the corresponding nursing evidence is reduced due to the data upload delay; when hand hygiene data is continuously uploaded in real time and is consistent with nursing location data, the credibility of the corresponding nursing evidence is increased; when the hand hygiene completion result in the nursing record conflicts with the nursing staff's location trajectory and the liquid status of the hand hygiene terminal, the validity of the current nursing evidence in the corresponding nursing risk analysis path is reduced.

[0017] The nursing evidence credibility fusion module performs dynamic weighted fusion based on the credibility of nursing evidence, the validity of current nursing evidence, and the patient risk probability of each nursing risk analysis path. It also combines the patient transmission risk generated by the nursing transmission time sequence analysis module and the patient's current susceptible exposure coefficient to generate the patient's hospital infection risk value.

[0018] The patient's current susceptible exposure coefficient is generated based on the duration of central venous catheter placement, urinary catheter placement, duration of mechanical ventilation, number of invasive nursing care sessions and number of wound care sessions on the day, so that the risk of nursing transmission can be adjusted according to the patient's current actual nursing exposure status.

[0019] When any nursing data source is disconnected, severely missing, or has obvious cross-source conflict, the contribution of the corresponding risk analysis path to the final hospital infection risk value is reduced, and other high-risk paths are not used to compensate for failed nursing evidence; when the patient's own condition has not shown obvious abnormalities, but there is a continuous nursing transmission chain and the patient's susceptible exposure status is increased, the patient's hospital infection risk value is increased.

[0020] The graded early warning and feedback update module generates graded early warning results based on the patient's hospital infection risk value, main nursing transmission chain, abnormal nursing objects, and risk duration, and outputs the nursing staff node, equipment node, environmental node, and high transmission contribution nursing contact edge corresponding to high-risk patients to the nursing terminal.

[0021] After nursing staff complete hand hygiene, disinfection of shared instruments, environmental disinfection, and verification of the nursing process, the graded early warning and feedback update module collects the nursing treatment results and feeds them back to the nursing contact dynamic map construction module and the nursing evidence credibility fusion module to regenerate the nursing contact dynamic map and the patient's hospital infection risk value.

[0022] This invention thus forms a complete technology chain of "multi-source nursing event collection - construction of nursing contact dynamic map - analysis of nursing transmission time sequence - multi-path risk analysis - fusion of dual credibility of nursing evidence - correction of patient susceptibility exposure - graded early warning - nursing treatment feedback".

[0023] The beneficial effects of this invention are as follows: (1) A dynamic nursing contact map is established based on nursing staff, patients, instruments and environmental objects. The transmission edge weight is dynamically determined according to the duration of nursing contact, the degree of exposure of nursing operation, the effective status of hand hygiene and the time distance of nursing contact. Then, the transmission status is analyzed for multiple consecutive nursing time windows. This can identify the continuous nursing transmission chain formed between patients-nursing staff-shared instruments-patients, and solve the problem that the risk of cross-bed nursing transmission cannot be found by analyzing patient indicators alone.

[0024] (2) Nursing evidence credibility is generated based on historical effective identification rate, current data missing, data delay and cross-source conflict respectively, and nursing evidence validity is generated based on current effective records and cross-validation bias. This allows the long-term reliability of the data source and the reliability of current nursing evidence to participate in risk fusion, which can reduce the impact of nursing record supplementation, location tag loss and hand hygiene equipment disconnection on hospital infection risk results.

[0025] (3) The risk probability of multiple pathways is analyzed together with the risk of nursing transmission and the patient’s current susceptible exposure status. The risk results are then reversed to nursing staff, patients, shared devices and environmental objects. After nursing staff complete hand hygiene, device disinfection and nursing process verification, the risk is re-mapped and updated to achieve a closed-loop hospital infection early warning from transmission structure identification, credible risk calculation to targeted nursing treatment. Attached Figure Description

[0026] The accompanying drawings are provided to further understand the present invention and form part of the specification. They are used together with the embodiments of the present invention to explain the invention and do not constitute a limitation thereof.

[0027] Figure 1 This is a diagram illustrating the overall architecture of the hospital infection risk early warning system based on multi-source nursing data fusion, as described in this invention. Figure 2 This is a dynamic diagram of nursing contact and a schematic diagram of its time evolution according to the present invention. Detailed Implementation

[0028] Example 1, see Figures 1-2 The present invention provides a hospital infection risk early warning system based on multi-source nursing data fusion, including a multi-source nursing event acquisition module, a nursing contact dynamic map construction module, a nursing transmission time sequence analysis module, a multi-path hospital infection risk analysis module, a nursing evidence credibility fusion module, and a hierarchical early warning and feedback update module.

[0029] This embodiment is applied to the night shift handover and centralized nursing work in a 24-bed comprehensive intensive care unit. The ward is equipped with 6 night shift nurses. The centralized nursing period is set from 19:30 to 21:00. During the corresponding period, they continuously perform sputum suction, central venous catheter maintenance, urine bag handling, infusion replacement, turning over, wound care, and vital sign recording. Nursing staff positioning tag receivers are set up in the patient bedside area, and hand sanitizer dispensing recording terminals are set up in the hand hygiene area. Shared treatment carts, bedside ultrasound equipment, sputum suction equipment, and respiratory support equipment are equipped with equipment identification tags.

[0030] The multi-source nursing event acquisition module collects nursing record data, nursing staff location data, hand hygiene data, invasive instrument data, patient status data, and environmental contact data. Among them, nursing record data includes nursing task type, task execution object, task start time, task end time, abnormal observation record, wound care record, and catheter maintenance record. Nursing staff location data includes nursing staff number, time of entering the bed area, time of leaving the bed area, and time spent on shared equipment. Hand hygiene data includes hand hygiene execution time, fluid discharge status, and nursing staff identity matching status.

[0031] Invasive device data includes central venous catheter indwelling time, urinary catheter indwelling time, duration of mechanical ventilation, device maintenance time, tubing replacement time, and number of unplanned procedures. Patient status data includes changes in body temperature, changes in drainage volume, wound status, respiratory support status, and records of infection-related nursing abnormalities. Environmental contact data includes bed rail contact events, treatment cart contact events, equipment control panel contact events, the time of most recent disinfection, and the number of times the device was shared across beds.

[0032] The nursing contact dynamic diagram construction module uses 15 minutes as the basic nursing time window. It establishes nursing staff, patients, instruments, and environment objects as nursing staff nodes, patient nodes, instrument nodes, and environment nodes, respectively. It also establishes directed nursing contact edges based on nursing staff-patient contact relationships, nursing staff-instrument operation relationships, instrument-patient use relationships, nursing staff-environment contact relationships, and environment-patient association relationships. The nursing contact dynamic diagram is regenerated after each basic nursing time window.

[0033] The nursing transmission time sequence analysis module receives nursing contact dynamic graphs corresponding to five consecutive basic nursing time windows. It performs structural aggregation on the neighborhood relationship of nodes in each time window according to the nursing contact transmission edge weights, and then inputs the states of the five consecutive graphs into the nursing time evolution memory unit in chronological order to generate the nursing transmission time sequence state and nursing contact transmission risk for each patient.

[0034] The multi-pathway infection risk analysis module establishes nursing record analysis path, nursing contact analysis path, invasive device analysis path, and patient status analysis path. The nursing record analysis path analyzes abnormal observation records, wound care records, and catheter maintenance records. The nursing contact analysis path analyzes the dynamic diagram of nursing contact and the time sequence of nursing transmission. The invasive device analysis path analyzes the status of catheter placement, mechanical ventilation, device maintenance, and unplanned operations. The patient status analysis path analyzes body temperature, drainage, wound, and infection-related nursing status. Each analysis path generates the patient's infection risk probability.

[0035] The nursing evidence credibility fusion module does not perform fixed averaging on the four analysis paths. Instead, it generates nursing evidence credibility based on the effective identification rate of each nursing data source in the retrospective of historical hospital-acquired infection events, the current data missing ratio, data upload delay, and cross-source conflict status. At the same time, it generates the current nursing evidence validity based on the effective record ratio in the current time window and the cross-validation bias between different data sources. Finally, it generates the patient hospital-acquired infection risk value by combining the nursing contact transmission risk and the patient's current susceptible exposure status.

[0036] The graded early warning and feedback update module generates risk alerts based on the patient's hospital infection risk value, the source of the nursing transmission chain, the abnormal nursing subjects, and the duration. In the low-risk state, the event chain is saved. In the high-risk state, key verification tasks are pushed to the responsible nurse and the on-duty nursing team leader. In the continuous high-risk state, the nursing staff node, patient node, equipment node, environmental node, and corresponding nursing contact edge that lead to the increase in risk are displayed synchronously. After the nursing staff completes hand hygiene replenishment, equipment replacement, environmental disinfection, and nursing record verification, the map is reconstructed and the patient's hospital infection risk value is updated.

[0037] In this embodiment, a nurse N1 performs nursing tasks sequentially in the order of patient P1, treatment cart E1, patient P2, shared ultrasound equipment D1, and patient P3. When patient P1 has infection-related nursing abnormalities, and nurse N1 leaves patient P1 without matching a hand hygiene event within the effective time range, and enters patient P2's bedside in a short period of time, the system does not only judge based on patient P2's own status, but increases the propagation intensity of the corresponding nursing contact edges from P1 to N1, from N1 to P2, and from N1 to shared equipment D1, and continues to transmit the corresponding propagation status to patient P3.

[0038] Through the above processing, this system reconstructs the independent events that were originally scattered in nursing records, hand hygiene equipment, positioning terminals, invasive instruments and patient monitoring systems into a nursing contact transmission structure that changes over time. It also dynamically determines the actual contribution of each risk analysis path based on the reliability of the data source itself, thus solving the problem that traditional multi-source nursing data cannot identify continuous nursing transmission chains across patients and devices after simple splicing.

[0039] Example 2: Based on all the above examples, the multi-source nursing event acquisition module specifically performs unified encoding of nursing events, timeline alignment, object identity binding, nursing task fragment generation, and multi-source event complete status check.

[0040] The unified coding process for nursing events involves writing the event number, data source number, nursing staff number, patient number, equipment number, environmental object number, event occurrence time, system reception time, event type, and event result for each nursing event. Data fields that do not correspond to a specific object are left empty, and no manual padding is used to replace the original missing data.

[0041] The timeline alignment process uses the hospital clock server time as a unified time reference. It calculates the time deviation for the collection of data from the nursing information system, hand hygiene terminal, positioning terminal, bedside monitoring equipment, and instrument management terminal, and maps each event to the corresponding 15-minute basic nursing time window.

[0042] The positioning terminal continuously records the time interval in which the nursing staff is in a certain bed area. The nursing record system records the start and end time of the nursing task. When the positioning interval and the nursing task interval overlap and the patient number is the same, a direct nursing contact event is generated. When the positioning interval exists but the nursing record is missing, a nursing contact event to be cross-validated is generated.

[0043] Hand hygiene events are matched with two adjacent nursing contacts based on the nurse's identity, location, and time of occurrence. When a nurse completes hand hygiene and generates a valid dispensing record after leaving the previous patient area and before entering the next patient area, the corresponding hand hygiene status is marked as valid. When the nursing record shows "Hand hygiene completed" but there is no corresponding record on the dispensing terminal and the nurse's location trajectory continuously enters the next patient area, the hand hygiene status in the nursing text is marked as a cross-source conflict status.

[0044] Device operation events are linked through three types of information: device label, nurse location, and nursing record. When a shared ultrasound device is moved from patient P2 to patient P3, a cross-bed shared device event is generated based on the device movement time, device disinfection record, and operating nurse.

[0045] Nursing task segments are divided into suctioning nursing segments, central venous catheter maintenance segments, urinary catheter nursing segments, wound care segments, infusion operation segments, turning nursing segments, and vital sign recording segments according to the nature of the task risk exposure. The duration of each nursing task segment and the adjacent nursing task subjects before and after it are recorded.

[0046] The data integrity check counts the number of nursing record data, nursing positioning data, hand hygiene data, invasive instrument data, and patient status data that should have been recorded and the number of records that were actually recorded within the current basic nursing time window, and generates the data missing ratio used for subsequent nursing evidence credibility calculation.

[0047] When the hand hygiene terminal is continuously disconnected for more than one basic nursing time window, the failure to receive a hand hygiene record will not be directly judged as a failure to perform hand hygiene. Instead, the hand hygiene data source will be marked as incompletely observable, and the credibility of the corresponding data source will be reduced during the credible fusion of nursing evidence.

[0048] When a nurse's location tag is temporarily lost, but the nursing task record, instrument operation log, and hand hygiene record form a consistent object chain, the system retains the nursing contact event and lowers the data source credibility level of the corresponding nursing contact event, without directly deleting the corresponding contact relationship.

[0049] Regarding parameter adjustments: During the intensive care period of the ICU night shift, a 15-minute basic nursing time window is used. During the routine daytime nursing period, the basic nursing time window can be adjusted to 30 minutes. When the density of nursing tasks increases continuously, the nursing event convergence cycle is shortened, but all data is kept mapped to a unified basic nursing time window.

[0050] Through the above processing, this system does not require all nursing data to be in a complete state during the collection phase. Instead, it retains information on missing data, upload delays, and cross-source conflicts, enabling subsequent algorithms to distinguish between two different situations: "no nursing action occurred" and "nursing action existed but the data source failed to record it reliably." This solves the problem of misjudging risks directly caused by anomalies in a single data source.

[0051] Example 3: Based on all the above examples, the nursing contact dynamic graph construction module specifically performs node state generation, nursing contact event merging, propagation edge type identification, nursing contact propagation edge weight calculation, and time window graph updating.

[0052] The nursing staff node records the number of nursing tasks, cross-bed nursing times, high-exposure nursing times, continuous nursing intervals, effective hand hygiene times, and shared device operations within the current 15 minutes. The patient node records infection-related nursing abnormalities, the number of invasive nursing procedures performed that day, device indwelling status, and current nursing frequency. The device node records the current patient using the device, the most recent maintenance time, the most recent disinfection time, the number of consecutive operations, and the number of cross-bed sharing procedures. The environment node records the most recent disinfection time, the number of current contacts, and the associated patient number.

[0053] When a nursing staff performs a direct nursing procedure on a patient, a nursing staff-patient contact edge is established. When a nursing staff operates a central venous catheter, urinary catheter, ventilator tubing, suction equipment, or shared ultrasound equipment, a nursing staff-instrument operation edge is established. When an instrument is connected to a patient, an instrument-patient use edge is established. When a nursing staff comes into contact with a treatment cart, bed rails, or equipment control panel, a nursing staff-environment contact edge is established. When an environmental object is located in the corresponding patient's bed area, an environment-patient association edge is established.

[0054] When multiple contacts occur with the same group of objects within the same basic nursing time window, instead of establishing multiple independent parallel edges, the corresponding nursing contact events are grouped into the same set of nursing contact events. The nursing contact transmission edge weight is calculated based on the duration of the operation, the degree of operation exposure, the effectiveness of hand hygiene, and the time from the current warning time.

[0055] Calculate the nursing contact propagation edge weight between object a and object b in the t-th basic nursing time window according to Formula 1:

[0056] In Formula 1, This represents the nursing contact propagation edge weight between object a and object b within the t-th basic nursing time window. This represents the set of nursing contact events that occur between object a and object b within the current basic nursing time window. This indicates the duration of the qth nursing contact. This indicates the reference contact duration for the corresponding nursing procedure.

[0057] In Formula 1, This represents the nursing exposure coefficient for the q-th nursing procedure. Suctioning, central venous catheter maintenance, and wound care correspond to higher nursing exposure coefficients, while routine vital sign recording corresponds to lower nursing exposure coefficients. This represents the hand hygiene effectiveness obtained by matching before and after the qth nursing contact, with a value ranging from zero to one. This indicates the time elapsed between the qth nursing contact and the current warning time. This represents the decay time constant of the effect of nursing contact.

[0058] The hand hygiene terminal generates complete and valid records, and the nursing staff's location trajectory can prove that hand hygiene occurred after the previous patient left and before the next patient entered, improving... When nursing records contain hand hygiene text but discharge records are missing and the equipment is in normal condition, the risk of infection decreases. When the hand hygiene terminal is disconnected, do not directly... Instead of adjusting to zero, the credibility contribution of the corresponding evidence source is reduced by the nursing evidence credibility fusion module.

[0059] Taking the example of nurse N1 entering patient P2 bed 1 minute after completing sputum suctioning for patient P1 to perform central venous catheter maintenance, when the duration of the two nursing operations is long, the nursing exposure coefficient is high and no effective hand hygiene is detected, the corresponding nursing contact transmission side weight remains high; when the interval between the two nursing operations is 40 minutes and effective hand hygiene is completed in between, the time decay term and the effective hand hygiene term together reduce the subsequent transmission side weight.

[0060] After environmental objects are disinfected, historical nursing contact edges are not deleted. Instead, the transmission status of the corresponding environmental nodes is reduced starting from the next basic nursing time window, so that the system can retain the time difference between "continuous contact that occurred before disinfection" and "new contact that occurred after disinfection".

[0061] Through the above processing, this system incorporates the duration of nursing contact, the degree of exposure to nursing procedures, the results of hand hygiene implementation, and the time distance into the dynamic graph edge weights, enabling nursing contact relationships to continuously change with actual nursing behaviors. This solves the problem that using a fixed nurse-patient relationship graph cannot reflect the risk of short-term cross-bed transmission during centralized nursing care.

[0062] Example 4: Based on all the above examples, the nursing transmission time sequence parsing module adopts the nursing contact transmission graph time sequence coding algorithm, which specifically includes continuous nursing graph sequence construction, node neighborhood transmission aggregation, nursing graph state compression, time evolution memory update, and patient transmission risk output.

[0063] The continuous nursing chart sequence construction process reads the current basic nursing time window and the four basic nursing time windows before it at the current warning time, forming five nursing contact dynamic charts arranged in chronological order. The five nursing contact dynamic charts cover the concentrated nursing process 75 minutes before the current warning time.

[0064] The node neighborhood propagation aggregation process uses the current nursing contact propagation edge weight as the neighborhood information influence strength. The patient node aggregates the status of the nursing staff nodes, device nodes and environment nodes connected to it. The nursing staff node aggregates the status of the most recently contacted patient, the device operated and the contact environment. The device node aggregates the status of the most recently used patient and the operating nursing staff.

[0065] Unlike directly splicing node features into a time series, node neighborhood propagation aggregation first reorganizes information according to the current nursing contact graph structure, so that when the same nurse enters different bed areas at different time windows, the corresponding patient node receives different neighborhood information, thus enabling changes in the nursing path to be included in subsequent time evolution analysis.

[0066] The nursing graph state compression process compresses the state of each patient node in each time window after neighborhood propagation into a patient graph state vector, while retaining the numbers of nursing staff nodes, instrument nodes, and environmental nodes that have a high risk contribution to the patient, so that the final risk result can be back-located to the propagation chain.

[0067] The temporal evolution memory update process inputs the patient graph state vector in the order of the five basic nursing time windows. The temporal evolution memory unit is equipped with a historical state retention gate, a current nursing state writing gate, and an invalid propagation state suppression gate. When the same nursing propagation chain appears continuously in multiple consecutive basic nursing time windows, the corresponding historical state is enhanced. When effective hand hygiene, instrument disinfection, or interruption of the nursing chain between patients occurs in the middle, the corresponding historical propagation state is suppressed.

[0068] If patient P2 does not show any obvious abnormalities in their own nursing status in the current time window, but there is a continuous contact structure between patient P1, nursing staff N1, treatment cart E1 and patient P2 in the previous two time windows, and patient P1 already has infection-related nursing abnormalities, the nursing transmission sequence analysis module can still retain the corresponding transmission information through the preceding graph state, instead of judging the risk based solely on patient P2's current data.

[0069] The nursing transmission timeline analysis module ultimately outputs the patient transmission risk. The main transmission chain, the number of transmission chain duration windows, the nursing staff node with the highest transmission contribution, and the device node with the highest transmission contribution are included, with patient transmission risk serving as the structural risk correction input for the nursing evidence credibility fusion module.

[0070] Regarding parameter adjustment: In centralized nursing scenarios, a time evolution sequence is established using five consecutive basic nursing time windows. As the duration of nursing handover increases, the number of historical time windows can be expanded. When the increase in the number of historical windows leads to an excessive impact from premature nursing events, the suppression intensity of distant historical states is increased to ensure that the most recent nursing contact remains the primary contributor.

[0071] Through the above processing, this algorithm first analyzes the transmission structure among nursing staff, patients, equipment and environment within the same time window, and then analyzes the changes in the transmission structure with multiple nursing time windows. Unlike the method of using only the patient's vital signs time series for prediction, this algorithm allows "who contacted whom, through what objects, and how the contact relationship continuously evolves" to be directly incorporated into the hospital infection risk analysis process.

[0072] Example 5: Based on all the above examples, the multi-path hospital infection risk analysis module establishes four independent nursing risk analysis paths. Each path only processes data corresponding to its own nursing evidence type and outputs the patient risk probability respectively.

[0073] The nursing record analysis path reads nursing abnormality observation records, wound care records, catheter maintenance records, sputum suction records and supplementary recording times, extracts infection-related nursing descriptions, catheter maintenance abnormalities, wound abnormalities, drainage abnormalities and record delay status, and generates nursing record risk probabilities.

[0074] The nursing contact analysis path reads the nursing transmission time sequence status, current nursing contact transmission edge weight, number of nursing staff associated with the patient, number of consecutive nursing care across beds, and continuous use status of shared equipment, and generates the nursing contact risk probability.

[0075] The invasive device analysis pathway reads the indwelling time of central venous catheter, indwelling time of urinary catheter, duration of mechanical ventilation, device maintenance interval, number of invasive nursing care and unplanned procedures per day, and generates the probability of invasive device risk.

[0076] The patient status analysis path reads changes in body temperature, drainage volume, wound status, respiratory support status, and infection-related nursing abnormality trends, and generates patient status risk probabilities.

[0077] The four analysis paths each store the risk probability, input event number, and main risk contribution items, and do not share prediction results within the paths. This maintains the independence of evidence among nursing records, nursing contacts, invasive instruments, and patient status, and prevents the same abnormal event from being repeatedly amplified through multiple paths, resulting in an inflated risk.

[0078] The historical reliability of nursing data sources is dynamically generated based on the retrospective results of historical hospital-acquired infection events and the data quality of the current time window.

[0079] The credibility of nursing evidence from the m-th type of nursing data source in the t-th basic nursing time window is calculated according to Formula 2:

[0080] In Formula 2, This represents the credibility of nursing evidence from the m-th type of nursing data source within the t-th basic nursing time window. This represents the effective identification rate of the m-th type of nursing data source obtained by retrospectively analyzing historical hospital infection events. This indicates the percentage of missing data within the current basic nursing care time window.

[0081] In Formula 2, This indicates the delay between the time the data is generated and the time the system receives it. This represents the maximum allowable effective delay for the m-th type of nursing data. This indicates the proportion of conflicts between the current nursing data source and other nursing data sources. This represents the time delay attenuation coefficient. This represents the cross-source conflict penalty coefficient.

[0082] When nursing records are entered retroactively after a nursing procedure, the data itself remains complete, but the Delta T_{m,t} increases, resulting in a decrease in the credibility of the corresponding nursing evidence. When hand hygiene terminals upload data in real time and operate continuously, and can be mutually verified with the location trajectory of nursing staff, the proportion of missing data and cross-source conflicts remains low, resulting in a high level of credibility of the corresponding nursing evidence.

[0083] Location data indicates that when nursing staff enter patient P3 bed, there is no patient P3 nursing task in the nursing record, and the shared device operation log shows that when the same nursing staff uses the device at this bed, the corresponding cross-source conflict is not simply deleted, but written into K_{m,t} to reduce the impact of conflict data sources on the final risk fusion.

[0084] Through the above processing, this system incorporates the long-term reliability of the data source in historical events into the credibility of nursing evidence along with the data missing, delay, and conflict in the current time window, thus resolving the problem of assigning the same weight to "a historically reliable data source that is currently offline" and "a data source with complete current data but a high long-term false alarm rate".

[0085] Example 6: Based on all the above examples, the nursing evidence credibility fusion module adopts the nursing evidence dual credibility fusion algorithm, which specifically performs current evidence validity calculation, susceptible exposure status generation, path risk normalization fusion, nursing transmission risk correction, and abnormal evidence suppression.

[0086] The validity of current evidence is not determined by the probability output of the analysis path itself, but by checking whether the corresponding nursing evidence in the current basic nursing time window can be cross-validated by other nursing data.

[0087] The cross-validation objects for the nursing record analysis path include nursing location data, device operation logs, and hand hygiene events; the cross-validation objects for the nursing contact analysis path include nursing records and device operation logs; the cross-validation objects for the invasive device analysis path include maintenance records, device status, and patient usage relationships; and the cross-validation objects for the patient status analysis path include nursing observation records and continuous status data.

[0088] Calculate the validity of current nursing evidence for the m-th nursing risk analysis path in the t-th basic nursing time window using Formula 3:

[0089] In Formula 3, This indicates the validity of current nursing evidence for the m-th nursing risk analysis path within the current basic nursing time window. This indicates the percentage of valid nursing records for this analysis path within the current basic nursing time window. This indicates the deviation between the current recorded results of this analysis path and the cross-validation results of other nursing data. This indicates the allowable cross-validation bias for the corresponding nursing data source. This indicates a positive number that prevents the denominator from being zero.

[0090] When the nursing record states "hand hygiene completed," and the nursing staff location record shows that the nursing staff went directly from patient P1 to patient P2, and the hand hygiene terminal did not generate a discharge record within the corresponding interval and the terminal was operating normally, the cross-validation deviation between the nursing record analysis path and other nursing data increased. Although the nursing record exists, its validity as current nursing evidence decreased.

[0091] The patient's current susceptible exposure status is generated based on the current nursing conditions. The patient's age is not directly used as the main correction parameter. Specifically, the indwelling time of the central venous catheter, the indwelling time of the urinary catheter, the duration of mechanical ventilation, the number of invasive nursing care and the number of wound care care on the same day are read. The exposure degree of each indicator relative to the corresponding nursing management reference range is combined to form the patient's current susceptible exposure coefficient.

[0092] Calculate the hospital infection risk value of patient p in the t-th basic nursing time window according to Formula 4:

[0093] In Formula 4, This represents the hospital infection risk value of patient p in the t-th basic nursing time window, and M represents the number of nursing risk analysis paths. This indicates the credibility of nursing evidence from the m-th type of nursing data source. This indicates the validity of the current nursing evidence for the m-th nursing risk analysis path. This represents the patient's hospital infection risk probability output by the m-th nursing risk analysis path.

[0094] In Formula 4, This indicates the patient transmission risk output by the nursing transmission timeline analysis module. This indicates the patient's current susceptibility exposure coefficient. This represents the correction factor for the risk of transmission through nursing care. This represents a positive number to prevent the total weight from being zero.

[0095] When a data source becomes completely invalid within the current basic nursing time window, the credibility of the corresponding nursing evidence or the validity of the current nursing evidence decreases. This analysis path automatically reduces its contribution to the final hospital infection risk value and does not artificially compensate for invalid evidence using other high-risk paths.

[0096] When the patient's own status pathway risk is low, but the risk of nursing transmission is high, and the patient has a long-term central venous catheter, mechanical ventilation, and multiple invasive nursing conditions, the nursing transmission risk modification item increases the final hospital infection risk, enabling the system to identify the early risk of the formation of nursing transmission chains before the patient's clinical status changes significantly.

[0097] When the risk of nursing transmission is high but the patient's current susceptible exposure status is low, and nursing staff perform effective hand hygiene and shared instruments are disinfected, the nursing contact transmission risk and the patient's current susceptible exposure correction are reduced together in the subsequent time window, preventing a historically high-risk contact from permanently maintaining a high-risk status.

[0098] Regarding parameter adjustments: after increasing the number of historical event backtracking samples, the effective identification rate of each nursing data source will be recalculated; after changing the hardware terminal of a certain type of data source, the data delay benchmark and cross-source conflict benchmark will be re-established; when the proportion of invasive instruments used in the ICU increases, the nursing transmission risk correction coefficient will be increased, but the dual-credibility weight calculation method of the four nursing risk analysis paths will remain unchanged.

[0099] Through the above processing, this system addresses the two questions of "whether the data source is trustworthy in the long term" and "whether the current nursing evidence is trustworthy." It then introduces the nursing transmission structure as an independent correction factor into patient risk, avoiding the use of simple averaging and fixed-weight fusion of multiple models. This reduces the errors caused by missing nursing data, delayed data entry, and cross-source conflicts in the early warning of hospital infection risks.

[0100] Example 7: This example is based on all the above examples and uses the night shift handover and centralized nursing process of a 24-bed comprehensive ICU to illustrate the actual operation of this system.

[0101] The concentrated nursing period begins at 19:30, with 6 nurses each taking on the main nursing tasks for 4 patients. The system establishes a basic nursing time window of 15 minutes and continuously generates nurse nodes, patient nodes, equipment nodes, and environment nodes.

[0102] At 19:42, patient P1 completed suctioning care. Nursing staff N1 stayed at patient P1's bed for 7 minutes. Patient P1 had a fever and abnormal airway secretions on the day of the incident. After leaving patient P1, nursing staff N1 did not pass through the hand hygiene area. At 19:44, N1 entered patient P2's bed and began central venous catheter dressing maintenance.

[0103] The system reads the bed movement trajectory, sputum suction nursing record, central venous catheter maintenance record and hand hygiene terminal status of nurse N1. It confirms that the hand hygiene terminal is normal but there is no corresponding fluid discharge event. According to Formula 1, it increases the nursing contact transmission edge weight between the nursing chain associated with patient P1 and the nursing chain associated with patient P2.

[0104] At 19:51, nurse N1 took the shared ultrasound device D1 from patient P2 bed and entered patient P3 bed at 19:55. After patient P2 used the device D1, no corresponding disinfection completion record was generated. The nursing contact dynamic diagram increases the nursing contact edges between nurse N1 and shared ultrasound device D1 and patient P3.

[0105] Within the current time window, no significant abnormalities have been observed in the body temperature, drainage status, and wound observation of patients P2 and P3. The traditional risk assessment based on individual patient indicators remains at a low level. This system, however, obtains the nursing transmission chain of "patient P1 - nursing staff N1 - patient P2 - shared ultrasound equipment D1 - patient P3" based on the dynamic diagram of multiple consecutive nursing contacts.

[0106] The nursing record analysis path outputs a low risk probability for patient P3, the nursing contact analysis path outputs a high risk probability, the invasive device analysis path identifies that patient P3 has had a central venous catheter in place for 6 days and has completed two invasive nursing procedures on the same day, and the patient status analysis path outputs a low to medium risk probability.

[0107] At this time, the nursing record system has a delay of about 18 minutes in supplementing the record. The credibility of the nursing evidence from the nursing record data source decreases according to Formula 2. Nursing location data, shared equipment operation data, and hand hygiene data are all uploaded in real time and form a consistent event chain. The credibility of the nursing evidence from the corresponding data source remains high.

[0108] At 20:02, nurse N1 supplemented the nursing records for patients P2 and P3. The nursing records showed that the nursing tasks had been completed, but there was a cross-validation bias between the hand hygiene completion field and the hand hygiene terminal and positioning trajectory. The validity of the current nursing evidence in the corresponding nursing record analysis path was reduced according to Formula 3.

[0109] The nursing evidence credibility fusion module calculates the patient's P3 hospital infection risk value based on the credibility of nursing evidence corresponding to the four paths, the validity of current nursing evidence, the probability of path risk, the patient transmission risk, and the patient's current susceptible exposure coefficient. After the risk value enters a high-risk state, the system pushes a targeted nursing verification task to the nursing staff N1 and the on-duty nursing team leader, which is "verifying hand hygiene - disinfection of shared ultrasound equipment - maintenance process of patient's P3 central venous catheter".

[0110] At 20:08, nursing staff completed supplementary hand hygiene procedures, shared ultrasound equipment D1 was disinfected and the disinfection completion time was recorded, and the nursing team leader completed the maintenance check of the central venous catheter for patient P3. After the next basic nursing time window was re-constructed, the nursing contact transmission margin between nursing staff N1 and subsequent patients decreased, and the environmental transmission status of shared ultrasound equipment D1 decreased.

[0111] The system saves the nursing contact dynamics, main transmission chains, risk probabilities of the four analysis paths, credibility of nursing evidence, validity of current nursing evidence, patient's current susceptible exposure coefficient, final hospital infection risk value, and nursing intervention events corresponding to risk reduction for each risk calculation, enabling risk warnings to be traced back to specific nursing staff, patients, equipment, environmental objects, and nursing time windows.

[0112] During the model training phase, training data was established using historical nursing events and risk events confirmed by hospital infection control review. The nursing contact propagation graph temporal coding algorithm mainly learns the connection relationships between different nodes and the changes in continuous time windows. The multi-path hospital infection risk analysis module trains nursing records, nursing contacts, invasive instruments, and patient status analysis paths respectively. The effective identification rate of historical data sources is generated through independent backtracking statistics, and the credibility of nursing evidence is not used as a free learning parameter of ordinary neural networks.

[0113] When a hospital replaces its hand hygiene terminals, positioning systems, and nursing information systems, the system first re-establishes benchmarks for data delays, missing data, and cross-validation biases, and then gradually updates the credibility of nursing evidence, without directly using the credibility parameters corresponding to the old equipment.

[0114] Through the above processing, this system forms a complete technology chain of "multi-source nursing event collection - dynamic nursing contact map construction - nursing transmission structure temporal analysis - four-path risk analysis - long-term credibility and current validity fusion - patient susceptibility exposure correction - targeted nursing early warning - post-treatment reconstruction", which transforms hospital infection early warning from the detection of abnormal indicators of a single patient to the joint analysis of the actual nursing contact transmission process and the quality of multi-source nursing evidence.

Claims

1. A hospital infection risk early warning system based on multi-source nursing data fusion, characterized in that, It includes a multi-source nursing event collection module, a nursing contact dynamic map construction module, a nursing transmission time sequence analysis module, a multi-path hospital infection risk analysis module, a nursing evidence credibility fusion module, and a graded early warning and feedback update module; The multi-source nursing event acquisition module collects nursing record data, nursing staff location data, hand hygiene data, invasive instrument data, patient status data, and environmental contact data. It performs timeline alignment, object identity binding, and unified encoding of nursing events on the collected data to generate multi-source nursing event data. The multi-source nursing event data is then sent to the nursing contact dynamic graph construction module and the multi-path hospital infection risk analysis module. The nursing contact dynamic graph construction module receives the multi-source nursing event data, establishes nursing staff nodes, patient nodes, instrument nodes, and environment nodes according to the continuous nursing time window, establishes nursing contact edges based on nursing contact between nursing staff and patients, nursing staff operation of instruments, usage relationship between instruments and patients, nursing staff contact with environmental objects, and association between environmental objects and patients, and generates nursing contact propagation edge weights based on nursing contact duration, nursing operation exposure status, hand hygiene effectiveness status, and nursing contact time distance to obtain the nursing contact dynamic graph corresponding to the continuous nursing time window, and sends the nursing contact dynamic graph to the nursing propagation timing parsing module; The nursing transmission time sequence parsing module receives the nursing contact dynamic graph corresponding to multiple consecutive nursing time windows, performs transmission aggregation on the neighborhood state of each node according to the nursing contact transmission edge weight, generates a patient graph state vector, performs time evolution state update according to the patient graph state vector in consecutive nursing time windows, generates patient transmission risk and main nursing transmission chain, and sends the patient transmission risk and main nursing transmission chain to the multi-path hospital infection risk analysis module and the nursing evidence credibility fusion module; The multi-path hospital infection risk analysis module receives the multi-source nursing event data, the patient transmission risk, and the main nursing transmission chain. It generates nursing record risk probability, nursing contact risk probability, invasive device risk probability, and patient status risk probability through nursing record analysis path, nursing contact analysis path, invasive device risk probability, and patient status risk probability, respectively, to obtain multi-path hospital infection risk results. The multi-path hospital infection risk results are then sent to the nursing evidence credibility fusion module. The nursing evidence credibility fusion module receives the multi-source nursing event data, the patient transmission risk, and the multi-path hospital infection risk results. It generates nursing evidence credibility based on the historical valid identification status, current data missing status, data reception delay status, and cross-source conflict status of each nursing data source. It generates current nursing evidence validity based on the current valid nursing record status and cross-source cross-validation status. It performs dynamic weighted fusion on the multi-path hospital infection risk results based on the nursing evidence credibility and the current nursing evidence validity. It also corrects the transmission risk based on the patient transmission risk and the patient's current susceptible exposure status to generate a patient hospital infection risk value. Finally, it sends the patient hospital infection risk value and the main nursing transmission chain to the hierarchical early warning and feedback update module. The graded early warning and feedback update module receives the patient's hospital infection risk value and the main nursing transmission chain. Based on the patient's hospital infection risk value, the source of the nursing transmission chain, and the risk persistence status, it generates graded early warning results, extracts the corresponding nursing staff nodes, patient nodes, equipment nodes, environmental nodes, and nursing contact edges, receives nursing treatment results, generates updated nursing status data, and sends the updated nursing status data to the nursing contact dynamic graph construction module and the nursing evidence credibility fusion module to update the nursing contact dynamic graph and the patient's hospital infection risk value.

2. The hospital infection risk early warning system based on multi-source nursing data fusion according to claim 1, characterized in that, The multi-source nursing event acquisition module writes each nursing event into the event number, data source number, nursing staff number, patient number, instrument number, environmental object number, event occurrence time, system reception time, event type, and event result; The nursing record data includes nursing task type, task recipient, task start time, task end time, abnormal observation record, wound care record, and catheter maintenance record; The nursing staff location data includes the nursing staff number, the time of entering the bed area, the time of leaving the bed area, and the time spent on the shared equipment; The hand hygiene data includes the hand hygiene execution time, fluid discharge status, and nursing staff identity matching status. Direct nursing contact events are generated based on the overlap between the time of the nursing task and the time interval of the nurse's bed location. Effective hand hygiene status is generated based on the hand hygiene discharge record between the time the nurse leaves the previous patient area and enters the next patient area.

3. The hospital infection risk early warning system based on multi-source nursing data fusion according to claim 2, characterized in that, In the nursing contact dynamic graph construction module, the nursing staff node records the number of nursing tasks, the number of cross-bed nursing care, the number of high-exposure nursing care, the continuous nursing interval and the number of effective hand hygienes; the patient node records infection-related nursing abnormalities, the number of invasive nursing care and the indwelling status of instruments; the instrument node records the current user, the most recent maintenance time, the most recent disinfection time and the number of cross-bed sharing; and the environment node records the most recent disinfection time, the number of contacts and the associated patients. Multiple nursing contact events occurring within the same nursing time window for the same group of objects are grouped into the same set of nursing contact events. The nursing contact propagation weights are generated based on the relationship between the contact duration and the reference duration of each nursing contact event, the degree of exposure to nursing procedures, the effectiveness of hand hygiene, and the time decay of nursing contact from the current warning time. After disinfection of the environment, the history of nursing contact before disinfection is retained, and the transmission status of the corresponding environmental nodes is reduced from the subsequent nursing time window.

4. The hospital infection risk early warning system based on multi-source nursing data fusion according to claim 3, characterized in that, The nursing propagation time sequence parsing module reads the nursing contact dynamic graphs corresponding to the current nursing time window and multiple historical nursing time windows, and performs neighborhood propagation aggregation based on the nursing staff nodes, instrument nodes, and environment nodes connected to the patient node and the corresponding nursing contact propagation edge weights to generate the patient graph state vectors corresponding to each nursing time window. Update the temporal evolution state corresponding to the patient graph state vector according to the time sequence of the nursing time window; When the same nursing transmission chain persists across multiple consecutive nursing time windows, it enhances the corresponding historical transmission status; when effective hand hygiene, instrument disinfection, and interruption of the nursing transmission chain are detected, it reduces the corresponding historical transmission status. The patient transmission risk is generated based on the time evolution status, and nursing staff nodes, equipment nodes, and environmental nodes that have reached the preset conditions for transmission contribution are extracted to form the main nursing transmission chain.

5. The hospital infection risk early warning system based on multi-source nursing data fusion according to claim 4, characterized in that, In the multi-path hospital infection risk analysis module, the nursing record analysis path generates the nursing record risk probability based on abnormal observation records, wound care records, catheter maintenance records, and nursing record delay status. The nursing contact analysis path generates the nursing contact risk probability based on the patient transmission risk, nursing contact transmission boundary weight, cross-bed continuous nursing status, and shared equipment continuous use status. The invasive device analysis pathway generates the invasive device risk probability based on central venous catheter indwelling time, urinary catheter indwelling time, duration of mechanical ventilation, device maintenance interval, number of invasive nursing care, and number of unplanned operations. The patient status analysis path generates the patient status risk probability based on changes in body temperature, drainage volume, wound status, respiratory support status, and infection-related nursing abnormality trends. Each nursing risk analysis path saves the corresponding input nursing events and risk contribution status.

6. The hospital infection risk early warning system based on multi-source nursing data fusion according to claim 5, characterized in that, The nursing evidence credible fusion module generates a historical credible state based on the effective identification status of the m-th type of nursing data source in the retrospective of historical hospital infection events, generates a data missing state based on the number of nursing records that should have been received and the number of nursing records that actually were received in the current nursing time window, generates a data delay state based on the nursing data generation time and the system reception time, and generates a cross-source conflict state based on the consistency between the corresponding nursing data and other nursing data. The credibility of the nursing evidence is generated based on the historical credibility status, the data missing status, the data delay status, and the cross-source conflict status. A continuous disconnection of the nursing data source reduces the credibility of the corresponding nursing evidence, while continuous real-time uploading of nursing data and consistency with other nursing data increases the credibility of the corresponding nursing evidence.

7. The hospital infection risk early warning system based on multi-source nursing data fusion according to claim 6, characterized in that, The nursing evidence credibility fusion module generates the validity of the current nursing evidence based on the proportion of valid nursing records in the corresponding nursing risk analysis path within the current nursing time window and the cross-validation deviation with other nursing data. When the hand hygiene completion status in the nursing record is inconsistent with the nurse's location trajectory and the liquid dispensing status of the hand hygiene terminal, the validity of the current nursing evidence corresponding to the nursing record analysis path is reduced. The fusion contribution of the nursing record risk probability, the nursing contact risk probability, the invasive device risk probability, and the patient state risk probability is determined based on the combination of the credibility of the nursing evidence and the validity of the current nursing evidence. The patient's current susceptible exposure status is generated based on the indwelling time of the central venous catheter, the indwelling time of the urinary catheter, the duration of mechanical ventilation, the number of invasive nursing care procedures and the number of wound care procedures performed on the same day. The dynamic weighted fusion result is then corrected based on the patient's transmission risk and the patient's current susceptible exposure status to obtain the patient's hospital infection risk value.

8. The hospital infection risk early warning system based on multi-source nursing data fusion according to claim 7, characterized in that, The graded early warning and feedback update module determines the patient's risk level based on the patient's hospital infection risk value, and determines the risk source nursing staff, risk-associated patients, risk-associated equipment, and risk-associated environmental objects based on the main nursing transmission chain. When a patient's hospital infection risk value reaches the high-risk condition, a nursing check task is generated, which includes hand hygiene check, shared device disinfection check, invasive device maintenance check, and nursing record check. The nursing treatment results are formed by collecting data on the completion status of hand hygiene, disinfection of shared instruments, and environmental disinfection, as well as the results of nursing process verification. The corresponding nursing contact transmission edge and node status are updated based on the nursing treatment results, and the nursing transmission time sequence analysis and nursing evidence credibility fusion are re-executed to obtain the post-treatment patient hospital infection risk value.

Images